?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
cputlb: Hoist tlb portions in tlb_flush_one_mmuidx_locked
[qemu/ar7.git] / migration / migration.c
blob990bff00c0b752af0233f3ce2d7fbd3f68dd05cb
1 /*
2 * QEMU live migration
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
14 */
15
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
21 #include "exec.h"
22 #include "fd.h"
23 #include "socket.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "rdma.h"
27 #include "ram.h"
28 #include "migration/global_state.h"
29 #include "migration/misc.h"
30 #include "migration.h"
31 #include "savevm.h"
32 #include "qemu-file-channel.h"
33 #include "qemu-file.h"
34 #include "migration/vmstate.h"
35 #include "block/block.h"
36 #include "qapi/error.h"
37 #include "qapi/clone-visitor.h"
38 #include "qapi/qapi-visit-sockets.h"
39 #include "qapi/qapi-commands-migration.h"
40 #include "qapi/qapi-events-migration.h"
41 #include "qapi/qmp/qerror.h"
42 #include "qapi/qmp/qnull.h"
43 #include "qemu/rcu.h"
44 #include "block.h"
45 #include "postcopy-ram.h"
46 #include "qemu/thread.h"
47 #include "trace.h"
48 #include "exec/target_page.h"
49 #include "io/channel-buffer.h"
50 #include "migration/colo.h"
51 #include "hw/boards.h"
52 #include "hw/qdev-properties.h"
53 #include "monitor/monitor.h"
54 #include "net/announce.h"
55 #include "qemu/queue.h"
56
57 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
58
59 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
60 * data. */
61 #define BUFFER_DELAY 100
62 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
63
64 /* Time in milliseconds we are allowed to stop the source,
65 * for sending the last part */
66 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
67
68 /* Maximum migrate downtime set to 2000 seconds */
69 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
70 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
71
72 /* Default compression thread count */
73 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
74 /* Default decompression thread count, usually decompression is at
75 * least 4 times as fast as compression.*/
76 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
77 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
78 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
79 /* Define default autoconverge cpu throttle migration parameters */
80 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
81 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
82 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
83
84 /* Migration XBZRLE default cache size */
85 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
86
87 /* The delay time (in ms) between two COLO checkpoints */
88 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
89 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
90
91 /* Background transfer rate for postcopy, 0 means unlimited, note
92 * that page requests can still exceed this limit.
93 */
94 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
95
96 /*
97 * Parameters for self_announce_delay giving a stream of RARP/ARP
98 * packets after migration.
99 */
100 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
101 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
102 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
103 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
104
105 static NotifierList migration_state_notifiers =
106 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
107
108 static bool deferred_incoming;
109
110 /* Messages sent on the return path from destination to source */
111 enum mig_rp_message_type {
112 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
113 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
114 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
115
116 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
117 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
118 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
119 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
120
121 MIG_RP_MSG_MAX
122 };
123
124 /* When we add fault tolerance, we could have several
125 migrations at once. For now we don't need to add
126 dynamic creation of migration */
127
128 static MigrationState *current_migration;
129 static MigrationIncomingState *current_incoming;
130
131 static bool migration_object_check(MigrationState *ms, Error **errp);
132 static int migration_maybe_pause(MigrationState *s,
133 int *current_active_state,
134 int new_state);
135 static void migrate_fd_cancel(MigrationState *s);
136
137 void migration_object_init(void)
138 {
139 MachineState *ms = MACHINE(qdev_get_machine());
140 Error *err = NULL;
141
142 /* This can only be called once. */
143 assert(!current_migration);
144 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
145
146 /*
147 * Init the migrate incoming object as well no matter whether
148 * we'll use it or not.
149 */
150 assert(!current_incoming);
151 current_incoming = g_new0(MigrationIncomingState, 1);
152 current_incoming->state = MIGRATION_STATUS_NONE;
153 current_incoming->postcopy_remote_fds =
154 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
155 qemu_mutex_init(&current_incoming->rp_mutex);
156 qemu_event_init(&current_incoming->main_thread_load_event, false);
157 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
158 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
159
160 init_dirty_bitmap_incoming_migration();
161
162 if (!migration_object_check(current_migration, &err)) {
163 error_report_err(err);
164 exit(1);
165 }
166
167 /*
168 * We cannot really do this in migration_instance_init() since at
169 * that time global properties are not yet applied, then this
170 * value will be definitely replaced by something else.
171 */
172 if (ms->enforce_config_section) {
173 current_migration->send_configuration = true;
174 }
175 }
176
177 void migration_shutdown(void)
178 {
179 /*
180 * Cancel the current migration - that will (eventually)
181 * stop the migration using this structure
182 */
183 migrate_fd_cancel(current_migration);
184 object_unref(OBJECT(current_migration));
185 }
186
187 /* For outgoing */
188 MigrationState *migrate_get_current(void)
189 {
190 /* This can only be called after the object created. */
191 assert(current_migration);
192 return current_migration;
193 }
194
195 MigrationIncomingState *migration_incoming_get_current(void)
196 {
197 assert(current_incoming);
198 return current_incoming;
199 }
200
201 void migration_incoming_state_destroy(void)
202 {
203 struct MigrationIncomingState *mis = migration_incoming_get_current();
204
205 if (mis->to_src_file) {
206 /* Tell source that we are done */
207 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
208 qemu_fclose(mis->to_src_file);
209 mis->to_src_file = NULL;
210 }
211
212 if (mis->from_src_file) {
213 qemu_fclose(mis->from_src_file);
214 mis->from_src_file = NULL;
215 }
216 if (mis->postcopy_remote_fds) {
217 g_array_free(mis->postcopy_remote_fds, TRUE);
218 mis->postcopy_remote_fds = NULL;
219 }
220
221 qemu_event_reset(&mis->main_thread_load_event);
222
223 if (mis->socket_address_list) {
224 qapi_free_SocketAddressList(mis->socket_address_list);
225 mis->socket_address_list = NULL;
226 }
227 }
228
229 static void migrate_generate_event(int new_state)
230 {
231 if (migrate_use_events()) {
232 qapi_event_send_migration(new_state);
233 }
234 }
235
236 static bool migrate_late_block_activate(void)
237 {
238 MigrationState *s;
239
240 s = migrate_get_current();
241
242 return s->enabled_capabilities[
243 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
244 }
245
246 /*
247 * Called on -incoming with a defer: uri.
248 * The migration can be started later after any parameters have been
249 * changed.
250 */
251 static void deferred_incoming_migration(Error **errp)
252 {
253 if (deferred_incoming) {
254 error_setg(errp, "Incoming migration already deferred");
255 }
256 deferred_incoming = true;
257 }
258
259 /*
260 * Send a message on the return channel back to the source
261 * of the migration.
262 */
263 static int migrate_send_rp_message(MigrationIncomingState *mis,
264 enum mig_rp_message_type message_type,
265 uint16_t len, void *data)
266 {
267 int ret = 0;
268
269 trace_migrate_send_rp_message((int)message_type, len);
270 qemu_mutex_lock(&mis->rp_mutex);
271
272 /*
273 * It's possible that the file handle got lost due to network
274 * failures.
275 */
276 if (!mis->to_src_file) {
277 ret = -EIO;
278 goto error;
279 }
280
281 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
282 qemu_put_be16(mis->to_src_file, len);
283 qemu_put_buffer(mis->to_src_file, data, len);
284 qemu_fflush(mis->to_src_file);
285
286 /* It's possible that qemu file got error during sending */
287 ret = qemu_file_get_error(mis->to_src_file);
288
289 error:
290 qemu_mutex_unlock(&mis->rp_mutex);
291 return ret;
292 }
293
294 /* Request a range of pages from the source VM at the given
295 * start address.
296 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
297 * as the last request (a name must have been given previously)
298 * Start: Address offset within the RB
299 * Len: Length in bytes required - must be a multiple of pagesize
300 */
301 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
302 ram_addr_t start, size_t len)
303 {
304 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
305 size_t msglen = 12; /* start + len */
306 enum mig_rp_message_type msg_type;
307
308 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
309 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
310
311 if (rbname) {
312 int rbname_len = strlen(rbname);
313 assert(rbname_len < 256);
314
315 bufc[msglen++] = rbname_len;
316 memcpy(bufc + msglen, rbname, rbname_len);
317 msglen += rbname_len;
318 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
319 } else {
320 msg_type = MIG_RP_MSG_REQ_PAGES;
321 }
322
323 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
324 }
325
326 static bool migration_colo_enabled;
327 bool migration_incoming_colo_enabled(void)
328 {
329 return migration_colo_enabled;
330 }
331
332 void migration_incoming_disable_colo(void)
333 {
334 migration_colo_enabled = false;
335 }
336
337 void migration_incoming_enable_colo(void)
338 {
339 migration_colo_enabled = true;
340 }
341
342 void migrate_add_address(SocketAddress *address)
343 {
344 MigrationIncomingState *mis = migration_incoming_get_current();
345 SocketAddressList *addrs;
346
347 addrs = g_new0(SocketAddressList, 1);
348 addrs->next = mis->socket_address_list;
349 mis->socket_address_list = addrs;
350 addrs->value = QAPI_CLONE(SocketAddress, address);
351 }
352
353 void qemu_start_incoming_migration(const char *uri, Error **errp)
354 {
355 const char *p;
356
357 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
358 if (!strcmp(uri, "defer")) {
359 deferred_incoming_migration(errp);
360 } else if (strstart(uri, "tcp:", &p)) {
361 tcp_start_incoming_migration(p, errp);
362 #ifdef CONFIG_RDMA
363 } else if (strstart(uri, "rdma:", &p)) {
364 rdma_start_incoming_migration(p, errp);
365 #endif
366 } else if (strstart(uri, "exec:", &p)) {
367 exec_start_incoming_migration(p, errp);
368 } else if (strstart(uri, "unix:", &p)) {
369 unix_start_incoming_migration(p, errp);
370 } else if (strstart(uri, "fd:", &p)) {
371 fd_start_incoming_migration(p, errp);
372 } else {
373 error_setg(errp, "unknown migration protocol: %s", uri);
374 }
375 }
376
377 static void process_incoming_migration_bh(void *opaque)
378 {
379 Error *local_err = NULL;
380 MigrationIncomingState *mis = opaque;
381
382 /* If capability late_block_activate is set:
383 * Only fire up the block code now if we're going to restart the
384 * VM, else 'cont' will do it.
385 * This causes file locking to happen; so we don't want it to happen
386 * unless we really are starting the VM.
387 */
388 if (!migrate_late_block_activate() ||
389 (autostart && (!global_state_received() ||
390 global_state_get_runstate() == RUN_STATE_RUNNING))) {
391 /* Make sure all file formats flush their mutable metadata.
392 * If we get an error here, just don't restart the VM yet. */
393 bdrv_invalidate_cache_all(&local_err);
394 if (local_err) {
395 error_report_err(local_err);
396 local_err = NULL;
397 autostart = false;
398 }
399 }
400
401 /*
402 * This must happen after all error conditions are dealt with and
403 * we're sure the VM is going to be running on this host.
404 */
405 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
406
407 if (multifd_load_cleanup(&local_err) != 0) {
408 error_report_err(local_err);
409 autostart = false;
410 }
411 /* If global state section was not received or we are in running
412 state, we need to obey autostart. Any other state is set with
413 runstate_set. */
414
415 dirty_bitmap_mig_before_vm_start();
416
417 if (!global_state_received() ||
418 global_state_get_runstate() == RUN_STATE_RUNNING) {
419 if (autostart) {
420 vm_start();
421 } else {
422 runstate_set(RUN_STATE_PAUSED);
423 }
424 } else if (migration_incoming_colo_enabled()) {
425 migration_incoming_disable_colo();
426 vm_start();
427 } else {
428 runstate_set(global_state_get_runstate());
429 }
430 /*
431 * This must happen after any state changes since as soon as an external
432 * observer sees this event they might start to prod at the VM assuming
433 * it's ready to use.
434 */
435 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
436 MIGRATION_STATUS_COMPLETED);
437 qemu_bh_delete(mis->bh);
438 migration_incoming_state_destroy();
439 }
440
441 static void process_incoming_migration_co(void *opaque)
442 {
443 MigrationIncomingState *mis = migration_incoming_get_current();
444 PostcopyState ps;
445 int ret;
446 Error *local_err = NULL;
447
448 assert(mis->from_src_file);
449 mis->migration_incoming_co = qemu_coroutine_self();
450 mis->largest_page_size = qemu_ram_pagesize_largest();
451 postcopy_state_set(POSTCOPY_INCOMING_NONE);
452 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
453 MIGRATION_STATUS_ACTIVE);
454 ret = qemu_loadvm_state(mis->from_src_file);
455
456 ps = postcopy_state_get();
457 trace_process_incoming_migration_co_end(ret, ps);
458 if (ps != POSTCOPY_INCOMING_NONE) {
459 if (ps == POSTCOPY_INCOMING_ADVISE) {
460 /*
461 * Where a migration had postcopy enabled (and thus went to advise)
462 * but managed to complete within the precopy period, we can use
463 * the normal exit.
464 */
465 postcopy_ram_incoming_cleanup(mis);
466 } else if (ret >= 0) {
467 /*
468 * Postcopy was started, cleanup should happen at the end of the
469 * postcopy thread.
470 */
471 trace_process_incoming_migration_co_postcopy_end_main();
472 return;
473 }
474 /* Else if something went wrong then just fall out of the normal exit */
475 }
476
477 /* we get COLO info, and know if we are in COLO mode */
478 if (!ret && migration_incoming_colo_enabled()) {
479 /* Make sure all file formats flush their mutable metadata */
480 bdrv_invalidate_cache_all(&local_err);
481 if (local_err) {
482 error_report_err(local_err);
483 goto fail;
484 }
485
486 if (colo_init_ram_cache() < 0) {
487 error_report("Init ram cache failed");
488 goto fail;
489 }
490
491 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
492 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
493 mis->have_colo_incoming_thread = true;
494 qemu_coroutine_yield();
495
496 /* Wait checkpoint incoming thread exit before free resource */
497 qemu_thread_join(&mis->colo_incoming_thread);
498 /* We hold the global iothread lock, so it is safe here */
499 colo_release_ram_cache();
500 }
501
502 if (ret < 0) {
503 error_report("load of migration failed: %s", strerror(-ret));
504 goto fail;
505 }
506 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
507 qemu_bh_schedule(mis->bh);
508 mis->migration_incoming_co = NULL;
509 return;
510 fail:
511 local_err = NULL;
512 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
513 MIGRATION_STATUS_FAILED);
514 qemu_fclose(mis->from_src_file);
515 if (multifd_load_cleanup(&local_err) != 0) {
516 error_report_err(local_err);
517 }
518 exit(EXIT_FAILURE);
519 }
520
521 static void migration_incoming_setup(QEMUFile *f)
522 {
523 MigrationIncomingState *mis = migration_incoming_get_current();
524
525 if (multifd_load_setup() != 0) {
526 /* We haven't been able to create multifd threads
527 nothing better to do */
528 exit(EXIT_FAILURE);
529 }
530
531 if (!mis->from_src_file) {
532 mis->from_src_file = f;
533 }
534 qemu_file_set_blocking(f, false);
535 }
536
537 void migration_incoming_process(void)
538 {
539 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
540 qemu_coroutine_enter(co);
541 }
542
543 /* Returns true if recovered from a paused migration, otherwise false */
544 static bool postcopy_try_recover(QEMUFile *f)
545 {
546 MigrationIncomingState *mis = migration_incoming_get_current();
547
548 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
549 /* Resumed from a paused postcopy migration */
550
551 mis->from_src_file = f;
552 /* Postcopy has standalone thread to do vm load */
553 qemu_file_set_blocking(f, true);
554
555 /* Re-configure the return path */
556 mis->to_src_file = qemu_file_get_return_path(f);
557
558 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
559 MIGRATION_STATUS_POSTCOPY_RECOVER);
560
561 /*
562 * Here, we only wake up the main loading thread (while the
563 * fault thread will still be waiting), so that we can receive
564 * commands from source now, and answer it if needed. The
565 * fault thread will be woken up afterwards until we are sure
566 * that source is ready to reply to page requests.
567 */
568 qemu_sem_post(&mis->postcopy_pause_sem_dst);
569 return true;
570 }
571
572 return false;
573 }
574
575 void migration_fd_process_incoming(QEMUFile *f)
576 {
577 if (postcopy_try_recover(f)) {
578 return;
579 }
580
581 migration_incoming_setup(f);
582 migration_incoming_process();
583 }
584
585 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
586 {
587 MigrationIncomingState *mis = migration_incoming_get_current();
588 bool start_migration;
589
590 if (!mis->from_src_file) {
591 /* The first connection (multifd may have multiple) */
592 QEMUFile *f = qemu_fopen_channel_input(ioc);
593
594 /* If it's a recovery, we're done */
595 if (postcopy_try_recover(f)) {
596 return;
597 }
598
599 migration_incoming_setup(f);
600
601 /*
602 * Common migration only needs one channel, so we can start
603 * right now. Multifd needs more than one channel, we wait.
604 */
605 start_migration = !migrate_use_multifd();
606 } else {
607 Error *local_err = NULL;
608 /* Multiple connections */
609 assert(migrate_use_multifd());
610 start_migration = multifd_recv_new_channel(ioc, &local_err);
611 if (local_err) {
612 error_propagate(errp, local_err);
613 return;
614 }
615 }
616
617 if (start_migration) {
618 migration_incoming_process();
619 }
620 }
621
622 /**
623 * @migration_has_all_channels: We have received all channels that we need
624 *
625 * Returns true when we have got connections to all the channels that
626 * we need for migration.
627 */
628 bool migration_has_all_channels(void)
629 {
630 MigrationIncomingState *mis = migration_incoming_get_current();
631 bool all_channels;
632
633 all_channels = multifd_recv_all_channels_created();
634
635 return all_channels && mis->from_src_file != NULL;
636 }
637
638 /*
639 * Send a 'SHUT' message on the return channel with the given value
640 * to indicate that we've finished with the RP. Non-0 value indicates
641 * error.
642 */
643 void migrate_send_rp_shut(MigrationIncomingState *mis,
644 uint32_t value)
645 {
646 uint32_t buf;
647
648 buf = cpu_to_be32(value);
649 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
650 }
651
652 /*
653 * Send a 'PONG' message on the return channel with the given value
654 * (normally in response to a 'PING')
655 */
656 void migrate_send_rp_pong(MigrationIncomingState *mis,
657 uint32_t value)
658 {
659 uint32_t buf;
660
661 buf = cpu_to_be32(value);
662 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
663 }
664
665 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
666 char *block_name)
667 {
668 char buf[512];
669 int len;
670 int64_t res;
671
672 /*
673 * First, we send the header part. It contains only the len of
674 * idstr, and the idstr itself.
675 */
676 len = strlen(block_name);
677 buf[0] = len;
678 memcpy(buf + 1, block_name, len);
679
680 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
681 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
682 __func__);
683 return;
684 }
685
686 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
687
688 /*
689 * Next, we dump the received bitmap to the stream.
690 *
691 * TODO: currently we are safe since we are the only one that is
692 * using the to_src_file handle (fault thread is still paused),
693 * and it's ok even not taking the mutex. However the best way is
694 * to take the lock before sending the message header, and release
695 * the lock after sending the bitmap.
696 */
697 qemu_mutex_lock(&mis->rp_mutex);
698 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
699 qemu_mutex_unlock(&mis->rp_mutex);
700
701 trace_migrate_send_rp_recv_bitmap(block_name, res);
702 }
703
704 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
705 {
706 uint32_t buf;
707
708 buf = cpu_to_be32(value);
709 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
710 }
711
712 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
713 {
714 MigrationCapabilityStatusList *head = NULL;
715 MigrationCapabilityStatusList *caps;
716 MigrationState *s = migrate_get_current();
717 int i;
718
719 caps = NULL; /* silence compiler warning */
720 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
721 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
722 if (i == MIGRATION_CAPABILITY_BLOCK) {
723 continue;
724 }
725 #endif
726 if (head == NULL) {
727 head = g_malloc0(sizeof(*caps));
728 caps = head;
729 } else {
730 caps->next = g_malloc0(sizeof(*caps));
731 caps = caps->next;
732 }
733 caps->value =
734 g_malloc(sizeof(*caps->value));
735 caps->value->capability = i;
736 caps->value->state = s->enabled_capabilities[i];
737 }
738
739 return head;
740 }
741
742 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
743 {
744 MigrationParameters *params;
745 MigrationState *s = migrate_get_current();
746
747 /* TODO use QAPI_CLONE() instead of duplicating it inline */
748 params = g_malloc0(sizeof(*params));
749 params->has_compress_level = true;
750 params->compress_level = s->parameters.compress_level;
751 params->has_compress_threads = true;
752 params->compress_threads = s->parameters.compress_threads;
753 params->has_compress_wait_thread = true;
754 params->compress_wait_thread = s->parameters.compress_wait_thread;
755 params->has_decompress_threads = true;
756 params->decompress_threads = s->parameters.decompress_threads;
757 params->has_cpu_throttle_initial = true;
758 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
759 params->has_cpu_throttle_increment = true;
760 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
761 params->has_tls_creds = true;
762 params->tls_creds = g_strdup(s->parameters.tls_creds);
763 params->has_tls_hostname = true;
764 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
765 params->has_tls_authz = true;
766 params->tls_authz = g_strdup(s->parameters.tls_authz);
767 params->has_max_bandwidth = true;
768 params->max_bandwidth = s->parameters.max_bandwidth;
769 params->has_downtime_limit = true;
770 params->downtime_limit = s->parameters.downtime_limit;
771 params->has_x_checkpoint_delay = true;
772 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
773 params->has_block_incremental = true;
774 params->block_incremental = s->parameters.block_incremental;
775 params->has_multifd_channels = true;
776 params->multifd_channels = s->parameters.multifd_channels;
777 params->has_xbzrle_cache_size = true;
778 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
779 params->has_max_postcopy_bandwidth = true;
780 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
781 params->has_max_cpu_throttle = true;
782 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
783 params->has_announce_initial = true;
784 params->announce_initial = s->parameters.announce_initial;
785 params->has_announce_max = true;
786 params->announce_max = s->parameters.announce_max;
787 params->has_announce_rounds = true;
788 params->announce_rounds = s->parameters.announce_rounds;
789 params->has_announce_step = true;
790 params->announce_step = s->parameters.announce_step;
791
792 return params;
793 }
794
795 AnnounceParameters *migrate_announce_params(void)
796 {
797 static AnnounceParameters ap;
798
799 MigrationState *s = migrate_get_current();
800
801 ap.initial = s->parameters.announce_initial;
802 ap.max = s->parameters.announce_max;
803 ap.rounds = s->parameters.announce_rounds;
804 ap.step = s->parameters.announce_step;
805
806 return &ap;
807 }
808
809 /*
810 * Return true if we're already in the middle of a migration
811 * (i.e. any of the active or setup states)
812 */
813 bool migration_is_setup_or_active(int state)
814 {
815 switch (state) {
816 case MIGRATION_STATUS_ACTIVE:
817 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
818 case MIGRATION_STATUS_POSTCOPY_PAUSED:
819 case MIGRATION_STATUS_POSTCOPY_RECOVER:
820 case MIGRATION_STATUS_SETUP:
821 case MIGRATION_STATUS_PRE_SWITCHOVER:
822 case MIGRATION_STATUS_DEVICE:
823 case MIGRATION_STATUS_WAIT_UNPLUG:
824 return true;
825
826 default:
827 return false;
828
829 }
830 }
831
832 static void populate_time_info(MigrationInfo *info, MigrationState *s)
833 {
834 info->has_status = true;
835 info->has_setup_time = true;
836 info->setup_time = s->setup_time;
837 if (s->state == MIGRATION_STATUS_COMPLETED) {
838 info->has_total_time = true;
839 info->total_time = s->total_time;
840 info->has_downtime = true;
841 info->downtime = s->downtime;
842 } else {
843 info->has_total_time = true;
844 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
845 s->start_time;
846 info->has_expected_downtime = true;
847 info->expected_downtime = s->expected_downtime;
848 }
849 }
850
851 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
852 {
853 info->has_ram = true;
854 info->ram = g_malloc0(sizeof(*info->ram));
855 info->ram->transferred = ram_counters.transferred;
856 info->ram->total = ram_bytes_total();
857 info->ram->duplicate = ram_counters.duplicate;
858 /* legacy value. It is not used anymore */
859 info->ram->skipped = 0;
860 info->ram->normal = ram_counters.normal;
861 info->ram->normal_bytes = ram_counters.normal *
862 qemu_target_page_size();
863 info->ram->mbps = s->mbps;
864 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
865 info->ram->postcopy_requests = ram_counters.postcopy_requests;
866 info->ram->page_size = qemu_target_page_size();
867 info->ram->multifd_bytes = ram_counters.multifd_bytes;
868 info->ram->pages_per_second = s->pages_per_second;
869
870 if (migrate_use_xbzrle()) {
871 info->has_xbzrle_cache = true;
872 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
873 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
874 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
875 info->xbzrle_cache->pages = xbzrle_counters.pages;
876 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
877 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
878 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
879 }
880
881 if (migrate_use_compression()) {
882 info->has_compression = true;
883 info->compression = g_malloc0(sizeof(*info->compression));
884 info->compression->pages = compression_counters.pages;
885 info->compression->busy = compression_counters.busy;
886 info->compression->busy_rate = compression_counters.busy_rate;
887 info->compression->compressed_size =
888 compression_counters.compressed_size;
889 info->compression->compression_rate =
890 compression_counters.compression_rate;
891 }
892
893 if (cpu_throttle_active()) {
894 info->has_cpu_throttle_percentage = true;
895 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
896 }
897
898 if (s->state != MIGRATION_STATUS_COMPLETED) {
899 info->ram->remaining = ram_bytes_remaining();
900 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
901 }
902 }
903
904 static void populate_disk_info(MigrationInfo *info)
905 {
906 if (blk_mig_active()) {
907 info->has_disk = true;
908 info->disk = g_malloc0(sizeof(*info->disk));
909 info->disk->transferred = blk_mig_bytes_transferred();
910 info->disk->remaining = blk_mig_bytes_remaining();
911 info->disk->total = blk_mig_bytes_total();
912 }
913 }
914
915 static void fill_source_migration_info(MigrationInfo *info)
916 {
917 MigrationState *s = migrate_get_current();
918
919 switch (s->state) {
920 case MIGRATION_STATUS_NONE:
921 /* no migration has happened ever */
922 /* do not overwrite destination migration status */
923 return;
924 break;
925 case MIGRATION_STATUS_SETUP:
926 info->has_status = true;
927 info->has_total_time = false;
928 break;
929 case MIGRATION_STATUS_ACTIVE:
930 case MIGRATION_STATUS_CANCELLING:
931 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
932 case MIGRATION_STATUS_PRE_SWITCHOVER:
933 case MIGRATION_STATUS_DEVICE:
934 case MIGRATION_STATUS_POSTCOPY_PAUSED:
935 case MIGRATION_STATUS_POSTCOPY_RECOVER:
936 /* TODO add some postcopy stats */
937 populate_time_info(info, s);
938 populate_ram_info(info, s);
939 populate_disk_info(info);
940 break;
941 case MIGRATION_STATUS_COLO:
942 info->has_status = true;
943 /* TODO: display COLO specific information (checkpoint info etc.) */
944 break;
945 case MIGRATION_STATUS_COMPLETED:
946 populate_time_info(info, s);
947 populate_ram_info(info, s);
948 break;
949 case MIGRATION_STATUS_FAILED:
950 info->has_status = true;
951 if (s->error) {
952 info->has_error_desc = true;
953 info->error_desc = g_strdup(error_get_pretty(s->error));
954 }
955 break;
956 case MIGRATION_STATUS_CANCELLED:
957 info->has_status = true;
958 break;
959 case MIGRATION_STATUS_WAIT_UNPLUG:
960 info->has_status = true;
961 break;
962 }
963 info->status = s->state;
964 }
965
966 /**
967 * @migration_caps_check - check capability validity
968 *
969 * @cap_list: old capability list, array of bool
970 * @params: new capabilities to be applied soon
971 * @errp: set *errp if the check failed, with reason
972 *
973 * Returns true if check passed, otherwise false.
974 */
975 static bool migrate_caps_check(bool *cap_list,
976 MigrationCapabilityStatusList *params,
977 Error **errp)
978 {
979 MigrationCapabilityStatusList *cap;
980 bool old_postcopy_cap;
981 MigrationIncomingState *mis = migration_incoming_get_current();
982
983 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
984
985 for (cap = params; cap; cap = cap->next) {
986 cap_list[cap->value->capability] = cap->value->state;
987 }
988
989 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
990 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
991 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
992 "block migration");
993 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
994 return false;
995 }
996 #endif
997
998 #ifndef CONFIG_REPLICATION
999 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1000 error_setg(errp, "QEMU compiled without replication module"
1001 " can't enable COLO");
1002 error_append_hint(errp, "Please enable replication before COLO.\n");
1003 return false;
1004 }
1005 #endif
1006
1007 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1008 /* This check is reasonably expensive, so only when it's being
1009 * set the first time, also it's only the destination that needs
1010 * special support.
1011 */
1012 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1013 !postcopy_ram_supported_by_host(mis)) {
1014 /* postcopy_ram_supported_by_host will have emitted a more
1015 * detailed message
1016 */
1017 error_setg(errp, "Postcopy is not supported");
1018 return false;
1019 }
1020
1021 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1022 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1023 return false;
1024 }
1025 }
1026
1027 return true;
1028 }
1029
1030 static void fill_destination_migration_info(MigrationInfo *info)
1031 {
1032 MigrationIncomingState *mis = migration_incoming_get_current();
1033
1034 if (mis->socket_address_list) {
1035 info->has_socket_address = true;
1036 info->socket_address =
1037 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1038 }
1039
1040 switch (mis->state) {
1041 case MIGRATION_STATUS_NONE:
1042 return;
1043 break;
1044 case MIGRATION_STATUS_SETUP:
1045 case MIGRATION_STATUS_CANCELLING:
1046 case MIGRATION_STATUS_CANCELLED:
1047 case MIGRATION_STATUS_ACTIVE:
1048 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1049 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1050 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1051 case MIGRATION_STATUS_FAILED:
1052 case MIGRATION_STATUS_COLO:
1053 info->has_status = true;
1054 break;
1055 case MIGRATION_STATUS_COMPLETED:
1056 info->has_status = true;
1057 fill_destination_postcopy_migration_info(info);
1058 break;
1059 }
1060 info->status = mis->state;
1061 }
1062
1063 MigrationInfo *qmp_query_migrate(Error **errp)
1064 {
1065 MigrationInfo *info = g_malloc0(sizeof(*info));
1066
1067 fill_destination_migration_info(info);
1068 fill_source_migration_info(info);
1069
1070 return info;
1071 }
1072
1073 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1074 Error **errp)
1075 {
1076 MigrationState *s = migrate_get_current();
1077 MigrationCapabilityStatusList *cap;
1078 bool cap_list[MIGRATION_CAPABILITY__MAX];
1079
1080 if (migration_is_setup_or_active(s->state)) {
1081 error_setg(errp, QERR_MIGRATION_ACTIVE);
1082 return;
1083 }
1084
1085 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1086 if (!migrate_caps_check(cap_list, params, errp)) {
1087 return;
1088 }
1089
1090 for (cap = params; cap; cap = cap->next) {
1091 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1092 }
1093 }
1094
1095 /*
1096 * Check whether the parameters are valid. Error will be put into errp
1097 * (if provided). Return true if valid, otherwise false.
1098 */
1099 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1100 {
1101 if (params->has_compress_level &&
1102 (params->compress_level > 9)) {
1103 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1104 "is invalid, it should be in the range of 0 to 9");
1105 return false;
1106 }
1107
1108 if (params->has_compress_threads && (params->compress_threads < 1)) {
1109 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1110 "compress_threads",
1111 "is invalid, it should be in the range of 1 to 255");
1112 return false;
1113 }
1114
1115 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1116 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1117 "decompress_threads",
1118 "is invalid, it should be in the range of 1 to 255");
1119 return false;
1120 }
1121
1122 if (params->has_cpu_throttle_initial &&
1123 (params->cpu_throttle_initial < 1 ||
1124 params->cpu_throttle_initial > 99)) {
1125 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1126 "cpu_throttle_initial",
1127 "an integer in the range of 1 to 99");
1128 return false;
1129 }
1130
1131 if (params->has_cpu_throttle_increment &&
1132 (params->cpu_throttle_increment < 1 ||
1133 params->cpu_throttle_increment > 99)) {
1134 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1135 "cpu_throttle_increment",
1136 "an integer in the range of 1 to 99");
1137 return false;
1138 }
1139
1140 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1141 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
1142 " range of 0 to %zu bytes/second", SIZE_MAX);
1143 return false;
1144 }
1145
1146 if (params->has_downtime_limit &&
1147 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1148 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1149 "the range of 0 to %d milliseconds",
1150 MAX_MIGRATE_DOWNTIME);
1151 return false;
1152 }
1153
1154 /* x_checkpoint_delay is now always positive */
1155
1156 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1157 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1158 "multifd_channels",
1159 "is invalid, it should be in the range of 1 to 255");
1160 return false;
1161 }
1162
1163 if (params->has_xbzrle_cache_size &&
1164 (params->xbzrle_cache_size < qemu_target_page_size() ||
1165 !is_power_of_2(params->xbzrle_cache_size))) {
1166 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1167 "xbzrle_cache_size",
1168 "is invalid, it should be bigger than target page size"
1169 " and a power of two");
1170 return false;
1171 }
1172
1173 if (params->has_max_cpu_throttle &&
1174 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1175 params->max_cpu_throttle > 99)) {
1176 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1177 "max_cpu_throttle",
1178 "an integer in the range of cpu_throttle_initial to 99");
1179 return false;
1180 }
1181
1182 if (params->has_announce_initial &&
1183 params->announce_initial > 100000) {
1184 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1185 "announce_initial",
1186 "is invalid, it must be less than 100000 ms");
1187 return false;
1188 }
1189 if (params->has_announce_max &&
1190 params->announce_max > 100000) {
1191 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1192 "announce_max",
1193 "is invalid, it must be less than 100000 ms");
1194 return false;
1195 }
1196 if (params->has_announce_rounds &&
1197 params->announce_rounds > 1000) {
1198 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1199 "announce_rounds",
1200 "is invalid, it must be in the range of 0 to 1000");
1201 return false;
1202 }
1203 if (params->has_announce_step &&
1204 (params->announce_step < 1 ||
1205 params->announce_step > 10000)) {
1206 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1207 "announce_step",
1208 "is invalid, it must be in the range of 1 to 10000 ms");
1209 return false;
1210 }
1211 return true;
1212 }
1213
1214 static void migrate_params_test_apply(MigrateSetParameters *params,
1215 MigrationParameters *dest)
1216 {
1217 *dest = migrate_get_current()->parameters;
1218
1219 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1220
1221 if (params->has_compress_level) {
1222 dest->compress_level = params->compress_level;
1223 }
1224
1225 if (params->has_compress_threads) {
1226 dest->compress_threads = params->compress_threads;
1227 }
1228
1229 if (params->has_compress_wait_thread) {
1230 dest->compress_wait_thread = params->compress_wait_thread;
1231 }
1232
1233 if (params->has_decompress_threads) {
1234 dest->decompress_threads = params->decompress_threads;
1235 }
1236
1237 if (params->has_cpu_throttle_initial) {
1238 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1239 }
1240
1241 if (params->has_cpu_throttle_increment) {
1242 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1243 }
1244
1245 if (params->has_tls_creds) {
1246 assert(params->tls_creds->type == QTYPE_QSTRING);
1247 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1248 }
1249
1250 if (params->has_tls_hostname) {
1251 assert(params->tls_hostname->type == QTYPE_QSTRING);
1252 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1253 }
1254
1255 if (params->has_max_bandwidth) {
1256 dest->max_bandwidth = params->max_bandwidth;
1257 }
1258
1259 if (params->has_downtime_limit) {
1260 dest->downtime_limit = params->downtime_limit;
1261 }
1262
1263 if (params->has_x_checkpoint_delay) {
1264 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1265 }
1266
1267 if (params->has_block_incremental) {
1268 dest->block_incremental = params->block_incremental;
1269 }
1270 if (params->has_multifd_channels) {
1271 dest->multifd_channels = params->multifd_channels;
1272 }
1273 if (params->has_xbzrle_cache_size) {
1274 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1275 }
1276 if (params->has_max_postcopy_bandwidth) {
1277 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1278 }
1279 if (params->has_max_cpu_throttle) {
1280 dest->max_cpu_throttle = params->max_cpu_throttle;
1281 }
1282 if (params->has_announce_initial) {
1283 dest->announce_initial = params->announce_initial;
1284 }
1285 if (params->has_announce_max) {
1286 dest->announce_max = params->announce_max;
1287 }
1288 if (params->has_announce_rounds) {
1289 dest->announce_rounds = params->announce_rounds;
1290 }
1291 if (params->has_announce_step) {
1292 dest->announce_step = params->announce_step;
1293 }
1294 }
1295
1296 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1297 {
1298 MigrationState *s = migrate_get_current();
1299
1300 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1301
1302 if (params->has_compress_level) {
1303 s->parameters.compress_level = params->compress_level;
1304 }
1305
1306 if (params->has_compress_threads) {
1307 s->parameters.compress_threads = params->compress_threads;
1308 }
1309
1310 if (params->has_compress_wait_thread) {
1311 s->parameters.compress_wait_thread = params->compress_wait_thread;
1312 }
1313
1314 if (params->has_decompress_threads) {
1315 s->parameters.decompress_threads = params->decompress_threads;
1316 }
1317
1318 if (params->has_cpu_throttle_initial) {
1319 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1320 }
1321
1322 if (params->has_cpu_throttle_increment) {
1323 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1324 }
1325
1326 if (params->has_tls_creds) {
1327 g_free(s->parameters.tls_creds);
1328 assert(params->tls_creds->type == QTYPE_QSTRING);
1329 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1330 }
1331
1332 if (params->has_tls_hostname) {
1333 g_free(s->parameters.tls_hostname);
1334 assert(params->tls_hostname->type == QTYPE_QSTRING);
1335 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1336 }
1337
1338 if (params->has_tls_authz) {
1339 g_free(s->parameters.tls_authz);
1340 assert(params->tls_authz->type == QTYPE_QSTRING);
1341 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1342 }
1343
1344 if (params->has_max_bandwidth) {
1345 s->parameters.max_bandwidth = params->max_bandwidth;
1346 if (s->to_dst_file && !migration_in_postcopy()) {
1347 qemu_file_set_rate_limit(s->to_dst_file,
1348 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1349 }
1350 }
1351
1352 if (params->has_downtime_limit) {
1353 s->parameters.downtime_limit = params->downtime_limit;
1354 }
1355
1356 if (params->has_x_checkpoint_delay) {
1357 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1358 if (migration_in_colo_state()) {
1359 colo_checkpoint_notify(s);
1360 }
1361 }
1362
1363 if (params->has_block_incremental) {
1364 s->parameters.block_incremental = params->block_incremental;
1365 }
1366 if (params->has_multifd_channels) {
1367 s->parameters.multifd_channels = params->multifd_channels;
1368 }
1369 if (params->has_xbzrle_cache_size) {
1370 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1371 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1372 }
1373 if (params->has_max_postcopy_bandwidth) {
1374 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1375 if (s->to_dst_file && migration_in_postcopy()) {
1376 qemu_file_set_rate_limit(s->to_dst_file,
1377 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1378 }
1379 }
1380 if (params->has_max_cpu_throttle) {
1381 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1382 }
1383 if (params->has_announce_initial) {
1384 s->parameters.announce_initial = params->announce_initial;
1385 }
1386 if (params->has_announce_max) {
1387 s->parameters.announce_max = params->announce_max;
1388 }
1389 if (params->has_announce_rounds) {
1390 s->parameters.announce_rounds = params->announce_rounds;
1391 }
1392 if (params->has_announce_step) {
1393 s->parameters.announce_step = params->announce_step;
1394 }
1395 }
1396
1397 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1398 {
1399 MigrationParameters tmp;
1400
1401 /* TODO Rewrite "" to null instead */
1402 if (params->has_tls_creds
1403 && params->tls_creds->type == QTYPE_QNULL) {
1404 qobject_unref(params->tls_creds->u.n);
1405 params->tls_creds->type = QTYPE_QSTRING;
1406 params->tls_creds->u.s = strdup("");
1407 }
1408 /* TODO Rewrite "" to null instead */
1409 if (params->has_tls_hostname
1410 && params->tls_hostname->type == QTYPE_QNULL) {
1411 qobject_unref(params->tls_hostname->u.n);
1412 params->tls_hostname->type = QTYPE_QSTRING;
1413 params->tls_hostname->u.s = strdup("");
1414 }
1415
1416 migrate_params_test_apply(params, &tmp);
1417
1418 if (!migrate_params_check(&tmp, errp)) {
1419 /* Invalid parameter */
1420 return;
1421 }
1422
1423 migrate_params_apply(params, errp);
1424 }
1425
1426
1427 void qmp_migrate_start_postcopy(Error **errp)
1428 {
1429 MigrationState *s = migrate_get_current();
1430
1431 if (!migrate_postcopy()) {
1432 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1433 " the start of migration");
1434 return;
1435 }
1436
1437 if (s->state == MIGRATION_STATUS_NONE) {
1438 error_setg(errp, "Postcopy must be started after migration has been"
1439 " started");
1440 return;
1441 }
1442 /*
1443 * we don't error if migration has finished since that would be racy
1444 * with issuing this command.
1445 */
1446 atomic_set(&s->start_postcopy, true);
1447 }
1448
1449 /* shared migration helpers */
1450
1451 void migrate_set_state(int *state, int old_state, int new_state)
1452 {
1453 assert(new_state < MIGRATION_STATUS__MAX);
1454 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1455 trace_migrate_set_state(MigrationStatus_str(new_state));
1456 migrate_generate_event(new_state);
1457 }
1458 }
1459
1460 static MigrationCapabilityStatusList *migrate_cap_add(
1461 MigrationCapabilityStatusList *list,
1462 MigrationCapability index,
1463 bool state)
1464 {
1465 MigrationCapabilityStatusList *cap;
1466
1467 cap = g_new0(MigrationCapabilityStatusList, 1);
1468 cap->value = g_new0(MigrationCapabilityStatus, 1);
1469 cap->value->capability = index;
1470 cap->value->state = state;
1471 cap->next = list;
1472
1473 return cap;
1474 }
1475
1476 void migrate_set_block_enabled(bool value, Error **errp)
1477 {
1478 MigrationCapabilityStatusList *cap;
1479
1480 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1481 qmp_migrate_set_capabilities(cap, errp);
1482 qapi_free_MigrationCapabilityStatusList(cap);
1483 }
1484
1485 static void migrate_set_block_incremental(MigrationState *s, bool value)
1486 {
1487 s->parameters.block_incremental = value;
1488 }
1489
1490 static void block_cleanup_parameters(MigrationState *s)
1491 {
1492 if (s->must_remove_block_options) {
1493 /* setting to false can never fail */
1494 migrate_set_block_enabled(false, &error_abort);
1495 migrate_set_block_incremental(s, false);
1496 s->must_remove_block_options = false;
1497 }
1498 }
1499
1500 static void migrate_fd_cleanup(MigrationState *s)
1501 {
1502 qemu_bh_delete(s->cleanup_bh);
1503 s->cleanup_bh = NULL;
1504
1505 qemu_savevm_state_cleanup();
1506
1507 if (s->to_dst_file) {
1508 QEMUFile *tmp;
1509
1510 trace_migrate_fd_cleanup();
1511 qemu_mutex_unlock_iothread();
1512 if (s->migration_thread_running) {
1513 qemu_thread_join(&s->thread);
1514 s->migration_thread_running = false;
1515 }
1516 qemu_mutex_lock_iothread();
1517
1518 multifd_save_cleanup();
1519 qemu_mutex_lock(&s->qemu_file_lock);
1520 tmp = s->to_dst_file;
1521 s->to_dst_file = NULL;
1522 qemu_mutex_unlock(&s->qemu_file_lock);
1523 /*
1524 * Close the file handle without the lock to make sure the
1525 * critical section won't block for long.
1526 */
1527 qemu_fclose(tmp);
1528 }
1529
1530 assert(!migration_is_active(s));
1531
1532 if (s->state == MIGRATION_STATUS_CANCELLING) {
1533 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1534 MIGRATION_STATUS_CANCELLED);
1535 }
1536
1537 if (s->error) {
1538 /* It is used on info migrate. We can't free it */
1539 error_report_err(error_copy(s->error));
1540 }
1541 notifier_list_notify(&migration_state_notifiers, s);
1542 block_cleanup_parameters(s);
1543 }
1544
1545 static void migrate_fd_cleanup_schedule(MigrationState *s)
1546 {
1547 /*
1548 * Ref the state for bh, because it may be called when
1549 * there're already no other refs
1550 */
1551 object_ref(OBJECT(s));
1552 qemu_bh_schedule(s->cleanup_bh);
1553 }
1554
1555 static void migrate_fd_cleanup_bh(void *opaque)
1556 {
1557 MigrationState *s = opaque;
1558 migrate_fd_cleanup(s);
1559 object_unref(OBJECT(s));
1560 }
1561
1562 void migrate_set_error(MigrationState *s, const Error *error)
1563 {
1564 qemu_mutex_lock(&s->error_mutex);
1565 if (!s->error) {
1566 s->error = error_copy(error);
1567 }
1568 qemu_mutex_unlock(&s->error_mutex);
1569 }
1570
1571 void migrate_fd_error(MigrationState *s, const Error *error)
1572 {
1573 trace_migrate_fd_error(error_get_pretty(error));
1574 assert(s->to_dst_file == NULL);
1575 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1576 MIGRATION_STATUS_FAILED);
1577 migrate_set_error(s, error);
1578 }
1579
1580 static void migrate_fd_cancel(MigrationState *s)
1581 {
1582 int old_state ;
1583 QEMUFile *f = migrate_get_current()->to_dst_file;
1584 trace_migrate_fd_cancel();
1585
1586 if (s->rp_state.from_dst_file) {
1587 /* shutdown the rp socket, so causing the rp thread to shutdown */
1588 qemu_file_shutdown(s->rp_state.from_dst_file);
1589 }
1590
1591 do {
1592 old_state = s->state;
1593 if (!migration_is_setup_or_active(old_state)) {
1594 break;
1595 }
1596 /* If the migration is paused, kick it out of the pause */
1597 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1598 qemu_sem_post(&s->pause_sem);
1599 }
1600 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1601 } while (s->state != MIGRATION_STATUS_CANCELLING);
1602
1603 /*
1604 * If we're unlucky the migration code might be stuck somewhere in a
1605 * send/write while the network has failed and is waiting to timeout;
1606 * if we've got shutdown(2) available then we can force it to quit.
1607 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1608 * called in a bh, so there is no race against this cancel.
1609 */
1610 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1611 qemu_file_shutdown(f);
1612 }
1613 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1614 Error *local_err = NULL;
1615
1616 bdrv_invalidate_cache_all(&local_err);
1617 if (local_err) {
1618 error_report_err(local_err);
1619 } else {
1620 s->block_inactive = false;
1621 }
1622 }
1623 }
1624
1625 void add_migration_state_change_notifier(Notifier *notify)
1626 {
1627 notifier_list_add(&migration_state_notifiers, notify);
1628 }
1629
1630 void remove_migration_state_change_notifier(Notifier *notify)
1631 {
1632 notifier_remove(notify);
1633 }
1634
1635 bool migration_in_setup(MigrationState *s)
1636 {
1637 return s->state == MIGRATION_STATUS_SETUP;
1638 }
1639
1640 bool migration_has_finished(MigrationState *s)
1641 {
1642 return s->state == MIGRATION_STATUS_COMPLETED;
1643 }
1644
1645 bool migration_has_failed(MigrationState *s)
1646 {
1647 return (s->state == MIGRATION_STATUS_CANCELLED ||
1648 s->state == MIGRATION_STATUS_FAILED);
1649 }
1650
1651 bool migration_in_postcopy(void)
1652 {
1653 MigrationState *s = migrate_get_current();
1654
1655 switch (s->state) {
1656 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1657 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1658 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1659 return true;
1660 default:
1661 return false;
1662 }
1663 }
1664
1665 bool migration_in_postcopy_after_devices(MigrationState *s)
1666 {
1667 return migration_in_postcopy() && s->postcopy_after_devices;
1668 }
1669
1670 bool migration_is_idle(void)
1671 {
1672 MigrationState *s = current_migration;
1673
1674 if (!s) {
1675 return true;
1676 }
1677
1678 switch (s->state) {
1679 case MIGRATION_STATUS_NONE:
1680 case MIGRATION_STATUS_CANCELLED:
1681 case MIGRATION_STATUS_COMPLETED:
1682 case MIGRATION_STATUS_FAILED:
1683 return true;
1684 case MIGRATION_STATUS_SETUP:
1685 case MIGRATION_STATUS_CANCELLING:
1686 case MIGRATION_STATUS_ACTIVE:
1687 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1688 case MIGRATION_STATUS_COLO:
1689 case MIGRATION_STATUS_PRE_SWITCHOVER:
1690 case MIGRATION_STATUS_DEVICE:
1691 case MIGRATION_STATUS_WAIT_UNPLUG:
1692 return false;
1693 case MIGRATION_STATUS__MAX:
1694 g_assert_not_reached();
1695 }
1696
1697 return false;
1698 }
1699
1700 bool migration_is_active(MigrationState *s)
1701 {
1702 return (s->state == MIGRATION_STATUS_ACTIVE ||
1703 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1704 }
1705
1706 void migrate_init(MigrationState *s)
1707 {
1708 /*
1709 * Reinitialise all migration state, except
1710 * parameters/capabilities that the user set, and
1711 * locks.
1712 */
1713 s->cleanup_bh = 0;
1714 s->to_dst_file = NULL;
1715 s->state = MIGRATION_STATUS_NONE;
1716 s->rp_state.from_dst_file = NULL;
1717 s->rp_state.error = false;
1718 s->mbps = 0.0;
1719 s->pages_per_second = 0.0;
1720 s->downtime = 0;
1721 s->expected_downtime = 0;
1722 s->setup_time = 0;
1723 s->start_postcopy = false;
1724 s->postcopy_after_devices = false;
1725 s->migration_thread_running = false;
1726 error_free(s->error);
1727 s->error = NULL;
1728
1729 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1730
1731 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1732 s->total_time = 0;
1733 s->vm_was_running = false;
1734 s->iteration_initial_bytes = 0;
1735 s->threshold_size = 0;
1736 }
1737
1738 static GSList *migration_blockers;
1739
1740 int migrate_add_blocker(Error *reason, Error **errp)
1741 {
1742 if (only_migratable) {
1743 error_propagate_prepend(errp, error_copy(reason),
1744 "disallowing migration blocker "
1745 "(--only-migratable) for: ");
1746 return -EACCES;
1747 }
1748
1749 if (migration_is_idle()) {
1750 migration_blockers = g_slist_prepend(migration_blockers, reason);
1751 return 0;
1752 }
1753
1754 error_propagate_prepend(errp, error_copy(reason),
1755 "disallowing migration blocker "
1756 "(migration in progress) for: ");
1757 return -EBUSY;
1758 }
1759
1760 void migrate_del_blocker(Error *reason)
1761 {
1762 migration_blockers = g_slist_remove(migration_blockers, reason);
1763 }
1764
1765 void qmp_migrate_incoming(const char *uri, Error **errp)
1766 {
1767 Error *local_err = NULL;
1768 static bool once = true;
1769
1770 if (!deferred_incoming) {
1771 error_setg(errp, "For use with '-incoming defer'");
1772 return;
1773 }
1774 if (!once) {
1775 error_setg(errp, "The incoming migration has already been started");
1776 return;
1777 }
1778
1779 qemu_start_incoming_migration(uri, &local_err);
1780
1781 if (local_err) {
1782 error_propagate(errp, local_err);
1783 return;
1784 }
1785
1786 once = false;
1787 }
1788
1789 void qmp_migrate_recover(const char *uri, Error **errp)
1790 {
1791 MigrationIncomingState *mis = migration_incoming_get_current();
1792
1793 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1794 error_setg(errp, "Migrate recover can only be run "
1795 "when postcopy is paused.");
1796 return;
1797 }
1798
1799 if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1800 false, true) == true) {
1801 error_setg(errp, "Migrate recovery is triggered already");
1802 return;
1803 }
1804
1805 /*
1806 * Note that this call will never start a real migration; it will
1807 * only re-setup the migration stream and poke existing migration
1808 * to continue using that newly established channel.
1809 */
1810 qemu_start_incoming_migration(uri, errp);
1811 }
1812
1813 void qmp_migrate_pause(Error **errp)
1814 {
1815 MigrationState *ms = migrate_get_current();
1816 MigrationIncomingState *mis = migration_incoming_get_current();
1817 int ret;
1818
1819 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1820 /* Source side, during postcopy */
1821 qemu_mutex_lock(&ms->qemu_file_lock);
1822 ret = qemu_file_shutdown(ms->to_dst_file);
1823 qemu_mutex_unlock(&ms->qemu_file_lock);
1824 if (ret) {
1825 error_setg(errp, "Failed to pause source migration");
1826 }
1827 return;
1828 }
1829
1830 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1831 ret = qemu_file_shutdown(mis->from_src_file);
1832 if (ret) {
1833 error_setg(errp, "Failed to pause destination migration");
1834 }
1835 return;
1836 }
1837
1838 error_setg(errp, "migrate-pause is currently only supported "
1839 "during postcopy-active state");
1840 }
1841
1842 bool migration_is_blocked(Error **errp)
1843 {
1844 if (qemu_savevm_state_blocked(errp)) {
1845 return true;
1846 }
1847
1848 if (migration_blockers) {
1849 error_propagate(errp, error_copy(migration_blockers->data));
1850 return true;
1851 }
1852
1853 return false;
1854 }
1855
1856 /* Returns true if continue to migrate, or false if error detected */
1857 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1858 bool resume, Error **errp)
1859 {
1860 Error *local_err = NULL;
1861
1862 if (resume) {
1863 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1864 error_setg(errp, "Cannot resume if there is no "
1865 "paused migration");
1866 return false;
1867 }
1868
1869 /*
1870 * Postcopy recovery won't work well with release-ram
1871 * capability since release-ram will drop the page buffer as
1872 * long as the page is put into the send buffer. So if there
1873 * is a network failure happened, any page buffers that have
1874 * not yet reached the destination VM but have already been
1875 * sent from the source VM will be lost forever. Let's refuse
1876 * the client from resuming such a postcopy migration.
1877 * Luckily release-ram was designed to only be used when src
1878 * and destination VMs are on the same host, so it should be
1879 * fine.
1880 */
1881 if (migrate_release_ram()) {
1882 error_setg(errp, "Postcopy recovery cannot work "
1883 "when release-ram capability is set");
1884 return false;
1885 }
1886
1887 /* This is a resume, skip init status */
1888 return true;
1889 }
1890
1891 if (migration_is_setup_or_active(s->state) ||
1892 s->state == MIGRATION_STATUS_CANCELLING ||
1893 s->state == MIGRATION_STATUS_COLO) {
1894 error_setg(errp, QERR_MIGRATION_ACTIVE);
1895 return false;
1896 }
1897
1898 if (runstate_check(RUN_STATE_INMIGRATE)) {
1899 error_setg(errp, "Guest is waiting for an incoming migration");
1900 return false;
1901 }
1902
1903 if (migration_is_blocked(errp)) {
1904 return false;
1905 }
1906
1907 if (blk || blk_inc) {
1908 if (migrate_use_block() || migrate_use_block_incremental()) {
1909 error_setg(errp, "Command options are incompatible with "
1910 "current migration capabilities");
1911 return false;
1912 }
1913 migrate_set_block_enabled(true, &local_err);
1914 if (local_err) {
1915 error_propagate(errp, local_err);
1916 return false;
1917 }
1918 s->must_remove_block_options = true;
1919 }
1920
1921 if (blk_inc) {
1922 migrate_set_block_incremental(s, true);
1923 }
1924
1925 migrate_init(s);
1926 /*
1927 * set ram_counters memory to zero for a
1928 * new migration
1929 */
1930 memset(&ram_counters, 0, sizeof(ram_counters));
1931
1932 return true;
1933 }
1934
1935 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1936 bool has_inc, bool inc, bool has_detach, bool detach,
1937 bool has_resume, bool resume, Error **errp)
1938 {
1939 Error *local_err = NULL;
1940 MigrationState *s = migrate_get_current();
1941 const char *p;
1942
1943 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1944 has_resume && resume, errp)) {
1945 /* Error detected, put into errp */
1946 return;
1947 }
1948
1949 if (strstart(uri, "tcp:", &p)) {
1950 tcp_start_outgoing_migration(s, p, &local_err);
1951 #ifdef CONFIG_RDMA
1952 } else if (strstart(uri, "rdma:", &p)) {
1953 rdma_start_outgoing_migration(s, p, &local_err);
1954 #endif
1955 } else if (strstart(uri, "exec:", &p)) {
1956 exec_start_outgoing_migration(s, p, &local_err);
1957 } else if (strstart(uri, "unix:", &p)) {
1958 unix_start_outgoing_migration(s, p, &local_err);
1959 } else if (strstart(uri, "fd:", &p)) {
1960 fd_start_outgoing_migration(s, p, &local_err);
1961 } else {
1962 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1963 "a valid migration protocol");
1964 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1965 MIGRATION_STATUS_FAILED);
1966 block_cleanup_parameters(s);
1967 return;
1968 }
1969
1970 if (local_err) {
1971 migrate_fd_error(s, local_err);
1972 error_propagate(errp, local_err);
1973 return;
1974 }
1975 }
1976
1977 void qmp_migrate_cancel(Error **errp)
1978 {
1979 migrate_fd_cancel(migrate_get_current());
1980 }
1981
1982 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1983 {
1984 MigrationState *s = migrate_get_current();
1985 if (s->state != state) {
1986 error_setg(errp, "Migration not in expected state: %s",
1987 MigrationStatus_str(s->state));
1988 return;
1989 }
1990 qemu_sem_post(&s->pause_sem);
1991 }
1992
1993 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1994 {
1995 MigrateSetParameters p = {
1996 .has_xbzrle_cache_size = true,
1997 .xbzrle_cache_size = value,
1998 };
1999
2000 qmp_migrate_set_parameters(&p, errp);
2001 }
2002
2003 int64_t qmp_query_migrate_cache_size(Error **errp)
2004 {
2005 return migrate_xbzrle_cache_size();
2006 }
2007
2008 void qmp_migrate_set_speed(int64_t value, Error **errp)
2009 {
2010 MigrateSetParameters p = {
2011 .has_max_bandwidth = true,
2012 .max_bandwidth = value,
2013 };
2014
2015 qmp_migrate_set_parameters(&p, errp);
2016 }
2017
2018 void qmp_migrate_set_downtime(double value, Error **errp)
2019 {
2020 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2021 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
2022 "the range of 0 to %d seconds",
2023 MAX_MIGRATE_DOWNTIME_SECONDS);
2024 return;
2025 }
2026
2027 value *= 1000; /* Convert to milliseconds */
2028
2029 MigrateSetParameters p = {
2030 .has_downtime_limit = true,
2031 .downtime_limit = (int64_t)value,
2032 };
2033
2034 qmp_migrate_set_parameters(&p, errp);
2035 }
2036
2037 bool migrate_release_ram(void)
2038 {
2039 MigrationState *s;
2040
2041 s = migrate_get_current();
2042
2043 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2044 }
2045
2046 bool migrate_postcopy_ram(void)
2047 {
2048 MigrationState *s;
2049
2050 s = migrate_get_current();
2051
2052 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2053 }
2054
2055 bool migrate_postcopy(void)
2056 {
2057 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2058 }
2059
2060 bool migrate_auto_converge(void)
2061 {
2062 MigrationState *s;
2063
2064 s = migrate_get_current();
2065
2066 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2067 }
2068
2069 bool migrate_zero_blocks(void)
2070 {
2071 MigrationState *s;
2072
2073 s = migrate_get_current();
2074
2075 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2076 }
2077
2078 bool migrate_postcopy_blocktime(void)
2079 {
2080 MigrationState *s;
2081
2082 s = migrate_get_current();
2083
2084 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2085 }
2086
2087 bool migrate_use_compression(void)
2088 {
2089 MigrationState *s;
2090
2091 s = migrate_get_current();
2092
2093 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2094 }
2095
2096 int migrate_compress_level(void)
2097 {
2098 MigrationState *s;
2099
2100 s = migrate_get_current();
2101
2102 return s->parameters.compress_level;
2103 }
2104
2105 int migrate_compress_threads(void)
2106 {
2107 MigrationState *s;
2108
2109 s = migrate_get_current();
2110
2111 return s->parameters.compress_threads;
2112 }
2113
2114 int migrate_compress_wait_thread(void)
2115 {
2116 MigrationState *s;
2117
2118 s = migrate_get_current();
2119
2120 return s->parameters.compress_wait_thread;
2121 }
2122
2123 int migrate_decompress_threads(void)
2124 {
2125 MigrationState *s;
2126
2127 s = migrate_get_current();
2128
2129 return s->parameters.decompress_threads;
2130 }
2131
2132 bool migrate_dirty_bitmaps(void)
2133 {
2134 MigrationState *s;
2135
2136 s = migrate_get_current();
2137
2138 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2139 }
2140
2141 bool migrate_ignore_shared(void)
2142 {
2143 MigrationState *s;
2144
2145 s = migrate_get_current();
2146
2147 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2148 }
2149
2150 bool migrate_validate_uuid(void)
2151 {
2152 MigrationState *s;
2153
2154 s = migrate_get_current();
2155
2156 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2157 }
2158
2159 bool migrate_use_events(void)
2160 {
2161 MigrationState *s;
2162
2163 s = migrate_get_current();
2164
2165 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2166 }
2167
2168 bool migrate_use_multifd(void)
2169 {
2170 MigrationState *s;
2171
2172 s = migrate_get_current();
2173
2174 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2175 }
2176
2177 bool migrate_pause_before_switchover(void)
2178 {
2179 MigrationState *s;
2180
2181 s = migrate_get_current();
2182
2183 return s->enabled_capabilities[
2184 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2185 }
2186
2187 int migrate_multifd_channels(void)
2188 {
2189 MigrationState *s;
2190
2191 s = migrate_get_current();
2192
2193 return s->parameters.multifd_channels;
2194 }
2195
2196 int migrate_use_xbzrle(void)
2197 {
2198 MigrationState *s;
2199
2200 s = migrate_get_current();
2201
2202 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2203 }
2204
2205 int64_t migrate_xbzrle_cache_size(void)
2206 {
2207 MigrationState *s;
2208
2209 s = migrate_get_current();
2210
2211 return s->parameters.xbzrle_cache_size;
2212 }
2213
2214 static int64_t migrate_max_postcopy_bandwidth(void)
2215 {
2216 MigrationState *s;
2217
2218 s = migrate_get_current();
2219
2220 return s->parameters.max_postcopy_bandwidth;
2221 }
2222
2223 bool migrate_use_block(void)
2224 {
2225 MigrationState *s;
2226
2227 s = migrate_get_current();
2228
2229 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2230 }
2231
2232 bool migrate_use_return_path(void)
2233 {
2234 MigrationState *s;
2235
2236 s = migrate_get_current();
2237
2238 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2239 }
2240
2241 bool migrate_use_block_incremental(void)
2242 {
2243 MigrationState *s;
2244
2245 s = migrate_get_current();
2246
2247 return s->parameters.block_incremental;
2248 }
2249
2250 /* migration thread support */
2251 /*
2252 * Something bad happened to the RP stream, mark an error
2253 * The caller shall print or trace something to indicate why
2254 */
2255 static void mark_source_rp_bad(MigrationState *s)
2256 {
2257 s->rp_state.error = true;
2258 }
2259
2260 static struct rp_cmd_args {
2261 ssize_t len; /* -1 = variable */
2262 const char *name;
2263 } rp_cmd_args[] = {
2264 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2265 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2266 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2267 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2268 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2269 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2270 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2271 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2272 };
2273
2274 /*
2275 * Process a request for pages received on the return path,
2276 * We're allowed to send more than requested (e.g. to round to our page size)
2277 * and we don't need to send pages that have already been sent.
2278 */
2279 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2280 ram_addr_t start, size_t len)
2281 {
2282 long our_host_ps = qemu_real_host_page_size;
2283
2284 trace_migrate_handle_rp_req_pages(rbname, start, len);
2285
2286 /*
2287 * Since we currently insist on matching page sizes, just sanity check
2288 * we're being asked for whole host pages.
2289 */
2290 if (start & (our_host_ps-1) ||
2291 (len & (our_host_ps-1))) {
2292 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2293 " len: %zd", __func__, start, len);
2294 mark_source_rp_bad(ms);
2295 return;
2296 }
2297
2298 if (ram_save_queue_pages(rbname, start, len)) {
2299 mark_source_rp_bad(ms);
2300 }
2301 }
2302
2303 /* Return true to retry, false to quit */
2304 static bool postcopy_pause_return_path_thread(MigrationState *s)
2305 {
2306 trace_postcopy_pause_return_path();
2307
2308 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2309
2310 trace_postcopy_pause_return_path_continued();
2311
2312 return true;
2313 }
2314
2315 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2316 {
2317 RAMBlock *block = qemu_ram_block_by_name(block_name);
2318
2319 if (!block) {
2320 error_report("%s: invalid block name '%s'", __func__, block_name);
2321 return -EINVAL;
2322 }
2323
2324 /* Fetch the received bitmap and refresh the dirty bitmap */
2325 return ram_dirty_bitmap_reload(s, block);
2326 }
2327
2328 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2329 {
2330 trace_source_return_path_thread_resume_ack(value);
2331
2332 if (value != MIGRATION_RESUME_ACK_VALUE) {
2333 error_report("%s: illegal resume_ack value %"PRIu32,
2334 __func__, value);
2335 return -1;
2336 }
2337
2338 /* Now both sides are active. */
2339 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2340 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2341
2342 /* Notify send thread that time to continue send pages */
2343 qemu_sem_post(&s->rp_state.rp_sem);
2344
2345 return 0;
2346 }
2347
2348 /*
2349 * Handles messages sent on the return path towards the source VM
2350 *
2351 */
2352 static void *source_return_path_thread(void *opaque)
2353 {
2354 MigrationState *ms = opaque;
2355 QEMUFile *rp = ms->rp_state.from_dst_file;
2356 uint16_t header_len, header_type;
2357 uint8_t buf[512];
2358 uint32_t tmp32, sibling_error;
2359 ram_addr_t start = 0; /* =0 to silence warning */
2360 size_t len = 0, expected_len;
2361 int res;
2362
2363 trace_source_return_path_thread_entry();
2364 rcu_register_thread();
2365
2366 retry:
2367 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2368 migration_is_setup_or_active(ms->state)) {
2369 trace_source_return_path_thread_loop_top();
2370 header_type = qemu_get_be16(rp);
2371 header_len = qemu_get_be16(rp);
2372
2373 if (qemu_file_get_error(rp)) {
2374 mark_source_rp_bad(ms);
2375 goto out;
2376 }
2377
2378 if (header_type >= MIG_RP_MSG_MAX ||
2379 header_type == MIG_RP_MSG_INVALID) {
2380 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2381 header_type, header_len);
2382 mark_source_rp_bad(ms);
2383 goto out;
2384 }
2385
2386 if ((rp_cmd_args[header_type].len != -1 &&
2387 header_len != rp_cmd_args[header_type].len) ||
2388 header_len > sizeof(buf)) {
2389 error_report("RP: Received '%s' message (0x%04x) with"
2390 "incorrect length %d expecting %zu",
2391 rp_cmd_args[header_type].name, header_type, header_len,
2392 (size_t)rp_cmd_args[header_type].len);
2393 mark_source_rp_bad(ms);
2394 goto out;
2395 }
2396
2397 /* We know we've got a valid header by this point */
2398 res = qemu_get_buffer(rp, buf, header_len);
2399 if (res != header_len) {
2400 error_report("RP: Failed reading data for message 0x%04x"
2401 " read %d expected %d",
2402 header_type, res, header_len);
2403 mark_source_rp_bad(ms);
2404 goto out;
2405 }
2406
2407 /* OK, we have the message and the data */
2408 switch (header_type) {
2409 case MIG_RP_MSG_SHUT:
2410 sibling_error = ldl_be_p(buf);
2411 trace_source_return_path_thread_shut(sibling_error);
2412 if (sibling_error) {
2413 error_report("RP: Sibling indicated error %d", sibling_error);
2414 mark_source_rp_bad(ms);
2415 }
2416 /*
2417 * We'll let the main thread deal with closing the RP
2418 * we could do a shutdown(2) on it, but we're the only user
2419 * anyway, so there's nothing gained.
2420 */
2421 goto out;
2422
2423 case MIG_RP_MSG_PONG:
2424 tmp32 = ldl_be_p(buf);
2425 trace_source_return_path_thread_pong(tmp32);
2426 break;
2427
2428 case MIG_RP_MSG_REQ_PAGES:
2429 start = ldq_be_p(buf);
2430 len = ldl_be_p(buf + 8);
2431 migrate_handle_rp_req_pages(ms, NULL, start, len);
2432 break;
2433
2434 case MIG_RP_MSG_REQ_PAGES_ID:
2435 expected_len = 12 + 1; /* header + termination */
2436
2437 if (header_len >= expected_len) {
2438 start = ldq_be_p(buf);
2439 len = ldl_be_p(buf + 8);
2440 /* Now we expect an idstr */
2441 tmp32 = buf[12]; /* Length of the following idstr */
2442 buf[13 + tmp32] = '\0';
2443 expected_len += tmp32;
2444 }
2445 if (header_len != expected_len) {
2446 error_report("RP: Req_Page_id with length %d expecting %zd",
2447 header_len, expected_len);
2448 mark_source_rp_bad(ms);
2449 goto out;
2450 }
2451 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2452 break;
2453
2454 case MIG_RP_MSG_RECV_BITMAP:
2455 if (header_len < 1) {
2456 error_report("%s: missing block name", __func__);
2457 mark_source_rp_bad(ms);
2458 goto out;
2459 }
2460 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2461 buf[buf[0] + 1] = '\0';
2462 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2463 mark_source_rp_bad(ms);
2464 goto out;
2465 }
2466 break;
2467
2468 case MIG_RP_MSG_RESUME_ACK:
2469 tmp32 = ldl_be_p(buf);
2470 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2471 mark_source_rp_bad(ms);
2472 goto out;
2473 }
2474 break;
2475
2476 default:
2477 break;
2478 }
2479 }
2480
2481 out:
2482 res = qemu_file_get_error(rp);
2483 if (res) {
2484 if (res == -EIO && migration_in_postcopy()) {
2485 /*
2486 * Maybe there is something we can do: it looks like a
2487 * network down issue, and we pause for a recovery.
2488 */
2489 if (postcopy_pause_return_path_thread(ms)) {
2490 /* Reload rp, reset the rest */
2491 if (rp != ms->rp_state.from_dst_file) {
2492 qemu_fclose(rp);
2493 rp = ms->rp_state.from_dst_file;
2494 }
2495 ms->rp_state.error = false;
2496 goto retry;
2497 }
2498 }
2499
2500 trace_source_return_path_thread_bad_end();
2501 mark_source_rp_bad(ms);
2502 }
2503
2504 trace_source_return_path_thread_end();
2505 ms->rp_state.from_dst_file = NULL;
2506 qemu_fclose(rp);
2507 rcu_unregister_thread();
2508 return NULL;
2509 }
2510
2511 static int open_return_path_on_source(MigrationState *ms,
2512 bool create_thread)
2513 {
2514
2515 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2516 if (!ms->rp_state.from_dst_file) {
2517 return -1;
2518 }
2519
2520 trace_open_return_path_on_source();
2521
2522 if (!create_thread) {
2523 /* We're done */
2524 return 0;
2525 }
2526
2527 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2528 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2529
2530 trace_open_return_path_on_source_continue();
2531
2532 return 0;
2533 }
2534
2535 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2536 static int await_return_path_close_on_source(MigrationState *ms)
2537 {
2538 /*
2539 * If this is a normal exit then the destination will send a SHUT and the
2540 * rp_thread will exit, however if there's an error we need to cause
2541 * it to exit.
2542 */
2543 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2544 /*
2545 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2546 * waiting for the destination.
2547 */
2548 qemu_file_shutdown(ms->rp_state.from_dst_file);
2549 mark_source_rp_bad(ms);
2550 }
2551 trace_await_return_path_close_on_source_joining();
2552 qemu_thread_join(&ms->rp_state.rp_thread);
2553 trace_await_return_path_close_on_source_close();
2554 return ms->rp_state.error;
2555 }
2556
2557 /*
2558 * Switch from normal iteration to postcopy
2559 * Returns non-0 on error
2560 */
2561 static int postcopy_start(MigrationState *ms)
2562 {
2563 int ret;
2564 QIOChannelBuffer *bioc;
2565 QEMUFile *fb;
2566 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2567 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2568 bool restart_block = false;
2569 int cur_state = MIGRATION_STATUS_ACTIVE;
2570 if (!migrate_pause_before_switchover()) {
2571 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2572 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2573 }
2574
2575 trace_postcopy_start();
2576 qemu_mutex_lock_iothread();
2577 trace_postcopy_start_set_run();
2578
2579 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2580 global_state_store();
2581 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2582 if (ret < 0) {
2583 goto fail;
2584 }
2585
2586 ret = migration_maybe_pause(ms, &cur_state,
2587 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2588 if (ret < 0) {
2589 goto fail;
2590 }
2591
2592 ret = bdrv_inactivate_all();
2593 if (ret < 0) {
2594 goto fail;
2595 }
2596 restart_block = true;
2597
2598 /*
2599 * Cause any non-postcopiable, but iterative devices to
2600 * send out their final data.
2601 */
2602 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2603
2604 /*
2605 * in Finish migrate and with the io-lock held everything should
2606 * be quiet, but we've potentially still got dirty pages and we
2607 * need to tell the destination to throw any pages it's already received
2608 * that are dirty
2609 */
2610 if (migrate_postcopy_ram()) {
2611 if (ram_postcopy_send_discard_bitmap(ms)) {
2612 error_report("postcopy send discard bitmap failed");
2613 goto fail;
2614 }
2615 }
2616
2617 /*
2618 * send rest of state - note things that are doing postcopy
2619 * will notice we're in POSTCOPY_ACTIVE and not actually
2620 * wrap their state up here
2621 */
2622 /* 0 max-postcopy-bandwidth means unlimited */
2623 if (!bandwidth) {
2624 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2625 } else {
2626 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2627 }
2628 if (migrate_postcopy_ram()) {
2629 /* Ping just for debugging, helps line traces up */
2630 qemu_savevm_send_ping(ms->to_dst_file, 2);
2631 }
2632
2633 /*
2634 * While loading the device state we may trigger page transfer
2635 * requests and the fd must be free to process those, and thus
2636 * the destination must read the whole device state off the fd before
2637 * it starts processing it. Unfortunately the ad-hoc migration format
2638 * doesn't allow the destination to know the size to read without fully
2639 * parsing it through each devices load-state code (especially the open
2640 * coded devices that use get/put).
2641 * So we wrap the device state up in a package with a length at the start;
2642 * to do this we use a qemu_buf to hold the whole of the device state.
2643 */
2644 bioc = qio_channel_buffer_new(4096);
2645 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2646 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2647 object_unref(OBJECT(bioc));
2648
2649 /*
2650 * Make sure the receiver can get incoming pages before we send the rest
2651 * of the state
2652 */
2653 qemu_savevm_send_postcopy_listen(fb);
2654
2655 qemu_savevm_state_complete_precopy(fb, false, false);
2656 if (migrate_postcopy_ram()) {
2657 qemu_savevm_send_ping(fb, 3);
2658 }
2659
2660 qemu_savevm_send_postcopy_run(fb);
2661
2662 /* <><> end of stuff going into the package */
2663
2664 /* Last point of recovery; as soon as we send the package the destination
2665 * can open devices and potentially start running.
2666 * Lets just check again we've not got any errors.
2667 */
2668 ret = qemu_file_get_error(ms->to_dst_file);
2669 if (ret) {
2670 error_report("postcopy_start: Migration stream errored (pre package)");
2671 goto fail_closefb;
2672 }
2673
2674 restart_block = false;
2675
2676 /* Now send that blob */
2677 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2678 goto fail_closefb;
2679 }
2680 qemu_fclose(fb);
2681
2682 /* Send a notify to give a chance for anything that needs to happen
2683 * at the transition to postcopy and after the device state; in particular
2684 * spice needs to trigger a transition now
2685 */
2686 ms->postcopy_after_devices = true;
2687 notifier_list_notify(&migration_state_notifiers, ms);
2688
2689 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2690
2691 qemu_mutex_unlock_iothread();
2692
2693 if (migrate_postcopy_ram()) {
2694 /*
2695 * Although this ping is just for debug, it could potentially be
2696 * used for getting a better measurement of downtime at the source.
2697 */
2698 qemu_savevm_send_ping(ms->to_dst_file, 4);
2699 }
2700
2701 if (migrate_release_ram()) {
2702 ram_postcopy_migrated_memory_release(ms);
2703 }
2704
2705 ret = qemu_file_get_error(ms->to_dst_file);
2706 if (ret) {
2707 error_report("postcopy_start: Migration stream errored");
2708 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2709 MIGRATION_STATUS_FAILED);
2710 }
2711
2712 return ret;
2713
2714 fail_closefb:
2715 qemu_fclose(fb);
2716 fail:
2717 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2718 MIGRATION_STATUS_FAILED);
2719 if (restart_block) {
2720 /* A failure happened early enough that we know the destination hasn't
2721 * accessed block devices, so we're safe to recover.
2722 */
2723 Error *local_err = NULL;
2724
2725 bdrv_invalidate_cache_all(&local_err);
2726 if (local_err) {
2727 error_report_err(local_err);
2728 }
2729 }
2730 qemu_mutex_unlock_iothread();
2731 return -1;
2732 }
2733
2734 /**
2735 * migration_maybe_pause: Pause if required to by
2736 * migrate_pause_before_switchover called with the iothread locked
2737 * Returns: 0 on success
2738 */
2739 static int migration_maybe_pause(MigrationState *s,
2740 int *current_active_state,
2741 int new_state)
2742 {
2743 if (!migrate_pause_before_switchover()) {
2744 return 0;
2745 }
2746
2747 /* Since leaving this state is not atomic with posting the semaphore
2748 * it's possible that someone could have issued multiple migrate_continue
2749 * and the semaphore is incorrectly positive at this point;
2750 * the docs say it's undefined to reinit a semaphore that's already
2751 * init'd, so use timedwait to eat up any existing posts.
2752 */
2753 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2754 /* This block intentionally left blank */
2755 }
2756
2757 qemu_mutex_unlock_iothread();
2758 migrate_set_state(&s->state, *current_active_state,
2759 MIGRATION_STATUS_PRE_SWITCHOVER);
2760 qemu_sem_wait(&s->pause_sem);
2761 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2762 new_state);
2763 *current_active_state = new_state;
2764 qemu_mutex_lock_iothread();
2765
2766 return s->state == new_state ? 0 : -EINVAL;
2767 }
2768
2769 /**
2770 * migration_completion: Used by migration_thread when there's not much left.
2771 * The caller 'breaks' the loop when this returns.
2772 *
2773 * @s: Current migration state
2774 */
2775 static void migration_completion(MigrationState *s)
2776 {
2777 int ret;
2778 int current_active_state = s->state;
2779
2780 if (s->state == MIGRATION_STATUS_ACTIVE) {
2781 qemu_mutex_lock_iothread();
2782 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2783 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2784 s->vm_was_running = runstate_is_running();
2785 ret = global_state_store();
2786
2787 if (!ret) {
2788 bool inactivate = !migrate_colo_enabled();
2789 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2790 if (ret >= 0) {
2791 ret = migration_maybe_pause(s, &current_active_state,
2792 MIGRATION_STATUS_DEVICE);
2793 }
2794 if (ret >= 0) {
2795 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2796 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2797 inactivate);
2798 }
2799 if (inactivate && ret >= 0) {
2800 s->block_inactive = true;
2801 }
2802 }
2803 qemu_mutex_unlock_iothread();
2804
2805 if (ret < 0) {
2806 goto fail;
2807 }
2808 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2809 trace_migration_completion_postcopy_end();
2810
2811 qemu_savevm_state_complete_postcopy(s->to_dst_file);
2812 trace_migration_completion_postcopy_end_after_complete();
2813 }
2814
2815 /*
2816 * If rp was opened we must clean up the thread before
2817 * cleaning everything else up (since if there are no failures
2818 * it will wait for the destination to send it's status in
2819 * a SHUT command).
2820 */
2821 if (s->rp_state.from_dst_file) {
2822 int rp_error;
2823 trace_migration_return_path_end_before();
2824 rp_error = await_return_path_close_on_source(s);
2825 trace_migration_return_path_end_after(rp_error);
2826 if (rp_error) {
2827 goto fail_invalidate;
2828 }
2829 }
2830
2831 if (qemu_file_get_error(s->to_dst_file)) {
2832 trace_migration_completion_file_err();
2833 goto fail_invalidate;
2834 }
2835
2836 if (!migrate_colo_enabled()) {
2837 migrate_set_state(&s->state, current_active_state,
2838 MIGRATION_STATUS_COMPLETED);
2839 }
2840
2841 return;
2842
2843 fail_invalidate:
2844 /* If not doing postcopy, vm_start() will be called: let's regain
2845 * control on images.
2846 */
2847 if (s->state == MIGRATION_STATUS_ACTIVE ||
2848 s->state == MIGRATION_STATUS_DEVICE) {
2849 Error *local_err = NULL;
2850
2851 qemu_mutex_lock_iothread();
2852 bdrv_invalidate_cache_all(&local_err);
2853 if (local_err) {
2854 error_report_err(local_err);
2855 } else {
2856 s->block_inactive = false;
2857 }
2858 qemu_mutex_unlock_iothread();
2859 }
2860
2861 fail:
2862 migrate_set_state(&s->state, current_active_state,
2863 MIGRATION_STATUS_FAILED);
2864 }
2865
2866 bool migrate_colo_enabled(void)
2867 {
2868 MigrationState *s = migrate_get_current();
2869 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2870 }
2871
2872 typedef enum MigThrError {
2873 /* No error detected */
2874 MIG_THR_ERR_NONE = 0,
2875 /* Detected error, but resumed successfully */
2876 MIG_THR_ERR_RECOVERED = 1,
2877 /* Detected fatal error, need to exit */
2878 MIG_THR_ERR_FATAL = 2,
2879 } MigThrError;
2880
2881 static int postcopy_resume_handshake(MigrationState *s)
2882 {
2883 qemu_savevm_send_postcopy_resume(s->to_dst_file);
2884
2885 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2886 qemu_sem_wait(&s->rp_state.rp_sem);
2887 }
2888
2889 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2890 return 0;
2891 }
2892
2893 return -1;
2894 }
2895
2896 /* Return zero if success, or <0 for error */
2897 static int postcopy_do_resume(MigrationState *s)
2898 {
2899 int ret;
2900
2901 /*
2902 * Call all the resume_prepare() hooks, so that modules can be
2903 * ready for the migration resume.
2904 */
2905 ret = qemu_savevm_state_resume_prepare(s);
2906 if (ret) {
2907 error_report("%s: resume_prepare() failure detected: %d",
2908 __func__, ret);
2909 return ret;
2910 }
2911
2912 /*
2913 * Last handshake with destination on the resume (destination will
2914 * switch to postcopy-active afterwards)
2915 */
2916 ret = postcopy_resume_handshake(s);
2917 if (ret) {
2918 error_report("%s: handshake failed: %d", __func__, ret);
2919 return ret;
2920 }
2921
2922 return 0;
2923 }
2924
2925 /*
2926 * We don't return until we are in a safe state to continue current
2927 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2928 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2929 */
2930 static MigThrError postcopy_pause(MigrationState *s)
2931 {
2932 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2933
2934 while (true) {
2935 QEMUFile *file;
2936
2937 migrate_set_state(&s->state, s->state,
2938 MIGRATION_STATUS_POSTCOPY_PAUSED);
2939
2940 /* Current channel is possibly broken. Release it. */
2941 assert(s->to_dst_file);
2942 qemu_mutex_lock(&s->qemu_file_lock);
2943 file = s->to_dst_file;
2944 s->to_dst_file = NULL;
2945 qemu_mutex_unlock(&s->qemu_file_lock);
2946
2947 qemu_file_shutdown(file);
2948 qemu_fclose(file);
2949
2950 error_report("Detected IO failure for postcopy. "
2951 "Migration paused.");
2952
2953 /*
2954 * We wait until things fixed up. Then someone will setup the
2955 * status back for us.
2956 */
2957 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2958 qemu_sem_wait(&s->postcopy_pause_sem);
2959 }
2960
2961 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2962 /* Woken up by a recover procedure. Give it a shot */
2963
2964 /*
2965 * Firstly, let's wake up the return path now, with a new
2966 * return path channel.
2967 */
2968 qemu_sem_post(&s->postcopy_pause_rp_sem);
2969
2970 /* Do the resume logic */
2971 if (postcopy_do_resume(s) == 0) {
2972 /* Let's continue! */
2973 trace_postcopy_pause_continued();
2974 return MIG_THR_ERR_RECOVERED;
2975 } else {
2976 /*
2977 * Something wrong happened during the recovery, let's
2978 * pause again. Pause is always better than throwing
2979 * data away.
2980 */
2981 continue;
2982 }
2983 } else {
2984 /* This is not right... Time to quit. */
2985 return MIG_THR_ERR_FATAL;
2986 }
2987 }
2988 }
2989
2990 static MigThrError migration_detect_error(MigrationState *s)
2991 {
2992 int ret;
2993 int state = s->state;
2994 Error *local_error = NULL;
2995
2996 if (state == MIGRATION_STATUS_CANCELLING ||
2997 state == MIGRATION_STATUS_CANCELLED) {
2998 /* End the migration, but don't set the state to failed */
2999 return MIG_THR_ERR_FATAL;
3000 }
3001
3002 /* Try to detect any file errors */
3003 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3004 if (!ret) {
3005 /* Everything is fine */
3006 assert(!local_error);
3007 return MIG_THR_ERR_NONE;
3008 }
3009
3010 if (local_error) {
3011 migrate_set_error(s, local_error);
3012 error_free(local_error);
3013 }
3014
3015 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3016 /*
3017 * For postcopy, we allow the network to be down for a
3018 * while. After that, it can be continued by a
3019 * recovery phase.
3020 */
3021 return postcopy_pause(s);
3022 } else {
3023 /*
3024 * For precopy (or postcopy with error outside IO), we fail
3025 * with no time.
3026 */
3027 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3028 trace_migration_thread_file_err();
3029
3030 /* Time to stop the migration, now. */
3031 return MIG_THR_ERR_FATAL;
3032 }
3033 }
3034
3035 /* How many bytes have we transferred since the beginning of the migration */
3036 static uint64_t migration_total_bytes(MigrationState *s)
3037 {
3038 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3039 }
3040
3041 static void migration_calculate_complete(MigrationState *s)
3042 {
3043 uint64_t bytes = migration_total_bytes(s);
3044 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3045 int64_t transfer_time;
3046
3047 s->total_time = end_time - s->start_time;
3048 if (!s->downtime) {
3049 /*
3050 * It's still not set, so we are precopy migration. For
3051 * postcopy, downtime is calculated during postcopy_start().
3052 */
3053 s->downtime = end_time - s->downtime_start;
3054 }
3055
3056 transfer_time = s->total_time - s->setup_time;
3057 if (transfer_time) {
3058 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3059 }
3060 }
3061
3062 static void update_iteration_initial_status(MigrationState *s)
3063 {
3064 /*
3065 * Update these three fields at the same time to avoid mismatch info lead
3066 * wrong speed calculation.
3067 */
3068 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3069 s->iteration_initial_bytes = migration_total_bytes(s);
3070 s->iteration_initial_pages = ram_get_total_transferred_pages();
3071 }
3072
3073 static void migration_update_counters(MigrationState *s,
3074 int64_t current_time)
3075 {
3076 uint64_t transferred, transferred_pages, time_spent;
3077 uint64_t current_bytes; /* bytes transferred since the beginning */
3078 double bandwidth;
3079
3080 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3081 return;
3082 }
3083
3084 current_bytes = migration_total_bytes(s);
3085 transferred = current_bytes - s->iteration_initial_bytes;
3086 time_spent = current_time - s->iteration_start_time;
3087 bandwidth = (double)transferred / time_spent;
3088 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3089
3090 s->mbps = (((double) transferred * 8.0) /
3091 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3092
3093 transferred_pages = ram_get_total_transferred_pages() -
3094 s->iteration_initial_pages;
3095 s->pages_per_second = (double) transferred_pages /
3096 (((double) time_spent / 1000.0));
3097
3098 /*
3099 * if we haven't sent anything, we don't want to
3100 * recalculate. 10000 is a small enough number for our purposes
3101 */
3102 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3103 s->expected_downtime = ram_counters.remaining / bandwidth;
3104 }
3105
3106 qemu_file_reset_rate_limit(s->to_dst_file);
3107
3108 update_iteration_initial_status(s);
3109
3110 trace_migrate_transferred(transferred, time_spent,
3111 bandwidth, s->threshold_size);
3112 }
3113
3114 /* Migration thread iteration status */
3115 typedef enum {
3116 MIG_ITERATE_RESUME, /* Resume current iteration */
3117 MIG_ITERATE_SKIP, /* Skip current iteration */
3118 MIG_ITERATE_BREAK, /* Break the loop */
3119 } MigIterateState;
3120
3121 /*
3122 * Return true if continue to the next iteration directly, false
3123 * otherwise.
3124 */
3125 static MigIterateState migration_iteration_run(MigrationState *s)
3126 {
3127 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3128 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3129
3130 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3131 &pend_compat, &pend_post);
3132 pending_size = pend_pre + pend_compat + pend_post;
3133
3134 trace_migrate_pending(pending_size, s->threshold_size,
3135 pend_pre, pend_compat, pend_post);
3136
3137 if (pending_size && pending_size >= s->threshold_size) {
3138 /* Still a significant amount to transfer */
3139 if (!in_postcopy && pend_pre <= s->threshold_size &&
3140 atomic_read(&s->start_postcopy)) {
3141 if (postcopy_start(s)) {
3142 error_report("%s: postcopy failed to start", __func__);
3143 }
3144 return MIG_ITERATE_SKIP;
3145 }
3146 /* Just another iteration step */
3147 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3148 } else {
3149 trace_migration_thread_low_pending(pending_size);
3150 migration_completion(s);
3151 return MIG_ITERATE_BREAK;
3152 }
3153
3154 return MIG_ITERATE_RESUME;
3155 }
3156
3157 static void migration_iteration_finish(MigrationState *s)
3158 {
3159 /* If we enabled cpu throttling for auto-converge, turn it off. */
3160 cpu_throttle_stop();
3161
3162 qemu_mutex_lock_iothread();
3163 switch (s->state) {
3164 case MIGRATION_STATUS_COMPLETED:
3165 migration_calculate_complete(s);
3166 runstate_set(RUN_STATE_POSTMIGRATE);
3167 break;
3168
3169 case MIGRATION_STATUS_ACTIVE:
3170 /*
3171 * We should really assert here, but since it's during
3172 * migration, let's try to reduce the usage of assertions.
3173 */
3174 if (!migrate_colo_enabled()) {
3175 error_report("%s: critical error: calling COLO code without "
3176 "COLO enabled", __func__);
3177 }
3178 migrate_start_colo_process(s);
3179 /*
3180 * Fixme: we will run VM in COLO no matter its old running state.
3181 * After exited COLO, we will keep running.
3182 */
3183 s->vm_was_running = true;
3184 /* Fallthrough */
3185 case MIGRATION_STATUS_FAILED:
3186 case MIGRATION_STATUS_CANCELLED:
3187 case MIGRATION_STATUS_CANCELLING:
3188 if (s->vm_was_running) {
3189 vm_start();
3190 } else {
3191 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3192 runstate_set(RUN_STATE_POSTMIGRATE);
3193 }
3194 }
3195 break;
3196
3197 default:
3198 /* Should not reach here, but if so, forgive the VM. */
3199 error_report("%s: Unknown ending state %d", __func__, s->state);
3200 break;
3201 }
3202 migrate_fd_cleanup_schedule(s);
3203 qemu_mutex_unlock_iothread();
3204 }
3205
3206 void migration_make_urgent_request(void)
3207 {
3208 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3209 }
3210
3211 void migration_consume_urgent_request(void)
3212 {
3213 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3214 }
3215
3216 /* Returns true if the rate limiting was broken by an urgent request */
3217 bool migration_rate_limit(void)
3218 {
3219 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3220 MigrationState *s = migrate_get_current();
3221
3222 bool urgent = false;
3223 migration_update_counters(s, now);
3224 if (qemu_file_rate_limit(s->to_dst_file)) {
3225 /*
3226 * Wait for a delay to do rate limiting OR
3227 * something urgent to post the semaphore.
3228 */
3229 int ms = s->iteration_start_time + BUFFER_DELAY - now;
3230 trace_migration_rate_limit_pre(ms);
3231 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3232 /*
3233 * We were woken by one or more urgent things but
3234 * the timedwait will have consumed one of them.
3235 * The service routine for the urgent wake will dec
3236 * the semaphore itself for each item it consumes,
3237 * so add this one we just eat back.
3238 */
3239 qemu_sem_post(&s->rate_limit_sem);
3240 urgent = true;
3241 }
3242 trace_migration_rate_limit_post(urgent);
3243 }
3244 return urgent;
3245 }
3246
3247 /*
3248 * Master migration thread on the source VM.
3249 * It drives the migration and pumps the data down the outgoing channel.
3250 */
3251 static void *migration_thread(void *opaque)
3252 {
3253 MigrationState *s = opaque;
3254 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3255 MigThrError thr_error;
3256 bool urgent = false;
3257
3258 rcu_register_thread();
3259
3260 object_ref(OBJECT(s));
3261 update_iteration_initial_status(s);
3262
3263 qemu_savevm_state_header(s->to_dst_file);
3264
3265 /*
3266 * If we opened the return path, we need to make sure dst has it
3267 * opened as well.
3268 */
3269 if (s->rp_state.from_dst_file) {
3270 /* Now tell the dest that it should open its end so it can reply */
3271 qemu_savevm_send_open_return_path(s->to_dst_file);
3272
3273 /* And do a ping that will make stuff easier to debug */
3274 qemu_savevm_send_ping(s->to_dst_file, 1);
3275 }
3276
3277 if (migrate_postcopy()) {
3278 /*
3279 * Tell the destination that we *might* want to do postcopy later;
3280 * if the other end can't do postcopy it should fail now, nice and
3281 * early.
3282 */
3283 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3284 }
3285
3286 if (migrate_colo_enabled()) {
3287 /* Notify migration destination that we enable COLO */
3288 qemu_savevm_send_colo_enable(s->to_dst_file);
3289 }
3290
3291 qemu_savevm_state_setup(s->to_dst_file);
3292
3293 if (qemu_savevm_nr_failover_devices()) {
3294 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3295 MIGRATION_STATUS_WAIT_UNPLUG);
3296
3297 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3298 qemu_savevm_state_guest_unplug_pending()) {
3299 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3300 }
3301
3302 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3303 MIGRATION_STATUS_ACTIVE);
3304 }
3305
3306 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3307 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3308 MIGRATION_STATUS_ACTIVE);
3309
3310 trace_migration_thread_setup_complete();
3311
3312 while (migration_is_active(s)) {
3313 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3314 MigIterateState iter_state = migration_iteration_run(s);
3315 if (iter_state == MIG_ITERATE_SKIP) {
3316 continue;
3317 } else if (iter_state == MIG_ITERATE_BREAK) {
3318 break;
3319 }
3320 }
3321
3322 /*
3323 * Try to detect any kind of failures, and see whether we
3324 * should stop the migration now.
3325 */
3326 thr_error = migration_detect_error(s);
3327 if (thr_error == MIG_THR_ERR_FATAL) {
3328 /* Stop migration */
3329 break;
3330 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3331 /*
3332 * Just recovered from a e.g. network failure, reset all
3333 * the local variables. This is important to avoid
3334 * breaking transferred_bytes and bandwidth calculation
3335 */
3336 update_iteration_initial_status(s);
3337 }
3338
3339 urgent = migration_rate_limit();
3340 }
3341
3342 trace_migration_thread_after_loop();
3343 migration_iteration_finish(s);
3344 object_unref(OBJECT(s));
3345 rcu_unregister_thread();
3346 return NULL;
3347 }
3348
3349 void migrate_fd_connect(MigrationState *s, Error *error_in)
3350 {
3351 int64_t rate_limit;
3352 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3353
3354 s->expected_downtime = s->parameters.downtime_limit;
3355 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3356 if (error_in) {
3357 migrate_fd_error(s, error_in);
3358 migrate_fd_cleanup(s);
3359 return;
3360 }
3361
3362 if (resume) {
3363 /* This is a resumed migration */
3364 rate_limit = s->parameters.max_postcopy_bandwidth /
3365 XFER_LIMIT_RATIO;
3366 } else {
3367 /* This is a fresh new migration */
3368 rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3369
3370 /* Notify before starting migration thread */
3371 notifier_list_notify(&migration_state_notifiers, s);
3372 }
3373
3374 qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3375 qemu_file_set_blocking(s->to_dst_file, true);
3376
3377 /*
3378 * Open the return path. For postcopy, it is used exclusively. For
3379 * precopy, only if user specified "return-path" capability would
3380 * QEMU uses the return path.
3381 */
3382 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3383 if (open_return_path_on_source(s, !resume)) {
3384 error_report("Unable to open return-path for postcopy");
3385 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3386 migrate_fd_cleanup(s);
3387 return;
3388 }
3389 }
3390
3391 if (resume) {
3392 /* Wakeup the main migration thread to do the recovery */
3393 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3394 MIGRATION_STATUS_POSTCOPY_RECOVER);
3395 qemu_sem_post(&s->postcopy_pause_sem);
3396 return;
3397 }
3398
3399 if (multifd_save_setup() != 0) {
3400 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3401 MIGRATION_STATUS_FAILED);
3402 migrate_fd_cleanup(s);
3403 return;
3404 }
3405 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3406 QEMU_THREAD_JOINABLE);
3407 s->migration_thread_running = true;
3408 }
3409
3410 void migration_global_dump(Monitor *mon)
3411 {
3412 MigrationState *ms = migrate_get_current();
3413
3414 monitor_printf(mon, "globals:\n");
3415 monitor_printf(mon, "store-global-state: %s\n",
3416 ms->store_global_state ? "on" : "off");
3417 monitor_printf(mon, "only-migratable: %s\n",
3418 only_migratable ? "on" : "off");
3419 monitor_printf(mon, "send-configuration: %s\n",
3420 ms->send_configuration ? "on" : "off");
3421 monitor_printf(mon, "send-section-footer: %s\n",
3422 ms->send_section_footer ? "on" : "off");
3423 monitor_printf(mon, "decompress-error-check: %s\n",
3424 ms->decompress_error_check ? "on" : "off");
3425 monitor_printf(mon, "clear-bitmap-shift: %u\n",
3426 ms->clear_bitmap_shift);
3427 }
3428
3429 #define DEFINE_PROP_MIG_CAP(name, x) \
3430 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3431
3432 static Property migration_properties[] = {
3433 DEFINE_PROP_BOOL("store-global-state", MigrationState,
3434 store_global_state, true),
3435 DEFINE_PROP_BOOL("send-configuration", MigrationState,
3436 send_configuration, true),
3437 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3438 send_section_footer, true),
3439 DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3440 decompress_error_check, true),
3441 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
3442 clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
3443
3444 /* Migration parameters */
3445 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3446 parameters.compress_level,
3447 DEFAULT_MIGRATE_COMPRESS_LEVEL),
3448 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3449 parameters.compress_threads,
3450 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3451 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3452 parameters.compress_wait_thread, true),
3453 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3454 parameters.decompress_threads,
3455 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3456 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3457 parameters.cpu_throttle_initial,
3458 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3459 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3460 parameters.cpu_throttle_increment,
3461 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3462 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3463 parameters.max_bandwidth, MAX_THROTTLE),
3464 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3465 parameters.downtime_limit,
3466 DEFAULT_MIGRATE_SET_DOWNTIME),
3467 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3468 parameters.x_checkpoint_delay,
3469 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3470 DEFINE_PROP_UINT8("multifd-channels", MigrationState,
3471 parameters.multifd_channels,
3472 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3473 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3474 parameters.xbzrle_cache_size,
3475 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3476 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3477 parameters.max_postcopy_bandwidth,
3478 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3479 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3480 parameters.max_cpu_throttle,
3481 DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3482 DEFINE_PROP_SIZE("announce-initial", MigrationState,
3483 parameters.announce_initial,
3484 DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
3485 DEFINE_PROP_SIZE("announce-max", MigrationState,
3486 parameters.announce_max,
3487 DEFAULT_MIGRATE_ANNOUNCE_MAX),
3488 DEFINE_PROP_SIZE("announce-rounds", MigrationState,
3489 parameters.announce_rounds,
3490 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
3491 DEFINE_PROP_SIZE("announce-step", MigrationState,
3492 parameters.announce_step,
3493 DEFAULT_MIGRATE_ANNOUNCE_STEP),
3494
3495 /* Migration capabilities */
3496 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3497 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3498 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3499 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3500 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3501 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3502 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3503 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3504 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3505 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3506 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3507 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD),
3508
3509 DEFINE_PROP_END_OF_LIST(),
3510 };
3511
3512 static void migration_class_init(ObjectClass *klass, void *data)
3513 {
3514 DeviceClass *dc = DEVICE_CLASS(klass);
3515
3516 dc->user_creatable = false;
3517 dc->props = migration_properties;
3518 }
3519
3520 static void migration_instance_finalize(Object *obj)
3521 {
3522 MigrationState *ms = MIGRATION_OBJ(obj);
3523 MigrationParameters *params = &ms->parameters;
3524
3525 qemu_mutex_destroy(&ms->error_mutex);
3526 qemu_mutex_destroy(&ms->qemu_file_lock);
3527 g_free(params->tls_hostname);
3528 g_free(params->tls_creds);
3529 qemu_sem_destroy(&ms->wait_unplug_sem);
3530 qemu_sem_destroy(&ms->rate_limit_sem);
3531 qemu_sem_destroy(&ms->pause_sem);
3532 qemu_sem_destroy(&ms->postcopy_pause_sem);
3533 qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3534 qemu_sem_destroy(&ms->rp_state.rp_sem);
3535 error_free(ms->error);
3536 }
3537
3538 static void migration_instance_init(Object *obj)
3539 {
3540 MigrationState *ms = MIGRATION_OBJ(obj);
3541 MigrationParameters *params = &ms->parameters;
3542
3543 ms->state = MIGRATION_STATUS_NONE;
3544 ms->mbps = -1;
3545 ms->pages_per_second = -1;
3546 qemu_sem_init(&ms->pause_sem, 0);
3547 qemu_mutex_init(&ms->error_mutex);
3548
3549 params->tls_hostname = g_strdup("");
3550 params->tls_creds = g_strdup("");
3551
3552 /* Set has_* up only for parameter checks */
3553 params->has_compress_level = true;
3554 params->has_compress_threads = true;
3555 params->has_decompress_threads = true;
3556 params->has_cpu_throttle_initial = true;
3557 params->has_cpu_throttle_increment = true;
3558 params->has_max_bandwidth = true;
3559 params->has_downtime_limit = true;
3560 params->has_x_checkpoint_delay = true;
3561 params->has_block_incremental = true;
3562 params->has_multifd_channels = true;
3563 params->has_xbzrle_cache_size = true;
3564 params->has_max_postcopy_bandwidth = true;
3565 params->has_max_cpu_throttle = true;
3566 params->has_announce_initial = true;
3567 params->has_announce_max = true;
3568 params->has_announce_rounds = true;
3569 params->has_announce_step = true;
3570
3571 qemu_sem_init(&ms->postcopy_pause_sem, 0);
3572 qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3573 qemu_sem_init(&ms->rp_state.rp_sem, 0);
3574 qemu_sem_init(&ms->rate_limit_sem, 0);
3575 qemu_sem_init(&ms->wait_unplug_sem, 0);
3576 qemu_mutex_init(&ms->qemu_file_lock);
3577 }
3578
3579 /*
3580 * Return true if check pass, false otherwise. Error will be put
3581 * inside errp if provided.
3582 */
3583 static bool migration_object_check(MigrationState *ms, Error **errp)
3584 {
3585 MigrationCapabilityStatusList *head = NULL;
3586 /* Assuming all off */
3587 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3588 int i;
3589
3590 if (!migrate_params_check(&ms->parameters, errp)) {
3591 return false;
3592 }
3593
3594 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3595 if (ms->enabled_capabilities[i]) {
3596 head = migrate_cap_add(head, i, true);
3597 }
3598 }
3599
3600 ret = migrate_caps_check(cap_list, head, errp);
3601
3602 /* It works with head == NULL */
3603 qapi_free_MigrationCapabilityStatusList(head);
3604
3605 return ret;
3606 }
3607
3608 static const TypeInfo migration_type = {
3609 .name = TYPE_MIGRATION,
3610 /*
3611 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3612 * not created using qdev_create(), it is not attached to the qdev
3613 * device tree, and it is never realized.
3614 *
3615 * TODO: Make this TYPE_OBJECT once QOM provides something like
3616 * TYPE_DEVICE's "-global" properties.
3617 */
3618 .parent = TYPE_DEVICE,
3619 .class_init = migration_class_init,
3620 .class_size = sizeof(MigrationClass),
3621 .instance_size = sizeof(MigrationState),
3622 .instance_init = migration_instance_init,
3623 .instance_finalize = migration_instance_finalize,
3624 };
3625
3626 static void register_migration_types(void)
3627 {
3628 type_register_static(&migration_type);
3629 }
3630
3631 type_init(register_migration_types);
AR7 emulation for QEMU