xref: /openbmc/qemu/migration/migration.c (revision 31cf4b97)
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 "migration/blocker.h"
20 #include "exec.h"
21 #include "fd.h"
22 #include "socket.h"
23 #include "rdma.h"
24 #include "ram.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
28 #include "savevm.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-commands-migration.h"
35 #include "qapi/qapi-events-migration.h"
36 #include "qapi/qmp/qerror.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qemu/rcu.h"
39 #include "block.h"
40 #include "postcopy-ram.h"
41 #include "qemu/thread.h"
42 #include "trace.h"
43 #include "exec/target_page.h"
44 #include "io/channel-buffer.h"
45 #include "migration/colo.h"
46 #include "hw/boards.h"
47 #include "monitor/monitor.h"
48 
49 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
50 
51 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
52  * data. */
53 #define BUFFER_DELAY     100
54 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
55 
56 /* Time in milliseconds we are allowed to stop the source,
57  * for sending the last part */
58 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
59 
60 /* Maximum migrate downtime set to 2000 seconds */
61 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
62 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
63 
64 /* Default compression thread count */
65 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
66 /* Default decompression thread count, usually decompression is at
67  * least 4 times as fast as compression.*/
68 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
69 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
70 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
71 /* Define default autoconverge cpu throttle migration parameters */
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
73 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
74 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
75 
76 /* Migration XBZRLE default cache size */
77 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
78 
79 /* The delay time (in ms) between two COLO checkpoints */
80 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
81 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
82 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
83 
84 /* Background transfer rate for postcopy, 0 means unlimited, note
85  * that page requests can still exceed this limit.
86  */
87 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
88 
89 static NotifierList migration_state_notifiers =
90     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
91 
92 static bool deferred_incoming;
93 
94 /* Messages sent on the return path from destination to source */
95 enum mig_rp_message_type {
96     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
97     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
98     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
99 
100     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
101     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
102     MIG_RP_MSG_RECV_BITMAP,  /* send recved_bitmap back to source */
103     MIG_RP_MSG_RESUME_ACK,   /* tell source that we are ready to resume */
104 
105     MIG_RP_MSG_MAX
106 };
107 
108 /* When we add fault tolerance, we could have several
109    migrations at once.  For now we don't need to add
110    dynamic creation of migration */
111 
112 static MigrationState *current_migration;
113 static MigrationIncomingState *current_incoming;
114 
115 static bool migration_object_check(MigrationState *ms, Error **errp);
116 static int migration_maybe_pause(MigrationState *s,
117                                  int *current_active_state,
118                                  int new_state);
119 
120 void migration_object_init(void)
121 {
122     MachineState *ms = MACHINE(qdev_get_machine());
123     Error *err = NULL;
124 
125     /* This can only be called once. */
126     assert(!current_migration);
127     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
128 
129     /*
130      * Init the migrate incoming object as well no matter whether
131      * we'll use it or not.
132      */
133     assert(!current_incoming);
134     current_incoming = g_new0(MigrationIncomingState, 1);
135     current_incoming->state = MIGRATION_STATUS_NONE;
136     current_incoming->postcopy_remote_fds =
137         g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
138     qemu_mutex_init(&current_incoming->rp_mutex);
139     qemu_event_init(&current_incoming->main_thread_load_event, false);
140     qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
141     qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
142 
143     init_dirty_bitmap_incoming_migration();
144 
145     if (!migration_object_check(current_migration, &err)) {
146         error_report_err(err);
147         exit(1);
148     }
149 
150     /*
151      * We cannot really do this in migration_instance_init() since at
152      * that time global properties are not yet applied, then this
153      * value will be definitely replaced by something else.
154      */
155     if (ms->enforce_config_section) {
156         current_migration->send_configuration = true;
157     }
158 }
159 
160 void migration_object_finalize(void)
161 {
162     object_unref(OBJECT(current_migration));
163 }
164 
165 /* For outgoing */
166 MigrationState *migrate_get_current(void)
167 {
168     /* This can only be called after the object created. */
169     assert(current_migration);
170     return current_migration;
171 }
172 
173 MigrationIncomingState *migration_incoming_get_current(void)
174 {
175     assert(current_incoming);
176     return current_incoming;
177 }
178 
179 void migration_incoming_state_destroy(void)
180 {
181     struct MigrationIncomingState *mis = migration_incoming_get_current();
182 
183     if (mis->to_src_file) {
184         /* Tell source that we are done */
185         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
186         qemu_fclose(mis->to_src_file);
187         mis->to_src_file = NULL;
188     }
189 
190     if (mis->from_src_file) {
191         qemu_fclose(mis->from_src_file);
192         mis->from_src_file = NULL;
193     }
194     if (mis->postcopy_remote_fds) {
195         g_array_free(mis->postcopy_remote_fds, TRUE);
196         mis->postcopy_remote_fds = NULL;
197     }
198 
199     qemu_event_reset(&mis->main_thread_load_event);
200 }
201 
202 static void migrate_generate_event(int new_state)
203 {
204     if (migrate_use_events()) {
205         qapi_event_send_migration(new_state);
206     }
207 }
208 
209 static bool migrate_late_block_activate(void)
210 {
211     MigrationState *s;
212 
213     s = migrate_get_current();
214 
215     return s->enabled_capabilities[
216         MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
217 }
218 
219 /*
220  * Called on -incoming with a defer: uri.
221  * The migration can be started later after any parameters have been
222  * changed.
223  */
224 static void deferred_incoming_migration(Error **errp)
225 {
226     if (deferred_incoming) {
227         error_setg(errp, "Incoming migration already deferred");
228     }
229     deferred_incoming = true;
230 }
231 
232 /*
233  * Send a message on the return channel back to the source
234  * of the migration.
235  */
236 static int migrate_send_rp_message(MigrationIncomingState *mis,
237                                    enum mig_rp_message_type message_type,
238                                    uint16_t len, void *data)
239 {
240     int ret = 0;
241 
242     trace_migrate_send_rp_message((int)message_type, len);
243     qemu_mutex_lock(&mis->rp_mutex);
244 
245     /*
246      * It's possible that the file handle got lost due to network
247      * failures.
248      */
249     if (!mis->to_src_file) {
250         ret = -EIO;
251         goto error;
252     }
253 
254     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
255     qemu_put_be16(mis->to_src_file, len);
256     qemu_put_buffer(mis->to_src_file, data, len);
257     qemu_fflush(mis->to_src_file);
258 
259     /* It's possible that qemu file got error during sending */
260     ret = qemu_file_get_error(mis->to_src_file);
261 
262 error:
263     qemu_mutex_unlock(&mis->rp_mutex);
264     return ret;
265 }
266 
267 /* Request a range of pages from the source VM at the given
268  * start address.
269  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
270  *           as the last request (a name must have been given previously)
271  *   Start: Address offset within the RB
272  *   Len: Length in bytes required - must be a multiple of pagesize
273  */
274 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
275                               ram_addr_t start, size_t len)
276 {
277     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
278     size_t msglen = 12; /* start + len */
279     enum mig_rp_message_type msg_type;
280 
281     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
282     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
283 
284     if (rbname) {
285         int rbname_len = strlen(rbname);
286         assert(rbname_len < 256);
287 
288         bufc[msglen++] = rbname_len;
289         memcpy(bufc + msglen, rbname, rbname_len);
290         msglen += rbname_len;
291         msg_type = MIG_RP_MSG_REQ_PAGES_ID;
292     } else {
293         msg_type = MIG_RP_MSG_REQ_PAGES;
294     }
295 
296     return migrate_send_rp_message(mis, msg_type, msglen, bufc);
297 }
298 
299 static bool migration_colo_enabled;
300 bool migration_incoming_colo_enabled(void)
301 {
302     return migration_colo_enabled;
303 }
304 
305 void migration_incoming_disable_colo(void)
306 {
307     migration_colo_enabled = false;
308 }
309 
310 void migration_incoming_enable_colo(void)
311 {
312     migration_colo_enabled = true;
313 }
314 
315 void qemu_start_incoming_migration(const char *uri, Error **errp)
316 {
317     const char *p;
318 
319     qapi_event_send_migration(MIGRATION_STATUS_SETUP);
320     if (!strcmp(uri, "defer")) {
321         deferred_incoming_migration(errp);
322     } else if (strstart(uri, "tcp:", &p)) {
323         tcp_start_incoming_migration(p, errp);
324 #ifdef CONFIG_RDMA
325     } else if (strstart(uri, "rdma:", &p)) {
326         rdma_start_incoming_migration(p, errp);
327 #endif
328     } else if (strstart(uri, "exec:", &p)) {
329         exec_start_incoming_migration(p, errp);
330     } else if (strstart(uri, "unix:", &p)) {
331         unix_start_incoming_migration(p, errp);
332     } else if (strstart(uri, "fd:", &p)) {
333         fd_start_incoming_migration(p, errp);
334     } else {
335         error_setg(errp, "unknown migration protocol: %s", uri);
336     }
337 }
338 
339 static void process_incoming_migration_bh(void *opaque)
340 {
341     Error *local_err = NULL;
342     MigrationIncomingState *mis = opaque;
343 
344     /* If capability late_block_activate is set:
345      * Only fire up the block code now if we're going to restart the
346      * VM, else 'cont' will do it.
347      * This causes file locking to happen; so we don't want it to happen
348      * unless we really are starting the VM.
349      */
350     if (!migrate_late_block_activate() ||
351          (autostart && (!global_state_received() ||
352             global_state_get_runstate() == RUN_STATE_RUNNING))) {
353         /* Make sure all file formats flush their mutable metadata.
354          * If we get an error here, just don't restart the VM yet. */
355         bdrv_invalidate_cache_all(&local_err);
356         if (local_err) {
357             error_report_err(local_err);
358             local_err = NULL;
359             autostart = false;
360         }
361     }
362 
363     /*
364      * This must happen after all error conditions are dealt with and
365      * we're sure the VM is going to be running on this host.
366      */
367     qemu_announce_self();
368 
369     if (multifd_load_cleanup(&local_err) != 0) {
370         error_report_err(local_err);
371         autostart = false;
372     }
373     /* If global state section was not received or we are in running
374        state, we need to obey autostart. Any other state is set with
375        runstate_set. */
376 
377     dirty_bitmap_mig_before_vm_start();
378 
379     if (!global_state_received() ||
380         global_state_get_runstate() == RUN_STATE_RUNNING) {
381         if (autostart) {
382             vm_start();
383         } else {
384             runstate_set(RUN_STATE_PAUSED);
385         }
386     } else {
387         runstate_set(global_state_get_runstate());
388     }
389     /*
390      * This must happen after any state changes since as soon as an external
391      * observer sees this event they might start to prod at the VM assuming
392      * it's ready to use.
393      */
394     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
395                       MIGRATION_STATUS_COMPLETED);
396     qemu_bh_delete(mis->bh);
397     migration_incoming_state_destroy();
398 }
399 
400 static void process_incoming_migration_co(void *opaque)
401 {
402     MigrationIncomingState *mis = migration_incoming_get_current();
403     PostcopyState ps;
404     int ret;
405     Error *local_err = NULL;
406 
407     assert(mis->from_src_file);
408     mis->migration_incoming_co = qemu_coroutine_self();
409     mis->largest_page_size = qemu_ram_pagesize_largest();
410     postcopy_state_set(POSTCOPY_INCOMING_NONE);
411     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
412                       MIGRATION_STATUS_ACTIVE);
413     ret = qemu_loadvm_state(mis->from_src_file);
414 
415     ps = postcopy_state_get();
416     trace_process_incoming_migration_co_end(ret, ps);
417     if (ps != POSTCOPY_INCOMING_NONE) {
418         if (ps == POSTCOPY_INCOMING_ADVISE) {
419             /*
420              * Where a migration had postcopy enabled (and thus went to advise)
421              * but managed to complete within the precopy period, we can use
422              * the normal exit.
423              */
424             postcopy_ram_incoming_cleanup(mis);
425         } else if (ret >= 0) {
426             /*
427              * Postcopy was started, cleanup should happen at the end of the
428              * postcopy thread.
429              */
430             trace_process_incoming_migration_co_postcopy_end_main();
431             return;
432         }
433         /* Else if something went wrong then just fall out of the normal exit */
434     }
435 
436     /* we get COLO info, and know if we are in COLO mode */
437     if (!ret && migration_incoming_colo_enabled()) {
438         /* Make sure all file formats flush their mutable metadata */
439         bdrv_invalidate_cache_all(&local_err);
440         if (local_err) {
441             migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
442                     MIGRATION_STATUS_FAILED);
443             error_report_err(local_err);
444             exit(EXIT_FAILURE);
445         }
446 
447         if (colo_init_ram_cache() < 0) {
448             error_report("Init ram cache failed");
449             exit(EXIT_FAILURE);
450         }
451 
452         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
453              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
454         mis->have_colo_incoming_thread = true;
455         qemu_coroutine_yield();
456 
457         /* Wait checkpoint incoming thread exit before free resource */
458         qemu_thread_join(&mis->colo_incoming_thread);
459         /* We hold the global iothread lock, so it is safe here */
460         colo_release_ram_cache();
461     }
462 
463     if (ret < 0) {
464         Error *local_err = NULL;
465 
466         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
467                           MIGRATION_STATUS_FAILED);
468         error_report("load of migration failed: %s", strerror(-ret));
469         qemu_fclose(mis->from_src_file);
470         if (multifd_load_cleanup(&local_err) != 0) {
471             error_report_err(local_err);
472         }
473         exit(EXIT_FAILURE);
474     }
475     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
476     qemu_bh_schedule(mis->bh);
477     mis->migration_incoming_co = NULL;
478 }
479 
480 static void migration_incoming_setup(QEMUFile *f)
481 {
482     MigrationIncomingState *mis = migration_incoming_get_current();
483 
484     if (multifd_load_setup() != 0) {
485         /* We haven't been able to create multifd threads
486            nothing better to do */
487         exit(EXIT_FAILURE);
488     }
489 
490     if (!mis->from_src_file) {
491         mis->from_src_file = f;
492     }
493     qemu_file_set_blocking(f, false);
494 }
495 
496 void migration_incoming_process(void)
497 {
498     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
499     qemu_coroutine_enter(co);
500 }
501 
502 /* Returns true if recovered from a paused migration, otherwise false */
503 static bool postcopy_try_recover(QEMUFile *f)
504 {
505     MigrationIncomingState *mis = migration_incoming_get_current();
506 
507     if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
508         /* Resumed from a paused postcopy migration */
509 
510         mis->from_src_file = f;
511         /* Postcopy has standalone thread to do vm load */
512         qemu_file_set_blocking(f, true);
513 
514         /* Re-configure the return path */
515         mis->to_src_file = qemu_file_get_return_path(f);
516 
517         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
518                           MIGRATION_STATUS_POSTCOPY_RECOVER);
519 
520         /*
521          * Here, we only wake up the main loading thread (while the
522          * fault thread will still be waiting), so that we can receive
523          * commands from source now, and answer it if needed. The
524          * fault thread will be woken up afterwards until we are sure
525          * that source is ready to reply to page requests.
526          */
527         qemu_sem_post(&mis->postcopy_pause_sem_dst);
528         return true;
529     }
530 
531     return false;
532 }
533 
534 void migration_fd_process_incoming(QEMUFile *f)
535 {
536     if (postcopy_try_recover(f)) {
537         return;
538     }
539 
540     migration_incoming_setup(f);
541     migration_incoming_process();
542 }
543 
544 void migration_ioc_process_incoming(QIOChannel *ioc)
545 {
546     MigrationIncomingState *mis = migration_incoming_get_current();
547     bool start_migration;
548 
549     if (!mis->from_src_file) {
550         /* The first connection (multifd may have multiple) */
551         QEMUFile *f = qemu_fopen_channel_input(ioc);
552 
553         /* If it's a recovery, we're done */
554         if (postcopy_try_recover(f)) {
555             return;
556         }
557 
558         migration_incoming_setup(f);
559 
560         /*
561          * Common migration only needs one channel, so we can start
562          * right now.  Multifd needs more than one channel, we wait.
563          */
564         start_migration = !migrate_use_multifd();
565     } else {
566         /* Multiple connections */
567         assert(migrate_use_multifd());
568         start_migration = multifd_recv_new_channel(ioc);
569     }
570 
571     if (start_migration) {
572         migration_incoming_process();
573     }
574 }
575 
576 /**
577  * @migration_has_all_channels: We have received all channels that we need
578  *
579  * Returns true when we have got connections to all the channels that
580  * we need for migration.
581  */
582 bool migration_has_all_channels(void)
583 {
584     MigrationIncomingState *mis = migration_incoming_get_current();
585     bool all_channels;
586 
587     all_channels = multifd_recv_all_channels_created();
588 
589     return all_channels && mis->from_src_file != NULL;
590 }
591 
592 /*
593  * Send a 'SHUT' message on the return channel with the given value
594  * to indicate that we've finished with the RP.  Non-0 value indicates
595  * error.
596  */
597 void migrate_send_rp_shut(MigrationIncomingState *mis,
598                           uint32_t value)
599 {
600     uint32_t buf;
601 
602     buf = cpu_to_be32(value);
603     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
604 }
605 
606 /*
607  * Send a 'PONG' message on the return channel with the given value
608  * (normally in response to a 'PING')
609  */
610 void migrate_send_rp_pong(MigrationIncomingState *mis,
611                           uint32_t value)
612 {
613     uint32_t buf;
614 
615     buf = cpu_to_be32(value);
616     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
617 }
618 
619 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
620                                  char *block_name)
621 {
622     char buf[512];
623     int len;
624     int64_t res;
625 
626     /*
627      * First, we send the header part. It contains only the len of
628      * idstr, and the idstr itself.
629      */
630     len = strlen(block_name);
631     buf[0] = len;
632     memcpy(buf + 1, block_name, len);
633 
634     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
635         error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
636                      __func__);
637         return;
638     }
639 
640     migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
641 
642     /*
643      * Next, we dump the received bitmap to the stream.
644      *
645      * TODO: currently we are safe since we are the only one that is
646      * using the to_src_file handle (fault thread is still paused),
647      * and it's ok even not taking the mutex. However the best way is
648      * to take the lock before sending the message header, and release
649      * the lock after sending the bitmap.
650      */
651     qemu_mutex_lock(&mis->rp_mutex);
652     res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
653     qemu_mutex_unlock(&mis->rp_mutex);
654 
655     trace_migrate_send_rp_recv_bitmap(block_name, res);
656 }
657 
658 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
659 {
660     uint32_t buf;
661 
662     buf = cpu_to_be32(value);
663     migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
664 }
665 
666 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
667 {
668     MigrationCapabilityStatusList *head = NULL;
669     MigrationCapabilityStatusList *caps;
670     MigrationState *s = migrate_get_current();
671     int i;
672 
673     caps = NULL; /* silence compiler warning */
674     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
675 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
676         if (i == MIGRATION_CAPABILITY_BLOCK) {
677             continue;
678         }
679 #endif
680         if (head == NULL) {
681             head = g_malloc0(sizeof(*caps));
682             caps = head;
683         } else {
684             caps->next = g_malloc0(sizeof(*caps));
685             caps = caps->next;
686         }
687         caps->value =
688             g_malloc(sizeof(*caps->value));
689         caps->value->capability = i;
690         caps->value->state = s->enabled_capabilities[i];
691     }
692 
693     return head;
694 }
695 
696 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
697 {
698     MigrationParameters *params;
699     MigrationState *s = migrate_get_current();
700 
701     /* TODO use QAPI_CLONE() instead of duplicating it inline */
702     params = g_malloc0(sizeof(*params));
703     params->has_compress_level = true;
704     params->compress_level = s->parameters.compress_level;
705     params->has_compress_threads = true;
706     params->compress_threads = s->parameters.compress_threads;
707     params->has_compress_wait_thread = true;
708     params->compress_wait_thread = s->parameters.compress_wait_thread;
709     params->has_decompress_threads = true;
710     params->decompress_threads = s->parameters.decompress_threads;
711     params->has_cpu_throttle_initial = true;
712     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
713     params->has_cpu_throttle_increment = true;
714     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
715     params->has_tls_creds = true;
716     params->tls_creds = g_strdup(s->parameters.tls_creds);
717     params->has_tls_hostname = true;
718     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
719     params->has_max_bandwidth = true;
720     params->max_bandwidth = s->parameters.max_bandwidth;
721     params->has_downtime_limit = true;
722     params->downtime_limit = s->parameters.downtime_limit;
723     params->has_x_checkpoint_delay = true;
724     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
725     params->has_block_incremental = true;
726     params->block_incremental = s->parameters.block_incremental;
727     params->has_x_multifd_channels = true;
728     params->x_multifd_channels = s->parameters.x_multifd_channels;
729     params->has_x_multifd_page_count = true;
730     params->x_multifd_page_count = s->parameters.x_multifd_page_count;
731     params->has_xbzrle_cache_size = true;
732     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
733     params->has_max_postcopy_bandwidth = true;
734     params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
735     params->has_max_cpu_throttle = true;
736     params->max_cpu_throttle = s->parameters.max_cpu_throttle;
737 
738     return params;
739 }
740 
741 /*
742  * Return true if we're already in the middle of a migration
743  * (i.e. any of the active or setup states)
744  */
745 bool migration_is_setup_or_active(int state)
746 {
747     switch (state) {
748     case MIGRATION_STATUS_ACTIVE:
749     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
750     case MIGRATION_STATUS_POSTCOPY_PAUSED:
751     case MIGRATION_STATUS_POSTCOPY_RECOVER:
752     case MIGRATION_STATUS_SETUP:
753     case MIGRATION_STATUS_PRE_SWITCHOVER:
754     case MIGRATION_STATUS_DEVICE:
755         return true;
756 
757     default:
758         return false;
759 
760     }
761 }
762 
763 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
764 {
765     info->has_ram = true;
766     info->ram = g_malloc0(sizeof(*info->ram));
767     info->ram->transferred = ram_counters.transferred;
768     info->ram->total = ram_bytes_total();
769     info->ram->duplicate = ram_counters.duplicate;
770     /* legacy value.  It is not used anymore */
771     info->ram->skipped = 0;
772     info->ram->normal = ram_counters.normal;
773     info->ram->normal_bytes = ram_counters.normal *
774         qemu_target_page_size();
775     info->ram->mbps = s->mbps;
776     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
777     info->ram->postcopy_requests = ram_counters.postcopy_requests;
778     info->ram->page_size = qemu_target_page_size();
779     info->ram->multifd_bytes = ram_counters.multifd_bytes;
780 
781     if (migrate_use_xbzrle()) {
782         info->has_xbzrle_cache = true;
783         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
784         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
785         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
786         info->xbzrle_cache->pages = xbzrle_counters.pages;
787         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
788         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
789         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
790     }
791 
792     if (migrate_use_compression()) {
793         info->has_compression = true;
794         info->compression = g_malloc0(sizeof(*info->compression));
795         info->compression->pages = compression_counters.pages;
796         info->compression->busy = compression_counters.busy;
797         info->compression->busy_rate = compression_counters.busy_rate;
798         info->compression->compressed_size =
799                                     compression_counters.compressed_size;
800         info->compression->compression_rate =
801                                     compression_counters.compression_rate;
802     }
803 
804     if (cpu_throttle_active()) {
805         info->has_cpu_throttle_percentage = true;
806         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
807     }
808 
809     if (s->state != MIGRATION_STATUS_COMPLETED) {
810         info->ram->remaining = ram_bytes_remaining();
811         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
812     }
813 }
814 
815 static void populate_disk_info(MigrationInfo *info)
816 {
817     if (blk_mig_active()) {
818         info->has_disk = true;
819         info->disk = g_malloc0(sizeof(*info->disk));
820         info->disk->transferred = blk_mig_bytes_transferred();
821         info->disk->remaining = blk_mig_bytes_remaining();
822         info->disk->total = blk_mig_bytes_total();
823     }
824 }
825 
826 static void fill_source_migration_info(MigrationInfo *info)
827 {
828     MigrationState *s = migrate_get_current();
829 
830     switch (s->state) {
831     case MIGRATION_STATUS_NONE:
832         /* no migration has happened ever */
833         /* do not overwrite destination migration status */
834         return;
835         break;
836     case MIGRATION_STATUS_SETUP:
837         info->has_status = true;
838         info->has_total_time = false;
839         break;
840     case MIGRATION_STATUS_ACTIVE:
841     case MIGRATION_STATUS_CANCELLING:
842     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
843     case MIGRATION_STATUS_PRE_SWITCHOVER:
844     case MIGRATION_STATUS_DEVICE:
845     case MIGRATION_STATUS_POSTCOPY_PAUSED:
846     case MIGRATION_STATUS_POSTCOPY_RECOVER:
847          /* TODO add some postcopy stats */
848         info->has_status = true;
849         info->has_total_time = true;
850         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
851             - s->start_time;
852         info->has_expected_downtime = true;
853         info->expected_downtime = s->expected_downtime;
854         info->has_setup_time = true;
855         info->setup_time = s->setup_time;
856 
857         populate_ram_info(info, s);
858         populate_disk_info(info);
859         break;
860     case MIGRATION_STATUS_COLO:
861         info->has_status = true;
862         /* TODO: display COLO specific information (checkpoint info etc.) */
863         break;
864     case MIGRATION_STATUS_COMPLETED:
865         info->has_status = true;
866         info->has_total_time = true;
867         info->total_time = s->total_time;
868         info->has_downtime = true;
869         info->downtime = s->downtime;
870         info->has_setup_time = true;
871         info->setup_time = s->setup_time;
872 
873         populate_ram_info(info, s);
874         break;
875     case MIGRATION_STATUS_FAILED:
876         info->has_status = true;
877         if (s->error) {
878             info->has_error_desc = true;
879             info->error_desc = g_strdup(error_get_pretty(s->error));
880         }
881         break;
882     case MIGRATION_STATUS_CANCELLED:
883         info->has_status = true;
884         break;
885     }
886     info->status = s->state;
887 }
888 
889 /**
890  * @migration_caps_check - check capability validity
891  *
892  * @cap_list: old capability list, array of bool
893  * @params: new capabilities to be applied soon
894  * @errp: set *errp if the check failed, with reason
895  *
896  * Returns true if check passed, otherwise false.
897  */
898 static bool migrate_caps_check(bool *cap_list,
899                                MigrationCapabilityStatusList *params,
900                                Error **errp)
901 {
902     MigrationCapabilityStatusList *cap;
903     bool old_postcopy_cap;
904     MigrationIncomingState *mis = migration_incoming_get_current();
905 
906     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
907 
908     for (cap = params; cap; cap = cap->next) {
909         cap_list[cap->value->capability] = cap->value->state;
910     }
911 
912 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
913     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
914         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
915                    "block migration");
916         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
917         return false;
918     }
919 #endif
920 
921 #ifndef CONFIG_REPLICATION
922     if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
923         error_setg(errp, "QEMU compiled without replication module"
924                    " can't enable COLO");
925         error_append_hint(errp, "Please enable replication before COLO.\n");
926         return false;
927     }
928 #endif
929 
930     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
931         if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
932             /* The decompression threads asynchronously write into RAM
933              * rather than use the atomic copies needed to avoid
934              * userfaulting.  It should be possible to fix the decompression
935              * threads for compatibility in future.
936              */
937             error_setg(errp, "Postcopy is not currently compatible "
938                        "with compression");
939             return false;
940         }
941 
942         /* This check is reasonably expensive, so only when it's being
943          * set the first time, also it's only the destination that needs
944          * special support.
945          */
946         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
947             !postcopy_ram_supported_by_host(mis)) {
948             /* postcopy_ram_supported_by_host will have emitted a more
949              * detailed message
950              */
951             error_setg(errp, "Postcopy is not supported");
952             return false;
953         }
954     }
955 
956     return true;
957 }
958 
959 static void fill_destination_migration_info(MigrationInfo *info)
960 {
961     MigrationIncomingState *mis = migration_incoming_get_current();
962 
963     switch (mis->state) {
964     case MIGRATION_STATUS_NONE:
965         return;
966         break;
967     case MIGRATION_STATUS_SETUP:
968     case MIGRATION_STATUS_CANCELLING:
969     case MIGRATION_STATUS_CANCELLED:
970     case MIGRATION_STATUS_ACTIVE:
971     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
972     case MIGRATION_STATUS_POSTCOPY_PAUSED:
973     case MIGRATION_STATUS_POSTCOPY_RECOVER:
974     case MIGRATION_STATUS_FAILED:
975     case MIGRATION_STATUS_COLO:
976         info->has_status = true;
977         break;
978     case MIGRATION_STATUS_COMPLETED:
979         info->has_status = true;
980         fill_destination_postcopy_migration_info(info);
981         break;
982     }
983     info->status = mis->state;
984 }
985 
986 MigrationInfo *qmp_query_migrate(Error **errp)
987 {
988     MigrationInfo *info = g_malloc0(sizeof(*info));
989 
990     fill_destination_migration_info(info);
991     fill_source_migration_info(info);
992 
993     return info;
994 }
995 
996 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
997                                   Error **errp)
998 {
999     MigrationState *s = migrate_get_current();
1000     MigrationCapabilityStatusList *cap;
1001     bool cap_list[MIGRATION_CAPABILITY__MAX];
1002 
1003     if (migration_is_setup_or_active(s->state)) {
1004         error_setg(errp, QERR_MIGRATION_ACTIVE);
1005         return;
1006     }
1007 
1008     memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1009     if (!migrate_caps_check(cap_list, params, errp)) {
1010         return;
1011     }
1012 
1013     for (cap = params; cap; cap = cap->next) {
1014         s->enabled_capabilities[cap->value->capability] = cap->value->state;
1015     }
1016 }
1017 
1018 /*
1019  * Check whether the parameters are valid. Error will be put into errp
1020  * (if provided). Return true if valid, otherwise false.
1021  */
1022 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1023 {
1024     if (params->has_compress_level &&
1025         (params->compress_level > 9)) {
1026         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1027                    "is invalid, it should be in the range of 0 to 9");
1028         return false;
1029     }
1030 
1031     if (params->has_compress_threads && (params->compress_threads < 1)) {
1032         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1033                    "compress_threads",
1034                    "is invalid, it should be in the range of 1 to 255");
1035         return false;
1036     }
1037 
1038     if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1039         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1040                    "decompress_threads",
1041                    "is invalid, it should be in the range of 1 to 255");
1042         return false;
1043     }
1044 
1045     if (params->has_cpu_throttle_initial &&
1046         (params->cpu_throttle_initial < 1 ||
1047          params->cpu_throttle_initial > 99)) {
1048         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1049                    "cpu_throttle_initial",
1050                    "an integer in the range of 1 to 99");
1051         return false;
1052     }
1053 
1054     if (params->has_cpu_throttle_increment &&
1055         (params->cpu_throttle_increment < 1 ||
1056          params->cpu_throttle_increment > 99)) {
1057         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1058                    "cpu_throttle_increment",
1059                    "an integer in the range of 1 to 99");
1060         return false;
1061     }
1062 
1063     if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1064         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
1065                          " range of 0 to %zu bytes/second", SIZE_MAX);
1066         return false;
1067     }
1068 
1069     if (params->has_downtime_limit &&
1070         (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1071         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1072                          "the range of 0 to %d milliseconds",
1073                          MAX_MIGRATE_DOWNTIME);
1074         return false;
1075     }
1076 
1077     /* x_checkpoint_delay is now always positive */
1078 
1079     if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
1080         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1081                    "multifd_channels",
1082                    "is invalid, it should be in the range of 1 to 255");
1083         return false;
1084     }
1085     if (params->has_x_multifd_page_count &&
1086         (params->x_multifd_page_count < 1 ||
1087          params->x_multifd_page_count > 10000)) {
1088         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1089                    "multifd_page_count",
1090                    "is invalid, it should be in the range of 1 to 10000");
1091         return false;
1092     }
1093 
1094     if (params->has_xbzrle_cache_size &&
1095         (params->xbzrle_cache_size < qemu_target_page_size() ||
1096          !is_power_of_2(params->xbzrle_cache_size))) {
1097         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1098                    "xbzrle_cache_size",
1099                    "is invalid, it should be bigger than target page size"
1100                    " and a power of two");
1101         return false;
1102     }
1103 
1104     if (params->has_max_cpu_throttle &&
1105         (params->max_cpu_throttle < params->cpu_throttle_initial ||
1106          params->max_cpu_throttle > 99)) {
1107         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1108                    "max_cpu_throttle",
1109                    "an integer in the range of cpu_throttle_initial to 99");
1110         return false;
1111     }
1112 
1113     return true;
1114 }
1115 
1116 static void migrate_params_test_apply(MigrateSetParameters *params,
1117                                       MigrationParameters *dest)
1118 {
1119     *dest = migrate_get_current()->parameters;
1120 
1121     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1122 
1123     if (params->has_compress_level) {
1124         dest->compress_level = params->compress_level;
1125     }
1126 
1127     if (params->has_compress_threads) {
1128         dest->compress_threads = params->compress_threads;
1129     }
1130 
1131     if (params->has_compress_wait_thread) {
1132         dest->compress_wait_thread = params->compress_wait_thread;
1133     }
1134 
1135     if (params->has_decompress_threads) {
1136         dest->decompress_threads = params->decompress_threads;
1137     }
1138 
1139     if (params->has_cpu_throttle_initial) {
1140         dest->cpu_throttle_initial = params->cpu_throttle_initial;
1141     }
1142 
1143     if (params->has_cpu_throttle_increment) {
1144         dest->cpu_throttle_increment = params->cpu_throttle_increment;
1145     }
1146 
1147     if (params->has_tls_creds) {
1148         assert(params->tls_creds->type == QTYPE_QSTRING);
1149         dest->tls_creds = g_strdup(params->tls_creds->u.s);
1150     }
1151 
1152     if (params->has_tls_hostname) {
1153         assert(params->tls_hostname->type == QTYPE_QSTRING);
1154         dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1155     }
1156 
1157     if (params->has_max_bandwidth) {
1158         dest->max_bandwidth = params->max_bandwidth;
1159     }
1160 
1161     if (params->has_downtime_limit) {
1162         dest->downtime_limit = params->downtime_limit;
1163     }
1164 
1165     if (params->has_x_checkpoint_delay) {
1166         dest->x_checkpoint_delay = params->x_checkpoint_delay;
1167     }
1168 
1169     if (params->has_block_incremental) {
1170         dest->block_incremental = params->block_incremental;
1171     }
1172     if (params->has_x_multifd_channels) {
1173         dest->x_multifd_channels = params->x_multifd_channels;
1174     }
1175     if (params->has_x_multifd_page_count) {
1176         dest->x_multifd_page_count = params->x_multifd_page_count;
1177     }
1178     if (params->has_xbzrle_cache_size) {
1179         dest->xbzrle_cache_size = params->xbzrle_cache_size;
1180     }
1181     if (params->has_max_postcopy_bandwidth) {
1182         dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1183     }
1184     if (params->has_max_cpu_throttle) {
1185         dest->max_cpu_throttle = params->max_cpu_throttle;
1186     }
1187 }
1188 
1189 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1190 {
1191     MigrationState *s = migrate_get_current();
1192 
1193     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1194 
1195     if (params->has_compress_level) {
1196         s->parameters.compress_level = params->compress_level;
1197     }
1198 
1199     if (params->has_compress_threads) {
1200         s->parameters.compress_threads = params->compress_threads;
1201     }
1202 
1203     if (params->has_compress_wait_thread) {
1204         s->parameters.compress_wait_thread = params->compress_wait_thread;
1205     }
1206 
1207     if (params->has_decompress_threads) {
1208         s->parameters.decompress_threads = params->decompress_threads;
1209     }
1210 
1211     if (params->has_cpu_throttle_initial) {
1212         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1213     }
1214 
1215     if (params->has_cpu_throttle_increment) {
1216         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1217     }
1218 
1219     if (params->has_tls_creds) {
1220         g_free(s->parameters.tls_creds);
1221         assert(params->tls_creds->type == QTYPE_QSTRING);
1222         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1223     }
1224 
1225     if (params->has_tls_hostname) {
1226         g_free(s->parameters.tls_hostname);
1227         assert(params->tls_hostname->type == QTYPE_QSTRING);
1228         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1229     }
1230 
1231     if (params->has_max_bandwidth) {
1232         s->parameters.max_bandwidth = params->max_bandwidth;
1233         if (s->to_dst_file) {
1234             qemu_file_set_rate_limit(s->to_dst_file,
1235                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1236         }
1237     }
1238 
1239     if (params->has_downtime_limit) {
1240         s->parameters.downtime_limit = params->downtime_limit;
1241     }
1242 
1243     if (params->has_x_checkpoint_delay) {
1244         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1245         if (migration_in_colo_state()) {
1246             colo_checkpoint_notify(s);
1247         }
1248     }
1249 
1250     if (params->has_block_incremental) {
1251         s->parameters.block_incremental = params->block_incremental;
1252     }
1253     if (params->has_x_multifd_channels) {
1254         s->parameters.x_multifd_channels = params->x_multifd_channels;
1255     }
1256     if (params->has_x_multifd_page_count) {
1257         s->parameters.x_multifd_page_count = params->x_multifd_page_count;
1258     }
1259     if (params->has_xbzrle_cache_size) {
1260         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1261         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1262     }
1263     if (params->has_max_postcopy_bandwidth) {
1264         s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1265     }
1266     if (params->has_max_cpu_throttle) {
1267         s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1268     }
1269 }
1270 
1271 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1272 {
1273     MigrationParameters tmp;
1274 
1275     /* TODO Rewrite "" to null instead */
1276     if (params->has_tls_creds
1277         && params->tls_creds->type == QTYPE_QNULL) {
1278         qobject_unref(params->tls_creds->u.n);
1279         params->tls_creds->type = QTYPE_QSTRING;
1280         params->tls_creds->u.s = strdup("");
1281     }
1282     /* TODO Rewrite "" to null instead */
1283     if (params->has_tls_hostname
1284         && params->tls_hostname->type == QTYPE_QNULL) {
1285         qobject_unref(params->tls_hostname->u.n);
1286         params->tls_hostname->type = QTYPE_QSTRING;
1287         params->tls_hostname->u.s = strdup("");
1288     }
1289 
1290     migrate_params_test_apply(params, &tmp);
1291 
1292     if (!migrate_params_check(&tmp, errp)) {
1293         /* Invalid parameter */
1294         return;
1295     }
1296 
1297     migrate_params_apply(params, errp);
1298 }
1299 
1300 
1301 void qmp_migrate_start_postcopy(Error **errp)
1302 {
1303     MigrationState *s = migrate_get_current();
1304 
1305     if (!migrate_postcopy()) {
1306         error_setg(errp, "Enable postcopy with migrate_set_capability before"
1307                          " the start of migration");
1308         return;
1309     }
1310 
1311     if (s->state == MIGRATION_STATUS_NONE) {
1312         error_setg(errp, "Postcopy must be started after migration has been"
1313                          " started");
1314         return;
1315     }
1316     /*
1317      * we don't error if migration has finished since that would be racy
1318      * with issuing this command.
1319      */
1320     atomic_set(&s->start_postcopy, true);
1321 }
1322 
1323 /* shared migration helpers */
1324 
1325 void migrate_set_state(int *state, int old_state, int new_state)
1326 {
1327     assert(new_state < MIGRATION_STATUS__MAX);
1328     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1329         trace_migrate_set_state(MigrationStatus_str(new_state));
1330         migrate_generate_event(new_state);
1331     }
1332 }
1333 
1334 static MigrationCapabilityStatusList *migrate_cap_add(
1335     MigrationCapabilityStatusList *list,
1336     MigrationCapability index,
1337     bool state)
1338 {
1339     MigrationCapabilityStatusList *cap;
1340 
1341     cap = g_new0(MigrationCapabilityStatusList, 1);
1342     cap->value = g_new0(MigrationCapabilityStatus, 1);
1343     cap->value->capability = index;
1344     cap->value->state = state;
1345     cap->next = list;
1346 
1347     return cap;
1348 }
1349 
1350 void migrate_set_block_enabled(bool value, Error **errp)
1351 {
1352     MigrationCapabilityStatusList *cap;
1353 
1354     cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1355     qmp_migrate_set_capabilities(cap, errp);
1356     qapi_free_MigrationCapabilityStatusList(cap);
1357 }
1358 
1359 static void migrate_set_block_incremental(MigrationState *s, bool value)
1360 {
1361     s->parameters.block_incremental = value;
1362 }
1363 
1364 static void block_cleanup_parameters(MigrationState *s)
1365 {
1366     if (s->must_remove_block_options) {
1367         /* setting to false can never fail */
1368         migrate_set_block_enabled(false, &error_abort);
1369         migrate_set_block_incremental(s, false);
1370         s->must_remove_block_options = false;
1371     }
1372 }
1373 
1374 static void migrate_fd_cleanup(void *opaque)
1375 {
1376     MigrationState *s = opaque;
1377 
1378     qemu_bh_delete(s->cleanup_bh);
1379     s->cleanup_bh = NULL;
1380 
1381     qemu_savevm_state_cleanup();
1382 
1383     if (s->to_dst_file) {
1384         Error *local_err = NULL;
1385         QEMUFile *tmp;
1386 
1387         trace_migrate_fd_cleanup();
1388         qemu_mutex_unlock_iothread();
1389         if (s->migration_thread_running) {
1390             qemu_thread_join(&s->thread);
1391             s->migration_thread_running = false;
1392         }
1393         qemu_mutex_lock_iothread();
1394 
1395         if (multifd_save_cleanup(&local_err) != 0) {
1396             error_report_err(local_err);
1397         }
1398         qemu_mutex_lock(&s->qemu_file_lock);
1399         tmp = s->to_dst_file;
1400         s->to_dst_file = NULL;
1401         qemu_mutex_unlock(&s->qemu_file_lock);
1402         /*
1403          * Close the file handle without the lock to make sure the
1404          * critical section won't block for long.
1405          */
1406         qemu_fclose(tmp);
1407     }
1408 
1409     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1410            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1411 
1412     if (s->state == MIGRATION_STATUS_CANCELLING) {
1413         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1414                           MIGRATION_STATUS_CANCELLED);
1415     }
1416 
1417     if (s->error) {
1418         /* It is used on info migrate.  We can't free it */
1419         error_report_err(error_copy(s->error));
1420     }
1421     notifier_list_notify(&migration_state_notifiers, s);
1422     block_cleanup_parameters(s);
1423 }
1424 
1425 void migrate_set_error(MigrationState *s, const Error *error)
1426 {
1427     qemu_mutex_lock(&s->error_mutex);
1428     if (!s->error) {
1429         s->error = error_copy(error);
1430     }
1431     qemu_mutex_unlock(&s->error_mutex);
1432 }
1433 
1434 void migrate_fd_error(MigrationState *s, const Error *error)
1435 {
1436     trace_migrate_fd_error(error_get_pretty(error));
1437     assert(s->to_dst_file == NULL);
1438     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1439                       MIGRATION_STATUS_FAILED);
1440     migrate_set_error(s, error);
1441 }
1442 
1443 static void migrate_fd_cancel(MigrationState *s)
1444 {
1445     int old_state ;
1446     QEMUFile *f = migrate_get_current()->to_dst_file;
1447     trace_migrate_fd_cancel();
1448 
1449     if (s->rp_state.from_dst_file) {
1450         /* shutdown the rp socket, so causing the rp thread to shutdown */
1451         qemu_file_shutdown(s->rp_state.from_dst_file);
1452     }
1453 
1454     do {
1455         old_state = s->state;
1456         if (!migration_is_setup_or_active(old_state)) {
1457             break;
1458         }
1459         /* If the migration is paused, kick it out of the pause */
1460         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1461             qemu_sem_post(&s->pause_sem);
1462         }
1463         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1464     } while (s->state != MIGRATION_STATUS_CANCELLING);
1465 
1466     /*
1467      * If we're unlucky the migration code might be stuck somewhere in a
1468      * send/write while the network has failed and is waiting to timeout;
1469      * if we've got shutdown(2) available then we can force it to quit.
1470      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1471      * called in a bh, so there is no race against this cancel.
1472      */
1473     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1474         qemu_file_shutdown(f);
1475     }
1476     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1477         Error *local_err = NULL;
1478 
1479         bdrv_invalidate_cache_all(&local_err);
1480         if (local_err) {
1481             error_report_err(local_err);
1482         } else {
1483             s->block_inactive = false;
1484         }
1485     }
1486 }
1487 
1488 void add_migration_state_change_notifier(Notifier *notify)
1489 {
1490     notifier_list_add(&migration_state_notifiers, notify);
1491 }
1492 
1493 void remove_migration_state_change_notifier(Notifier *notify)
1494 {
1495     notifier_remove(notify);
1496 }
1497 
1498 bool migration_in_setup(MigrationState *s)
1499 {
1500     return s->state == MIGRATION_STATUS_SETUP;
1501 }
1502 
1503 bool migration_has_finished(MigrationState *s)
1504 {
1505     return s->state == MIGRATION_STATUS_COMPLETED;
1506 }
1507 
1508 bool migration_has_failed(MigrationState *s)
1509 {
1510     return (s->state == MIGRATION_STATUS_CANCELLED ||
1511             s->state == MIGRATION_STATUS_FAILED);
1512 }
1513 
1514 bool migration_in_postcopy(void)
1515 {
1516     MigrationState *s = migrate_get_current();
1517 
1518     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1519 }
1520 
1521 bool migration_in_postcopy_after_devices(MigrationState *s)
1522 {
1523     return migration_in_postcopy() && s->postcopy_after_devices;
1524 }
1525 
1526 bool migration_is_idle(void)
1527 {
1528     MigrationState *s = migrate_get_current();
1529 
1530     switch (s->state) {
1531     case MIGRATION_STATUS_NONE:
1532     case MIGRATION_STATUS_CANCELLED:
1533     case MIGRATION_STATUS_COMPLETED:
1534     case MIGRATION_STATUS_FAILED:
1535         return true;
1536     case MIGRATION_STATUS_SETUP:
1537     case MIGRATION_STATUS_CANCELLING:
1538     case MIGRATION_STATUS_ACTIVE:
1539     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1540     case MIGRATION_STATUS_COLO:
1541     case MIGRATION_STATUS_PRE_SWITCHOVER:
1542     case MIGRATION_STATUS_DEVICE:
1543         return false;
1544     case MIGRATION_STATUS__MAX:
1545         g_assert_not_reached();
1546     }
1547 
1548     return false;
1549 }
1550 
1551 void migrate_init(MigrationState *s)
1552 {
1553     /*
1554      * Reinitialise all migration state, except
1555      * parameters/capabilities that the user set, and
1556      * locks.
1557      */
1558     s->bytes_xfer = 0;
1559     s->xfer_limit = 0;
1560     s->cleanup_bh = 0;
1561     s->to_dst_file = NULL;
1562     s->state = MIGRATION_STATUS_NONE;
1563     s->rp_state.from_dst_file = NULL;
1564     s->rp_state.error = false;
1565     s->mbps = 0.0;
1566     s->downtime = 0;
1567     s->expected_downtime = 0;
1568     s->setup_time = 0;
1569     s->start_postcopy = false;
1570     s->postcopy_after_devices = false;
1571     s->migration_thread_running = false;
1572     error_free(s->error);
1573     s->error = NULL;
1574 
1575     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1576 
1577     s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1578     s->total_time = 0;
1579     s->vm_was_running = false;
1580     s->iteration_initial_bytes = 0;
1581     s->threshold_size = 0;
1582 }
1583 
1584 static GSList *migration_blockers;
1585 
1586 int migrate_add_blocker(Error *reason, Error **errp)
1587 {
1588     if (migrate_get_current()->only_migratable) {
1589         error_propagate_prepend(errp, error_copy(reason),
1590                                 "disallowing migration blocker "
1591                                 "(--only_migratable) for: ");
1592         return -EACCES;
1593     }
1594 
1595     if (migration_is_idle()) {
1596         migration_blockers = g_slist_prepend(migration_blockers, reason);
1597         return 0;
1598     }
1599 
1600     error_propagate_prepend(errp, error_copy(reason),
1601                             "disallowing migration blocker "
1602                             "(migration in progress) for: ");
1603     return -EBUSY;
1604 }
1605 
1606 void migrate_del_blocker(Error *reason)
1607 {
1608     migration_blockers = g_slist_remove(migration_blockers, reason);
1609 }
1610 
1611 void qmp_migrate_incoming(const char *uri, Error **errp)
1612 {
1613     Error *local_err = NULL;
1614     static bool once = true;
1615 
1616     if (!deferred_incoming) {
1617         error_setg(errp, "For use with '-incoming defer'");
1618         return;
1619     }
1620     if (!once) {
1621         error_setg(errp, "The incoming migration has already been started");
1622     }
1623 
1624     qemu_start_incoming_migration(uri, &local_err);
1625 
1626     if (local_err) {
1627         error_propagate(errp, local_err);
1628         return;
1629     }
1630 
1631     once = false;
1632 }
1633 
1634 void qmp_migrate_recover(const char *uri, Error **errp)
1635 {
1636     MigrationIncomingState *mis = migration_incoming_get_current();
1637 
1638     if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1639         error_setg(errp, "Migrate recover can only be run "
1640                    "when postcopy is paused.");
1641         return;
1642     }
1643 
1644     if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1645                        false, true) == true) {
1646         error_setg(errp, "Migrate recovery is triggered already");
1647         return;
1648     }
1649 
1650     /*
1651      * Note that this call will never start a real migration; it will
1652      * only re-setup the migration stream and poke existing migration
1653      * to continue using that newly established channel.
1654      */
1655     qemu_start_incoming_migration(uri, errp);
1656 }
1657 
1658 void qmp_migrate_pause(Error **errp)
1659 {
1660     MigrationState *ms = migrate_get_current();
1661     MigrationIncomingState *mis = migration_incoming_get_current();
1662     int ret;
1663 
1664     if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1665         /* Source side, during postcopy */
1666         qemu_mutex_lock(&ms->qemu_file_lock);
1667         ret = qemu_file_shutdown(ms->to_dst_file);
1668         qemu_mutex_unlock(&ms->qemu_file_lock);
1669         if (ret) {
1670             error_setg(errp, "Failed to pause source migration");
1671         }
1672         return;
1673     }
1674 
1675     if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1676         ret = qemu_file_shutdown(mis->from_src_file);
1677         if (ret) {
1678             error_setg(errp, "Failed to pause destination migration");
1679         }
1680         return;
1681     }
1682 
1683     error_setg(errp, "migrate-pause is currently only supported "
1684                "during postcopy-active state");
1685 }
1686 
1687 bool migration_is_blocked(Error **errp)
1688 {
1689     if (qemu_savevm_state_blocked(errp)) {
1690         return true;
1691     }
1692 
1693     if (migration_blockers) {
1694         error_propagate(errp, error_copy(migration_blockers->data));
1695         return true;
1696     }
1697 
1698     return false;
1699 }
1700 
1701 /* Returns true if continue to migrate, or false if error detected */
1702 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1703                             bool resume, Error **errp)
1704 {
1705     Error *local_err = NULL;
1706 
1707     if (resume) {
1708         if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1709             error_setg(errp, "Cannot resume if there is no "
1710                        "paused migration");
1711             return false;
1712         }
1713 
1714         /*
1715          * Postcopy recovery won't work well with release-ram
1716          * capability since release-ram will drop the page buffer as
1717          * long as the page is put into the send buffer.  So if there
1718          * is a network failure happened, any page buffers that have
1719          * not yet reached the destination VM but have already been
1720          * sent from the source VM will be lost forever.  Let's refuse
1721          * the client from resuming such a postcopy migration.
1722          * Luckily release-ram was designed to only be used when src
1723          * and destination VMs are on the same host, so it should be
1724          * fine.
1725          */
1726         if (migrate_release_ram()) {
1727             error_setg(errp, "Postcopy recovery cannot work "
1728                        "when release-ram capability is set");
1729             return false;
1730         }
1731 
1732         /* This is a resume, skip init status */
1733         return true;
1734     }
1735 
1736     if (migration_is_setup_or_active(s->state) ||
1737         s->state == MIGRATION_STATUS_CANCELLING ||
1738         s->state == MIGRATION_STATUS_COLO) {
1739         error_setg(errp, QERR_MIGRATION_ACTIVE);
1740         return false;
1741     }
1742 
1743     if (runstate_check(RUN_STATE_INMIGRATE)) {
1744         error_setg(errp, "Guest is waiting for an incoming migration");
1745         return false;
1746     }
1747 
1748     if (migration_is_blocked(errp)) {
1749         return false;
1750     }
1751 
1752     if (blk || blk_inc) {
1753         if (migrate_use_block() || migrate_use_block_incremental()) {
1754             error_setg(errp, "Command options are incompatible with "
1755                        "current migration capabilities");
1756             return false;
1757         }
1758         migrate_set_block_enabled(true, &local_err);
1759         if (local_err) {
1760             error_propagate(errp, local_err);
1761             return false;
1762         }
1763         s->must_remove_block_options = true;
1764     }
1765 
1766     if (blk_inc) {
1767         migrate_set_block_incremental(s, true);
1768     }
1769 
1770     migrate_init(s);
1771 
1772     return true;
1773 }
1774 
1775 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1776                  bool has_inc, bool inc, bool has_detach, bool detach,
1777                  bool has_resume, bool resume, Error **errp)
1778 {
1779     Error *local_err = NULL;
1780     MigrationState *s = migrate_get_current();
1781     const char *p;
1782 
1783     if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1784                          has_resume && resume, errp)) {
1785         /* Error detected, put into errp */
1786         return;
1787     }
1788 
1789     if (strstart(uri, "tcp:", &p)) {
1790         tcp_start_outgoing_migration(s, p, &local_err);
1791 #ifdef CONFIG_RDMA
1792     } else if (strstart(uri, "rdma:", &p)) {
1793         rdma_start_outgoing_migration(s, p, &local_err);
1794 #endif
1795     } else if (strstart(uri, "exec:", &p)) {
1796         exec_start_outgoing_migration(s, p, &local_err);
1797     } else if (strstart(uri, "unix:", &p)) {
1798         unix_start_outgoing_migration(s, p, &local_err);
1799     } else if (strstart(uri, "fd:", &p)) {
1800         fd_start_outgoing_migration(s, p, &local_err);
1801     } else {
1802         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1803                    "a valid migration protocol");
1804         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1805                           MIGRATION_STATUS_FAILED);
1806         block_cleanup_parameters(s);
1807         return;
1808     }
1809 
1810     if (local_err) {
1811         migrate_fd_error(s, local_err);
1812         error_propagate(errp, local_err);
1813         return;
1814     }
1815 }
1816 
1817 void qmp_migrate_cancel(Error **errp)
1818 {
1819     migrate_fd_cancel(migrate_get_current());
1820 }
1821 
1822 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1823 {
1824     MigrationState *s = migrate_get_current();
1825     if (s->state != state) {
1826         error_setg(errp,  "Migration not in expected state: %s",
1827                    MigrationStatus_str(s->state));
1828         return;
1829     }
1830     qemu_sem_post(&s->pause_sem);
1831 }
1832 
1833 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1834 {
1835     MigrateSetParameters p = {
1836         .has_xbzrle_cache_size = true,
1837         .xbzrle_cache_size = value,
1838     };
1839 
1840     qmp_migrate_set_parameters(&p, errp);
1841 }
1842 
1843 int64_t qmp_query_migrate_cache_size(Error **errp)
1844 {
1845     return migrate_xbzrle_cache_size();
1846 }
1847 
1848 void qmp_migrate_set_speed(int64_t value, Error **errp)
1849 {
1850     MigrateSetParameters p = {
1851         .has_max_bandwidth = true,
1852         .max_bandwidth = value,
1853     };
1854 
1855     qmp_migrate_set_parameters(&p, errp);
1856 }
1857 
1858 void qmp_migrate_set_downtime(double value, Error **errp)
1859 {
1860     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1861         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1862                          "the range of 0 to %d seconds",
1863                          MAX_MIGRATE_DOWNTIME_SECONDS);
1864         return;
1865     }
1866 
1867     value *= 1000; /* Convert to milliseconds */
1868     value = MAX(0, MIN(INT64_MAX, value));
1869 
1870     MigrateSetParameters p = {
1871         .has_downtime_limit = true,
1872         .downtime_limit = value,
1873     };
1874 
1875     qmp_migrate_set_parameters(&p, errp);
1876 }
1877 
1878 bool migrate_release_ram(void)
1879 {
1880     MigrationState *s;
1881 
1882     s = migrate_get_current();
1883 
1884     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1885 }
1886 
1887 bool migrate_postcopy_ram(void)
1888 {
1889     MigrationState *s;
1890 
1891     s = migrate_get_current();
1892 
1893     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1894 }
1895 
1896 bool migrate_postcopy(void)
1897 {
1898     return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1899 }
1900 
1901 bool migrate_auto_converge(void)
1902 {
1903     MigrationState *s;
1904 
1905     s = migrate_get_current();
1906 
1907     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1908 }
1909 
1910 bool migrate_zero_blocks(void)
1911 {
1912     MigrationState *s;
1913 
1914     s = migrate_get_current();
1915 
1916     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1917 }
1918 
1919 bool migrate_postcopy_blocktime(void)
1920 {
1921     MigrationState *s;
1922 
1923     s = migrate_get_current();
1924 
1925     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
1926 }
1927 
1928 bool migrate_use_compression(void)
1929 {
1930     MigrationState *s;
1931 
1932     s = migrate_get_current();
1933 
1934     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1935 }
1936 
1937 int migrate_compress_level(void)
1938 {
1939     MigrationState *s;
1940 
1941     s = migrate_get_current();
1942 
1943     return s->parameters.compress_level;
1944 }
1945 
1946 int migrate_compress_threads(void)
1947 {
1948     MigrationState *s;
1949 
1950     s = migrate_get_current();
1951 
1952     return s->parameters.compress_threads;
1953 }
1954 
1955 int migrate_compress_wait_thread(void)
1956 {
1957     MigrationState *s;
1958 
1959     s = migrate_get_current();
1960 
1961     return s->parameters.compress_wait_thread;
1962 }
1963 
1964 int migrate_decompress_threads(void)
1965 {
1966     MigrationState *s;
1967 
1968     s = migrate_get_current();
1969 
1970     return s->parameters.decompress_threads;
1971 }
1972 
1973 bool migrate_dirty_bitmaps(void)
1974 {
1975     MigrationState *s;
1976 
1977     s = migrate_get_current();
1978 
1979     return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
1980 }
1981 
1982 bool migrate_use_events(void)
1983 {
1984     MigrationState *s;
1985 
1986     s = migrate_get_current();
1987 
1988     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1989 }
1990 
1991 bool migrate_use_multifd(void)
1992 {
1993     MigrationState *s;
1994 
1995     s = migrate_get_current();
1996 
1997     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1998 }
1999 
2000 bool migrate_pause_before_switchover(void)
2001 {
2002     MigrationState *s;
2003 
2004     s = migrate_get_current();
2005 
2006     return s->enabled_capabilities[
2007         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2008 }
2009 
2010 int migrate_multifd_channels(void)
2011 {
2012     MigrationState *s;
2013 
2014     s = migrate_get_current();
2015 
2016     return s->parameters.x_multifd_channels;
2017 }
2018 
2019 int migrate_multifd_page_count(void)
2020 {
2021     MigrationState *s;
2022 
2023     s = migrate_get_current();
2024 
2025     return s->parameters.x_multifd_page_count;
2026 }
2027 
2028 int migrate_use_xbzrle(void)
2029 {
2030     MigrationState *s;
2031 
2032     s = migrate_get_current();
2033 
2034     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2035 }
2036 
2037 int64_t migrate_xbzrle_cache_size(void)
2038 {
2039     MigrationState *s;
2040 
2041     s = migrate_get_current();
2042 
2043     return s->parameters.xbzrle_cache_size;
2044 }
2045 
2046 static int64_t migrate_max_postcopy_bandwidth(void)
2047 {
2048     MigrationState *s;
2049 
2050     s = migrate_get_current();
2051 
2052     return s->parameters.max_postcopy_bandwidth;
2053 }
2054 
2055 bool migrate_use_block(void)
2056 {
2057     MigrationState *s;
2058 
2059     s = migrate_get_current();
2060 
2061     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2062 }
2063 
2064 bool migrate_use_return_path(void)
2065 {
2066     MigrationState *s;
2067 
2068     s = migrate_get_current();
2069 
2070     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2071 }
2072 
2073 bool migrate_use_block_incremental(void)
2074 {
2075     MigrationState *s;
2076 
2077     s = migrate_get_current();
2078 
2079     return s->parameters.block_incremental;
2080 }
2081 
2082 /* migration thread support */
2083 /*
2084  * Something bad happened to the RP stream, mark an error
2085  * The caller shall print or trace something to indicate why
2086  */
2087 static void mark_source_rp_bad(MigrationState *s)
2088 {
2089     s->rp_state.error = true;
2090 }
2091 
2092 static struct rp_cmd_args {
2093     ssize_t     len; /* -1 = variable */
2094     const char *name;
2095 } rp_cmd_args[] = {
2096     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
2097     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
2098     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
2099     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
2100     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
2101     [MIG_RP_MSG_RECV_BITMAP]    = { .len = -1, .name = "RECV_BITMAP" },
2102     [MIG_RP_MSG_RESUME_ACK]     = { .len =  4, .name = "RESUME_ACK" },
2103     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
2104 };
2105 
2106 /*
2107  * Process a request for pages received on the return path,
2108  * We're allowed to send more than requested (e.g. to round to our page size)
2109  * and we don't need to send pages that have already been sent.
2110  */
2111 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2112                                        ram_addr_t start, size_t len)
2113 {
2114     long our_host_ps = getpagesize();
2115 
2116     trace_migrate_handle_rp_req_pages(rbname, start, len);
2117 
2118     /*
2119      * Since we currently insist on matching page sizes, just sanity check
2120      * we're being asked for whole host pages.
2121      */
2122     if (start & (our_host_ps-1) ||
2123        (len & (our_host_ps-1))) {
2124         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2125                      " len: %zd", __func__, start, len);
2126         mark_source_rp_bad(ms);
2127         return;
2128     }
2129 
2130     if (ram_save_queue_pages(rbname, start, len)) {
2131         mark_source_rp_bad(ms);
2132     }
2133 }
2134 
2135 /* Return true to retry, false to quit */
2136 static bool postcopy_pause_return_path_thread(MigrationState *s)
2137 {
2138     trace_postcopy_pause_return_path();
2139 
2140     qemu_sem_wait(&s->postcopy_pause_rp_sem);
2141 
2142     trace_postcopy_pause_return_path_continued();
2143 
2144     return true;
2145 }
2146 
2147 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2148 {
2149     RAMBlock *block = qemu_ram_block_by_name(block_name);
2150 
2151     if (!block) {
2152         error_report("%s: invalid block name '%s'", __func__, block_name);
2153         return -EINVAL;
2154     }
2155 
2156     /* Fetch the received bitmap and refresh the dirty bitmap */
2157     return ram_dirty_bitmap_reload(s, block);
2158 }
2159 
2160 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2161 {
2162     trace_source_return_path_thread_resume_ack(value);
2163 
2164     if (value != MIGRATION_RESUME_ACK_VALUE) {
2165         error_report("%s: illegal resume_ack value %"PRIu32,
2166                      __func__, value);
2167         return -1;
2168     }
2169 
2170     /* Now both sides are active. */
2171     migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2172                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2173 
2174     /* Notify send thread that time to continue send pages */
2175     qemu_sem_post(&s->rp_state.rp_sem);
2176 
2177     return 0;
2178 }
2179 
2180 /*
2181  * Handles messages sent on the return path towards the source VM
2182  *
2183  */
2184 static void *source_return_path_thread(void *opaque)
2185 {
2186     MigrationState *ms = opaque;
2187     QEMUFile *rp = ms->rp_state.from_dst_file;
2188     uint16_t header_len, header_type;
2189     uint8_t buf[512];
2190     uint32_t tmp32, sibling_error;
2191     ram_addr_t start = 0; /* =0 to silence warning */
2192     size_t  len = 0, expected_len;
2193     int res;
2194 
2195     trace_source_return_path_thread_entry();
2196     rcu_register_thread();
2197 
2198 retry:
2199     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2200            migration_is_setup_or_active(ms->state)) {
2201         trace_source_return_path_thread_loop_top();
2202         header_type = qemu_get_be16(rp);
2203         header_len = qemu_get_be16(rp);
2204 
2205         if (qemu_file_get_error(rp)) {
2206             mark_source_rp_bad(ms);
2207             goto out;
2208         }
2209 
2210         if (header_type >= MIG_RP_MSG_MAX ||
2211             header_type == MIG_RP_MSG_INVALID) {
2212             error_report("RP: Received invalid message 0x%04x length 0x%04x",
2213                     header_type, header_len);
2214             mark_source_rp_bad(ms);
2215             goto out;
2216         }
2217 
2218         if ((rp_cmd_args[header_type].len != -1 &&
2219             header_len != rp_cmd_args[header_type].len) ||
2220             header_len > sizeof(buf)) {
2221             error_report("RP: Received '%s' message (0x%04x) with"
2222                     "incorrect length %d expecting %zu",
2223                     rp_cmd_args[header_type].name, header_type, header_len,
2224                     (size_t)rp_cmd_args[header_type].len);
2225             mark_source_rp_bad(ms);
2226             goto out;
2227         }
2228 
2229         /* We know we've got a valid header by this point */
2230         res = qemu_get_buffer(rp, buf, header_len);
2231         if (res != header_len) {
2232             error_report("RP: Failed reading data for message 0x%04x"
2233                          " read %d expected %d",
2234                          header_type, res, header_len);
2235             mark_source_rp_bad(ms);
2236             goto out;
2237         }
2238 
2239         /* OK, we have the message and the data */
2240         switch (header_type) {
2241         case MIG_RP_MSG_SHUT:
2242             sibling_error = ldl_be_p(buf);
2243             trace_source_return_path_thread_shut(sibling_error);
2244             if (sibling_error) {
2245                 error_report("RP: Sibling indicated error %d", sibling_error);
2246                 mark_source_rp_bad(ms);
2247             }
2248             /*
2249              * We'll let the main thread deal with closing the RP
2250              * we could do a shutdown(2) on it, but we're the only user
2251              * anyway, so there's nothing gained.
2252              */
2253             goto out;
2254 
2255         case MIG_RP_MSG_PONG:
2256             tmp32 = ldl_be_p(buf);
2257             trace_source_return_path_thread_pong(tmp32);
2258             break;
2259 
2260         case MIG_RP_MSG_REQ_PAGES:
2261             start = ldq_be_p(buf);
2262             len = ldl_be_p(buf + 8);
2263             migrate_handle_rp_req_pages(ms, NULL, start, len);
2264             break;
2265 
2266         case MIG_RP_MSG_REQ_PAGES_ID:
2267             expected_len = 12 + 1; /* header + termination */
2268 
2269             if (header_len >= expected_len) {
2270                 start = ldq_be_p(buf);
2271                 len = ldl_be_p(buf + 8);
2272                 /* Now we expect an idstr */
2273                 tmp32 = buf[12]; /* Length of the following idstr */
2274                 buf[13 + tmp32] = '\0';
2275                 expected_len += tmp32;
2276             }
2277             if (header_len != expected_len) {
2278                 error_report("RP: Req_Page_id with length %d expecting %zd",
2279                         header_len, expected_len);
2280                 mark_source_rp_bad(ms);
2281                 goto out;
2282             }
2283             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2284             break;
2285 
2286         case MIG_RP_MSG_RECV_BITMAP:
2287             if (header_len < 1) {
2288                 error_report("%s: missing block name", __func__);
2289                 mark_source_rp_bad(ms);
2290                 goto out;
2291             }
2292             /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2293             buf[buf[0] + 1] = '\0';
2294             if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2295                 mark_source_rp_bad(ms);
2296                 goto out;
2297             }
2298             break;
2299 
2300         case MIG_RP_MSG_RESUME_ACK:
2301             tmp32 = ldl_be_p(buf);
2302             if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2303                 mark_source_rp_bad(ms);
2304                 goto out;
2305             }
2306             break;
2307 
2308         default:
2309             break;
2310         }
2311     }
2312 
2313 out:
2314     res = qemu_file_get_error(rp);
2315     if (res) {
2316         if (res == -EIO) {
2317             /*
2318              * Maybe there is something we can do: it looks like a
2319              * network down issue, and we pause for a recovery.
2320              */
2321             if (postcopy_pause_return_path_thread(ms)) {
2322                 /* Reload rp, reset the rest */
2323                 if (rp != ms->rp_state.from_dst_file) {
2324                     qemu_fclose(rp);
2325                     rp = ms->rp_state.from_dst_file;
2326                 }
2327                 ms->rp_state.error = false;
2328                 goto retry;
2329             }
2330         }
2331 
2332         trace_source_return_path_thread_bad_end();
2333         mark_source_rp_bad(ms);
2334     }
2335 
2336     trace_source_return_path_thread_end();
2337     ms->rp_state.from_dst_file = NULL;
2338     qemu_fclose(rp);
2339     rcu_unregister_thread();
2340     return NULL;
2341 }
2342 
2343 static int open_return_path_on_source(MigrationState *ms,
2344                                       bool create_thread)
2345 {
2346 
2347     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2348     if (!ms->rp_state.from_dst_file) {
2349         return -1;
2350     }
2351 
2352     trace_open_return_path_on_source();
2353 
2354     if (!create_thread) {
2355         /* We're done */
2356         return 0;
2357     }
2358 
2359     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2360                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2361 
2362     trace_open_return_path_on_source_continue();
2363 
2364     return 0;
2365 }
2366 
2367 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2368 static int await_return_path_close_on_source(MigrationState *ms)
2369 {
2370     /*
2371      * If this is a normal exit then the destination will send a SHUT and the
2372      * rp_thread will exit, however if there's an error we need to cause
2373      * it to exit.
2374      */
2375     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2376         /*
2377          * shutdown(2), if we have it, will cause it to unblock if it's stuck
2378          * waiting for the destination.
2379          */
2380         qemu_file_shutdown(ms->rp_state.from_dst_file);
2381         mark_source_rp_bad(ms);
2382     }
2383     trace_await_return_path_close_on_source_joining();
2384     qemu_thread_join(&ms->rp_state.rp_thread);
2385     trace_await_return_path_close_on_source_close();
2386     return ms->rp_state.error;
2387 }
2388 
2389 /*
2390  * Switch from normal iteration to postcopy
2391  * Returns non-0 on error
2392  */
2393 static int postcopy_start(MigrationState *ms)
2394 {
2395     int ret;
2396     QIOChannelBuffer *bioc;
2397     QEMUFile *fb;
2398     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2399     int64_t bandwidth = migrate_max_postcopy_bandwidth();
2400     bool restart_block = false;
2401     int cur_state = MIGRATION_STATUS_ACTIVE;
2402     if (!migrate_pause_before_switchover()) {
2403         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2404                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
2405     }
2406 
2407     trace_postcopy_start();
2408     qemu_mutex_lock_iothread();
2409     trace_postcopy_start_set_run();
2410 
2411     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2412     global_state_store();
2413     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2414     if (ret < 0) {
2415         goto fail;
2416     }
2417 
2418     ret = migration_maybe_pause(ms, &cur_state,
2419                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2420     if (ret < 0) {
2421         goto fail;
2422     }
2423 
2424     ret = bdrv_inactivate_all();
2425     if (ret < 0) {
2426         goto fail;
2427     }
2428     restart_block = true;
2429 
2430     /*
2431      * Cause any non-postcopiable, but iterative devices to
2432      * send out their final data.
2433      */
2434     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2435 
2436     /*
2437      * in Finish migrate and with the io-lock held everything should
2438      * be quiet, but we've potentially still got dirty pages and we
2439      * need to tell the destination to throw any pages it's already received
2440      * that are dirty
2441      */
2442     if (migrate_postcopy_ram()) {
2443         if (ram_postcopy_send_discard_bitmap(ms)) {
2444             error_report("postcopy send discard bitmap failed");
2445             goto fail;
2446         }
2447     }
2448 
2449     /*
2450      * send rest of state - note things that are doing postcopy
2451      * will notice we're in POSTCOPY_ACTIVE and not actually
2452      * wrap their state up here
2453      */
2454     /* 0 max-postcopy-bandwidth means unlimited */
2455     if (!bandwidth) {
2456         qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2457     } else {
2458         qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2459     }
2460     if (migrate_postcopy_ram()) {
2461         /* Ping just for debugging, helps line traces up */
2462         qemu_savevm_send_ping(ms->to_dst_file, 2);
2463     }
2464 
2465     /*
2466      * While loading the device state we may trigger page transfer
2467      * requests and the fd must be free to process those, and thus
2468      * the destination must read the whole device state off the fd before
2469      * it starts processing it.  Unfortunately the ad-hoc migration format
2470      * doesn't allow the destination to know the size to read without fully
2471      * parsing it through each devices load-state code (especially the open
2472      * coded devices that use get/put).
2473      * So we wrap the device state up in a package with a length at the start;
2474      * to do this we use a qemu_buf to hold the whole of the device state.
2475      */
2476     bioc = qio_channel_buffer_new(4096);
2477     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2478     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2479     object_unref(OBJECT(bioc));
2480 
2481     /*
2482      * Make sure the receiver can get incoming pages before we send the rest
2483      * of the state
2484      */
2485     qemu_savevm_send_postcopy_listen(fb);
2486 
2487     qemu_savevm_state_complete_precopy(fb, false, false);
2488     if (migrate_postcopy_ram()) {
2489         qemu_savevm_send_ping(fb, 3);
2490     }
2491 
2492     qemu_savevm_send_postcopy_run(fb);
2493 
2494     /* <><> end of stuff going into the package */
2495 
2496     /* Last point of recovery; as soon as we send the package the destination
2497      * can open devices and potentially start running.
2498      * Lets just check again we've not got any errors.
2499      */
2500     ret = qemu_file_get_error(ms->to_dst_file);
2501     if (ret) {
2502         error_report("postcopy_start: Migration stream errored (pre package)");
2503         goto fail_closefb;
2504     }
2505 
2506     restart_block = false;
2507 
2508     /* Now send that blob */
2509     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2510         goto fail_closefb;
2511     }
2512     qemu_fclose(fb);
2513 
2514     /* Send a notify to give a chance for anything that needs to happen
2515      * at the transition to postcopy and after the device state; in particular
2516      * spice needs to trigger a transition now
2517      */
2518     ms->postcopy_after_devices = true;
2519     notifier_list_notify(&migration_state_notifiers, ms);
2520 
2521     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2522 
2523     qemu_mutex_unlock_iothread();
2524 
2525     if (migrate_postcopy_ram()) {
2526         /*
2527          * Although this ping is just for debug, it could potentially be
2528          * used for getting a better measurement of downtime at the source.
2529          */
2530         qemu_savevm_send_ping(ms->to_dst_file, 4);
2531     }
2532 
2533     if (migrate_release_ram()) {
2534         ram_postcopy_migrated_memory_release(ms);
2535     }
2536 
2537     ret = qemu_file_get_error(ms->to_dst_file);
2538     if (ret) {
2539         error_report("postcopy_start: Migration stream errored");
2540         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2541                               MIGRATION_STATUS_FAILED);
2542     }
2543 
2544     return ret;
2545 
2546 fail_closefb:
2547     qemu_fclose(fb);
2548 fail:
2549     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2550                           MIGRATION_STATUS_FAILED);
2551     if (restart_block) {
2552         /* A failure happened early enough that we know the destination hasn't
2553          * accessed block devices, so we're safe to recover.
2554          */
2555         Error *local_err = NULL;
2556 
2557         bdrv_invalidate_cache_all(&local_err);
2558         if (local_err) {
2559             error_report_err(local_err);
2560         }
2561     }
2562     qemu_mutex_unlock_iothread();
2563     return -1;
2564 }
2565 
2566 /**
2567  * migration_maybe_pause: Pause if required to by
2568  * migrate_pause_before_switchover called with the iothread locked
2569  * Returns: 0 on success
2570  */
2571 static int migration_maybe_pause(MigrationState *s,
2572                                  int *current_active_state,
2573                                  int new_state)
2574 {
2575     if (!migrate_pause_before_switchover()) {
2576         return 0;
2577     }
2578 
2579     /* Since leaving this state is not atomic with posting the semaphore
2580      * it's possible that someone could have issued multiple migrate_continue
2581      * and the semaphore is incorrectly positive at this point;
2582      * the docs say it's undefined to reinit a semaphore that's already
2583      * init'd, so use timedwait to eat up any existing posts.
2584      */
2585     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2586         /* This block intentionally left blank */
2587     }
2588 
2589     qemu_mutex_unlock_iothread();
2590     migrate_set_state(&s->state, *current_active_state,
2591                       MIGRATION_STATUS_PRE_SWITCHOVER);
2592     qemu_sem_wait(&s->pause_sem);
2593     migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2594                       new_state);
2595     *current_active_state = new_state;
2596     qemu_mutex_lock_iothread();
2597 
2598     return s->state == new_state ? 0 : -EINVAL;
2599 }
2600 
2601 /**
2602  * migration_completion: Used by migration_thread when there's not much left.
2603  *   The caller 'breaks' the loop when this returns.
2604  *
2605  * @s: Current migration state
2606  */
2607 static void migration_completion(MigrationState *s)
2608 {
2609     int ret;
2610     int current_active_state = s->state;
2611 
2612     if (s->state == MIGRATION_STATUS_ACTIVE) {
2613         qemu_mutex_lock_iothread();
2614         s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2615         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2616         s->vm_was_running = runstate_is_running();
2617         ret = global_state_store();
2618 
2619         if (!ret) {
2620             bool inactivate = !migrate_colo_enabled();
2621             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2622             if (ret >= 0) {
2623                 ret = migration_maybe_pause(s, &current_active_state,
2624                                             MIGRATION_STATUS_DEVICE);
2625             }
2626             if (ret >= 0) {
2627                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2628                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2629                                                          inactivate);
2630             }
2631             if (inactivate && ret >= 0) {
2632                 s->block_inactive = true;
2633             }
2634         }
2635         qemu_mutex_unlock_iothread();
2636 
2637         if (ret < 0) {
2638             goto fail;
2639         }
2640     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2641         trace_migration_completion_postcopy_end();
2642 
2643         qemu_savevm_state_complete_postcopy(s->to_dst_file);
2644         trace_migration_completion_postcopy_end_after_complete();
2645     }
2646 
2647     /*
2648      * If rp was opened we must clean up the thread before
2649      * cleaning everything else up (since if there are no failures
2650      * it will wait for the destination to send it's status in
2651      * a SHUT command).
2652      */
2653     if (s->rp_state.from_dst_file) {
2654         int rp_error;
2655         trace_migration_return_path_end_before();
2656         rp_error = await_return_path_close_on_source(s);
2657         trace_migration_return_path_end_after(rp_error);
2658         if (rp_error) {
2659             goto fail_invalidate;
2660         }
2661     }
2662 
2663     if (qemu_file_get_error(s->to_dst_file)) {
2664         trace_migration_completion_file_err();
2665         goto fail_invalidate;
2666     }
2667 
2668     if (!migrate_colo_enabled()) {
2669         migrate_set_state(&s->state, current_active_state,
2670                           MIGRATION_STATUS_COMPLETED);
2671     }
2672 
2673     return;
2674 
2675 fail_invalidate:
2676     /* If not doing postcopy, vm_start() will be called: let's regain
2677      * control on images.
2678      */
2679     if (s->state == MIGRATION_STATUS_ACTIVE ||
2680         s->state == MIGRATION_STATUS_DEVICE) {
2681         Error *local_err = NULL;
2682 
2683         qemu_mutex_lock_iothread();
2684         bdrv_invalidate_cache_all(&local_err);
2685         if (local_err) {
2686             error_report_err(local_err);
2687         } else {
2688             s->block_inactive = false;
2689         }
2690         qemu_mutex_unlock_iothread();
2691     }
2692 
2693 fail:
2694     migrate_set_state(&s->state, current_active_state,
2695                       MIGRATION_STATUS_FAILED);
2696 }
2697 
2698 bool migrate_colo_enabled(void)
2699 {
2700     MigrationState *s = migrate_get_current();
2701     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2702 }
2703 
2704 typedef enum MigThrError {
2705     /* No error detected */
2706     MIG_THR_ERR_NONE = 0,
2707     /* Detected error, but resumed successfully */
2708     MIG_THR_ERR_RECOVERED = 1,
2709     /* Detected fatal error, need to exit */
2710     MIG_THR_ERR_FATAL = 2,
2711 } MigThrError;
2712 
2713 static int postcopy_resume_handshake(MigrationState *s)
2714 {
2715     qemu_savevm_send_postcopy_resume(s->to_dst_file);
2716 
2717     while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2718         qemu_sem_wait(&s->rp_state.rp_sem);
2719     }
2720 
2721     if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2722         return 0;
2723     }
2724 
2725     return -1;
2726 }
2727 
2728 /* Return zero if success, or <0 for error */
2729 static int postcopy_do_resume(MigrationState *s)
2730 {
2731     int ret;
2732 
2733     /*
2734      * Call all the resume_prepare() hooks, so that modules can be
2735      * ready for the migration resume.
2736      */
2737     ret = qemu_savevm_state_resume_prepare(s);
2738     if (ret) {
2739         error_report("%s: resume_prepare() failure detected: %d",
2740                      __func__, ret);
2741         return ret;
2742     }
2743 
2744     /*
2745      * Last handshake with destination on the resume (destination will
2746      * switch to postcopy-active afterwards)
2747      */
2748     ret = postcopy_resume_handshake(s);
2749     if (ret) {
2750         error_report("%s: handshake failed: %d", __func__, ret);
2751         return ret;
2752     }
2753 
2754     return 0;
2755 }
2756 
2757 /*
2758  * We don't return until we are in a safe state to continue current
2759  * postcopy migration.  Returns MIG_THR_ERR_RECOVERED if recovered, or
2760  * MIG_THR_ERR_FATAL if unrecovery failure happened.
2761  */
2762 static MigThrError postcopy_pause(MigrationState *s)
2763 {
2764     assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2765 
2766     while (true) {
2767         QEMUFile *file;
2768 
2769         migrate_set_state(&s->state, s->state,
2770                           MIGRATION_STATUS_POSTCOPY_PAUSED);
2771 
2772         /* Current channel is possibly broken. Release it. */
2773         assert(s->to_dst_file);
2774         qemu_mutex_lock(&s->qemu_file_lock);
2775         file = s->to_dst_file;
2776         s->to_dst_file = NULL;
2777         qemu_mutex_unlock(&s->qemu_file_lock);
2778 
2779         qemu_file_shutdown(file);
2780         qemu_fclose(file);
2781 
2782         error_report("Detected IO failure for postcopy. "
2783                      "Migration paused.");
2784 
2785         /*
2786          * We wait until things fixed up. Then someone will setup the
2787          * status back for us.
2788          */
2789         while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2790             qemu_sem_wait(&s->postcopy_pause_sem);
2791         }
2792 
2793         if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2794             /* Woken up by a recover procedure. Give it a shot */
2795 
2796             /*
2797              * Firstly, let's wake up the return path now, with a new
2798              * return path channel.
2799              */
2800             qemu_sem_post(&s->postcopy_pause_rp_sem);
2801 
2802             /* Do the resume logic */
2803             if (postcopy_do_resume(s) == 0) {
2804                 /* Let's continue! */
2805                 trace_postcopy_pause_continued();
2806                 return MIG_THR_ERR_RECOVERED;
2807             } else {
2808                 /*
2809                  * Something wrong happened during the recovery, let's
2810                  * pause again. Pause is always better than throwing
2811                  * data away.
2812                  */
2813                 continue;
2814             }
2815         } else {
2816             /* This is not right... Time to quit. */
2817             return MIG_THR_ERR_FATAL;
2818         }
2819     }
2820 }
2821 
2822 static MigThrError migration_detect_error(MigrationState *s)
2823 {
2824     int ret;
2825 
2826     /* Try to detect any file errors */
2827     ret = qemu_file_get_error(s->to_dst_file);
2828 
2829     if (!ret) {
2830         /* Everything is fine */
2831         return MIG_THR_ERR_NONE;
2832     }
2833 
2834     if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
2835         /*
2836          * For postcopy, we allow the network to be down for a
2837          * while. After that, it can be continued by a
2838          * recovery phase.
2839          */
2840         return postcopy_pause(s);
2841     } else {
2842         /*
2843          * For precopy (or postcopy with error outside IO), we fail
2844          * with no time.
2845          */
2846         migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
2847         trace_migration_thread_file_err();
2848 
2849         /* Time to stop the migration, now. */
2850         return MIG_THR_ERR_FATAL;
2851     }
2852 }
2853 
2854 /* How many bytes have we transferred since the beggining of the migration */
2855 static uint64_t migration_total_bytes(MigrationState *s)
2856 {
2857     return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
2858 }
2859 
2860 static void migration_calculate_complete(MigrationState *s)
2861 {
2862     uint64_t bytes = migration_total_bytes(s);
2863     int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2864     int64_t transfer_time;
2865 
2866     s->total_time = end_time - s->start_time;
2867     if (!s->downtime) {
2868         /*
2869          * It's still not set, so we are precopy migration.  For
2870          * postcopy, downtime is calculated during postcopy_start().
2871          */
2872         s->downtime = end_time - s->downtime_start;
2873     }
2874 
2875     transfer_time = s->total_time - s->setup_time;
2876     if (transfer_time) {
2877         s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
2878     }
2879 }
2880 
2881 static void migration_update_counters(MigrationState *s,
2882                                       int64_t current_time)
2883 {
2884     uint64_t transferred, time_spent;
2885     uint64_t current_bytes; /* bytes transferred since the beginning */
2886     double bandwidth;
2887 
2888     if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2889         return;
2890     }
2891 
2892     current_bytes = migration_total_bytes(s);
2893     transferred = current_bytes - s->iteration_initial_bytes;
2894     time_spent = current_time - s->iteration_start_time;
2895     bandwidth = (double)transferred / time_spent;
2896     s->threshold_size = bandwidth * s->parameters.downtime_limit;
2897 
2898     s->mbps = (((double) transferred * 8.0) /
2899                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2900 
2901     /*
2902      * if we haven't sent anything, we don't want to
2903      * recalculate. 10000 is a small enough number for our purposes
2904      */
2905     if (ram_counters.dirty_pages_rate && transferred > 10000) {
2906         s->expected_downtime = ram_counters.remaining / bandwidth;
2907     }
2908 
2909     qemu_file_reset_rate_limit(s->to_dst_file);
2910 
2911     s->iteration_start_time = current_time;
2912     s->iteration_initial_bytes = current_bytes;
2913 
2914     trace_migrate_transferred(transferred, time_spent,
2915                               bandwidth, s->threshold_size);
2916 }
2917 
2918 /* Migration thread iteration status */
2919 typedef enum {
2920     MIG_ITERATE_RESUME,         /* Resume current iteration */
2921     MIG_ITERATE_SKIP,           /* Skip current iteration */
2922     MIG_ITERATE_BREAK,          /* Break the loop */
2923 } MigIterateState;
2924 
2925 /*
2926  * Return true if continue to the next iteration directly, false
2927  * otherwise.
2928  */
2929 static MigIterateState migration_iteration_run(MigrationState *s)
2930 {
2931     uint64_t pending_size, pend_pre, pend_compat, pend_post;
2932     bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2933 
2934     qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
2935                               &pend_compat, &pend_post);
2936     pending_size = pend_pre + pend_compat + pend_post;
2937 
2938     trace_migrate_pending(pending_size, s->threshold_size,
2939                           pend_pre, pend_compat, pend_post);
2940 
2941     if (pending_size && pending_size >= s->threshold_size) {
2942         /* Still a significant amount to transfer */
2943         if (migrate_postcopy() && !in_postcopy &&
2944             pend_pre <= s->threshold_size &&
2945             atomic_read(&s->start_postcopy)) {
2946             if (postcopy_start(s)) {
2947                 error_report("%s: postcopy failed to start", __func__);
2948             }
2949             return MIG_ITERATE_SKIP;
2950         }
2951         /* Just another iteration step */
2952         qemu_savevm_state_iterate(s->to_dst_file,
2953             s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2954     } else {
2955         trace_migration_thread_low_pending(pending_size);
2956         migration_completion(s);
2957         return MIG_ITERATE_BREAK;
2958     }
2959 
2960     return MIG_ITERATE_RESUME;
2961 }
2962 
2963 static void migration_iteration_finish(MigrationState *s)
2964 {
2965     /* If we enabled cpu throttling for auto-converge, turn it off. */
2966     cpu_throttle_stop();
2967 
2968     qemu_mutex_lock_iothread();
2969     switch (s->state) {
2970     case MIGRATION_STATUS_COMPLETED:
2971         migration_calculate_complete(s);
2972         runstate_set(RUN_STATE_POSTMIGRATE);
2973         break;
2974 
2975     case MIGRATION_STATUS_ACTIVE:
2976         /*
2977          * We should really assert here, but since it's during
2978          * migration, let's try to reduce the usage of assertions.
2979          */
2980         if (!migrate_colo_enabled()) {
2981             error_report("%s: critical error: calling COLO code without "
2982                          "COLO enabled", __func__);
2983         }
2984         migrate_start_colo_process(s);
2985         /*
2986          * Fixme: we will run VM in COLO no matter its old running state.
2987          * After exited COLO, we will keep running.
2988          */
2989         s->vm_was_running = true;
2990         /* Fallthrough */
2991     case MIGRATION_STATUS_FAILED:
2992     case MIGRATION_STATUS_CANCELLED:
2993     case MIGRATION_STATUS_CANCELLING:
2994         if (s->vm_was_running) {
2995             vm_start();
2996         } else {
2997             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2998                 runstate_set(RUN_STATE_POSTMIGRATE);
2999             }
3000         }
3001         break;
3002 
3003     default:
3004         /* Should not reach here, but if so, forgive the VM. */
3005         error_report("%s: Unknown ending state %d", __func__, s->state);
3006         break;
3007     }
3008     qemu_bh_schedule(s->cleanup_bh);
3009     qemu_mutex_unlock_iothread();
3010 }
3011 
3012 void migration_make_urgent_request(void)
3013 {
3014     qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3015 }
3016 
3017 void migration_consume_urgent_request(void)
3018 {
3019     qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3020 }
3021 
3022 /*
3023  * Master migration thread on the source VM.
3024  * It drives the migration and pumps the data down the outgoing channel.
3025  */
3026 static void *migration_thread(void *opaque)
3027 {
3028     MigrationState *s = opaque;
3029     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3030     MigThrError thr_error;
3031     bool urgent = false;
3032 
3033     rcu_register_thread();
3034 
3035     s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3036 
3037     qemu_savevm_state_header(s->to_dst_file);
3038 
3039     /*
3040      * If we opened the return path, we need to make sure dst has it
3041      * opened as well.
3042      */
3043     if (s->rp_state.from_dst_file) {
3044         /* Now tell the dest that it should open its end so it can reply */
3045         qemu_savevm_send_open_return_path(s->to_dst_file);
3046 
3047         /* And do a ping that will make stuff easier to debug */
3048         qemu_savevm_send_ping(s->to_dst_file, 1);
3049     }
3050 
3051     if (migrate_postcopy()) {
3052         /*
3053          * Tell the destination that we *might* want to do postcopy later;
3054          * if the other end can't do postcopy it should fail now, nice and
3055          * early.
3056          */
3057         qemu_savevm_send_postcopy_advise(s->to_dst_file);
3058     }
3059 
3060     if (migrate_colo_enabled()) {
3061         /* Notify migration destination that we enable COLO */
3062         qemu_savevm_send_colo_enable(s->to_dst_file);
3063     }
3064 
3065     qemu_savevm_state_setup(s->to_dst_file);
3066 
3067     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3068     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3069                       MIGRATION_STATUS_ACTIVE);
3070 
3071     trace_migration_thread_setup_complete();
3072 
3073     while (s->state == MIGRATION_STATUS_ACTIVE ||
3074            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3075         int64_t current_time;
3076 
3077         if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3078             MigIterateState iter_state = migration_iteration_run(s);
3079             if (iter_state == MIG_ITERATE_SKIP) {
3080                 continue;
3081             } else if (iter_state == MIG_ITERATE_BREAK) {
3082                 break;
3083             }
3084         }
3085 
3086         /*
3087          * Try to detect any kind of failures, and see whether we
3088          * should stop the migration now.
3089          */
3090         thr_error = migration_detect_error(s);
3091         if (thr_error == MIG_THR_ERR_FATAL) {
3092             /* Stop migration */
3093             break;
3094         } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3095             /*
3096              * Just recovered from a e.g. network failure, reset all
3097              * the local variables. This is important to avoid
3098              * breaking transferred_bytes and bandwidth calculation
3099              */
3100             s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3101             s->iteration_initial_bytes = 0;
3102         }
3103 
3104         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3105 
3106         migration_update_counters(s, current_time);
3107 
3108         urgent = false;
3109         if (qemu_file_rate_limit(s->to_dst_file)) {
3110             /* Wait for a delay to do rate limiting OR
3111              * something urgent to post the semaphore.
3112              */
3113             int ms = s->iteration_start_time + BUFFER_DELAY - current_time;
3114             trace_migration_thread_ratelimit_pre(ms);
3115             if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3116                 /* We were worken by one or more urgent things but
3117                  * the timedwait will have consumed one of them.
3118                  * The service routine for the urgent wake will dec
3119                  * the semaphore itself for each item it consumes,
3120                  * so add this one we just eat back.
3121                  */
3122                 qemu_sem_post(&s->rate_limit_sem);
3123                 urgent = true;
3124             }
3125             trace_migration_thread_ratelimit_post(urgent);
3126         }
3127     }
3128 
3129     trace_migration_thread_after_loop();
3130     migration_iteration_finish(s);
3131     rcu_unregister_thread();
3132     return NULL;
3133 }
3134 
3135 void migrate_fd_connect(MigrationState *s, Error *error_in)
3136 {
3137     int64_t rate_limit;
3138     bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3139 
3140     s->expected_downtime = s->parameters.downtime_limit;
3141     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
3142     if (error_in) {
3143         migrate_fd_error(s, error_in);
3144         migrate_fd_cleanup(s);
3145         return;
3146     }
3147 
3148     if (resume) {
3149         /* This is a resumed migration */
3150         rate_limit = INT64_MAX;
3151     } else {
3152         /* This is a fresh new migration */
3153         rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3154 
3155         /* Notify before starting migration thread */
3156         notifier_list_notify(&migration_state_notifiers, s);
3157     }
3158 
3159     qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3160     qemu_file_set_blocking(s->to_dst_file, true);
3161 
3162     /*
3163      * Open the return path. For postcopy, it is used exclusively. For
3164      * precopy, only if user specified "return-path" capability would
3165      * QEMU uses the return path.
3166      */
3167     if (migrate_postcopy_ram() || migrate_use_return_path()) {
3168         if (open_return_path_on_source(s, !resume)) {
3169             error_report("Unable to open return-path for postcopy");
3170             migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3171             migrate_fd_cleanup(s);
3172             return;
3173         }
3174     }
3175 
3176     if (resume) {
3177         /* Wakeup the main migration thread to do the recovery */
3178         migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3179                           MIGRATION_STATUS_POSTCOPY_RECOVER);
3180         qemu_sem_post(&s->postcopy_pause_sem);
3181         return;
3182     }
3183 
3184     if (multifd_save_setup() != 0) {
3185         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3186                           MIGRATION_STATUS_FAILED);
3187         migrate_fd_cleanup(s);
3188         return;
3189     }
3190     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3191                        QEMU_THREAD_JOINABLE);
3192     s->migration_thread_running = true;
3193 }
3194 
3195 void migration_global_dump(Monitor *mon)
3196 {
3197     MigrationState *ms = migrate_get_current();
3198 
3199     monitor_printf(mon, "globals:\n");
3200     monitor_printf(mon, "store-global-state: %s\n",
3201                    ms->store_global_state ? "on" : "off");
3202     monitor_printf(mon, "only-migratable: %s\n",
3203                    ms->only_migratable ? "on" : "off");
3204     monitor_printf(mon, "send-configuration: %s\n",
3205                    ms->send_configuration ? "on" : "off");
3206     monitor_printf(mon, "send-section-footer: %s\n",
3207                    ms->send_section_footer ? "on" : "off");
3208     monitor_printf(mon, "decompress-error-check: %s\n",
3209                    ms->decompress_error_check ? "on" : "off");
3210 }
3211 
3212 #define DEFINE_PROP_MIG_CAP(name, x)             \
3213     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3214 
3215 static Property migration_properties[] = {
3216     DEFINE_PROP_BOOL("store-global-state", MigrationState,
3217                      store_global_state, true),
3218     DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
3219     DEFINE_PROP_BOOL("send-configuration", MigrationState,
3220                      send_configuration, true),
3221     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3222                      send_section_footer, true),
3223     DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3224                       decompress_error_check, true),
3225 
3226     /* Migration parameters */
3227     DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3228                       parameters.compress_level,
3229                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
3230     DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3231                       parameters.compress_threads,
3232                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3233     DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3234                       parameters.compress_wait_thread, true),
3235     DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3236                       parameters.decompress_threads,
3237                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3238     DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3239                       parameters.cpu_throttle_initial,
3240                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3241     DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3242                       parameters.cpu_throttle_increment,
3243                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3244     DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3245                       parameters.max_bandwidth, MAX_THROTTLE),
3246     DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3247                       parameters.downtime_limit,
3248                       DEFAULT_MIGRATE_SET_DOWNTIME),
3249     DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3250                       parameters.x_checkpoint_delay,
3251                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3252     DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
3253                       parameters.x_multifd_channels,
3254                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3255     DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
3256                       parameters.x_multifd_page_count,
3257                       DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
3258     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3259                       parameters.xbzrle_cache_size,
3260                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3261     DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3262                       parameters.max_postcopy_bandwidth,
3263                       DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3264     DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3265                       parameters.max_cpu_throttle,
3266                       DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3267 
3268     /* Migration capabilities */
3269     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3270     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3271     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3272     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3273     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3274     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3275     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3276     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3277     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3278     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3279     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3280     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
3281 
3282     DEFINE_PROP_END_OF_LIST(),
3283 };
3284 
3285 static void migration_class_init(ObjectClass *klass, void *data)
3286 {
3287     DeviceClass *dc = DEVICE_CLASS(klass);
3288 
3289     dc->user_creatable = false;
3290     dc->props = migration_properties;
3291 }
3292 
3293 static void migration_instance_finalize(Object *obj)
3294 {
3295     MigrationState *ms = MIGRATION_OBJ(obj);
3296     MigrationParameters *params = &ms->parameters;
3297 
3298     qemu_mutex_destroy(&ms->error_mutex);
3299     qemu_mutex_destroy(&ms->qemu_file_lock);
3300     g_free(params->tls_hostname);
3301     g_free(params->tls_creds);
3302     qemu_sem_destroy(&ms->rate_limit_sem);
3303     qemu_sem_destroy(&ms->pause_sem);
3304     qemu_sem_destroy(&ms->postcopy_pause_sem);
3305     qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3306     qemu_sem_destroy(&ms->rp_state.rp_sem);
3307     error_free(ms->error);
3308 }
3309 
3310 static void migration_instance_init(Object *obj)
3311 {
3312     MigrationState *ms = MIGRATION_OBJ(obj);
3313     MigrationParameters *params = &ms->parameters;
3314 
3315     ms->state = MIGRATION_STATUS_NONE;
3316     ms->mbps = -1;
3317     qemu_sem_init(&ms->pause_sem, 0);
3318     qemu_mutex_init(&ms->error_mutex);
3319 
3320     params->tls_hostname = g_strdup("");
3321     params->tls_creds = g_strdup("");
3322 
3323     /* Set has_* up only for parameter checks */
3324     params->has_compress_level = true;
3325     params->has_compress_threads = true;
3326     params->has_decompress_threads = true;
3327     params->has_cpu_throttle_initial = true;
3328     params->has_cpu_throttle_increment = true;
3329     params->has_max_bandwidth = true;
3330     params->has_downtime_limit = true;
3331     params->has_x_checkpoint_delay = true;
3332     params->has_block_incremental = true;
3333     params->has_x_multifd_channels = true;
3334     params->has_x_multifd_page_count = true;
3335     params->has_xbzrle_cache_size = true;
3336     params->has_max_postcopy_bandwidth = true;
3337     params->has_max_cpu_throttle = true;
3338 
3339     qemu_sem_init(&ms->postcopy_pause_sem, 0);
3340     qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3341     qemu_sem_init(&ms->rp_state.rp_sem, 0);
3342     qemu_sem_init(&ms->rate_limit_sem, 0);
3343     qemu_mutex_init(&ms->qemu_file_lock);
3344 }
3345 
3346 /*
3347  * Return true if check pass, false otherwise. Error will be put
3348  * inside errp if provided.
3349  */
3350 static bool migration_object_check(MigrationState *ms, Error **errp)
3351 {
3352     MigrationCapabilityStatusList *head = NULL;
3353     /* Assuming all off */
3354     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3355     int i;
3356 
3357     if (!migrate_params_check(&ms->parameters, errp)) {
3358         return false;
3359     }
3360 
3361     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3362         if (ms->enabled_capabilities[i]) {
3363             head = migrate_cap_add(head, i, true);
3364         }
3365     }
3366 
3367     ret = migrate_caps_check(cap_list, head, errp);
3368 
3369     /* It works with head == NULL */
3370     qapi_free_MigrationCapabilityStatusList(head);
3371 
3372     return ret;
3373 }
3374 
3375 static const TypeInfo migration_type = {
3376     .name = TYPE_MIGRATION,
3377     /*
3378      * NOTE: TYPE_MIGRATION is not really a device, as the object is
3379      * not created using qdev_create(), it is not attached to the qdev
3380      * device tree, and it is never realized.
3381      *
3382      * TODO: Make this TYPE_OBJECT once QOM provides something like
3383      * TYPE_DEVICE's "-global" properties.
3384      */
3385     .parent = TYPE_DEVICE,
3386     .class_init = migration_class_init,
3387     .class_size = sizeof(MigrationClass),
3388     .instance_size = sizeof(MigrationState),
3389     .instance_init = migration_instance_init,
3390     .instance_finalize = migration_instance_finalize,
3391 };
3392 
3393 static void register_migration_types(void)
3394 {
3395     type_register_static(&migration_type);
3396 }
3397 
3398 type_init(register_migration_types);
3399