xref: /openbmc/qemu/migration/migration.c (revision fd990e86)
1 /*
2  * QEMU live migration
3  *
4  * Copyright IBM, Corp. 2008
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  * Contributions after 2012-01-13 are licensed under the terms of the
13  * GNU GPL, version 2 or (at your option) any later version.
14  */
15 
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
21 #include "exec.h"
22 #include "fd.h"
23 #include "socket.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "rdma.h"
27 #include "ram.h"
28 #include "migration/global_state.h"
29 #include "migration/misc.h"
30 #include "migration.h"
31 #include "savevm.h"
32 #include "qemu-file-channel.h"
33 #include "qemu-file.h"
34 #include "migration/vmstate.h"
35 #include "block/block.h"
36 #include "qapi/error.h"
37 #include "qapi/clone-visitor.h"
38 #include "qapi/qapi-visit-sockets.h"
39 #include "qapi/qapi-commands-migration.h"
40 #include "qapi/qapi-events-migration.h"
41 #include "qapi/qmp/qerror.h"
42 #include "qapi/qmp/qnull.h"
43 #include "qemu/rcu.h"
44 #include "block.h"
45 #include "postcopy-ram.h"
46 #include "qemu/thread.h"
47 #include "trace.h"
48 #include "exec/target_page.h"
49 #include "io/channel-buffer.h"
50 #include "migration/colo.h"
51 #include "hw/boards.h"
52 #include "hw/qdev-properties.h"
53 #include "monitor/monitor.h"
54 #include "net/announce.h"
55 #include "qemu/queue.h"
56 #include "multifd.h"
57 
58 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
59 
60 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
61  * data. */
62 #define BUFFER_DELAY     100
63 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
64 
65 /* Time in milliseconds we are allowed to stop the source,
66  * for sending the last part */
67 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
68 
69 /* Maximum migrate downtime set to 2000 seconds */
70 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
71 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
72 
73 /* Default compression thread count */
74 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
75 /* Default decompression thread count, usually decompression is at
76  * least 4 times as fast as compression.*/
77 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
78 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
79 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
80 /* Define default autoconverge cpu throttle migration parameters */
81 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
82 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
83 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
84 
85 /* Migration XBZRLE default cache size */
86 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
87 
88 /* The delay time (in ms) between two COLO checkpoints */
89 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
90 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
91 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
92 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
93 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
94 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
95 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
96 
97 /* Background transfer rate for postcopy, 0 means unlimited, note
98  * that page requests can still exceed this limit.
99  */
100 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
101 
102 /*
103  * Parameters for self_announce_delay giving a stream of RARP/ARP
104  * packets after migration.
105  */
106 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL  50
107 #define DEFAULT_MIGRATE_ANNOUNCE_MAX     550
108 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS    5
109 #define DEFAULT_MIGRATE_ANNOUNCE_STEP    100
110 
111 static NotifierList migration_state_notifiers =
112     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
113 
114 static bool deferred_incoming;
115 
116 /* Messages sent on the return path from destination to source */
117 enum mig_rp_message_type {
118     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
119     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
120     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
121 
122     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
123     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
124     MIG_RP_MSG_RECV_BITMAP,  /* send recved_bitmap back to source */
125     MIG_RP_MSG_RESUME_ACK,   /* tell source that we are ready to resume */
126 
127     MIG_RP_MSG_MAX
128 };
129 
130 /* When we add fault tolerance, we could have several
131    migrations at once.  For now we don't need to add
132    dynamic creation of migration */
133 
134 static MigrationState *current_migration;
135 static MigrationIncomingState *current_incoming;
136 
137 static bool migration_object_check(MigrationState *ms, Error **errp);
138 static int migration_maybe_pause(MigrationState *s,
139                                  int *current_active_state,
140                                  int new_state);
141 static void migrate_fd_cancel(MigrationState *s);
142 
143 void migration_object_init(void)
144 {
145     MachineState *ms = MACHINE(qdev_get_machine());
146     Error *err = NULL;
147 
148     /* This can only be called once. */
149     assert(!current_migration);
150     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
151 
152     /*
153      * Init the migrate incoming object as well no matter whether
154      * we'll use it or not.
155      */
156     assert(!current_incoming);
157     current_incoming = g_new0(MigrationIncomingState, 1);
158     current_incoming->state = MIGRATION_STATUS_NONE;
159     current_incoming->postcopy_remote_fds =
160         g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
161     qemu_mutex_init(&current_incoming->rp_mutex);
162     qemu_event_init(&current_incoming->main_thread_load_event, false);
163     qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
164     qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
165 
166     init_dirty_bitmap_incoming_migration();
167 
168     if (!migration_object_check(current_migration, &err)) {
169         error_report_err(err);
170         exit(1);
171     }
172 
173     /*
174      * We cannot really do this in migration_instance_init() since at
175      * that time global properties are not yet applied, then this
176      * value will be definitely replaced by something else.
177      */
178     if (ms->enforce_config_section) {
179         current_migration->send_configuration = true;
180     }
181 }
182 
183 void migration_shutdown(void)
184 {
185     /*
186      * Cancel the current migration - that will (eventually)
187      * stop the migration using this structure
188      */
189     migrate_fd_cancel(current_migration);
190     object_unref(OBJECT(current_migration));
191 }
192 
193 /* For outgoing */
194 MigrationState *migrate_get_current(void)
195 {
196     /* This can only be called after the object created. */
197     assert(current_migration);
198     return current_migration;
199 }
200 
201 MigrationIncomingState *migration_incoming_get_current(void)
202 {
203     assert(current_incoming);
204     return current_incoming;
205 }
206 
207 void migration_incoming_state_destroy(void)
208 {
209     struct MigrationIncomingState *mis = migration_incoming_get_current();
210 
211     if (mis->to_src_file) {
212         /* Tell source that we are done */
213         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
214         qemu_fclose(mis->to_src_file);
215         mis->to_src_file = NULL;
216     }
217 
218     if (mis->from_src_file) {
219         qemu_fclose(mis->from_src_file);
220         mis->from_src_file = NULL;
221     }
222     if (mis->postcopy_remote_fds) {
223         g_array_free(mis->postcopy_remote_fds, TRUE);
224         mis->postcopy_remote_fds = NULL;
225     }
226 
227     qemu_event_reset(&mis->main_thread_load_event);
228 
229     if (mis->socket_address_list) {
230         qapi_free_SocketAddressList(mis->socket_address_list);
231         mis->socket_address_list = NULL;
232     }
233 }
234 
235 static void migrate_generate_event(int new_state)
236 {
237     if (migrate_use_events()) {
238         qapi_event_send_migration(new_state);
239     }
240 }
241 
242 static bool migrate_late_block_activate(void)
243 {
244     MigrationState *s;
245 
246     s = migrate_get_current();
247 
248     return s->enabled_capabilities[
249         MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
250 }
251 
252 /*
253  * Called on -incoming with a defer: uri.
254  * The migration can be started later after any parameters have been
255  * changed.
256  */
257 static void deferred_incoming_migration(Error **errp)
258 {
259     if (deferred_incoming) {
260         error_setg(errp, "Incoming migration already deferred");
261     }
262     deferred_incoming = true;
263 }
264 
265 /*
266  * Send a message on the return channel back to the source
267  * of the migration.
268  */
269 static int migrate_send_rp_message(MigrationIncomingState *mis,
270                                    enum mig_rp_message_type message_type,
271                                    uint16_t len, void *data)
272 {
273     int ret = 0;
274 
275     trace_migrate_send_rp_message((int)message_type, len);
276     qemu_mutex_lock(&mis->rp_mutex);
277 
278     /*
279      * It's possible that the file handle got lost due to network
280      * failures.
281      */
282     if (!mis->to_src_file) {
283         ret = -EIO;
284         goto error;
285     }
286 
287     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
288     qemu_put_be16(mis->to_src_file, len);
289     qemu_put_buffer(mis->to_src_file, data, len);
290     qemu_fflush(mis->to_src_file);
291 
292     /* It's possible that qemu file got error during sending */
293     ret = qemu_file_get_error(mis->to_src_file);
294 
295 error:
296     qemu_mutex_unlock(&mis->rp_mutex);
297     return ret;
298 }
299 
300 /* Request a range of pages from the source VM at the given
301  * start address.
302  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
303  *           as the last request (a name must have been given previously)
304  *   Start: Address offset within the RB
305  *   Len: Length in bytes required - must be a multiple of pagesize
306  */
307 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
308                               ram_addr_t start, size_t len)
309 {
310     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
311     size_t msglen = 12; /* start + len */
312     enum mig_rp_message_type msg_type;
313 
314     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
315     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
316 
317     if (rbname) {
318         int rbname_len = strlen(rbname);
319         assert(rbname_len < 256);
320 
321         bufc[msglen++] = rbname_len;
322         memcpy(bufc + msglen, rbname, rbname_len);
323         msglen += rbname_len;
324         msg_type = MIG_RP_MSG_REQ_PAGES_ID;
325     } else {
326         msg_type = MIG_RP_MSG_REQ_PAGES;
327     }
328 
329     return migrate_send_rp_message(mis, msg_type, msglen, bufc);
330 }
331 
332 static bool migration_colo_enabled;
333 bool migration_incoming_colo_enabled(void)
334 {
335     return migration_colo_enabled;
336 }
337 
338 void migration_incoming_disable_colo(void)
339 {
340     migration_colo_enabled = false;
341 }
342 
343 void migration_incoming_enable_colo(void)
344 {
345     migration_colo_enabled = true;
346 }
347 
348 void migrate_add_address(SocketAddress *address)
349 {
350     MigrationIncomingState *mis = migration_incoming_get_current();
351     SocketAddressList *addrs;
352 
353     addrs = g_new0(SocketAddressList, 1);
354     addrs->next = mis->socket_address_list;
355     mis->socket_address_list = addrs;
356     addrs->value = QAPI_CLONE(SocketAddress, address);
357 }
358 
359 void qemu_start_incoming_migration(const char *uri, Error **errp)
360 {
361     const char *p;
362 
363     qapi_event_send_migration(MIGRATION_STATUS_SETUP);
364     if (!strcmp(uri, "defer")) {
365         deferred_incoming_migration(errp);
366     } else if (strstart(uri, "tcp:", &p)) {
367         tcp_start_incoming_migration(p, errp);
368 #ifdef CONFIG_RDMA
369     } else if (strstart(uri, "rdma:", &p)) {
370         rdma_start_incoming_migration(p, errp);
371 #endif
372     } else if (strstart(uri, "exec:", &p)) {
373         exec_start_incoming_migration(p, errp);
374     } else if (strstart(uri, "unix:", &p)) {
375         unix_start_incoming_migration(p, errp);
376     } else if (strstart(uri, "fd:", &p)) {
377         fd_start_incoming_migration(p, errp);
378     } else {
379         error_setg(errp, "unknown migration protocol: %s", uri);
380     }
381 }
382 
383 static void process_incoming_migration_bh(void *opaque)
384 {
385     Error *local_err = NULL;
386     MigrationIncomingState *mis = opaque;
387 
388     /* If capability late_block_activate is set:
389      * Only fire up the block code now if we're going to restart the
390      * VM, else 'cont' will do it.
391      * This causes file locking to happen; so we don't want it to happen
392      * unless we really are starting the VM.
393      */
394     if (!migrate_late_block_activate() ||
395          (autostart && (!global_state_received() ||
396             global_state_get_runstate() == RUN_STATE_RUNNING))) {
397         /* Make sure all file formats flush their mutable metadata.
398          * If we get an error here, just don't restart the VM yet. */
399         bdrv_invalidate_cache_all(&local_err);
400         if (local_err) {
401             error_report_err(local_err);
402             local_err = NULL;
403             autostart = false;
404         }
405     }
406 
407     /*
408      * This must happen after all error conditions are dealt with and
409      * we're sure the VM is going to be running on this host.
410      */
411     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
412 
413     if (multifd_load_cleanup(&local_err) != 0) {
414         error_report_err(local_err);
415         autostart = false;
416     }
417     /* If global state section was not received or we are in running
418        state, we need to obey autostart. Any other state is set with
419        runstate_set. */
420 
421     dirty_bitmap_mig_before_vm_start();
422 
423     if (!global_state_received() ||
424         global_state_get_runstate() == RUN_STATE_RUNNING) {
425         if (autostart) {
426             vm_start();
427         } else {
428             runstate_set(RUN_STATE_PAUSED);
429         }
430     } else if (migration_incoming_colo_enabled()) {
431         migration_incoming_disable_colo();
432         vm_start();
433     } else {
434         runstate_set(global_state_get_runstate());
435     }
436     /*
437      * This must happen after any state changes since as soon as an external
438      * observer sees this event they might start to prod at the VM assuming
439      * it's ready to use.
440      */
441     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
442                       MIGRATION_STATUS_COMPLETED);
443     qemu_bh_delete(mis->bh);
444     migration_incoming_state_destroy();
445 }
446 
447 static void process_incoming_migration_co(void *opaque)
448 {
449     MigrationIncomingState *mis = migration_incoming_get_current();
450     PostcopyState ps;
451     int ret;
452     Error *local_err = NULL;
453 
454     assert(mis->from_src_file);
455     mis->migration_incoming_co = qemu_coroutine_self();
456     mis->largest_page_size = qemu_ram_pagesize_largest();
457     postcopy_state_set(POSTCOPY_INCOMING_NONE);
458     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
459                       MIGRATION_STATUS_ACTIVE);
460     ret = qemu_loadvm_state(mis->from_src_file);
461 
462     ps = postcopy_state_get();
463     trace_process_incoming_migration_co_end(ret, ps);
464     if (ps != POSTCOPY_INCOMING_NONE) {
465         if (ps == POSTCOPY_INCOMING_ADVISE) {
466             /*
467              * Where a migration had postcopy enabled (and thus went to advise)
468              * but managed to complete within the precopy period, we can use
469              * the normal exit.
470              */
471             postcopy_ram_incoming_cleanup(mis);
472         } else if (ret >= 0) {
473             /*
474              * Postcopy was started, cleanup should happen at the end of the
475              * postcopy thread.
476              */
477             trace_process_incoming_migration_co_postcopy_end_main();
478             return;
479         }
480         /* Else if something went wrong then just fall out of the normal exit */
481     }
482 
483     /* we get COLO info, and know if we are in COLO mode */
484     if (!ret && migration_incoming_colo_enabled()) {
485         /* Make sure all file formats flush their mutable metadata */
486         bdrv_invalidate_cache_all(&local_err);
487         if (local_err) {
488             error_report_err(local_err);
489             goto fail;
490         }
491 
492         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
493              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
494         mis->have_colo_incoming_thread = true;
495         qemu_coroutine_yield();
496 
497         /* Wait checkpoint incoming thread exit before free resource */
498         qemu_thread_join(&mis->colo_incoming_thread);
499         /* We hold the global iothread lock, so it is safe here */
500         colo_release_ram_cache();
501     }
502 
503     if (ret < 0) {
504         error_report("load of migration failed: %s", strerror(-ret));
505         goto fail;
506     }
507     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
508     qemu_bh_schedule(mis->bh);
509     mis->migration_incoming_co = NULL;
510     return;
511 fail:
512     local_err = NULL;
513     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
514                       MIGRATION_STATUS_FAILED);
515     qemu_fclose(mis->from_src_file);
516     if (multifd_load_cleanup(&local_err) != 0) {
517         error_report_err(local_err);
518     }
519     exit(EXIT_FAILURE);
520 }
521 
522 /**
523  * @migration_incoming_setup: Setup incoming migration
524  *
525  * Returns 0 for no error or 1 for error
526  *
527  * @f: file for main migration channel
528  * @errp: where to put errors
529  */
530 static int migration_incoming_setup(QEMUFile *f, Error **errp)
531 {
532     MigrationIncomingState *mis = migration_incoming_get_current();
533     Error *local_err = NULL;
534 
535     if (multifd_load_setup(&local_err) != 0) {
536         /* We haven't been able to create multifd threads
537            nothing better to do */
538         error_report_err(local_err);
539         exit(EXIT_FAILURE);
540     }
541 
542     if (!mis->from_src_file) {
543         mis->from_src_file = f;
544     }
545     qemu_file_set_blocking(f, false);
546     return 0;
547 }
548 
549 void migration_incoming_process(void)
550 {
551     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
552     qemu_coroutine_enter(co);
553 }
554 
555 /* Returns true if recovered from a paused migration, otherwise false */
556 static bool postcopy_try_recover(QEMUFile *f)
557 {
558     MigrationIncomingState *mis = migration_incoming_get_current();
559 
560     if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
561         /* Resumed from a paused postcopy migration */
562 
563         mis->from_src_file = f;
564         /* Postcopy has standalone thread to do vm load */
565         qemu_file_set_blocking(f, true);
566 
567         /* Re-configure the return path */
568         mis->to_src_file = qemu_file_get_return_path(f);
569 
570         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
571                           MIGRATION_STATUS_POSTCOPY_RECOVER);
572 
573         /*
574          * Here, we only wake up the main loading thread (while the
575          * fault thread will still be waiting), so that we can receive
576          * commands from source now, and answer it if needed. The
577          * fault thread will be woken up afterwards until we are sure
578          * that source is ready to reply to page requests.
579          */
580         qemu_sem_post(&mis->postcopy_pause_sem_dst);
581         return true;
582     }
583 
584     return false;
585 }
586 
587 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
588 {
589     Error *local_err = NULL;
590 
591     if (postcopy_try_recover(f)) {
592         return;
593     }
594 
595     if (migration_incoming_setup(f, &local_err)) {
596         if (local_err) {
597             error_propagate(errp, local_err);
598         }
599         return;
600     }
601     migration_incoming_process();
602 }
603 
604 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
605 {
606     MigrationIncomingState *mis = migration_incoming_get_current();
607     Error *local_err = NULL;
608     bool start_migration;
609 
610     if (!mis->from_src_file) {
611         /* The first connection (multifd may have multiple) */
612         QEMUFile *f = qemu_fopen_channel_input(ioc);
613 
614         /* If it's a recovery, we're done */
615         if (postcopy_try_recover(f)) {
616             return;
617         }
618 
619         if (migration_incoming_setup(f, &local_err)) {
620             if (local_err) {
621                 error_propagate(errp, local_err);
622             }
623             return;
624         }
625 
626         /*
627          * Common migration only needs one channel, so we can start
628          * right now.  Multifd needs more than one channel, we wait.
629          */
630         start_migration = !migrate_use_multifd();
631     } else {
632         /* Multiple connections */
633         assert(migrate_use_multifd());
634         start_migration = multifd_recv_new_channel(ioc, &local_err);
635         if (local_err) {
636             error_propagate(errp, local_err);
637             return;
638         }
639     }
640 
641     if (start_migration) {
642         migration_incoming_process();
643     }
644 }
645 
646 /**
647  * @migration_has_all_channels: We have received all channels that we need
648  *
649  * Returns true when we have got connections to all the channels that
650  * we need for migration.
651  */
652 bool migration_has_all_channels(void)
653 {
654     MigrationIncomingState *mis = migration_incoming_get_current();
655     bool all_channels;
656 
657     all_channels = multifd_recv_all_channels_created();
658 
659     return all_channels && mis->from_src_file != NULL;
660 }
661 
662 /*
663  * Send a 'SHUT' message on the return channel with the given value
664  * to indicate that we've finished with the RP.  Non-0 value indicates
665  * error.
666  */
667 void migrate_send_rp_shut(MigrationIncomingState *mis,
668                           uint32_t value)
669 {
670     uint32_t buf;
671 
672     buf = cpu_to_be32(value);
673     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
674 }
675 
676 /*
677  * Send a 'PONG' message on the return channel with the given value
678  * (normally in response to a 'PING')
679  */
680 void migrate_send_rp_pong(MigrationIncomingState *mis,
681                           uint32_t value)
682 {
683     uint32_t buf;
684 
685     buf = cpu_to_be32(value);
686     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
687 }
688 
689 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
690                                  char *block_name)
691 {
692     char buf[512];
693     int len;
694     int64_t res;
695 
696     /*
697      * First, we send the header part. It contains only the len of
698      * idstr, and the idstr itself.
699      */
700     len = strlen(block_name);
701     buf[0] = len;
702     memcpy(buf + 1, block_name, len);
703 
704     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
705         error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
706                      __func__);
707         return;
708     }
709 
710     migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
711 
712     /*
713      * Next, we dump the received bitmap to the stream.
714      *
715      * TODO: currently we are safe since we are the only one that is
716      * using the to_src_file handle (fault thread is still paused),
717      * and it's ok even not taking the mutex. However the best way is
718      * to take the lock before sending the message header, and release
719      * the lock after sending the bitmap.
720      */
721     qemu_mutex_lock(&mis->rp_mutex);
722     res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
723     qemu_mutex_unlock(&mis->rp_mutex);
724 
725     trace_migrate_send_rp_recv_bitmap(block_name, res);
726 }
727 
728 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
729 {
730     uint32_t buf;
731 
732     buf = cpu_to_be32(value);
733     migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
734 }
735 
736 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
737 {
738     MigrationCapabilityStatusList *head = NULL;
739     MigrationCapabilityStatusList *caps;
740     MigrationState *s = migrate_get_current();
741     int i;
742 
743     caps = NULL; /* silence compiler warning */
744     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
745 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
746         if (i == MIGRATION_CAPABILITY_BLOCK) {
747             continue;
748         }
749 #endif
750         if (head == NULL) {
751             head = g_malloc0(sizeof(*caps));
752             caps = head;
753         } else {
754             caps->next = g_malloc0(sizeof(*caps));
755             caps = caps->next;
756         }
757         caps->value =
758             g_malloc(sizeof(*caps->value));
759         caps->value->capability = i;
760         caps->value->state = s->enabled_capabilities[i];
761     }
762 
763     return head;
764 }
765 
766 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
767 {
768     MigrationParameters *params;
769     MigrationState *s = migrate_get_current();
770 
771     /* TODO use QAPI_CLONE() instead of duplicating it inline */
772     params = g_malloc0(sizeof(*params));
773     params->has_compress_level = true;
774     params->compress_level = s->parameters.compress_level;
775     params->has_compress_threads = true;
776     params->compress_threads = s->parameters.compress_threads;
777     params->has_compress_wait_thread = true;
778     params->compress_wait_thread = s->parameters.compress_wait_thread;
779     params->has_decompress_threads = true;
780     params->decompress_threads = s->parameters.decompress_threads;
781     params->has_cpu_throttle_initial = true;
782     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
783     params->has_cpu_throttle_increment = true;
784     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
785     params->has_tls_creds = true;
786     params->tls_creds = g_strdup(s->parameters.tls_creds);
787     params->has_tls_hostname = true;
788     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
789     params->has_tls_authz = true;
790     params->tls_authz = g_strdup(s->parameters.tls_authz);
791     params->has_max_bandwidth = true;
792     params->max_bandwidth = s->parameters.max_bandwidth;
793     params->has_downtime_limit = true;
794     params->downtime_limit = s->parameters.downtime_limit;
795     params->has_x_checkpoint_delay = true;
796     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
797     params->has_block_incremental = true;
798     params->block_incremental = s->parameters.block_incremental;
799     params->has_multifd_channels = true;
800     params->multifd_channels = s->parameters.multifd_channels;
801     params->has_multifd_compression = true;
802     params->multifd_compression = s->parameters.multifd_compression;
803     params->has_multifd_zlib_level = true;
804     params->multifd_zlib_level = s->parameters.multifd_zlib_level;
805     params->has_multifd_zstd_level = true;
806     params->multifd_zstd_level = s->parameters.multifd_zstd_level;
807     params->has_xbzrle_cache_size = true;
808     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
809     params->has_max_postcopy_bandwidth = true;
810     params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
811     params->has_max_cpu_throttle = true;
812     params->max_cpu_throttle = s->parameters.max_cpu_throttle;
813     params->has_announce_initial = true;
814     params->announce_initial = s->parameters.announce_initial;
815     params->has_announce_max = true;
816     params->announce_max = s->parameters.announce_max;
817     params->has_announce_rounds = true;
818     params->announce_rounds = s->parameters.announce_rounds;
819     params->has_announce_step = true;
820     params->announce_step = s->parameters.announce_step;
821 
822     return params;
823 }
824 
825 AnnounceParameters *migrate_announce_params(void)
826 {
827     static AnnounceParameters ap;
828 
829     MigrationState *s = migrate_get_current();
830 
831     ap.initial = s->parameters.announce_initial;
832     ap.max = s->parameters.announce_max;
833     ap.rounds = s->parameters.announce_rounds;
834     ap.step = s->parameters.announce_step;
835 
836     return &ap;
837 }
838 
839 /*
840  * Return true if we're already in the middle of a migration
841  * (i.e. any of the active or setup states)
842  */
843 bool migration_is_setup_or_active(int state)
844 {
845     switch (state) {
846     case MIGRATION_STATUS_ACTIVE:
847     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
848     case MIGRATION_STATUS_POSTCOPY_PAUSED:
849     case MIGRATION_STATUS_POSTCOPY_RECOVER:
850     case MIGRATION_STATUS_SETUP:
851     case MIGRATION_STATUS_PRE_SWITCHOVER:
852     case MIGRATION_STATUS_DEVICE:
853     case MIGRATION_STATUS_WAIT_UNPLUG:
854         return true;
855 
856     default:
857         return false;
858 
859     }
860 }
861 
862 bool migration_is_running(int state)
863 {
864     switch (state) {
865     case MIGRATION_STATUS_ACTIVE:
866     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
867     case MIGRATION_STATUS_POSTCOPY_PAUSED:
868     case MIGRATION_STATUS_POSTCOPY_RECOVER:
869     case MIGRATION_STATUS_SETUP:
870     case MIGRATION_STATUS_PRE_SWITCHOVER:
871     case MIGRATION_STATUS_DEVICE:
872     case MIGRATION_STATUS_WAIT_UNPLUG:
873     case MIGRATION_STATUS_CANCELLING:
874         return true;
875 
876     default:
877         return false;
878 
879     }
880 }
881 
882 static void populate_time_info(MigrationInfo *info, MigrationState *s)
883 {
884     info->has_status = true;
885     info->has_setup_time = true;
886     info->setup_time = s->setup_time;
887     if (s->state == MIGRATION_STATUS_COMPLETED) {
888         info->has_total_time = true;
889         info->total_time = s->total_time;
890         info->has_downtime = true;
891         info->downtime = s->downtime;
892     } else {
893         info->has_total_time = true;
894         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
895                            s->start_time;
896         info->has_expected_downtime = true;
897         info->expected_downtime = s->expected_downtime;
898     }
899 }
900 
901 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
902 {
903     info->has_ram = true;
904     info->ram = g_malloc0(sizeof(*info->ram));
905     info->ram->transferred = ram_counters.transferred;
906     info->ram->total = ram_bytes_total();
907     info->ram->duplicate = ram_counters.duplicate;
908     /* legacy value.  It is not used anymore */
909     info->ram->skipped = 0;
910     info->ram->normal = ram_counters.normal;
911     info->ram->normal_bytes = ram_counters.normal *
912         qemu_target_page_size();
913     info->ram->mbps = s->mbps;
914     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
915     info->ram->postcopy_requests = ram_counters.postcopy_requests;
916     info->ram->page_size = qemu_target_page_size();
917     info->ram->multifd_bytes = ram_counters.multifd_bytes;
918     info->ram->pages_per_second = s->pages_per_second;
919 
920     if (migrate_use_xbzrle()) {
921         info->has_xbzrle_cache = true;
922         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
923         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
924         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
925         info->xbzrle_cache->pages = xbzrle_counters.pages;
926         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
927         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
928         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
929     }
930 
931     if (migrate_use_compression()) {
932         info->has_compression = true;
933         info->compression = g_malloc0(sizeof(*info->compression));
934         info->compression->pages = compression_counters.pages;
935         info->compression->busy = compression_counters.busy;
936         info->compression->busy_rate = compression_counters.busy_rate;
937         info->compression->compressed_size =
938                                     compression_counters.compressed_size;
939         info->compression->compression_rate =
940                                     compression_counters.compression_rate;
941     }
942 
943     if (cpu_throttle_active()) {
944         info->has_cpu_throttle_percentage = true;
945         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
946     }
947 
948     if (s->state != MIGRATION_STATUS_COMPLETED) {
949         info->ram->remaining = ram_bytes_remaining();
950         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
951     }
952 }
953 
954 static void populate_disk_info(MigrationInfo *info)
955 {
956     if (blk_mig_active()) {
957         info->has_disk = true;
958         info->disk = g_malloc0(sizeof(*info->disk));
959         info->disk->transferred = blk_mig_bytes_transferred();
960         info->disk->remaining = blk_mig_bytes_remaining();
961         info->disk->total = blk_mig_bytes_total();
962     }
963 }
964 
965 static void fill_source_migration_info(MigrationInfo *info)
966 {
967     MigrationState *s = migrate_get_current();
968 
969     switch (s->state) {
970     case MIGRATION_STATUS_NONE:
971         /* no migration has happened ever */
972         /* do not overwrite destination migration status */
973         return;
974         break;
975     case MIGRATION_STATUS_SETUP:
976         info->has_status = true;
977         info->has_total_time = false;
978         break;
979     case MIGRATION_STATUS_ACTIVE:
980     case MIGRATION_STATUS_CANCELLING:
981     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
982     case MIGRATION_STATUS_PRE_SWITCHOVER:
983     case MIGRATION_STATUS_DEVICE:
984     case MIGRATION_STATUS_POSTCOPY_PAUSED:
985     case MIGRATION_STATUS_POSTCOPY_RECOVER:
986         /* TODO add some postcopy stats */
987         populate_time_info(info, s);
988         populate_ram_info(info, s);
989         populate_disk_info(info);
990         break;
991     case MIGRATION_STATUS_COLO:
992         info->has_status = true;
993         /* TODO: display COLO specific information (checkpoint info etc.) */
994         break;
995     case MIGRATION_STATUS_COMPLETED:
996         populate_time_info(info, s);
997         populate_ram_info(info, s);
998         break;
999     case MIGRATION_STATUS_FAILED:
1000         info->has_status = true;
1001         if (s->error) {
1002             info->has_error_desc = true;
1003             info->error_desc = g_strdup(error_get_pretty(s->error));
1004         }
1005         break;
1006     case MIGRATION_STATUS_CANCELLED:
1007         info->has_status = true;
1008         break;
1009     case MIGRATION_STATUS_WAIT_UNPLUG:
1010         info->has_status = true;
1011         break;
1012     }
1013     info->status = s->state;
1014 }
1015 
1016 /**
1017  * @migration_caps_check - check capability validity
1018  *
1019  * @cap_list: old capability list, array of bool
1020  * @params: new capabilities to be applied soon
1021  * @errp: set *errp if the check failed, with reason
1022  *
1023  * Returns true if check passed, otherwise false.
1024  */
1025 static bool migrate_caps_check(bool *cap_list,
1026                                MigrationCapabilityStatusList *params,
1027                                Error **errp)
1028 {
1029     MigrationCapabilityStatusList *cap;
1030     bool old_postcopy_cap;
1031     MigrationIncomingState *mis = migration_incoming_get_current();
1032 
1033     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1034 
1035     for (cap = params; cap; cap = cap->next) {
1036         cap_list[cap->value->capability] = cap->value->state;
1037     }
1038 
1039 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1040     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1041         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1042                    "block migration");
1043         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1044         return false;
1045     }
1046 #endif
1047 
1048 #ifndef CONFIG_REPLICATION
1049     if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1050         error_setg(errp, "QEMU compiled without replication module"
1051                    " can't enable COLO");
1052         error_append_hint(errp, "Please enable replication before COLO.\n");
1053         return false;
1054     }
1055 #endif
1056 
1057     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1058         /* This check is reasonably expensive, so only when it's being
1059          * set the first time, also it's only the destination that needs
1060          * special support.
1061          */
1062         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1063             !postcopy_ram_supported_by_host(mis)) {
1064             /* postcopy_ram_supported_by_host will have emitted a more
1065              * detailed message
1066              */
1067             error_setg(errp, "Postcopy is not supported");
1068             return false;
1069         }
1070 
1071         if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1072             error_setg(errp, "Postcopy is not compatible with ignore-shared");
1073             return false;
1074         }
1075     }
1076 
1077     return true;
1078 }
1079 
1080 static void fill_destination_migration_info(MigrationInfo *info)
1081 {
1082     MigrationIncomingState *mis = migration_incoming_get_current();
1083 
1084     if (mis->socket_address_list) {
1085         info->has_socket_address = true;
1086         info->socket_address =
1087             QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1088     }
1089 
1090     switch (mis->state) {
1091     case MIGRATION_STATUS_NONE:
1092         return;
1093         break;
1094     case MIGRATION_STATUS_SETUP:
1095     case MIGRATION_STATUS_CANCELLING:
1096     case MIGRATION_STATUS_CANCELLED:
1097     case MIGRATION_STATUS_ACTIVE:
1098     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1099     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1100     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1101     case MIGRATION_STATUS_FAILED:
1102     case MIGRATION_STATUS_COLO:
1103         info->has_status = true;
1104         break;
1105     case MIGRATION_STATUS_COMPLETED:
1106         info->has_status = true;
1107         fill_destination_postcopy_migration_info(info);
1108         break;
1109     }
1110     info->status = mis->state;
1111 }
1112 
1113 MigrationInfo *qmp_query_migrate(Error **errp)
1114 {
1115     MigrationInfo *info = g_malloc0(sizeof(*info));
1116 
1117     fill_destination_migration_info(info);
1118     fill_source_migration_info(info);
1119 
1120     return info;
1121 }
1122 
1123 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1124                                   Error **errp)
1125 {
1126     MigrationState *s = migrate_get_current();
1127     MigrationCapabilityStatusList *cap;
1128     bool cap_list[MIGRATION_CAPABILITY__MAX];
1129 
1130     if (migration_is_running(s->state)) {
1131         error_setg(errp, QERR_MIGRATION_ACTIVE);
1132         return;
1133     }
1134 
1135     memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1136     if (!migrate_caps_check(cap_list, params, errp)) {
1137         return;
1138     }
1139 
1140     for (cap = params; cap; cap = cap->next) {
1141         s->enabled_capabilities[cap->value->capability] = cap->value->state;
1142     }
1143 }
1144 
1145 /*
1146  * Check whether the parameters are valid. Error will be put into errp
1147  * (if provided). Return true if valid, otherwise false.
1148  */
1149 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1150 {
1151     if (params->has_compress_level &&
1152         (params->compress_level > 9)) {
1153         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1154                    "is invalid, it should be in the range of 0 to 9");
1155         return false;
1156     }
1157 
1158     if (params->has_compress_threads && (params->compress_threads < 1)) {
1159         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1160                    "compress_threads",
1161                    "is invalid, it should be in the range of 1 to 255");
1162         return false;
1163     }
1164 
1165     if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1166         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1167                    "decompress_threads",
1168                    "is invalid, it should be in the range of 1 to 255");
1169         return false;
1170     }
1171 
1172     if (params->has_cpu_throttle_initial &&
1173         (params->cpu_throttle_initial < 1 ||
1174          params->cpu_throttle_initial > 99)) {
1175         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1176                    "cpu_throttle_initial",
1177                    "an integer in the range of 1 to 99");
1178         return false;
1179     }
1180 
1181     if (params->has_cpu_throttle_increment &&
1182         (params->cpu_throttle_increment < 1 ||
1183          params->cpu_throttle_increment > 99)) {
1184         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1185                    "cpu_throttle_increment",
1186                    "an integer in the range of 1 to 99");
1187         return false;
1188     }
1189 
1190     if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1191         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
1192                          " range of 0 to %zu bytes/second", SIZE_MAX);
1193         return false;
1194     }
1195 
1196     if (params->has_downtime_limit &&
1197         (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1198         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1199                          "the range of 0 to %d milliseconds",
1200                          MAX_MIGRATE_DOWNTIME);
1201         return false;
1202     }
1203 
1204     /* x_checkpoint_delay is now always positive */
1205 
1206     if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1207         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1208                    "multifd_channels",
1209                    "is invalid, it should be in the range of 1 to 255");
1210         return false;
1211     }
1212 
1213     if (params->has_multifd_zlib_level &&
1214         (params->multifd_zlib_level > 9)) {
1215         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1216                    "is invalid, it should be in the range of 0 to 9");
1217         return false;
1218     }
1219 
1220     if (params->has_multifd_zstd_level &&
1221         (params->multifd_zstd_level > 20)) {
1222         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1223                    "is invalid, it should be in the range of 0 to 20");
1224         return false;
1225     }
1226 
1227     if (params->has_xbzrle_cache_size &&
1228         (params->xbzrle_cache_size < qemu_target_page_size() ||
1229          !is_power_of_2(params->xbzrle_cache_size))) {
1230         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1231                    "xbzrle_cache_size",
1232                    "is invalid, it should be bigger than target page size"
1233                    " and a power of two");
1234         return false;
1235     }
1236 
1237     if (params->has_max_cpu_throttle &&
1238         (params->max_cpu_throttle < params->cpu_throttle_initial ||
1239          params->max_cpu_throttle > 99)) {
1240         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1241                    "max_cpu_throttle",
1242                    "an integer in the range of cpu_throttle_initial to 99");
1243         return false;
1244     }
1245 
1246     if (params->has_announce_initial &&
1247         params->announce_initial > 100000) {
1248         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1249                    "announce_initial",
1250                    "is invalid, it must be less than 100000 ms");
1251         return false;
1252     }
1253     if (params->has_announce_max &&
1254         params->announce_max > 100000) {
1255         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1256                    "announce_max",
1257                    "is invalid, it must be less than 100000 ms");
1258        return false;
1259     }
1260     if (params->has_announce_rounds &&
1261         params->announce_rounds > 1000) {
1262         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1263                    "announce_rounds",
1264                    "is invalid, it must be in the range of 0 to 1000");
1265        return false;
1266     }
1267     if (params->has_announce_step &&
1268         (params->announce_step < 1 ||
1269         params->announce_step > 10000)) {
1270         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1271                    "announce_step",
1272                    "is invalid, it must be in the range of 1 to 10000 ms");
1273        return false;
1274     }
1275     return true;
1276 }
1277 
1278 static void migrate_params_test_apply(MigrateSetParameters *params,
1279                                       MigrationParameters *dest)
1280 {
1281     *dest = migrate_get_current()->parameters;
1282 
1283     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1284 
1285     if (params->has_compress_level) {
1286         dest->compress_level = params->compress_level;
1287     }
1288 
1289     if (params->has_compress_threads) {
1290         dest->compress_threads = params->compress_threads;
1291     }
1292 
1293     if (params->has_compress_wait_thread) {
1294         dest->compress_wait_thread = params->compress_wait_thread;
1295     }
1296 
1297     if (params->has_decompress_threads) {
1298         dest->decompress_threads = params->decompress_threads;
1299     }
1300 
1301     if (params->has_cpu_throttle_initial) {
1302         dest->cpu_throttle_initial = params->cpu_throttle_initial;
1303     }
1304 
1305     if (params->has_cpu_throttle_increment) {
1306         dest->cpu_throttle_increment = params->cpu_throttle_increment;
1307     }
1308 
1309     if (params->has_tls_creds) {
1310         assert(params->tls_creds->type == QTYPE_QSTRING);
1311         dest->tls_creds = g_strdup(params->tls_creds->u.s);
1312     }
1313 
1314     if (params->has_tls_hostname) {
1315         assert(params->tls_hostname->type == QTYPE_QSTRING);
1316         dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1317     }
1318 
1319     if (params->has_max_bandwidth) {
1320         dest->max_bandwidth = params->max_bandwidth;
1321     }
1322 
1323     if (params->has_downtime_limit) {
1324         dest->downtime_limit = params->downtime_limit;
1325     }
1326 
1327     if (params->has_x_checkpoint_delay) {
1328         dest->x_checkpoint_delay = params->x_checkpoint_delay;
1329     }
1330 
1331     if (params->has_block_incremental) {
1332         dest->block_incremental = params->block_incremental;
1333     }
1334     if (params->has_multifd_channels) {
1335         dest->multifd_channels = params->multifd_channels;
1336     }
1337     if (params->has_multifd_compression) {
1338         dest->multifd_compression = params->multifd_compression;
1339     }
1340     if (params->has_xbzrle_cache_size) {
1341         dest->xbzrle_cache_size = params->xbzrle_cache_size;
1342     }
1343     if (params->has_max_postcopy_bandwidth) {
1344         dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1345     }
1346     if (params->has_max_cpu_throttle) {
1347         dest->max_cpu_throttle = params->max_cpu_throttle;
1348     }
1349     if (params->has_announce_initial) {
1350         dest->announce_initial = params->announce_initial;
1351     }
1352     if (params->has_announce_max) {
1353         dest->announce_max = params->announce_max;
1354     }
1355     if (params->has_announce_rounds) {
1356         dest->announce_rounds = params->announce_rounds;
1357     }
1358     if (params->has_announce_step) {
1359         dest->announce_step = params->announce_step;
1360     }
1361 }
1362 
1363 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1364 {
1365     MigrationState *s = migrate_get_current();
1366 
1367     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1368 
1369     if (params->has_compress_level) {
1370         s->parameters.compress_level = params->compress_level;
1371     }
1372 
1373     if (params->has_compress_threads) {
1374         s->parameters.compress_threads = params->compress_threads;
1375     }
1376 
1377     if (params->has_compress_wait_thread) {
1378         s->parameters.compress_wait_thread = params->compress_wait_thread;
1379     }
1380 
1381     if (params->has_decompress_threads) {
1382         s->parameters.decompress_threads = params->decompress_threads;
1383     }
1384 
1385     if (params->has_cpu_throttle_initial) {
1386         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1387     }
1388 
1389     if (params->has_cpu_throttle_increment) {
1390         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1391     }
1392 
1393     if (params->has_tls_creds) {
1394         g_free(s->parameters.tls_creds);
1395         assert(params->tls_creds->type == QTYPE_QSTRING);
1396         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1397     }
1398 
1399     if (params->has_tls_hostname) {
1400         g_free(s->parameters.tls_hostname);
1401         assert(params->tls_hostname->type == QTYPE_QSTRING);
1402         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1403     }
1404 
1405     if (params->has_tls_authz) {
1406         g_free(s->parameters.tls_authz);
1407         assert(params->tls_authz->type == QTYPE_QSTRING);
1408         s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1409     }
1410 
1411     if (params->has_max_bandwidth) {
1412         s->parameters.max_bandwidth = params->max_bandwidth;
1413         if (s->to_dst_file && !migration_in_postcopy()) {
1414             qemu_file_set_rate_limit(s->to_dst_file,
1415                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1416         }
1417     }
1418 
1419     if (params->has_downtime_limit) {
1420         s->parameters.downtime_limit = params->downtime_limit;
1421     }
1422 
1423     if (params->has_x_checkpoint_delay) {
1424         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1425         if (migration_in_colo_state()) {
1426             colo_checkpoint_notify(s);
1427         }
1428     }
1429 
1430     if (params->has_block_incremental) {
1431         s->parameters.block_incremental = params->block_incremental;
1432     }
1433     if (params->has_multifd_channels) {
1434         s->parameters.multifd_channels = params->multifd_channels;
1435     }
1436     if (params->has_multifd_compression) {
1437         s->parameters.multifd_compression = params->multifd_compression;
1438     }
1439     if (params->has_xbzrle_cache_size) {
1440         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1441         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1442     }
1443     if (params->has_max_postcopy_bandwidth) {
1444         s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1445         if (s->to_dst_file && migration_in_postcopy()) {
1446             qemu_file_set_rate_limit(s->to_dst_file,
1447                     s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1448         }
1449     }
1450     if (params->has_max_cpu_throttle) {
1451         s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1452     }
1453     if (params->has_announce_initial) {
1454         s->parameters.announce_initial = params->announce_initial;
1455     }
1456     if (params->has_announce_max) {
1457         s->parameters.announce_max = params->announce_max;
1458     }
1459     if (params->has_announce_rounds) {
1460         s->parameters.announce_rounds = params->announce_rounds;
1461     }
1462     if (params->has_announce_step) {
1463         s->parameters.announce_step = params->announce_step;
1464     }
1465 }
1466 
1467 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1468 {
1469     MigrationParameters tmp;
1470 
1471     /* TODO Rewrite "" to null instead */
1472     if (params->has_tls_creds
1473         && params->tls_creds->type == QTYPE_QNULL) {
1474         qobject_unref(params->tls_creds->u.n);
1475         params->tls_creds->type = QTYPE_QSTRING;
1476         params->tls_creds->u.s = strdup("");
1477     }
1478     /* TODO Rewrite "" to null instead */
1479     if (params->has_tls_hostname
1480         && params->tls_hostname->type == QTYPE_QNULL) {
1481         qobject_unref(params->tls_hostname->u.n);
1482         params->tls_hostname->type = QTYPE_QSTRING;
1483         params->tls_hostname->u.s = strdup("");
1484     }
1485 
1486     migrate_params_test_apply(params, &tmp);
1487 
1488     if (!migrate_params_check(&tmp, errp)) {
1489         /* Invalid parameter */
1490         return;
1491     }
1492 
1493     migrate_params_apply(params, errp);
1494 }
1495 
1496 
1497 void qmp_migrate_start_postcopy(Error **errp)
1498 {
1499     MigrationState *s = migrate_get_current();
1500 
1501     if (!migrate_postcopy()) {
1502         error_setg(errp, "Enable postcopy with migrate_set_capability before"
1503                          " the start of migration");
1504         return;
1505     }
1506 
1507     if (s->state == MIGRATION_STATUS_NONE) {
1508         error_setg(errp, "Postcopy must be started after migration has been"
1509                          " started");
1510         return;
1511     }
1512     /*
1513      * we don't error if migration has finished since that would be racy
1514      * with issuing this command.
1515      */
1516     atomic_set(&s->start_postcopy, true);
1517 }
1518 
1519 /* shared migration helpers */
1520 
1521 void migrate_set_state(int *state, int old_state, int new_state)
1522 {
1523     assert(new_state < MIGRATION_STATUS__MAX);
1524     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1525         trace_migrate_set_state(MigrationStatus_str(new_state));
1526         migrate_generate_event(new_state);
1527     }
1528 }
1529 
1530 static MigrationCapabilityStatusList *migrate_cap_add(
1531     MigrationCapabilityStatusList *list,
1532     MigrationCapability index,
1533     bool state)
1534 {
1535     MigrationCapabilityStatusList *cap;
1536 
1537     cap = g_new0(MigrationCapabilityStatusList, 1);
1538     cap->value = g_new0(MigrationCapabilityStatus, 1);
1539     cap->value->capability = index;
1540     cap->value->state = state;
1541     cap->next = list;
1542 
1543     return cap;
1544 }
1545 
1546 void migrate_set_block_enabled(bool value, Error **errp)
1547 {
1548     MigrationCapabilityStatusList *cap;
1549 
1550     cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1551     qmp_migrate_set_capabilities(cap, errp);
1552     qapi_free_MigrationCapabilityStatusList(cap);
1553 }
1554 
1555 static void migrate_set_block_incremental(MigrationState *s, bool value)
1556 {
1557     s->parameters.block_incremental = value;
1558 }
1559 
1560 static void block_cleanup_parameters(MigrationState *s)
1561 {
1562     if (s->must_remove_block_options) {
1563         /* setting to false can never fail */
1564         migrate_set_block_enabled(false, &error_abort);
1565         migrate_set_block_incremental(s, false);
1566         s->must_remove_block_options = false;
1567     }
1568 }
1569 
1570 static void migrate_fd_cleanup(MigrationState *s)
1571 {
1572     qemu_bh_delete(s->cleanup_bh);
1573     s->cleanup_bh = NULL;
1574 
1575     qemu_savevm_state_cleanup();
1576 
1577     if (s->to_dst_file) {
1578         QEMUFile *tmp;
1579 
1580         trace_migrate_fd_cleanup();
1581         qemu_mutex_unlock_iothread();
1582         if (s->migration_thread_running) {
1583             qemu_thread_join(&s->thread);
1584             s->migration_thread_running = false;
1585         }
1586         qemu_mutex_lock_iothread();
1587 
1588         multifd_save_cleanup();
1589         qemu_mutex_lock(&s->qemu_file_lock);
1590         tmp = s->to_dst_file;
1591         s->to_dst_file = NULL;
1592         qemu_mutex_unlock(&s->qemu_file_lock);
1593         /*
1594          * Close the file handle without the lock to make sure the
1595          * critical section won't block for long.
1596          */
1597         qemu_fclose(tmp);
1598     }
1599 
1600     assert(!migration_is_active(s));
1601 
1602     if (s->state == MIGRATION_STATUS_CANCELLING) {
1603         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1604                           MIGRATION_STATUS_CANCELLED);
1605     }
1606 
1607     if (s->error) {
1608         /* It is used on info migrate.  We can't free it */
1609         error_report_err(error_copy(s->error));
1610     }
1611     notifier_list_notify(&migration_state_notifiers, s);
1612     block_cleanup_parameters(s);
1613 }
1614 
1615 static void migrate_fd_cleanup_schedule(MigrationState *s)
1616 {
1617     /*
1618      * Ref the state for bh, because it may be called when
1619      * there're already no other refs
1620      */
1621     object_ref(OBJECT(s));
1622     qemu_bh_schedule(s->cleanup_bh);
1623 }
1624 
1625 static void migrate_fd_cleanup_bh(void *opaque)
1626 {
1627     MigrationState *s = opaque;
1628     migrate_fd_cleanup(s);
1629     object_unref(OBJECT(s));
1630 }
1631 
1632 void migrate_set_error(MigrationState *s, const Error *error)
1633 {
1634     qemu_mutex_lock(&s->error_mutex);
1635     if (!s->error) {
1636         s->error = error_copy(error);
1637     }
1638     qemu_mutex_unlock(&s->error_mutex);
1639 }
1640 
1641 void migrate_fd_error(MigrationState *s, const Error *error)
1642 {
1643     trace_migrate_fd_error(error_get_pretty(error));
1644     assert(s->to_dst_file == NULL);
1645     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1646                       MIGRATION_STATUS_FAILED);
1647     migrate_set_error(s, error);
1648 }
1649 
1650 static void migrate_fd_cancel(MigrationState *s)
1651 {
1652     int old_state ;
1653     QEMUFile *f = migrate_get_current()->to_dst_file;
1654     trace_migrate_fd_cancel();
1655 
1656     if (s->rp_state.from_dst_file) {
1657         /* shutdown the rp socket, so causing the rp thread to shutdown */
1658         qemu_file_shutdown(s->rp_state.from_dst_file);
1659     }
1660 
1661     do {
1662         old_state = s->state;
1663         if (!migration_is_running(old_state)) {
1664             break;
1665         }
1666         /* If the migration is paused, kick it out of the pause */
1667         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1668             qemu_sem_post(&s->pause_sem);
1669         }
1670         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1671     } while (s->state != MIGRATION_STATUS_CANCELLING);
1672 
1673     /*
1674      * If we're unlucky the migration code might be stuck somewhere in a
1675      * send/write while the network has failed and is waiting to timeout;
1676      * if we've got shutdown(2) available then we can force it to quit.
1677      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1678      * called in a bh, so there is no race against this cancel.
1679      */
1680     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1681         qemu_file_shutdown(f);
1682     }
1683     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1684         Error *local_err = NULL;
1685 
1686         bdrv_invalidate_cache_all(&local_err);
1687         if (local_err) {
1688             error_report_err(local_err);
1689         } else {
1690             s->block_inactive = false;
1691         }
1692     }
1693 }
1694 
1695 void add_migration_state_change_notifier(Notifier *notify)
1696 {
1697     notifier_list_add(&migration_state_notifiers, notify);
1698 }
1699 
1700 void remove_migration_state_change_notifier(Notifier *notify)
1701 {
1702     notifier_remove(notify);
1703 }
1704 
1705 bool migration_in_setup(MigrationState *s)
1706 {
1707     return s->state == MIGRATION_STATUS_SETUP;
1708 }
1709 
1710 bool migration_has_finished(MigrationState *s)
1711 {
1712     return s->state == MIGRATION_STATUS_COMPLETED;
1713 }
1714 
1715 bool migration_has_failed(MigrationState *s)
1716 {
1717     return (s->state == MIGRATION_STATUS_CANCELLED ||
1718             s->state == MIGRATION_STATUS_FAILED);
1719 }
1720 
1721 bool migration_in_postcopy(void)
1722 {
1723     MigrationState *s = migrate_get_current();
1724 
1725     switch (s->state) {
1726     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1727     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1728     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1729         return true;
1730     default:
1731         return false;
1732     }
1733 }
1734 
1735 bool migration_in_postcopy_after_devices(MigrationState *s)
1736 {
1737     return migration_in_postcopy() && s->postcopy_after_devices;
1738 }
1739 
1740 bool migration_is_idle(void)
1741 {
1742     MigrationState *s = current_migration;
1743 
1744     if (!s) {
1745         return true;
1746     }
1747 
1748     switch (s->state) {
1749     case MIGRATION_STATUS_NONE:
1750     case MIGRATION_STATUS_CANCELLED:
1751     case MIGRATION_STATUS_COMPLETED:
1752     case MIGRATION_STATUS_FAILED:
1753         return true;
1754     case MIGRATION_STATUS_SETUP:
1755     case MIGRATION_STATUS_CANCELLING:
1756     case MIGRATION_STATUS_ACTIVE:
1757     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1758     case MIGRATION_STATUS_COLO:
1759     case MIGRATION_STATUS_PRE_SWITCHOVER:
1760     case MIGRATION_STATUS_DEVICE:
1761     case MIGRATION_STATUS_WAIT_UNPLUG:
1762         return false;
1763     case MIGRATION_STATUS__MAX:
1764         g_assert_not_reached();
1765     }
1766 
1767     return false;
1768 }
1769 
1770 bool migration_is_active(MigrationState *s)
1771 {
1772     return (s->state == MIGRATION_STATUS_ACTIVE ||
1773             s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1774 }
1775 
1776 void migrate_init(MigrationState *s)
1777 {
1778     /*
1779      * Reinitialise all migration state, except
1780      * parameters/capabilities that the user set, and
1781      * locks.
1782      */
1783     s->cleanup_bh = 0;
1784     s->to_dst_file = NULL;
1785     s->state = MIGRATION_STATUS_NONE;
1786     s->rp_state.from_dst_file = NULL;
1787     s->rp_state.error = false;
1788     s->mbps = 0.0;
1789     s->pages_per_second = 0.0;
1790     s->downtime = 0;
1791     s->expected_downtime = 0;
1792     s->setup_time = 0;
1793     s->start_postcopy = false;
1794     s->postcopy_after_devices = false;
1795     s->migration_thread_running = false;
1796     error_free(s->error);
1797     s->error = NULL;
1798 
1799     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1800 
1801     s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1802     s->total_time = 0;
1803     s->vm_was_running = false;
1804     s->iteration_initial_bytes = 0;
1805     s->threshold_size = 0;
1806 }
1807 
1808 static GSList *migration_blockers;
1809 
1810 int migrate_add_blocker(Error *reason, Error **errp)
1811 {
1812     if (only_migratable) {
1813         error_propagate_prepend(errp, error_copy(reason),
1814                                 "disallowing migration blocker "
1815                                 "(--only-migratable) for: ");
1816         return -EACCES;
1817     }
1818 
1819     if (migration_is_idle()) {
1820         migration_blockers = g_slist_prepend(migration_blockers, reason);
1821         return 0;
1822     }
1823 
1824     error_propagate_prepend(errp, error_copy(reason),
1825                             "disallowing migration blocker "
1826                             "(migration in progress) for: ");
1827     return -EBUSY;
1828 }
1829 
1830 void migrate_del_blocker(Error *reason)
1831 {
1832     migration_blockers = g_slist_remove(migration_blockers, reason);
1833 }
1834 
1835 void qmp_migrate_incoming(const char *uri, Error **errp)
1836 {
1837     Error *local_err = NULL;
1838     static bool once = true;
1839 
1840     if (!deferred_incoming) {
1841         error_setg(errp, "For use with '-incoming defer'");
1842         return;
1843     }
1844     if (!once) {
1845         error_setg(errp, "The incoming migration has already been started");
1846         return;
1847     }
1848 
1849     qemu_start_incoming_migration(uri, &local_err);
1850 
1851     if (local_err) {
1852         error_propagate(errp, local_err);
1853         return;
1854     }
1855 
1856     once = false;
1857 }
1858 
1859 void qmp_migrate_recover(const char *uri, Error **errp)
1860 {
1861     MigrationIncomingState *mis = migration_incoming_get_current();
1862 
1863     if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1864         error_setg(errp, "Migrate recover can only be run "
1865                    "when postcopy is paused.");
1866         return;
1867     }
1868 
1869     if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1870                        false, true) == true) {
1871         error_setg(errp, "Migrate recovery is triggered already");
1872         return;
1873     }
1874 
1875     /*
1876      * Note that this call will never start a real migration; it will
1877      * only re-setup the migration stream and poke existing migration
1878      * to continue using that newly established channel.
1879      */
1880     qemu_start_incoming_migration(uri, errp);
1881 }
1882 
1883 void qmp_migrate_pause(Error **errp)
1884 {
1885     MigrationState *ms = migrate_get_current();
1886     MigrationIncomingState *mis = migration_incoming_get_current();
1887     int ret;
1888 
1889     if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1890         /* Source side, during postcopy */
1891         qemu_mutex_lock(&ms->qemu_file_lock);
1892         ret = qemu_file_shutdown(ms->to_dst_file);
1893         qemu_mutex_unlock(&ms->qemu_file_lock);
1894         if (ret) {
1895             error_setg(errp, "Failed to pause source migration");
1896         }
1897         return;
1898     }
1899 
1900     if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1901         ret = qemu_file_shutdown(mis->from_src_file);
1902         if (ret) {
1903             error_setg(errp, "Failed to pause destination migration");
1904         }
1905         return;
1906     }
1907 
1908     error_setg(errp, "migrate-pause is currently only supported "
1909                "during postcopy-active state");
1910 }
1911 
1912 bool migration_is_blocked(Error **errp)
1913 {
1914     if (qemu_savevm_state_blocked(errp)) {
1915         return true;
1916     }
1917 
1918     if (migration_blockers) {
1919         error_propagate(errp, error_copy(migration_blockers->data));
1920         return true;
1921     }
1922 
1923     return false;
1924 }
1925 
1926 /* Returns true if continue to migrate, or false if error detected */
1927 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1928                             bool resume, Error **errp)
1929 {
1930     Error *local_err = NULL;
1931 
1932     if (resume) {
1933         if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1934             error_setg(errp, "Cannot resume if there is no "
1935                        "paused migration");
1936             return false;
1937         }
1938 
1939         /*
1940          * Postcopy recovery won't work well with release-ram
1941          * capability since release-ram will drop the page buffer as
1942          * long as the page is put into the send buffer.  So if there
1943          * is a network failure happened, any page buffers that have
1944          * not yet reached the destination VM but have already been
1945          * sent from the source VM will be lost forever.  Let's refuse
1946          * the client from resuming such a postcopy migration.
1947          * Luckily release-ram was designed to only be used when src
1948          * and destination VMs are on the same host, so it should be
1949          * fine.
1950          */
1951         if (migrate_release_ram()) {
1952             error_setg(errp, "Postcopy recovery cannot work "
1953                        "when release-ram capability is set");
1954             return false;
1955         }
1956 
1957         /* This is a resume, skip init status */
1958         return true;
1959     }
1960 
1961     if (migration_is_running(s->state)) {
1962         error_setg(errp, QERR_MIGRATION_ACTIVE);
1963         return false;
1964     }
1965 
1966     if (runstate_check(RUN_STATE_INMIGRATE)) {
1967         error_setg(errp, "Guest is waiting for an incoming migration");
1968         return false;
1969     }
1970 
1971     if (migration_is_blocked(errp)) {
1972         return false;
1973     }
1974 
1975     if (blk || blk_inc) {
1976         if (migrate_use_block() || migrate_use_block_incremental()) {
1977             error_setg(errp, "Command options are incompatible with "
1978                        "current migration capabilities");
1979             return false;
1980         }
1981         migrate_set_block_enabled(true, &local_err);
1982         if (local_err) {
1983             error_propagate(errp, local_err);
1984             return false;
1985         }
1986         s->must_remove_block_options = true;
1987     }
1988 
1989     if (blk_inc) {
1990         migrate_set_block_incremental(s, true);
1991     }
1992 
1993     migrate_init(s);
1994     /*
1995      * set ram_counters memory to zero for a
1996      * new migration
1997      */
1998     memset(&ram_counters, 0, sizeof(ram_counters));
1999 
2000     return true;
2001 }
2002 
2003 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2004                  bool has_inc, bool inc, bool has_detach, bool detach,
2005                  bool has_resume, bool resume, Error **errp)
2006 {
2007     Error *local_err = NULL;
2008     MigrationState *s = migrate_get_current();
2009     const char *p;
2010 
2011     if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2012                          has_resume && resume, errp)) {
2013         /* Error detected, put into errp */
2014         return;
2015     }
2016 
2017     if (strstart(uri, "tcp:", &p)) {
2018         tcp_start_outgoing_migration(s, p, &local_err);
2019 #ifdef CONFIG_RDMA
2020     } else if (strstart(uri, "rdma:", &p)) {
2021         rdma_start_outgoing_migration(s, p, &local_err);
2022 #endif
2023     } else if (strstart(uri, "exec:", &p)) {
2024         exec_start_outgoing_migration(s, p, &local_err);
2025     } else if (strstart(uri, "unix:", &p)) {
2026         unix_start_outgoing_migration(s, p, &local_err);
2027     } else if (strstart(uri, "fd:", &p)) {
2028         fd_start_outgoing_migration(s, p, &local_err);
2029     } else {
2030         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2031                    "a valid migration protocol");
2032         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2033                           MIGRATION_STATUS_FAILED);
2034         block_cleanup_parameters(s);
2035         return;
2036     }
2037 
2038     if (local_err) {
2039         migrate_fd_error(s, local_err);
2040         error_propagate(errp, local_err);
2041         return;
2042     }
2043 }
2044 
2045 void qmp_migrate_cancel(Error **errp)
2046 {
2047     migrate_fd_cancel(migrate_get_current());
2048 }
2049 
2050 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2051 {
2052     MigrationState *s = migrate_get_current();
2053     if (s->state != state) {
2054         error_setg(errp,  "Migration not in expected state: %s",
2055                    MigrationStatus_str(s->state));
2056         return;
2057     }
2058     qemu_sem_post(&s->pause_sem);
2059 }
2060 
2061 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
2062 {
2063     MigrateSetParameters p = {
2064         .has_xbzrle_cache_size = true,
2065         .xbzrle_cache_size = value,
2066     };
2067 
2068     qmp_migrate_set_parameters(&p, errp);
2069 }
2070 
2071 int64_t qmp_query_migrate_cache_size(Error **errp)
2072 {
2073     return migrate_xbzrle_cache_size();
2074 }
2075 
2076 void qmp_migrate_set_speed(int64_t value, Error **errp)
2077 {
2078     MigrateSetParameters p = {
2079         .has_max_bandwidth = true,
2080         .max_bandwidth = value,
2081     };
2082 
2083     qmp_migrate_set_parameters(&p, errp);
2084 }
2085 
2086 void qmp_migrate_set_downtime(double value, Error **errp)
2087 {
2088     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2089         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
2090                          "the range of 0 to %d seconds",
2091                          MAX_MIGRATE_DOWNTIME_SECONDS);
2092         return;
2093     }
2094 
2095     value *= 1000; /* Convert to milliseconds */
2096 
2097     MigrateSetParameters p = {
2098         .has_downtime_limit = true,
2099         .downtime_limit = (int64_t)value,
2100     };
2101 
2102     qmp_migrate_set_parameters(&p, errp);
2103 }
2104 
2105 bool migrate_release_ram(void)
2106 {
2107     MigrationState *s;
2108 
2109     s = migrate_get_current();
2110 
2111     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2112 }
2113 
2114 bool migrate_postcopy_ram(void)
2115 {
2116     MigrationState *s;
2117 
2118     s = migrate_get_current();
2119 
2120     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2121 }
2122 
2123 bool migrate_postcopy(void)
2124 {
2125     return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2126 }
2127 
2128 bool migrate_auto_converge(void)
2129 {
2130     MigrationState *s;
2131 
2132     s = migrate_get_current();
2133 
2134     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2135 }
2136 
2137 bool migrate_zero_blocks(void)
2138 {
2139     MigrationState *s;
2140 
2141     s = migrate_get_current();
2142 
2143     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2144 }
2145 
2146 bool migrate_postcopy_blocktime(void)
2147 {
2148     MigrationState *s;
2149 
2150     s = migrate_get_current();
2151 
2152     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2153 }
2154 
2155 bool migrate_use_compression(void)
2156 {
2157     MigrationState *s;
2158 
2159     s = migrate_get_current();
2160 
2161     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2162 }
2163 
2164 int migrate_compress_level(void)
2165 {
2166     MigrationState *s;
2167 
2168     s = migrate_get_current();
2169 
2170     return s->parameters.compress_level;
2171 }
2172 
2173 int migrate_compress_threads(void)
2174 {
2175     MigrationState *s;
2176 
2177     s = migrate_get_current();
2178 
2179     return s->parameters.compress_threads;
2180 }
2181 
2182 int migrate_compress_wait_thread(void)
2183 {
2184     MigrationState *s;
2185 
2186     s = migrate_get_current();
2187 
2188     return s->parameters.compress_wait_thread;
2189 }
2190 
2191 int migrate_decompress_threads(void)
2192 {
2193     MigrationState *s;
2194 
2195     s = migrate_get_current();
2196 
2197     return s->parameters.decompress_threads;
2198 }
2199 
2200 bool migrate_dirty_bitmaps(void)
2201 {
2202     MigrationState *s;
2203 
2204     s = migrate_get_current();
2205 
2206     return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2207 }
2208 
2209 bool migrate_ignore_shared(void)
2210 {
2211     MigrationState *s;
2212 
2213     s = migrate_get_current();
2214 
2215     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2216 }
2217 
2218 bool migrate_validate_uuid(void)
2219 {
2220     MigrationState *s;
2221 
2222     s = migrate_get_current();
2223 
2224     return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2225 }
2226 
2227 bool migrate_use_events(void)
2228 {
2229     MigrationState *s;
2230 
2231     s = migrate_get_current();
2232 
2233     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2234 }
2235 
2236 bool migrate_use_multifd(void)
2237 {
2238     MigrationState *s;
2239 
2240     s = migrate_get_current();
2241 
2242     return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2243 }
2244 
2245 bool migrate_pause_before_switchover(void)
2246 {
2247     MigrationState *s;
2248 
2249     s = migrate_get_current();
2250 
2251     return s->enabled_capabilities[
2252         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2253 }
2254 
2255 int migrate_multifd_channels(void)
2256 {
2257     MigrationState *s;
2258 
2259     s = migrate_get_current();
2260 
2261     return s->parameters.multifd_channels;
2262 }
2263 
2264 MultiFDCompression migrate_multifd_compression(void)
2265 {
2266     MigrationState *s;
2267 
2268     s = migrate_get_current();
2269 
2270     return s->parameters.multifd_compression;
2271 }
2272 
2273 int migrate_multifd_zlib_level(void)
2274 {
2275     MigrationState *s;
2276 
2277     s = migrate_get_current();
2278 
2279     return s->parameters.multifd_zlib_level;
2280 }
2281 
2282 int migrate_multifd_zstd_level(void)
2283 {
2284     MigrationState *s;
2285 
2286     s = migrate_get_current();
2287 
2288     return s->parameters.multifd_zstd_level;
2289 }
2290 
2291 int migrate_use_xbzrle(void)
2292 {
2293     MigrationState *s;
2294 
2295     s = migrate_get_current();
2296 
2297     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2298 }
2299 
2300 int64_t migrate_xbzrle_cache_size(void)
2301 {
2302     MigrationState *s;
2303 
2304     s = migrate_get_current();
2305 
2306     return s->parameters.xbzrle_cache_size;
2307 }
2308 
2309 static int64_t migrate_max_postcopy_bandwidth(void)
2310 {
2311     MigrationState *s;
2312 
2313     s = migrate_get_current();
2314 
2315     return s->parameters.max_postcopy_bandwidth;
2316 }
2317 
2318 bool migrate_use_block(void)
2319 {
2320     MigrationState *s;
2321 
2322     s = migrate_get_current();
2323 
2324     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2325 }
2326 
2327 bool migrate_use_return_path(void)
2328 {
2329     MigrationState *s;
2330 
2331     s = migrate_get_current();
2332 
2333     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2334 }
2335 
2336 bool migrate_use_block_incremental(void)
2337 {
2338     MigrationState *s;
2339 
2340     s = migrate_get_current();
2341 
2342     return s->parameters.block_incremental;
2343 }
2344 
2345 /* migration thread support */
2346 /*
2347  * Something bad happened to the RP stream, mark an error
2348  * The caller shall print or trace something to indicate why
2349  */
2350 static void mark_source_rp_bad(MigrationState *s)
2351 {
2352     s->rp_state.error = true;
2353 }
2354 
2355 static struct rp_cmd_args {
2356     ssize_t     len; /* -1 = variable */
2357     const char *name;
2358 } rp_cmd_args[] = {
2359     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
2360     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
2361     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
2362     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
2363     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
2364     [MIG_RP_MSG_RECV_BITMAP]    = { .len = -1, .name = "RECV_BITMAP" },
2365     [MIG_RP_MSG_RESUME_ACK]     = { .len =  4, .name = "RESUME_ACK" },
2366     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
2367 };
2368 
2369 /*
2370  * Process a request for pages received on the return path,
2371  * We're allowed to send more than requested (e.g. to round to our page size)
2372  * and we don't need to send pages that have already been sent.
2373  */
2374 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2375                                        ram_addr_t start, size_t len)
2376 {
2377     long our_host_ps = qemu_real_host_page_size;
2378 
2379     trace_migrate_handle_rp_req_pages(rbname, start, len);
2380 
2381     /*
2382      * Since we currently insist on matching page sizes, just sanity check
2383      * we're being asked for whole host pages.
2384      */
2385     if (start & (our_host_ps-1) ||
2386        (len & (our_host_ps-1))) {
2387         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2388                      " len: %zd", __func__, start, len);
2389         mark_source_rp_bad(ms);
2390         return;
2391     }
2392 
2393     if (ram_save_queue_pages(rbname, start, len)) {
2394         mark_source_rp_bad(ms);
2395     }
2396 }
2397 
2398 /* Return true to retry, false to quit */
2399 static bool postcopy_pause_return_path_thread(MigrationState *s)
2400 {
2401     trace_postcopy_pause_return_path();
2402 
2403     qemu_sem_wait(&s->postcopy_pause_rp_sem);
2404 
2405     trace_postcopy_pause_return_path_continued();
2406 
2407     return true;
2408 }
2409 
2410 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2411 {
2412     RAMBlock *block = qemu_ram_block_by_name(block_name);
2413 
2414     if (!block) {
2415         error_report("%s: invalid block name '%s'", __func__, block_name);
2416         return -EINVAL;
2417     }
2418 
2419     /* Fetch the received bitmap and refresh the dirty bitmap */
2420     return ram_dirty_bitmap_reload(s, block);
2421 }
2422 
2423 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2424 {
2425     trace_source_return_path_thread_resume_ack(value);
2426 
2427     if (value != MIGRATION_RESUME_ACK_VALUE) {
2428         error_report("%s: illegal resume_ack value %"PRIu32,
2429                      __func__, value);
2430         return -1;
2431     }
2432 
2433     /* Now both sides are active. */
2434     migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2435                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2436 
2437     /* Notify send thread that time to continue send pages */
2438     qemu_sem_post(&s->rp_state.rp_sem);
2439 
2440     return 0;
2441 }
2442 
2443 /*
2444  * Handles messages sent on the return path towards the source VM
2445  *
2446  */
2447 static void *source_return_path_thread(void *opaque)
2448 {
2449     MigrationState *ms = opaque;
2450     QEMUFile *rp = ms->rp_state.from_dst_file;
2451     uint16_t header_len, header_type;
2452     uint8_t buf[512];
2453     uint32_t tmp32, sibling_error;
2454     ram_addr_t start = 0; /* =0 to silence warning */
2455     size_t  len = 0, expected_len;
2456     int res;
2457 
2458     trace_source_return_path_thread_entry();
2459     rcu_register_thread();
2460 
2461 retry:
2462     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2463            migration_is_setup_or_active(ms->state)) {
2464         trace_source_return_path_thread_loop_top();
2465         header_type = qemu_get_be16(rp);
2466         header_len = qemu_get_be16(rp);
2467 
2468         if (qemu_file_get_error(rp)) {
2469             mark_source_rp_bad(ms);
2470             goto out;
2471         }
2472 
2473         if (header_type >= MIG_RP_MSG_MAX ||
2474             header_type == MIG_RP_MSG_INVALID) {
2475             error_report("RP: Received invalid message 0x%04x length 0x%04x",
2476                     header_type, header_len);
2477             mark_source_rp_bad(ms);
2478             goto out;
2479         }
2480 
2481         if ((rp_cmd_args[header_type].len != -1 &&
2482             header_len != rp_cmd_args[header_type].len) ||
2483             header_len > sizeof(buf)) {
2484             error_report("RP: Received '%s' message (0x%04x) with"
2485                     "incorrect length %d expecting %zu",
2486                     rp_cmd_args[header_type].name, header_type, header_len,
2487                     (size_t)rp_cmd_args[header_type].len);
2488             mark_source_rp_bad(ms);
2489             goto out;
2490         }
2491 
2492         /* We know we've got a valid header by this point */
2493         res = qemu_get_buffer(rp, buf, header_len);
2494         if (res != header_len) {
2495             error_report("RP: Failed reading data for message 0x%04x"
2496                          " read %d expected %d",
2497                          header_type, res, header_len);
2498             mark_source_rp_bad(ms);
2499             goto out;
2500         }
2501 
2502         /* OK, we have the message and the data */
2503         switch (header_type) {
2504         case MIG_RP_MSG_SHUT:
2505             sibling_error = ldl_be_p(buf);
2506             trace_source_return_path_thread_shut(sibling_error);
2507             if (sibling_error) {
2508                 error_report("RP: Sibling indicated error %d", sibling_error);
2509                 mark_source_rp_bad(ms);
2510             }
2511             /*
2512              * We'll let the main thread deal with closing the RP
2513              * we could do a shutdown(2) on it, but we're the only user
2514              * anyway, so there's nothing gained.
2515              */
2516             goto out;
2517 
2518         case MIG_RP_MSG_PONG:
2519             tmp32 = ldl_be_p(buf);
2520             trace_source_return_path_thread_pong(tmp32);
2521             break;
2522 
2523         case MIG_RP_MSG_REQ_PAGES:
2524             start = ldq_be_p(buf);
2525             len = ldl_be_p(buf + 8);
2526             migrate_handle_rp_req_pages(ms, NULL, start, len);
2527             break;
2528 
2529         case MIG_RP_MSG_REQ_PAGES_ID:
2530             expected_len = 12 + 1; /* header + termination */
2531 
2532             if (header_len >= expected_len) {
2533                 start = ldq_be_p(buf);
2534                 len = ldl_be_p(buf + 8);
2535                 /* Now we expect an idstr */
2536                 tmp32 = buf[12]; /* Length of the following idstr */
2537                 buf[13 + tmp32] = '\0';
2538                 expected_len += tmp32;
2539             }
2540             if (header_len != expected_len) {
2541                 error_report("RP: Req_Page_id with length %d expecting %zd",
2542                         header_len, expected_len);
2543                 mark_source_rp_bad(ms);
2544                 goto out;
2545             }
2546             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2547             break;
2548 
2549         case MIG_RP_MSG_RECV_BITMAP:
2550             if (header_len < 1) {
2551                 error_report("%s: missing block name", __func__);
2552                 mark_source_rp_bad(ms);
2553                 goto out;
2554             }
2555             /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2556             buf[buf[0] + 1] = '\0';
2557             if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2558                 mark_source_rp_bad(ms);
2559                 goto out;
2560             }
2561             break;
2562 
2563         case MIG_RP_MSG_RESUME_ACK:
2564             tmp32 = ldl_be_p(buf);
2565             if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2566                 mark_source_rp_bad(ms);
2567                 goto out;
2568             }
2569             break;
2570 
2571         default:
2572             break;
2573         }
2574     }
2575 
2576 out:
2577     res = qemu_file_get_error(rp);
2578     if (res) {
2579         if (res == -EIO && migration_in_postcopy()) {
2580             /*
2581              * Maybe there is something we can do: it looks like a
2582              * network down issue, and we pause for a recovery.
2583              */
2584             if (postcopy_pause_return_path_thread(ms)) {
2585                 /* Reload rp, reset the rest */
2586                 if (rp != ms->rp_state.from_dst_file) {
2587                     qemu_fclose(rp);
2588                     rp = ms->rp_state.from_dst_file;
2589                 }
2590                 ms->rp_state.error = false;
2591                 goto retry;
2592             }
2593         }
2594 
2595         trace_source_return_path_thread_bad_end();
2596         mark_source_rp_bad(ms);
2597     }
2598 
2599     trace_source_return_path_thread_end();
2600     ms->rp_state.from_dst_file = NULL;
2601     qemu_fclose(rp);
2602     rcu_unregister_thread();
2603     return NULL;
2604 }
2605 
2606 static int open_return_path_on_source(MigrationState *ms,
2607                                       bool create_thread)
2608 {
2609 
2610     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2611     if (!ms->rp_state.from_dst_file) {
2612         return -1;
2613     }
2614 
2615     trace_open_return_path_on_source();
2616 
2617     if (!create_thread) {
2618         /* We're done */
2619         return 0;
2620     }
2621 
2622     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2623                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2624 
2625     trace_open_return_path_on_source_continue();
2626 
2627     return 0;
2628 }
2629 
2630 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2631 static int await_return_path_close_on_source(MigrationState *ms)
2632 {
2633     /*
2634      * If this is a normal exit then the destination will send a SHUT and the
2635      * rp_thread will exit, however if there's an error we need to cause
2636      * it to exit.
2637      */
2638     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2639         /*
2640          * shutdown(2), if we have it, will cause it to unblock if it's stuck
2641          * waiting for the destination.
2642          */
2643         qemu_file_shutdown(ms->rp_state.from_dst_file);
2644         mark_source_rp_bad(ms);
2645     }
2646     trace_await_return_path_close_on_source_joining();
2647     qemu_thread_join(&ms->rp_state.rp_thread);
2648     trace_await_return_path_close_on_source_close();
2649     return ms->rp_state.error;
2650 }
2651 
2652 /*
2653  * Switch from normal iteration to postcopy
2654  * Returns non-0 on error
2655  */
2656 static int postcopy_start(MigrationState *ms)
2657 {
2658     int ret;
2659     QIOChannelBuffer *bioc;
2660     QEMUFile *fb;
2661     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2662     int64_t bandwidth = migrate_max_postcopy_bandwidth();
2663     bool restart_block = false;
2664     int cur_state = MIGRATION_STATUS_ACTIVE;
2665     if (!migrate_pause_before_switchover()) {
2666         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2667                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
2668     }
2669 
2670     trace_postcopy_start();
2671     qemu_mutex_lock_iothread();
2672     trace_postcopy_start_set_run();
2673 
2674     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2675     global_state_store();
2676     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2677     if (ret < 0) {
2678         goto fail;
2679     }
2680 
2681     ret = migration_maybe_pause(ms, &cur_state,
2682                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2683     if (ret < 0) {
2684         goto fail;
2685     }
2686 
2687     ret = bdrv_inactivate_all();
2688     if (ret < 0) {
2689         goto fail;
2690     }
2691     restart_block = true;
2692 
2693     /*
2694      * Cause any non-postcopiable, but iterative devices to
2695      * send out their final data.
2696      */
2697     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2698 
2699     /*
2700      * in Finish migrate and with the io-lock held everything should
2701      * be quiet, but we've potentially still got dirty pages and we
2702      * need to tell the destination to throw any pages it's already received
2703      * that are dirty
2704      */
2705     if (migrate_postcopy_ram()) {
2706         if (ram_postcopy_send_discard_bitmap(ms)) {
2707             error_report("postcopy send discard bitmap failed");
2708             goto fail;
2709         }
2710     }
2711 
2712     /*
2713      * send rest of state - note things that are doing postcopy
2714      * will notice we're in POSTCOPY_ACTIVE and not actually
2715      * wrap their state up here
2716      */
2717     /* 0 max-postcopy-bandwidth means unlimited */
2718     if (!bandwidth) {
2719         qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2720     } else {
2721         qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2722     }
2723     if (migrate_postcopy_ram()) {
2724         /* Ping just for debugging, helps line traces up */
2725         qemu_savevm_send_ping(ms->to_dst_file, 2);
2726     }
2727 
2728     /*
2729      * While loading the device state we may trigger page transfer
2730      * requests and the fd must be free to process those, and thus
2731      * the destination must read the whole device state off the fd before
2732      * it starts processing it.  Unfortunately the ad-hoc migration format
2733      * doesn't allow the destination to know the size to read without fully
2734      * parsing it through each devices load-state code (especially the open
2735      * coded devices that use get/put).
2736      * So we wrap the device state up in a package with a length at the start;
2737      * to do this we use a qemu_buf to hold the whole of the device state.
2738      */
2739     bioc = qio_channel_buffer_new(4096);
2740     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2741     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2742     object_unref(OBJECT(bioc));
2743 
2744     /*
2745      * Make sure the receiver can get incoming pages before we send the rest
2746      * of the state
2747      */
2748     qemu_savevm_send_postcopy_listen(fb);
2749 
2750     qemu_savevm_state_complete_precopy(fb, false, false);
2751     if (migrate_postcopy_ram()) {
2752         qemu_savevm_send_ping(fb, 3);
2753     }
2754 
2755     qemu_savevm_send_postcopy_run(fb);
2756 
2757     /* <><> end of stuff going into the package */
2758 
2759     /* Last point of recovery; as soon as we send the package the destination
2760      * can open devices and potentially start running.
2761      * Lets just check again we've not got any errors.
2762      */
2763     ret = qemu_file_get_error(ms->to_dst_file);
2764     if (ret) {
2765         error_report("postcopy_start: Migration stream errored (pre package)");
2766         goto fail_closefb;
2767     }
2768 
2769     restart_block = false;
2770 
2771     /* Now send that blob */
2772     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2773         goto fail_closefb;
2774     }
2775     qemu_fclose(fb);
2776 
2777     /* Send a notify to give a chance for anything that needs to happen
2778      * at the transition to postcopy and after the device state; in particular
2779      * spice needs to trigger a transition now
2780      */
2781     ms->postcopy_after_devices = true;
2782     notifier_list_notify(&migration_state_notifiers, ms);
2783 
2784     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2785 
2786     qemu_mutex_unlock_iothread();
2787 
2788     if (migrate_postcopy_ram()) {
2789         /*
2790          * Although this ping is just for debug, it could potentially be
2791          * used for getting a better measurement of downtime at the source.
2792          */
2793         qemu_savevm_send_ping(ms->to_dst_file, 4);
2794     }
2795 
2796     if (migrate_release_ram()) {
2797         ram_postcopy_migrated_memory_release(ms);
2798     }
2799 
2800     ret = qemu_file_get_error(ms->to_dst_file);
2801     if (ret) {
2802         error_report("postcopy_start: Migration stream errored");
2803         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2804                               MIGRATION_STATUS_FAILED);
2805     }
2806 
2807     return ret;
2808 
2809 fail_closefb:
2810     qemu_fclose(fb);
2811 fail:
2812     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2813                           MIGRATION_STATUS_FAILED);
2814     if (restart_block) {
2815         /* A failure happened early enough that we know the destination hasn't
2816          * accessed block devices, so we're safe to recover.
2817          */
2818         Error *local_err = NULL;
2819 
2820         bdrv_invalidate_cache_all(&local_err);
2821         if (local_err) {
2822             error_report_err(local_err);
2823         }
2824     }
2825     qemu_mutex_unlock_iothread();
2826     return -1;
2827 }
2828 
2829 /**
2830  * migration_maybe_pause: Pause if required to by
2831  * migrate_pause_before_switchover called with the iothread locked
2832  * Returns: 0 on success
2833  */
2834 static int migration_maybe_pause(MigrationState *s,
2835                                  int *current_active_state,
2836                                  int new_state)
2837 {
2838     if (!migrate_pause_before_switchover()) {
2839         return 0;
2840     }
2841 
2842     /* Since leaving this state is not atomic with posting the semaphore
2843      * it's possible that someone could have issued multiple migrate_continue
2844      * and the semaphore is incorrectly positive at this point;
2845      * the docs say it's undefined to reinit a semaphore that's already
2846      * init'd, so use timedwait to eat up any existing posts.
2847      */
2848     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2849         /* This block intentionally left blank */
2850     }
2851 
2852     /*
2853      * If the migration is cancelled when it is in the completion phase,
2854      * the migration state is set to MIGRATION_STATUS_CANCELLING.
2855      * So we don't need to wait a semaphore, otherwise we would always
2856      * wait for the 'pause_sem' semaphore.
2857      */
2858     if (s->state != MIGRATION_STATUS_CANCELLING) {
2859         qemu_mutex_unlock_iothread();
2860         migrate_set_state(&s->state, *current_active_state,
2861                           MIGRATION_STATUS_PRE_SWITCHOVER);
2862         qemu_sem_wait(&s->pause_sem);
2863         migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2864                           new_state);
2865         *current_active_state = new_state;
2866         qemu_mutex_lock_iothread();
2867     }
2868 
2869     return s->state == new_state ? 0 : -EINVAL;
2870 }
2871 
2872 /**
2873  * migration_completion: Used by migration_thread when there's not much left.
2874  *   The caller 'breaks' the loop when this returns.
2875  *
2876  * @s: Current migration state
2877  */
2878 static void migration_completion(MigrationState *s)
2879 {
2880     int ret;
2881     int current_active_state = s->state;
2882 
2883     if (s->state == MIGRATION_STATUS_ACTIVE) {
2884         qemu_mutex_lock_iothread();
2885         s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2886         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2887         s->vm_was_running = runstate_is_running();
2888         ret = global_state_store();
2889 
2890         if (!ret) {
2891             bool inactivate = !migrate_colo_enabled();
2892             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2893             if (ret >= 0) {
2894                 ret = migration_maybe_pause(s, &current_active_state,
2895                                             MIGRATION_STATUS_DEVICE);
2896             }
2897             if (ret >= 0) {
2898                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2899                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2900                                                          inactivate);
2901             }
2902             if (inactivate && ret >= 0) {
2903                 s->block_inactive = true;
2904             }
2905         }
2906         qemu_mutex_unlock_iothread();
2907 
2908         if (ret < 0) {
2909             goto fail;
2910         }
2911     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2912         trace_migration_completion_postcopy_end();
2913 
2914         qemu_savevm_state_complete_postcopy(s->to_dst_file);
2915         trace_migration_completion_postcopy_end_after_complete();
2916     }
2917 
2918     /*
2919      * If rp was opened we must clean up the thread before
2920      * cleaning everything else up (since if there are no failures
2921      * it will wait for the destination to send it's status in
2922      * a SHUT command).
2923      */
2924     if (s->rp_state.from_dst_file) {
2925         int rp_error;
2926         trace_migration_return_path_end_before();
2927         rp_error = await_return_path_close_on_source(s);
2928         trace_migration_return_path_end_after(rp_error);
2929         if (rp_error) {
2930             goto fail_invalidate;
2931         }
2932     }
2933 
2934     if (qemu_file_get_error(s->to_dst_file)) {
2935         trace_migration_completion_file_err();
2936         goto fail_invalidate;
2937     }
2938 
2939     if (!migrate_colo_enabled()) {
2940         migrate_set_state(&s->state, current_active_state,
2941                           MIGRATION_STATUS_COMPLETED);
2942     }
2943 
2944     return;
2945 
2946 fail_invalidate:
2947     /* If not doing postcopy, vm_start() will be called: let's regain
2948      * control on images.
2949      */
2950     if (s->state == MIGRATION_STATUS_ACTIVE ||
2951         s->state == MIGRATION_STATUS_DEVICE) {
2952         Error *local_err = NULL;
2953 
2954         qemu_mutex_lock_iothread();
2955         bdrv_invalidate_cache_all(&local_err);
2956         if (local_err) {
2957             error_report_err(local_err);
2958         } else {
2959             s->block_inactive = false;
2960         }
2961         qemu_mutex_unlock_iothread();
2962     }
2963 
2964 fail:
2965     migrate_set_state(&s->state, current_active_state,
2966                       MIGRATION_STATUS_FAILED);
2967 }
2968 
2969 bool migrate_colo_enabled(void)
2970 {
2971     MigrationState *s = migrate_get_current();
2972     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2973 }
2974 
2975 typedef enum MigThrError {
2976     /* No error detected */
2977     MIG_THR_ERR_NONE = 0,
2978     /* Detected error, but resumed successfully */
2979     MIG_THR_ERR_RECOVERED = 1,
2980     /* Detected fatal error, need to exit */
2981     MIG_THR_ERR_FATAL = 2,
2982 } MigThrError;
2983 
2984 static int postcopy_resume_handshake(MigrationState *s)
2985 {
2986     qemu_savevm_send_postcopy_resume(s->to_dst_file);
2987 
2988     while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2989         qemu_sem_wait(&s->rp_state.rp_sem);
2990     }
2991 
2992     if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2993         return 0;
2994     }
2995 
2996     return -1;
2997 }
2998 
2999 /* Return zero if success, or <0 for error */
3000 static int postcopy_do_resume(MigrationState *s)
3001 {
3002     int ret;
3003 
3004     /*
3005      * Call all the resume_prepare() hooks, so that modules can be
3006      * ready for the migration resume.
3007      */
3008     ret = qemu_savevm_state_resume_prepare(s);
3009     if (ret) {
3010         error_report("%s: resume_prepare() failure detected: %d",
3011                      __func__, ret);
3012         return ret;
3013     }
3014 
3015     /*
3016      * Last handshake with destination on the resume (destination will
3017      * switch to postcopy-active afterwards)
3018      */
3019     ret = postcopy_resume_handshake(s);
3020     if (ret) {
3021         error_report("%s: handshake failed: %d", __func__, ret);
3022         return ret;
3023     }
3024 
3025     return 0;
3026 }
3027 
3028 /*
3029  * We don't return until we are in a safe state to continue current
3030  * postcopy migration.  Returns MIG_THR_ERR_RECOVERED if recovered, or
3031  * MIG_THR_ERR_FATAL if unrecovery failure happened.
3032  */
3033 static MigThrError postcopy_pause(MigrationState *s)
3034 {
3035     assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3036 
3037     while (true) {
3038         QEMUFile *file;
3039 
3040         migrate_set_state(&s->state, s->state,
3041                           MIGRATION_STATUS_POSTCOPY_PAUSED);
3042 
3043         /* Current channel is possibly broken. Release it. */
3044         assert(s->to_dst_file);
3045         qemu_mutex_lock(&s->qemu_file_lock);
3046         file = s->to_dst_file;
3047         s->to_dst_file = NULL;
3048         qemu_mutex_unlock(&s->qemu_file_lock);
3049 
3050         qemu_file_shutdown(file);
3051         qemu_fclose(file);
3052 
3053         error_report("Detected IO failure for postcopy. "
3054                      "Migration paused.");
3055 
3056         /*
3057          * We wait until things fixed up. Then someone will setup the
3058          * status back for us.
3059          */
3060         while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3061             qemu_sem_wait(&s->postcopy_pause_sem);
3062         }
3063 
3064         if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3065             /* Woken up by a recover procedure. Give it a shot */
3066 
3067             /*
3068              * Firstly, let's wake up the return path now, with a new
3069              * return path channel.
3070              */
3071             qemu_sem_post(&s->postcopy_pause_rp_sem);
3072 
3073             /* Do the resume logic */
3074             if (postcopy_do_resume(s) == 0) {
3075                 /* Let's continue! */
3076                 trace_postcopy_pause_continued();
3077                 return MIG_THR_ERR_RECOVERED;
3078             } else {
3079                 /*
3080                  * Something wrong happened during the recovery, let's
3081                  * pause again. Pause is always better than throwing
3082                  * data away.
3083                  */
3084                 continue;
3085             }
3086         } else {
3087             /* This is not right... Time to quit. */
3088             return MIG_THR_ERR_FATAL;
3089         }
3090     }
3091 }
3092 
3093 static MigThrError migration_detect_error(MigrationState *s)
3094 {
3095     int ret;
3096     int state = s->state;
3097     Error *local_error = NULL;
3098 
3099     if (state == MIGRATION_STATUS_CANCELLING ||
3100         state == MIGRATION_STATUS_CANCELLED) {
3101         /* End the migration, but don't set the state to failed */
3102         return MIG_THR_ERR_FATAL;
3103     }
3104 
3105     /* Try to detect any file errors */
3106     ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3107     if (!ret) {
3108         /* Everything is fine */
3109         assert(!local_error);
3110         return MIG_THR_ERR_NONE;
3111     }
3112 
3113     if (local_error) {
3114         migrate_set_error(s, local_error);
3115         error_free(local_error);
3116     }
3117 
3118     if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3119         /*
3120          * For postcopy, we allow the network to be down for a
3121          * while. After that, it can be continued by a
3122          * recovery phase.
3123          */
3124         return postcopy_pause(s);
3125     } else {
3126         /*
3127          * For precopy (or postcopy with error outside IO), we fail
3128          * with no time.
3129          */
3130         migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3131         trace_migration_thread_file_err();
3132 
3133         /* Time to stop the migration, now. */
3134         return MIG_THR_ERR_FATAL;
3135     }
3136 }
3137 
3138 /* How many bytes have we transferred since the beginning of the migration */
3139 static uint64_t migration_total_bytes(MigrationState *s)
3140 {
3141     return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3142 }
3143 
3144 static void migration_calculate_complete(MigrationState *s)
3145 {
3146     uint64_t bytes = migration_total_bytes(s);
3147     int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3148     int64_t transfer_time;
3149 
3150     s->total_time = end_time - s->start_time;
3151     if (!s->downtime) {
3152         /*
3153          * It's still not set, so we are precopy migration.  For
3154          * postcopy, downtime is calculated during postcopy_start().
3155          */
3156         s->downtime = end_time - s->downtime_start;
3157     }
3158 
3159     transfer_time = s->total_time - s->setup_time;
3160     if (transfer_time) {
3161         s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3162     }
3163 }
3164 
3165 static void update_iteration_initial_status(MigrationState *s)
3166 {
3167     /*
3168      * Update these three fields at the same time to avoid mismatch info lead
3169      * wrong speed calculation.
3170      */
3171     s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3172     s->iteration_initial_bytes = migration_total_bytes(s);
3173     s->iteration_initial_pages = ram_get_total_transferred_pages();
3174 }
3175 
3176 static void migration_update_counters(MigrationState *s,
3177                                       int64_t current_time)
3178 {
3179     uint64_t transferred, transferred_pages, time_spent;
3180     uint64_t current_bytes; /* bytes transferred since the beginning */
3181     double bandwidth;
3182 
3183     if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3184         return;
3185     }
3186 
3187     current_bytes = migration_total_bytes(s);
3188     transferred = current_bytes - s->iteration_initial_bytes;
3189     time_spent = current_time - s->iteration_start_time;
3190     bandwidth = (double)transferred / time_spent;
3191     s->threshold_size = bandwidth * s->parameters.downtime_limit;
3192 
3193     s->mbps = (((double) transferred * 8.0) /
3194                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3195 
3196     transferred_pages = ram_get_total_transferred_pages() -
3197                             s->iteration_initial_pages;
3198     s->pages_per_second = (double) transferred_pages /
3199                              (((double) time_spent / 1000.0));
3200 
3201     /*
3202      * if we haven't sent anything, we don't want to
3203      * recalculate. 10000 is a small enough number for our purposes
3204      */
3205     if (ram_counters.dirty_pages_rate && transferred > 10000) {
3206         s->expected_downtime = ram_counters.remaining / bandwidth;
3207     }
3208 
3209     qemu_file_reset_rate_limit(s->to_dst_file);
3210 
3211     update_iteration_initial_status(s);
3212 
3213     trace_migrate_transferred(transferred, time_spent,
3214                               bandwidth, s->threshold_size);
3215 }
3216 
3217 /* Migration thread iteration status */
3218 typedef enum {
3219     MIG_ITERATE_RESUME,         /* Resume current iteration */
3220     MIG_ITERATE_SKIP,           /* Skip current iteration */
3221     MIG_ITERATE_BREAK,          /* Break the loop */
3222 } MigIterateState;
3223 
3224 /*
3225  * Return true if continue to the next iteration directly, false
3226  * otherwise.
3227  */
3228 static MigIterateState migration_iteration_run(MigrationState *s)
3229 {
3230     uint64_t pending_size, pend_pre, pend_compat, pend_post;
3231     bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3232 
3233     qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3234                               &pend_compat, &pend_post);
3235     pending_size = pend_pre + pend_compat + pend_post;
3236 
3237     trace_migrate_pending(pending_size, s->threshold_size,
3238                           pend_pre, pend_compat, pend_post);
3239 
3240     if (pending_size && pending_size >= s->threshold_size) {
3241         /* Still a significant amount to transfer */
3242         if (!in_postcopy && pend_pre <= s->threshold_size &&
3243             atomic_read(&s->start_postcopy)) {
3244             if (postcopy_start(s)) {
3245                 error_report("%s: postcopy failed to start", __func__);
3246             }
3247             return MIG_ITERATE_SKIP;
3248         }
3249         /* Just another iteration step */
3250         qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3251     } else {
3252         trace_migration_thread_low_pending(pending_size);
3253         migration_completion(s);
3254         return MIG_ITERATE_BREAK;
3255     }
3256 
3257     return MIG_ITERATE_RESUME;
3258 }
3259 
3260 static void migration_iteration_finish(MigrationState *s)
3261 {
3262     /* If we enabled cpu throttling for auto-converge, turn it off. */
3263     cpu_throttle_stop();
3264 
3265     qemu_mutex_lock_iothread();
3266     switch (s->state) {
3267     case MIGRATION_STATUS_COMPLETED:
3268         migration_calculate_complete(s);
3269         runstate_set(RUN_STATE_POSTMIGRATE);
3270         break;
3271 
3272     case MIGRATION_STATUS_ACTIVE:
3273         /*
3274          * We should really assert here, but since it's during
3275          * migration, let's try to reduce the usage of assertions.
3276          */
3277         if (!migrate_colo_enabled()) {
3278             error_report("%s: critical error: calling COLO code without "
3279                          "COLO enabled", __func__);
3280         }
3281         migrate_start_colo_process(s);
3282         /*
3283          * Fixme: we will run VM in COLO no matter its old running state.
3284          * After exited COLO, we will keep running.
3285          */
3286         s->vm_was_running = true;
3287         /* Fallthrough */
3288     case MIGRATION_STATUS_FAILED:
3289     case MIGRATION_STATUS_CANCELLED:
3290     case MIGRATION_STATUS_CANCELLING:
3291         if (s->vm_was_running) {
3292             vm_start();
3293         } else {
3294             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3295                 runstate_set(RUN_STATE_POSTMIGRATE);
3296             }
3297         }
3298         break;
3299 
3300     default:
3301         /* Should not reach here, but if so, forgive the VM. */
3302         error_report("%s: Unknown ending state %d", __func__, s->state);
3303         break;
3304     }
3305     migrate_fd_cleanup_schedule(s);
3306     qemu_mutex_unlock_iothread();
3307 }
3308 
3309 void migration_make_urgent_request(void)
3310 {
3311     qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3312 }
3313 
3314 void migration_consume_urgent_request(void)
3315 {
3316     qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3317 }
3318 
3319 /* Returns true if the rate limiting was broken by an urgent request */
3320 bool migration_rate_limit(void)
3321 {
3322     int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3323     MigrationState *s = migrate_get_current();
3324 
3325     bool urgent = false;
3326     migration_update_counters(s, now);
3327     if (qemu_file_rate_limit(s->to_dst_file)) {
3328         /*
3329          * Wait for a delay to do rate limiting OR
3330          * something urgent to post the semaphore.
3331          */
3332         int ms = s->iteration_start_time + BUFFER_DELAY - now;
3333         trace_migration_rate_limit_pre(ms);
3334         if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3335             /*
3336              * We were woken by one or more urgent things but
3337              * the timedwait will have consumed one of them.
3338              * The service routine for the urgent wake will dec
3339              * the semaphore itself for each item it consumes,
3340              * so add this one we just eat back.
3341              */
3342             qemu_sem_post(&s->rate_limit_sem);
3343             urgent = true;
3344         }
3345         trace_migration_rate_limit_post(urgent);
3346     }
3347     return urgent;
3348 }
3349 
3350 /*
3351  * Master migration thread on the source VM.
3352  * It drives the migration and pumps the data down the outgoing channel.
3353  */
3354 static void *migration_thread(void *opaque)
3355 {
3356     MigrationState *s = opaque;
3357     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3358     MigThrError thr_error;
3359     bool urgent = false;
3360 
3361     rcu_register_thread();
3362 
3363     object_ref(OBJECT(s));
3364     update_iteration_initial_status(s);
3365 
3366     qemu_savevm_state_header(s->to_dst_file);
3367 
3368     /*
3369      * If we opened the return path, we need to make sure dst has it
3370      * opened as well.
3371      */
3372     if (s->rp_state.from_dst_file) {
3373         /* Now tell the dest that it should open its end so it can reply */
3374         qemu_savevm_send_open_return_path(s->to_dst_file);
3375 
3376         /* And do a ping that will make stuff easier to debug */
3377         qemu_savevm_send_ping(s->to_dst_file, 1);
3378     }
3379 
3380     if (migrate_postcopy()) {
3381         /*
3382          * Tell the destination that we *might* want to do postcopy later;
3383          * if the other end can't do postcopy it should fail now, nice and
3384          * early.
3385          */
3386         qemu_savevm_send_postcopy_advise(s->to_dst_file);
3387     }
3388 
3389     if (migrate_colo_enabled()) {
3390         /* Notify migration destination that we enable COLO */
3391         qemu_savevm_send_colo_enable(s->to_dst_file);
3392     }
3393 
3394     qemu_savevm_state_setup(s->to_dst_file);
3395 
3396     if (qemu_savevm_state_guest_unplug_pending()) {
3397         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3398                           MIGRATION_STATUS_WAIT_UNPLUG);
3399 
3400         while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3401                qemu_savevm_state_guest_unplug_pending()) {
3402             qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3403         }
3404 
3405         migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3406                 MIGRATION_STATUS_ACTIVE);
3407     }
3408 
3409     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3410     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3411                       MIGRATION_STATUS_ACTIVE);
3412 
3413     trace_migration_thread_setup_complete();
3414 
3415     while (migration_is_active(s)) {
3416         if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3417             MigIterateState iter_state = migration_iteration_run(s);
3418             if (iter_state == MIG_ITERATE_SKIP) {
3419                 continue;
3420             } else if (iter_state == MIG_ITERATE_BREAK) {
3421                 break;
3422             }
3423         }
3424 
3425         /*
3426          * Try to detect any kind of failures, and see whether we
3427          * should stop the migration now.
3428          */
3429         thr_error = migration_detect_error(s);
3430         if (thr_error == MIG_THR_ERR_FATAL) {
3431             /* Stop migration */
3432             break;
3433         } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3434             /*
3435              * Just recovered from a e.g. network failure, reset all
3436              * the local variables. This is important to avoid
3437              * breaking transferred_bytes and bandwidth calculation
3438              */
3439             update_iteration_initial_status(s);
3440         }
3441 
3442         urgent = migration_rate_limit();
3443     }
3444 
3445     trace_migration_thread_after_loop();
3446     migration_iteration_finish(s);
3447     object_unref(OBJECT(s));
3448     rcu_unregister_thread();
3449     return NULL;
3450 }
3451 
3452 void migrate_fd_connect(MigrationState *s, Error *error_in)
3453 {
3454     Error *local_err = NULL;
3455     int64_t rate_limit;
3456     bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3457 
3458     s->expected_downtime = s->parameters.downtime_limit;
3459     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3460     if (error_in) {
3461         migrate_fd_error(s, error_in);
3462         migrate_fd_cleanup(s);
3463         return;
3464     }
3465 
3466     if (resume) {
3467         /* This is a resumed migration */
3468         rate_limit = s->parameters.max_postcopy_bandwidth /
3469             XFER_LIMIT_RATIO;
3470     } else {
3471         /* This is a fresh new migration */
3472         rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3473 
3474         /* Notify before starting migration thread */
3475         notifier_list_notify(&migration_state_notifiers, s);
3476     }
3477 
3478     qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3479     qemu_file_set_blocking(s->to_dst_file, true);
3480 
3481     /*
3482      * Open the return path. For postcopy, it is used exclusively. For
3483      * precopy, only if user specified "return-path" capability would
3484      * QEMU uses the return path.
3485      */
3486     if (migrate_postcopy_ram() || migrate_use_return_path()) {
3487         if (open_return_path_on_source(s, !resume)) {
3488             error_report("Unable to open return-path for postcopy");
3489             migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3490             migrate_fd_cleanup(s);
3491             return;
3492         }
3493     }
3494 
3495     if (resume) {
3496         /* Wakeup the main migration thread to do the recovery */
3497         migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3498                           MIGRATION_STATUS_POSTCOPY_RECOVER);
3499         qemu_sem_post(&s->postcopy_pause_sem);
3500         return;
3501     }
3502 
3503     if (multifd_save_setup(&local_err) != 0) {
3504         error_report_err(local_err);
3505         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3506                           MIGRATION_STATUS_FAILED);
3507         migrate_fd_cleanup(s);
3508         return;
3509     }
3510     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3511                        QEMU_THREAD_JOINABLE);
3512     s->migration_thread_running = true;
3513 }
3514 
3515 void migration_global_dump(Monitor *mon)
3516 {
3517     MigrationState *ms = migrate_get_current();
3518 
3519     monitor_printf(mon, "globals:\n");
3520     monitor_printf(mon, "store-global-state: %s\n",
3521                    ms->store_global_state ? "on" : "off");
3522     monitor_printf(mon, "only-migratable: %s\n",
3523                    only_migratable ? "on" : "off");
3524     monitor_printf(mon, "send-configuration: %s\n",
3525                    ms->send_configuration ? "on" : "off");
3526     monitor_printf(mon, "send-section-footer: %s\n",
3527                    ms->send_section_footer ? "on" : "off");
3528     monitor_printf(mon, "decompress-error-check: %s\n",
3529                    ms->decompress_error_check ? "on" : "off");
3530     monitor_printf(mon, "clear-bitmap-shift: %u\n",
3531                    ms->clear_bitmap_shift);
3532 }
3533 
3534 #define DEFINE_PROP_MIG_CAP(name, x)             \
3535     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3536 
3537 static Property migration_properties[] = {
3538     DEFINE_PROP_BOOL("store-global-state", MigrationState,
3539                      store_global_state, true),
3540     DEFINE_PROP_BOOL("send-configuration", MigrationState,
3541                      send_configuration, true),
3542     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3543                      send_section_footer, true),
3544     DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3545                       decompress_error_check, true),
3546     DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
3547                       clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
3548 
3549     /* Migration parameters */
3550     DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3551                       parameters.compress_level,
3552                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
3553     DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3554                       parameters.compress_threads,
3555                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3556     DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3557                       parameters.compress_wait_thread, true),
3558     DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3559                       parameters.decompress_threads,
3560                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3561     DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3562                       parameters.cpu_throttle_initial,
3563                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3564     DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3565                       parameters.cpu_throttle_increment,
3566                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3567     DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3568                       parameters.max_bandwidth, MAX_THROTTLE),
3569     DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3570                       parameters.downtime_limit,
3571                       DEFAULT_MIGRATE_SET_DOWNTIME),
3572     DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3573                       parameters.x_checkpoint_delay,
3574                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3575     DEFINE_PROP_UINT8("multifd-channels", MigrationState,
3576                       parameters.multifd_channels,
3577                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3578     DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState,
3579                       parameters.multifd_compression,
3580                       DEFAULT_MIGRATE_MULTIFD_COMPRESSION),
3581     DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState,
3582                       parameters.multifd_zlib_level,
3583                       DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL),
3584     DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState,
3585                       parameters.multifd_zstd_level,
3586                       DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL),
3587     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3588                       parameters.xbzrle_cache_size,
3589                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3590     DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3591                       parameters.max_postcopy_bandwidth,
3592                       DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3593     DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3594                       parameters.max_cpu_throttle,
3595                       DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3596     DEFINE_PROP_SIZE("announce-initial", MigrationState,
3597                       parameters.announce_initial,
3598                       DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
3599     DEFINE_PROP_SIZE("announce-max", MigrationState,
3600                       parameters.announce_max,
3601                       DEFAULT_MIGRATE_ANNOUNCE_MAX),
3602     DEFINE_PROP_SIZE("announce-rounds", MigrationState,
3603                       parameters.announce_rounds,
3604                       DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
3605     DEFINE_PROP_SIZE("announce-step", MigrationState,
3606                       parameters.announce_step,
3607                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
3608 
3609     /* Migration capabilities */
3610     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3611     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3612     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3613     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3614     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3615     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3616     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3617     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3618     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3619     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3620     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3621     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD),
3622 
3623     DEFINE_PROP_END_OF_LIST(),
3624 };
3625 
3626 static void migration_class_init(ObjectClass *klass, void *data)
3627 {
3628     DeviceClass *dc = DEVICE_CLASS(klass);
3629 
3630     dc->user_creatable = false;
3631     device_class_set_props(dc, migration_properties);
3632 }
3633 
3634 static void migration_instance_finalize(Object *obj)
3635 {
3636     MigrationState *ms = MIGRATION_OBJ(obj);
3637     MigrationParameters *params = &ms->parameters;
3638 
3639     qemu_mutex_destroy(&ms->error_mutex);
3640     qemu_mutex_destroy(&ms->qemu_file_lock);
3641     g_free(params->tls_hostname);
3642     g_free(params->tls_creds);
3643     qemu_sem_destroy(&ms->wait_unplug_sem);
3644     qemu_sem_destroy(&ms->rate_limit_sem);
3645     qemu_sem_destroy(&ms->pause_sem);
3646     qemu_sem_destroy(&ms->postcopy_pause_sem);
3647     qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3648     qemu_sem_destroy(&ms->rp_state.rp_sem);
3649     error_free(ms->error);
3650 }
3651 
3652 static void migration_instance_init(Object *obj)
3653 {
3654     MigrationState *ms = MIGRATION_OBJ(obj);
3655     MigrationParameters *params = &ms->parameters;
3656 
3657     ms->state = MIGRATION_STATUS_NONE;
3658     ms->mbps = -1;
3659     ms->pages_per_second = -1;
3660     qemu_sem_init(&ms->pause_sem, 0);
3661     qemu_mutex_init(&ms->error_mutex);
3662 
3663     params->tls_hostname = g_strdup("");
3664     params->tls_creds = g_strdup("");
3665 
3666     /* Set has_* up only for parameter checks */
3667     params->has_compress_level = true;
3668     params->has_compress_threads = true;
3669     params->has_decompress_threads = true;
3670     params->has_cpu_throttle_initial = true;
3671     params->has_cpu_throttle_increment = true;
3672     params->has_max_bandwidth = true;
3673     params->has_downtime_limit = true;
3674     params->has_x_checkpoint_delay = true;
3675     params->has_block_incremental = true;
3676     params->has_multifd_channels = true;
3677     params->has_multifd_compression = true;
3678     params->has_multifd_zlib_level = true;
3679     params->has_multifd_zstd_level = true;
3680     params->has_xbzrle_cache_size = true;
3681     params->has_max_postcopy_bandwidth = true;
3682     params->has_max_cpu_throttle = true;
3683     params->has_announce_initial = true;
3684     params->has_announce_max = true;
3685     params->has_announce_rounds = true;
3686     params->has_announce_step = true;
3687 
3688     qemu_sem_init(&ms->postcopy_pause_sem, 0);
3689     qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3690     qemu_sem_init(&ms->rp_state.rp_sem, 0);
3691     qemu_sem_init(&ms->rate_limit_sem, 0);
3692     qemu_sem_init(&ms->wait_unplug_sem, 0);
3693     qemu_mutex_init(&ms->qemu_file_lock);
3694 }
3695 
3696 /*
3697  * Return true if check pass, false otherwise. Error will be put
3698  * inside errp if provided.
3699  */
3700 static bool migration_object_check(MigrationState *ms, Error **errp)
3701 {
3702     MigrationCapabilityStatusList *head = NULL;
3703     /* Assuming all off */
3704     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3705     int i;
3706 
3707     if (!migrate_params_check(&ms->parameters, errp)) {
3708         return false;
3709     }
3710 
3711     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3712         if (ms->enabled_capabilities[i]) {
3713             head = migrate_cap_add(head, i, true);
3714         }
3715     }
3716 
3717     ret = migrate_caps_check(cap_list, head, errp);
3718 
3719     /* It works with head == NULL */
3720     qapi_free_MigrationCapabilityStatusList(head);
3721 
3722     return ret;
3723 }
3724 
3725 static const TypeInfo migration_type = {
3726     .name = TYPE_MIGRATION,
3727     /*
3728      * NOTE: TYPE_MIGRATION is not really a device, as the object is
3729      * not created using qdev_create(), it is not attached to the qdev
3730      * device tree, and it is never realized.
3731      *
3732      * TODO: Make this TYPE_OBJECT once QOM provides something like
3733      * TYPE_DEVICE's "-global" properties.
3734      */
3735     .parent = TYPE_DEVICE,
3736     .class_init = migration_class_init,
3737     .class_size = sizeof(MigrationClass),
3738     .instance_size = sizeof(MigrationState),
3739     .instance_init = migration_instance_init,
3740     .instance_finalize = migration_instance_finalize,
3741 };
3742 
3743 static void register_migration_types(void)
3744 {
3745     type_register_static(&migration_type);
3746 }
3747 
3748 type_init(register_migration_types);
3749