xref: /openbmc/qemu/migration/migration.c (revision f96c6a87)
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_THROTTLE_TRIGGER_THRESHOLD 50
82 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
83 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
84 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
85 
86 /* Migration XBZRLE default cache size */
87 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
88 
89 /* The delay time (in ms) between two COLO checkpoints */
90 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
91 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
92 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
93 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
94 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
95 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
96 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
97 
98 /* Background transfer rate for postcopy, 0 means unlimited, note
99  * that page requests can still exceed this limit.
100  */
101 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
102 
103 /*
104  * Parameters for self_announce_delay giving a stream of RARP/ARP
105  * packets after migration.
106  */
107 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL  50
108 #define DEFAULT_MIGRATE_ANNOUNCE_MAX     550
109 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS    5
110 #define DEFAULT_MIGRATE_ANNOUNCE_STEP    100
111 
112 static NotifierList migration_state_notifiers =
113     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
114 
115 static bool deferred_incoming;
116 
117 /* Messages sent on the return path from destination to source */
118 enum mig_rp_message_type {
119     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
120     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
121     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
122 
123     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
124     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
125     MIG_RP_MSG_RECV_BITMAP,  /* send recved_bitmap back to source */
126     MIG_RP_MSG_RESUME_ACK,   /* tell source that we are ready to resume */
127 
128     MIG_RP_MSG_MAX
129 };
130 
131 /* When we add fault tolerance, we could have several
132    migrations at once.  For now we don't need to add
133    dynamic creation of migration */
134 
135 static MigrationState *current_migration;
136 static MigrationIncomingState *current_incoming;
137 
138 static bool migration_object_check(MigrationState *ms, Error **errp);
139 static int migration_maybe_pause(MigrationState *s,
140                                  int *current_active_state,
141                                  int new_state);
142 static void migrate_fd_cancel(MigrationState *s);
143 
144 void migration_object_init(void)
145 {
146     MachineState *ms = MACHINE(qdev_get_machine());
147     Error *err = NULL;
148 
149     /* This can only be called once. */
150     assert(!current_migration);
151     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
152 
153     /*
154      * Init the migrate incoming object as well no matter whether
155      * we'll use it or not.
156      */
157     assert(!current_incoming);
158     current_incoming = g_new0(MigrationIncomingState, 1);
159     current_incoming->state = MIGRATION_STATUS_NONE;
160     current_incoming->postcopy_remote_fds =
161         g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
162     qemu_mutex_init(&current_incoming->rp_mutex);
163     qemu_event_init(&current_incoming->main_thread_load_event, false);
164     qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
165     qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
166 
167     init_dirty_bitmap_incoming_migration();
168 
169     if (!migration_object_check(current_migration, &err)) {
170         error_report_err(err);
171         exit(1);
172     }
173 
174     /*
175      * We cannot really do this in migration_instance_init() since at
176      * that time global properties are not yet applied, then this
177      * value will be definitely replaced by something else.
178      */
179     if (ms->enforce_config_section) {
180         current_migration->send_configuration = true;
181     }
182 }
183 
184 void migration_shutdown(void)
185 {
186     /*
187      * Cancel the current migration - that will (eventually)
188      * stop the migration using this structure
189      */
190     migrate_fd_cancel(current_migration);
191     object_unref(OBJECT(current_migration));
192 }
193 
194 /* For outgoing */
195 MigrationState *migrate_get_current(void)
196 {
197     /* This can only be called after the object created. */
198     assert(current_migration);
199     return current_migration;
200 }
201 
202 MigrationIncomingState *migration_incoming_get_current(void)
203 {
204     assert(current_incoming);
205     return current_incoming;
206 }
207 
208 void migration_incoming_state_destroy(void)
209 {
210     struct MigrationIncomingState *mis = migration_incoming_get_current();
211 
212     if (mis->to_src_file) {
213         /* Tell source that we are done */
214         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
215         qemu_fclose(mis->to_src_file);
216         mis->to_src_file = NULL;
217     }
218 
219     if (mis->from_src_file) {
220         qemu_fclose(mis->from_src_file);
221         mis->from_src_file = NULL;
222     }
223     if (mis->postcopy_remote_fds) {
224         g_array_free(mis->postcopy_remote_fds, TRUE);
225         mis->postcopy_remote_fds = NULL;
226     }
227 
228     qemu_event_reset(&mis->main_thread_load_event);
229 
230     if (mis->socket_address_list) {
231         qapi_free_SocketAddressList(mis->socket_address_list);
232         mis->socket_address_list = NULL;
233     }
234 }
235 
236 static void migrate_generate_event(int new_state)
237 {
238     if (migrate_use_events()) {
239         qapi_event_send_migration(new_state);
240     }
241 }
242 
243 static bool migrate_late_block_activate(void)
244 {
245     MigrationState *s;
246 
247     s = migrate_get_current();
248 
249     return s->enabled_capabilities[
250         MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
251 }
252 
253 /*
254  * Called on -incoming with a defer: uri.
255  * The migration can be started later after any parameters have been
256  * changed.
257  */
258 static void deferred_incoming_migration(Error **errp)
259 {
260     if (deferred_incoming) {
261         error_setg(errp, "Incoming migration already deferred");
262     }
263     deferred_incoming = true;
264 }
265 
266 /*
267  * Send a message on the return channel back to the source
268  * of the migration.
269  */
270 static int migrate_send_rp_message(MigrationIncomingState *mis,
271                                    enum mig_rp_message_type message_type,
272                                    uint16_t len, void *data)
273 {
274     int ret = 0;
275 
276     trace_migrate_send_rp_message((int)message_type, len);
277     qemu_mutex_lock(&mis->rp_mutex);
278 
279     /*
280      * It's possible that the file handle got lost due to network
281      * failures.
282      */
283     if (!mis->to_src_file) {
284         ret = -EIO;
285         goto error;
286     }
287 
288     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
289     qemu_put_be16(mis->to_src_file, len);
290     qemu_put_buffer(mis->to_src_file, data, len);
291     qemu_fflush(mis->to_src_file);
292 
293     /* It's possible that qemu file got error during sending */
294     ret = qemu_file_get_error(mis->to_src_file);
295 
296 error:
297     qemu_mutex_unlock(&mis->rp_mutex);
298     return ret;
299 }
300 
301 /* Request a range of pages from the source VM at the given
302  * start address.
303  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
304  *           as the last request (a name must have been given previously)
305  *   Start: Address offset within the RB
306  *   Len: Length in bytes required - must be a multiple of pagesize
307  */
308 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
309                               ram_addr_t start, size_t len)
310 {
311     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
312     size_t msglen = 12; /* start + len */
313     enum mig_rp_message_type msg_type;
314 
315     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
316     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
317 
318     if (rbname) {
319         int rbname_len = strlen(rbname);
320         assert(rbname_len < 256);
321 
322         bufc[msglen++] = rbname_len;
323         memcpy(bufc + msglen, rbname, rbname_len);
324         msglen += rbname_len;
325         msg_type = MIG_RP_MSG_REQ_PAGES_ID;
326     } else {
327         msg_type = MIG_RP_MSG_REQ_PAGES;
328     }
329 
330     return migrate_send_rp_message(mis, msg_type, msglen, bufc);
331 }
332 
333 static bool migration_colo_enabled;
334 bool migration_incoming_colo_enabled(void)
335 {
336     return migration_colo_enabled;
337 }
338 
339 void migration_incoming_disable_colo(void)
340 {
341     migration_colo_enabled = false;
342 }
343 
344 void migration_incoming_enable_colo(void)
345 {
346     migration_colo_enabled = true;
347 }
348 
349 void migrate_add_address(SocketAddress *address)
350 {
351     MigrationIncomingState *mis = migration_incoming_get_current();
352     SocketAddressList *addrs;
353 
354     addrs = g_new0(SocketAddressList, 1);
355     addrs->next = mis->socket_address_list;
356     mis->socket_address_list = addrs;
357     addrs->value = QAPI_CLONE(SocketAddress, address);
358 }
359 
360 void qemu_start_incoming_migration(const char *uri, Error **errp)
361 {
362     const char *p;
363 
364     qapi_event_send_migration(MIGRATION_STATUS_SETUP);
365     if (!strcmp(uri, "defer")) {
366         deferred_incoming_migration(errp);
367     } else if (strstart(uri, "tcp:", &p)) {
368         tcp_start_incoming_migration(p, errp);
369 #ifdef CONFIG_RDMA
370     } else if (strstart(uri, "rdma:", &p)) {
371         rdma_start_incoming_migration(p, errp);
372 #endif
373     } else if (strstart(uri, "exec:", &p)) {
374         exec_start_incoming_migration(p, errp);
375     } else if (strstart(uri, "unix:", &p)) {
376         unix_start_incoming_migration(p, errp);
377     } else if (strstart(uri, "fd:", &p)) {
378         fd_start_incoming_migration(p, errp);
379     } else {
380         error_setg(errp, "unknown migration protocol: %s", uri);
381     }
382 }
383 
384 static void process_incoming_migration_bh(void *opaque)
385 {
386     Error *local_err = NULL;
387     MigrationIncomingState *mis = opaque;
388 
389     /* If capability late_block_activate is set:
390      * Only fire up the block code now if we're going to restart the
391      * VM, else 'cont' will do it.
392      * This causes file locking to happen; so we don't want it to happen
393      * unless we really are starting the VM.
394      */
395     if (!migrate_late_block_activate() ||
396          (autostart && (!global_state_received() ||
397             global_state_get_runstate() == RUN_STATE_RUNNING))) {
398         /* Make sure all file formats flush their mutable metadata.
399          * If we get an error here, just don't restart the VM yet. */
400         bdrv_invalidate_cache_all(&local_err);
401         if (local_err) {
402             error_report_err(local_err);
403             local_err = NULL;
404             autostart = false;
405         }
406     }
407 
408     /*
409      * This must happen after all error conditions are dealt with and
410      * we're sure the VM is going to be running on this host.
411      */
412     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
413 
414     if (multifd_load_cleanup(&local_err) != 0) {
415         error_report_err(local_err);
416         autostart = false;
417     }
418     /* If global state section was not received or we are in running
419        state, we need to obey autostart. Any other state is set with
420        runstate_set. */
421 
422     dirty_bitmap_mig_before_vm_start();
423 
424     if (!global_state_received() ||
425         global_state_get_runstate() == RUN_STATE_RUNNING) {
426         if (autostart) {
427             vm_start();
428         } else {
429             runstate_set(RUN_STATE_PAUSED);
430         }
431     } else if (migration_incoming_colo_enabled()) {
432         migration_incoming_disable_colo();
433         vm_start();
434     } else {
435         runstate_set(global_state_get_runstate());
436     }
437     /*
438      * This must happen after any state changes since as soon as an external
439      * observer sees this event they might start to prod at the VM assuming
440      * it's ready to use.
441      */
442     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
443                       MIGRATION_STATUS_COMPLETED);
444     qemu_bh_delete(mis->bh);
445     migration_incoming_state_destroy();
446 }
447 
448 static void process_incoming_migration_co(void *opaque)
449 {
450     MigrationIncomingState *mis = migration_incoming_get_current();
451     PostcopyState ps;
452     int ret;
453     Error *local_err = NULL;
454 
455     assert(mis->from_src_file);
456     mis->migration_incoming_co = qemu_coroutine_self();
457     mis->largest_page_size = qemu_ram_pagesize_largest();
458     postcopy_state_set(POSTCOPY_INCOMING_NONE);
459     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
460                       MIGRATION_STATUS_ACTIVE);
461     ret = qemu_loadvm_state(mis->from_src_file);
462 
463     ps = postcopy_state_get();
464     trace_process_incoming_migration_co_end(ret, ps);
465     if (ps != POSTCOPY_INCOMING_NONE) {
466         if (ps == POSTCOPY_INCOMING_ADVISE) {
467             /*
468              * Where a migration had postcopy enabled (and thus went to advise)
469              * but managed to complete within the precopy period, we can use
470              * the normal exit.
471              */
472             postcopy_ram_incoming_cleanup(mis);
473         } else if (ret >= 0) {
474             /*
475              * Postcopy was started, cleanup should happen at the end of the
476              * postcopy thread.
477              */
478             trace_process_incoming_migration_co_postcopy_end_main();
479             return;
480         }
481         /* Else if something went wrong then just fall out of the normal exit */
482     }
483 
484     /* we get COLO info, and know if we are in COLO mode */
485     if (!ret && migration_incoming_colo_enabled()) {
486         /* Make sure all file formats flush their mutable metadata */
487         bdrv_invalidate_cache_all(&local_err);
488         if (local_err) {
489             error_report_err(local_err);
490             goto fail;
491         }
492 
493         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
494              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
495         mis->have_colo_incoming_thread = true;
496         qemu_coroutine_yield();
497 
498         /* Wait checkpoint incoming thread exit before free resource */
499         qemu_thread_join(&mis->colo_incoming_thread);
500         /* We hold the global iothread lock, so it is safe here */
501         colo_release_ram_cache();
502     }
503 
504     if (ret < 0) {
505         error_report("load of migration failed: %s", strerror(-ret));
506         goto fail;
507     }
508     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
509     qemu_bh_schedule(mis->bh);
510     mis->migration_incoming_co = NULL;
511     return;
512 fail:
513     local_err = NULL;
514     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
515                       MIGRATION_STATUS_FAILED);
516     qemu_fclose(mis->from_src_file);
517     if (multifd_load_cleanup(&local_err) != 0) {
518         error_report_err(local_err);
519     }
520     exit(EXIT_FAILURE);
521 }
522 
523 /**
524  * @migration_incoming_setup: Setup incoming migration
525  *
526  * Returns 0 for no error or 1 for error
527  *
528  * @f: file for main migration channel
529  * @errp: where to put errors
530  */
531 static int migration_incoming_setup(QEMUFile *f, Error **errp)
532 {
533     MigrationIncomingState *mis = migration_incoming_get_current();
534     Error *local_err = NULL;
535 
536     if (multifd_load_setup(&local_err) != 0) {
537         /* We haven't been able to create multifd threads
538            nothing better to do */
539         error_report_err(local_err);
540         exit(EXIT_FAILURE);
541     }
542 
543     if (!mis->from_src_file) {
544         mis->from_src_file = f;
545     }
546     qemu_file_set_blocking(f, false);
547     return 0;
548 }
549 
550 void migration_incoming_process(void)
551 {
552     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
553     qemu_coroutine_enter(co);
554 }
555 
556 /* Returns true if recovered from a paused migration, otherwise false */
557 static bool postcopy_try_recover(QEMUFile *f)
558 {
559     MigrationIncomingState *mis = migration_incoming_get_current();
560 
561     if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
562         /* Resumed from a paused postcopy migration */
563 
564         mis->from_src_file = f;
565         /* Postcopy has standalone thread to do vm load */
566         qemu_file_set_blocking(f, true);
567 
568         /* Re-configure the return path */
569         mis->to_src_file = qemu_file_get_return_path(f);
570 
571         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
572                           MIGRATION_STATUS_POSTCOPY_RECOVER);
573 
574         /*
575          * Here, we only wake up the main loading thread (while the
576          * fault thread will still be waiting), so that we can receive
577          * commands from source now, and answer it if needed. The
578          * fault thread will be woken up afterwards until we are sure
579          * that source is ready to reply to page requests.
580          */
581         qemu_sem_post(&mis->postcopy_pause_sem_dst);
582         return true;
583     }
584 
585     return false;
586 }
587 
588 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
589 {
590     Error *local_err = NULL;
591 
592     if (postcopy_try_recover(f)) {
593         return;
594     }
595 
596     if (migration_incoming_setup(f, &local_err)) {
597         if (local_err) {
598             error_propagate(errp, local_err);
599         }
600         return;
601     }
602     migration_incoming_process();
603 }
604 
605 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
606 {
607     MigrationIncomingState *mis = migration_incoming_get_current();
608     Error *local_err = NULL;
609     bool start_migration;
610 
611     if (!mis->from_src_file) {
612         /* The first connection (multifd may have multiple) */
613         QEMUFile *f = qemu_fopen_channel_input(ioc);
614 
615         /* If it's a recovery, we're done */
616         if (postcopy_try_recover(f)) {
617             return;
618         }
619 
620         if (migration_incoming_setup(f, &local_err)) {
621             if (local_err) {
622                 error_propagate(errp, local_err);
623             }
624             return;
625         }
626 
627         /*
628          * Common migration only needs one channel, so we can start
629          * right now.  Multifd needs more than one channel, we wait.
630          */
631         start_migration = !migrate_use_multifd();
632     } else {
633         /* Multiple connections */
634         assert(migrate_use_multifd());
635         start_migration = multifd_recv_new_channel(ioc, &local_err);
636         if (local_err) {
637             error_propagate(errp, local_err);
638             return;
639         }
640     }
641 
642     if (start_migration) {
643         migration_incoming_process();
644     }
645 }
646 
647 /**
648  * @migration_has_all_channels: We have received all channels that we need
649  *
650  * Returns true when we have got connections to all the channels that
651  * we need for migration.
652  */
653 bool migration_has_all_channels(void)
654 {
655     MigrationIncomingState *mis = migration_incoming_get_current();
656     bool all_channels;
657 
658     all_channels = multifd_recv_all_channels_created();
659 
660     return all_channels && mis->from_src_file != NULL;
661 }
662 
663 /*
664  * Send a 'SHUT' message on the return channel with the given value
665  * to indicate that we've finished with the RP.  Non-0 value indicates
666  * error.
667  */
668 void migrate_send_rp_shut(MigrationIncomingState *mis,
669                           uint32_t value)
670 {
671     uint32_t buf;
672 
673     buf = cpu_to_be32(value);
674     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
675 }
676 
677 /*
678  * Send a 'PONG' message on the return channel with the given value
679  * (normally in response to a 'PING')
680  */
681 void migrate_send_rp_pong(MigrationIncomingState *mis,
682                           uint32_t value)
683 {
684     uint32_t buf;
685 
686     buf = cpu_to_be32(value);
687     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
688 }
689 
690 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
691                                  char *block_name)
692 {
693     char buf[512];
694     int len;
695     int64_t res;
696 
697     /*
698      * First, we send the header part. It contains only the len of
699      * idstr, and the idstr itself.
700      */
701     len = strlen(block_name);
702     buf[0] = len;
703     memcpy(buf + 1, block_name, len);
704 
705     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
706         error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
707                      __func__);
708         return;
709     }
710 
711     migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
712 
713     /*
714      * Next, we dump the received bitmap to the stream.
715      *
716      * TODO: currently we are safe since we are the only one that is
717      * using the to_src_file handle (fault thread is still paused),
718      * and it's ok even not taking the mutex. However the best way is
719      * to take the lock before sending the message header, and release
720      * the lock after sending the bitmap.
721      */
722     qemu_mutex_lock(&mis->rp_mutex);
723     res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
724     qemu_mutex_unlock(&mis->rp_mutex);
725 
726     trace_migrate_send_rp_recv_bitmap(block_name, res);
727 }
728 
729 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
730 {
731     uint32_t buf;
732 
733     buf = cpu_to_be32(value);
734     migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
735 }
736 
737 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
738 {
739     MigrationCapabilityStatusList *head = NULL;
740     MigrationCapabilityStatusList *caps;
741     MigrationState *s = migrate_get_current();
742     int i;
743 
744     caps = NULL; /* silence compiler warning */
745     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
746 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
747         if (i == MIGRATION_CAPABILITY_BLOCK) {
748             continue;
749         }
750 #endif
751         if (head == NULL) {
752             head = g_malloc0(sizeof(*caps));
753             caps = head;
754         } else {
755             caps->next = g_malloc0(sizeof(*caps));
756             caps = caps->next;
757         }
758         caps->value =
759             g_malloc(sizeof(*caps->value));
760         caps->value->capability = i;
761         caps->value->state = s->enabled_capabilities[i];
762     }
763 
764     return head;
765 }
766 
767 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
768 {
769     MigrationParameters *params;
770     MigrationState *s = migrate_get_current();
771 
772     /* TODO use QAPI_CLONE() instead of duplicating it inline */
773     params = g_malloc0(sizeof(*params));
774     params->has_compress_level = true;
775     params->compress_level = s->parameters.compress_level;
776     params->has_compress_threads = true;
777     params->compress_threads = s->parameters.compress_threads;
778     params->has_compress_wait_thread = true;
779     params->compress_wait_thread = s->parameters.compress_wait_thread;
780     params->has_decompress_threads = true;
781     params->decompress_threads = s->parameters.decompress_threads;
782     params->has_throttle_trigger_threshold = true;
783     params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
784     params->has_cpu_throttle_initial = true;
785     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
786     params->has_cpu_throttle_increment = true;
787     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
788     params->has_tls_creds = true;
789     params->tls_creds = g_strdup(s->parameters.tls_creds);
790     params->has_tls_hostname = true;
791     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
792     params->has_tls_authz = true;
793     params->tls_authz = g_strdup(s->parameters.tls_authz ?
794                                  s->parameters.tls_authz : "");
795     params->has_max_bandwidth = true;
796     params->max_bandwidth = s->parameters.max_bandwidth;
797     params->has_downtime_limit = true;
798     params->downtime_limit = s->parameters.downtime_limit;
799     params->has_x_checkpoint_delay = true;
800     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
801     params->has_block_incremental = true;
802     params->block_incremental = s->parameters.block_incremental;
803     params->has_multifd_channels = true;
804     params->multifd_channels = s->parameters.multifd_channels;
805     params->has_multifd_compression = true;
806     params->multifd_compression = s->parameters.multifd_compression;
807     params->has_multifd_zlib_level = true;
808     params->multifd_zlib_level = s->parameters.multifd_zlib_level;
809     params->has_multifd_zstd_level = true;
810     params->multifd_zstd_level = s->parameters.multifd_zstd_level;
811     params->has_xbzrle_cache_size = true;
812     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
813     params->has_max_postcopy_bandwidth = true;
814     params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
815     params->has_max_cpu_throttle = true;
816     params->max_cpu_throttle = s->parameters.max_cpu_throttle;
817     params->has_announce_initial = true;
818     params->announce_initial = s->parameters.announce_initial;
819     params->has_announce_max = true;
820     params->announce_max = s->parameters.announce_max;
821     params->has_announce_rounds = true;
822     params->announce_rounds = s->parameters.announce_rounds;
823     params->has_announce_step = true;
824     params->announce_step = s->parameters.announce_step;
825 
826     return params;
827 }
828 
829 AnnounceParameters *migrate_announce_params(void)
830 {
831     static AnnounceParameters ap;
832 
833     MigrationState *s = migrate_get_current();
834 
835     ap.initial = s->parameters.announce_initial;
836     ap.max = s->parameters.announce_max;
837     ap.rounds = s->parameters.announce_rounds;
838     ap.step = s->parameters.announce_step;
839 
840     return &ap;
841 }
842 
843 /*
844  * Return true if we're already in the middle of a migration
845  * (i.e. any of the active or setup states)
846  */
847 bool migration_is_setup_or_active(int state)
848 {
849     switch (state) {
850     case MIGRATION_STATUS_ACTIVE:
851     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
852     case MIGRATION_STATUS_POSTCOPY_PAUSED:
853     case MIGRATION_STATUS_POSTCOPY_RECOVER:
854     case MIGRATION_STATUS_SETUP:
855     case MIGRATION_STATUS_PRE_SWITCHOVER:
856     case MIGRATION_STATUS_DEVICE:
857     case MIGRATION_STATUS_WAIT_UNPLUG:
858     case MIGRATION_STATUS_COLO:
859         return true;
860 
861     default:
862         return false;
863 
864     }
865 }
866 
867 bool migration_is_running(int state)
868 {
869     switch (state) {
870     case MIGRATION_STATUS_ACTIVE:
871     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
872     case MIGRATION_STATUS_POSTCOPY_PAUSED:
873     case MIGRATION_STATUS_POSTCOPY_RECOVER:
874     case MIGRATION_STATUS_SETUP:
875     case MIGRATION_STATUS_PRE_SWITCHOVER:
876     case MIGRATION_STATUS_DEVICE:
877     case MIGRATION_STATUS_WAIT_UNPLUG:
878     case MIGRATION_STATUS_CANCELLING:
879         return true;
880 
881     default:
882         return false;
883 
884     }
885 }
886 
887 static void populate_time_info(MigrationInfo *info, MigrationState *s)
888 {
889     info->has_status = true;
890     info->has_setup_time = true;
891     info->setup_time = s->setup_time;
892     if (s->state == MIGRATION_STATUS_COMPLETED) {
893         info->has_total_time = true;
894         info->total_time = s->total_time;
895         info->has_downtime = true;
896         info->downtime = s->downtime;
897     } else {
898         info->has_total_time = true;
899         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
900                            s->start_time;
901         info->has_expected_downtime = true;
902         info->expected_downtime = s->expected_downtime;
903     }
904 }
905 
906 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
907 {
908     info->has_ram = true;
909     info->ram = g_malloc0(sizeof(*info->ram));
910     info->ram->transferred = ram_counters.transferred;
911     info->ram->total = ram_bytes_total();
912     info->ram->duplicate = ram_counters.duplicate;
913     /* legacy value.  It is not used anymore */
914     info->ram->skipped = 0;
915     info->ram->normal = ram_counters.normal;
916     info->ram->normal_bytes = ram_counters.normal *
917         qemu_target_page_size();
918     info->ram->mbps = s->mbps;
919     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
920     info->ram->postcopy_requests = ram_counters.postcopy_requests;
921     info->ram->page_size = qemu_target_page_size();
922     info->ram->multifd_bytes = ram_counters.multifd_bytes;
923     info->ram->pages_per_second = s->pages_per_second;
924 
925     if (migrate_use_xbzrle()) {
926         info->has_xbzrle_cache = true;
927         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
928         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
929         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
930         info->xbzrle_cache->pages = xbzrle_counters.pages;
931         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
932         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
933         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
934     }
935 
936     if (migrate_use_compression()) {
937         info->has_compression = true;
938         info->compression = g_malloc0(sizeof(*info->compression));
939         info->compression->pages = compression_counters.pages;
940         info->compression->busy = compression_counters.busy;
941         info->compression->busy_rate = compression_counters.busy_rate;
942         info->compression->compressed_size =
943                                     compression_counters.compressed_size;
944         info->compression->compression_rate =
945                                     compression_counters.compression_rate;
946     }
947 
948     if (cpu_throttle_active()) {
949         info->has_cpu_throttle_percentage = true;
950         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
951     }
952 
953     if (s->state != MIGRATION_STATUS_COMPLETED) {
954         info->ram->remaining = ram_bytes_remaining();
955         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
956     }
957 }
958 
959 static void populate_disk_info(MigrationInfo *info)
960 {
961     if (blk_mig_active()) {
962         info->has_disk = true;
963         info->disk = g_malloc0(sizeof(*info->disk));
964         info->disk->transferred = blk_mig_bytes_transferred();
965         info->disk->remaining = blk_mig_bytes_remaining();
966         info->disk->total = blk_mig_bytes_total();
967     }
968 }
969 
970 static void fill_source_migration_info(MigrationInfo *info)
971 {
972     MigrationState *s = migrate_get_current();
973 
974     switch (s->state) {
975     case MIGRATION_STATUS_NONE:
976         /* no migration has happened ever */
977         /* do not overwrite destination migration status */
978         return;
979         break;
980     case MIGRATION_STATUS_SETUP:
981         info->has_status = true;
982         info->has_total_time = false;
983         break;
984     case MIGRATION_STATUS_ACTIVE:
985     case MIGRATION_STATUS_CANCELLING:
986     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
987     case MIGRATION_STATUS_PRE_SWITCHOVER:
988     case MIGRATION_STATUS_DEVICE:
989     case MIGRATION_STATUS_POSTCOPY_PAUSED:
990     case MIGRATION_STATUS_POSTCOPY_RECOVER:
991         /* TODO add some postcopy stats */
992         populate_time_info(info, s);
993         populate_ram_info(info, s);
994         populate_disk_info(info);
995         break;
996     case MIGRATION_STATUS_COLO:
997         info->has_status = true;
998         /* TODO: display COLO specific information (checkpoint info etc.) */
999         break;
1000     case MIGRATION_STATUS_COMPLETED:
1001         populate_time_info(info, s);
1002         populate_ram_info(info, s);
1003         break;
1004     case MIGRATION_STATUS_FAILED:
1005         info->has_status = true;
1006         if (s->error) {
1007             info->has_error_desc = true;
1008             info->error_desc = g_strdup(error_get_pretty(s->error));
1009         }
1010         break;
1011     case MIGRATION_STATUS_CANCELLED:
1012         info->has_status = true;
1013         break;
1014     case MIGRATION_STATUS_WAIT_UNPLUG:
1015         info->has_status = true;
1016         break;
1017     }
1018     info->status = s->state;
1019 }
1020 
1021 /**
1022  * @migration_caps_check - check capability validity
1023  *
1024  * @cap_list: old capability list, array of bool
1025  * @params: new capabilities to be applied soon
1026  * @errp: set *errp if the check failed, with reason
1027  *
1028  * Returns true if check passed, otherwise false.
1029  */
1030 static bool migrate_caps_check(bool *cap_list,
1031                                MigrationCapabilityStatusList *params,
1032                                Error **errp)
1033 {
1034     MigrationCapabilityStatusList *cap;
1035     bool old_postcopy_cap;
1036     MigrationIncomingState *mis = migration_incoming_get_current();
1037 
1038     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1039 
1040     for (cap = params; cap; cap = cap->next) {
1041         cap_list[cap->value->capability] = cap->value->state;
1042     }
1043 
1044 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1045     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1046         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1047                    "block migration");
1048         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1049         return false;
1050     }
1051 #endif
1052 
1053 #ifndef CONFIG_REPLICATION
1054     if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1055         error_setg(errp, "QEMU compiled without replication module"
1056                    " can't enable COLO");
1057         error_append_hint(errp, "Please enable replication before COLO.\n");
1058         return false;
1059     }
1060 #endif
1061 
1062     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1063         /* This check is reasonably expensive, so only when it's being
1064          * set the first time, also it's only the destination that needs
1065          * special support.
1066          */
1067         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1068             !postcopy_ram_supported_by_host(mis)) {
1069             /* postcopy_ram_supported_by_host will have emitted a more
1070              * detailed message
1071              */
1072             error_setg(errp, "Postcopy is not supported");
1073             return false;
1074         }
1075 
1076         if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1077             error_setg(errp, "Postcopy is not compatible with ignore-shared");
1078             return false;
1079         }
1080     }
1081 
1082     return true;
1083 }
1084 
1085 static void fill_destination_migration_info(MigrationInfo *info)
1086 {
1087     MigrationIncomingState *mis = migration_incoming_get_current();
1088 
1089     if (mis->socket_address_list) {
1090         info->has_socket_address = true;
1091         info->socket_address =
1092             QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1093     }
1094 
1095     switch (mis->state) {
1096     case MIGRATION_STATUS_NONE:
1097         return;
1098         break;
1099     case MIGRATION_STATUS_SETUP:
1100     case MIGRATION_STATUS_CANCELLING:
1101     case MIGRATION_STATUS_CANCELLED:
1102     case MIGRATION_STATUS_ACTIVE:
1103     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1104     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1105     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1106     case MIGRATION_STATUS_FAILED:
1107     case MIGRATION_STATUS_COLO:
1108         info->has_status = true;
1109         break;
1110     case MIGRATION_STATUS_COMPLETED:
1111         info->has_status = true;
1112         fill_destination_postcopy_migration_info(info);
1113         break;
1114     }
1115     info->status = mis->state;
1116 }
1117 
1118 MigrationInfo *qmp_query_migrate(Error **errp)
1119 {
1120     MigrationInfo *info = g_malloc0(sizeof(*info));
1121 
1122     fill_destination_migration_info(info);
1123     fill_source_migration_info(info);
1124 
1125     return info;
1126 }
1127 
1128 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1129                                   Error **errp)
1130 {
1131     MigrationState *s = migrate_get_current();
1132     MigrationCapabilityStatusList *cap;
1133     bool cap_list[MIGRATION_CAPABILITY__MAX];
1134 
1135     if (migration_is_running(s->state)) {
1136         error_setg(errp, QERR_MIGRATION_ACTIVE);
1137         return;
1138     }
1139 
1140     memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1141     if (!migrate_caps_check(cap_list, params, errp)) {
1142         return;
1143     }
1144 
1145     for (cap = params; cap; cap = cap->next) {
1146         s->enabled_capabilities[cap->value->capability] = cap->value->state;
1147     }
1148 }
1149 
1150 /*
1151  * Check whether the parameters are valid. Error will be put into errp
1152  * (if provided). Return true if valid, otherwise false.
1153  */
1154 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1155 {
1156     if (params->has_compress_level &&
1157         (params->compress_level > 9)) {
1158         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1159                    "is invalid, it should be in the range of 0 to 9");
1160         return false;
1161     }
1162 
1163     if (params->has_compress_threads && (params->compress_threads < 1)) {
1164         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1165                    "compress_threads",
1166                    "is invalid, it should be in the range of 1 to 255");
1167         return false;
1168     }
1169 
1170     if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1171         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1172                    "decompress_threads",
1173                    "is invalid, it should be in the range of 1 to 255");
1174         return false;
1175     }
1176 
1177     if (params->has_throttle_trigger_threshold &&
1178         (params->throttle_trigger_threshold < 1 ||
1179          params->throttle_trigger_threshold > 100)) {
1180         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1181                    "throttle_trigger_threshold",
1182                    "an integer in the range of 1 to 100");
1183         return false;
1184     }
1185 
1186     if (params->has_cpu_throttle_initial &&
1187         (params->cpu_throttle_initial < 1 ||
1188          params->cpu_throttle_initial > 99)) {
1189         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1190                    "cpu_throttle_initial",
1191                    "an integer in the range of 1 to 99");
1192         return false;
1193     }
1194 
1195     if (params->has_cpu_throttle_increment &&
1196         (params->cpu_throttle_increment < 1 ||
1197          params->cpu_throttle_increment > 99)) {
1198         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1199                    "cpu_throttle_increment",
1200                    "an integer in the range of 1 to 99");
1201         return false;
1202     }
1203 
1204     if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1205         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1206                    "max_bandwidth",
1207                    "an integer in the range of 0 to "stringify(SIZE_MAX)
1208                    " bytes/second");
1209         return false;
1210     }
1211 
1212     if (params->has_downtime_limit &&
1213         (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1214         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1215                    "downtime_limit",
1216                    "an integer in the range of 0 to "
1217                     stringify(MAX_MIGRATE_DOWNTIME)" milliseconds");
1218         return false;
1219     }
1220 
1221     /* x_checkpoint_delay is now always positive */
1222 
1223     if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1224         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1225                    "multifd_channels",
1226                    "is invalid, it should be in the range of 1 to 255");
1227         return false;
1228     }
1229 
1230     if (params->has_multifd_zlib_level &&
1231         (params->multifd_zlib_level > 9)) {
1232         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1233                    "is invalid, it should be in the range of 0 to 9");
1234         return false;
1235     }
1236 
1237     if (params->has_multifd_zstd_level &&
1238         (params->multifd_zstd_level > 20)) {
1239         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1240                    "is invalid, it should be in the range of 0 to 20");
1241         return false;
1242     }
1243 
1244     if (params->has_xbzrle_cache_size &&
1245         (params->xbzrle_cache_size < qemu_target_page_size() ||
1246          !is_power_of_2(params->xbzrle_cache_size))) {
1247         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1248                    "xbzrle_cache_size",
1249                    "is invalid, it should be bigger than target page size"
1250                    " and a power of 2");
1251         return false;
1252     }
1253 
1254     if (params->has_max_cpu_throttle &&
1255         (params->max_cpu_throttle < params->cpu_throttle_initial ||
1256          params->max_cpu_throttle > 99)) {
1257         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1258                    "max_cpu_throttle",
1259                    "an integer in the range of cpu_throttle_initial to 99");
1260         return false;
1261     }
1262 
1263     if (params->has_announce_initial &&
1264         params->announce_initial > 100000) {
1265         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1266                    "announce_initial",
1267                    "is invalid, it must be less than 100000 ms");
1268         return false;
1269     }
1270     if (params->has_announce_max &&
1271         params->announce_max > 100000) {
1272         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1273                    "announce_max",
1274                    "is invalid, it must be less than 100000 ms");
1275        return false;
1276     }
1277     if (params->has_announce_rounds &&
1278         params->announce_rounds > 1000) {
1279         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1280                    "announce_rounds",
1281                    "is invalid, it must be in the range of 0 to 1000");
1282        return false;
1283     }
1284     if (params->has_announce_step &&
1285         (params->announce_step < 1 ||
1286         params->announce_step > 10000)) {
1287         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1288                    "announce_step",
1289                    "is invalid, it must be in the range of 1 to 10000 ms");
1290        return false;
1291     }
1292     return true;
1293 }
1294 
1295 static void migrate_params_test_apply(MigrateSetParameters *params,
1296                                       MigrationParameters *dest)
1297 {
1298     *dest = migrate_get_current()->parameters;
1299 
1300     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1301 
1302     if (params->has_compress_level) {
1303         dest->compress_level = params->compress_level;
1304     }
1305 
1306     if (params->has_compress_threads) {
1307         dest->compress_threads = params->compress_threads;
1308     }
1309 
1310     if (params->has_compress_wait_thread) {
1311         dest->compress_wait_thread = params->compress_wait_thread;
1312     }
1313 
1314     if (params->has_decompress_threads) {
1315         dest->decompress_threads = params->decompress_threads;
1316     }
1317 
1318     if (params->has_throttle_trigger_threshold) {
1319         dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
1320     }
1321 
1322     if (params->has_cpu_throttle_initial) {
1323         dest->cpu_throttle_initial = params->cpu_throttle_initial;
1324     }
1325 
1326     if (params->has_cpu_throttle_increment) {
1327         dest->cpu_throttle_increment = params->cpu_throttle_increment;
1328     }
1329 
1330     if (params->has_tls_creds) {
1331         assert(params->tls_creds->type == QTYPE_QSTRING);
1332         dest->tls_creds = g_strdup(params->tls_creds->u.s);
1333     }
1334 
1335     if (params->has_tls_hostname) {
1336         assert(params->tls_hostname->type == QTYPE_QSTRING);
1337         dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1338     }
1339 
1340     if (params->has_max_bandwidth) {
1341         dest->max_bandwidth = params->max_bandwidth;
1342     }
1343 
1344     if (params->has_downtime_limit) {
1345         dest->downtime_limit = params->downtime_limit;
1346     }
1347 
1348     if (params->has_x_checkpoint_delay) {
1349         dest->x_checkpoint_delay = params->x_checkpoint_delay;
1350     }
1351 
1352     if (params->has_block_incremental) {
1353         dest->block_incremental = params->block_incremental;
1354     }
1355     if (params->has_multifd_channels) {
1356         dest->multifd_channels = params->multifd_channels;
1357     }
1358     if (params->has_multifd_compression) {
1359         dest->multifd_compression = params->multifd_compression;
1360     }
1361     if (params->has_xbzrle_cache_size) {
1362         dest->xbzrle_cache_size = params->xbzrle_cache_size;
1363     }
1364     if (params->has_max_postcopy_bandwidth) {
1365         dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1366     }
1367     if (params->has_max_cpu_throttle) {
1368         dest->max_cpu_throttle = params->max_cpu_throttle;
1369     }
1370     if (params->has_announce_initial) {
1371         dest->announce_initial = params->announce_initial;
1372     }
1373     if (params->has_announce_max) {
1374         dest->announce_max = params->announce_max;
1375     }
1376     if (params->has_announce_rounds) {
1377         dest->announce_rounds = params->announce_rounds;
1378     }
1379     if (params->has_announce_step) {
1380         dest->announce_step = params->announce_step;
1381     }
1382 }
1383 
1384 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1385 {
1386     MigrationState *s = migrate_get_current();
1387 
1388     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1389 
1390     if (params->has_compress_level) {
1391         s->parameters.compress_level = params->compress_level;
1392     }
1393 
1394     if (params->has_compress_threads) {
1395         s->parameters.compress_threads = params->compress_threads;
1396     }
1397 
1398     if (params->has_compress_wait_thread) {
1399         s->parameters.compress_wait_thread = params->compress_wait_thread;
1400     }
1401 
1402     if (params->has_decompress_threads) {
1403         s->parameters.decompress_threads = params->decompress_threads;
1404     }
1405 
1406     if (params->has_throttle_trigger_threshold) {
1407         s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
1408     }
1409 
1410     if (params->has_cpu_throttle_initial) {
1411         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1412     }
1413 
1414     if (params->has_cpu_throttle_increment) {
1415         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1416     }
1417 
1418     if (params->has_tls_creds) {
1419         g_free(s->parameters.tls_creds);
1420         assert(params->tls_creds->type == QTYPE_QSTRING);
1421         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1422     }
1423 
1424     if (params->has_tls_hostname) {
1425         g_free(s->parameters.tls_hostname);
1426         assert(params->tls_hostname->type == QTYPE_QSTRING);
1427         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1428     }
1429 
1430     if (params->has_tls_authz) {
1431         g_free(s->parameters.tls_authz);
1432         assert(params->tls_authz->type == QTYPE_QSTRING);
1433         s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1434     }
1435 
1436     if (params->has_max_bandwidth) {
1437         s->parameters.max_bandwidth = params->max_bandwidth;
1438         if (s->to_dst_file && !migration_in_postcopy()) {
1439             qemu_file_set_rate_limit(s->to_dst_file,
1440                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1441         }
1442     }
1443 
1444     if (params->has_downtime_limit) {
1445         s->parameters.downtime_limit = params->downtime_limit;
1446     }
1447 
1448     if (params->has_x_checkpoint_delay) {
1449         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1450         if (migration_in_colo_state()) {
1451             colo_checkpoint_notify(s);
1452         }
1453     }
1454 
1455     if (params->has_block_incremental) {
1456         s->parameters.block_incremental = params->block_incremental;
1457     }
1458     if (params->has_multifd_channels) {
1459         s->parameters.multifd_channels = params->multifd_channels;
1460     }
1461     if (params->has_multifd_compression) {
1462         s->parameters.multifd_compression = params->multifd_compression;
1463     }
1464     if (params->has_xbzrle_cache_size) {
1465         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1466         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1467     }
1468     if (params->has_max_postcopy_bandwidth) {
1469         s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1470         if (s->to_dst_file && migration_in_postcopy()) {
1471             qemu_file_set_rate_limit(s->to_dst_file,
1472                     s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1473         }
1474     }
1475     if (params->has_max_cpu_throttle) {
1476         s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1477     }
1478     if (params->has_announce_initial) {
1479         s->parameters.announce_initial = params->announce_initial;
1480     }
1481     if (params->has_announce_max) {
1482         s->parameters.announce_max = params->announce_max;
1483     }
1484     if (params->has_announce_rounds) {
1485         s->parameters.announce_rounds = params->announce_rounds;
1486     }
1487     if (params->has_announce_step) {
1488         s->parameters.announce_step = params->announce_step;
1489     }
1490 }
1491 
1492 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1493 {
1494     MigrationParameters tmp;
1495 
1496     /* TODO Rewrite "" to null instead */
1497     if (params->has_tls_creds
1498         && params->tls_creds->type == QTYPE_QNULL) {
1499         qobject_unref(params->tls_creds->u.n);
1500         params->tls_creds->type = QTYPE_QSTRING;
1501         params->tls_creds->u.s = strdup("");
1502     }
1503     /* TODO Rewrite "" to null instead */
1504     if (params->has_tls_hostname
1505         && params->tls_hostname->type == QTYPE_QNULL) {
1506         qobject_unref(params->tls_hostname->u.n);
1507         params->tls_hostname->type = QTYPE_QSTRING;
1508         params->tls_hostname->u.s = strdup("");
1509     }
1510 
1511     migrate_params_test_apply(params, &tmp);
1512 
1513     if (!migrate_params_check(&tmp, errp)) {
1514         /* Invalid parameter */
1515         return;
1516     }
1517 
1518     migrate_params_apply(params, errp);
1519 }
1520 
1521 
1522 void qmp_migrate_start_postcopy(Error **errp)
1523 {
1524     MigrationState *s = migrate_get_current();
1525 
1526     if (!migrate_postcopy()) {
1527         error_setg(errp, "Enable postcopy with migrate_set_capability before"
1528                          " the start of migration");
1529         return;
1530     }
1531 
1532     if (s->state == MIGRATION_STATUS_NONE) {
1533         error_setg(errp, "Postcopy must be started after migration has been"
1534                          " started");
1535         return;
1536     }
1537     /*
1538      * we don't error if migration has finished since that would be racy
1539      * with issuing this command.
1540      */
1541     atomic_set(&s->start_postcopy, true);
1542 }
1543 
1544 /* shared migration helpers */
1545 
1546 void migrate_set_state(int *state, int old_state, int new_state)
1547 {
1548     assert(new_state < MIGRATION_STATUS__MAX);
1549     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1550         trace_migrate_set_state(MigrationStatus_str(new_state));
1551         migrate_generate_event(new_state);
1552     }
1553 }
1554 
1555 static MigrationCapabilityStatusList *migrate_cap_add(
1556     MigrationCapabilityStatusList *list,
1557     MigrationCapability index,
1558     bool state)
1559 {
1560     MigrationCapabilityStatusList *cap;
1561 
1562     cap = g_new0(MigrationCapabilityStatusList, 1);
1563     cap->value = g_new0(MigrationCapabilityStatus, 1);
1564     cap->value->capability = index;
1565     cap->value->state = state;
1566     cap->next = list;
1567 
1568     return cap;
1569 }
1570 
1571 void migrate_set_block_enabled(bool value, Error **errp)
1572 {
1573     MigrationCapabilityStatusList *cap;
1574 
1575     cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1576     qmp_migrate_set_capabilities(cap, errp);
1577     qapi_free_MigrationCapabilityStatusList(cap);
1578 }
1579 
1580 static void migrate_set_block_incremental(MigrationState *s, bool value)
1581 {
1582     s->parameters.block_incremental = value;
1583 }
1584 
1585 static void block_cleanup_parameters(MigrationState *s)
1586 {
1587     if (s->must_remove_block_options) {
1588         /* setting to false can never fail */
1589         migrate_set_block_enabled(false, &error_abort);
1590         migrate_set_block_incremental(s, false);
1591         s->must_remove_block_options = false;
1592     }
1593 }
1594 
1595 static void migrate_fd_cleanup(MigrationState *s)
1596 {
1597     qemu_bh_delete(s->cleanup_bh);
1598     s->cleanup_bh = NULL;
1599 
1600     qemu_savevm_state_cleanup();
1601 
1602     if (s->to_dst_file) {
1603         QEMUFile *tmp;
1604 
1605         trace_migrate_fd_cleanup();
1606         qemu_mutex_unlock_iothread();
1607         if (s->migration_thread_running) {
1608             qemu_thread_join(&s->thread);
1609             s->migration_thread_running = false;
1610         }
1611         qemu_mutex_lock_iothread();
1612 
1613         multifd_save_cleanup();
1614         qemu_mutex_lock(&s->qemu_file_lock);
1615         tmp = s->to_dst_file;
1616         s->to_dst_file = NULL;
1617         qemu_mutex_unlock(&s->qemu_file_lock);
1618         /*
1619          * Close the file handle without the lock to make sure the
1620          * critical section won't block for long.
1621          */
1622         qemu_fclose(tmp);
1623     }
1624 
1625     assert(!migration_is_active(s));
1626 
1627     if (s->state == MIGRATION_STATUS_CANCELLING) {
1628         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1629                           MIGRATION_STATUS_CANCELLED);
1630     }
1631 
1632     if (s->error) {
1633         /* It is used on info migrate.  We can't free it */
1634         error_report_err(error_copy(s->error));
1635     }
1636     notifier_list_notify(&migration_state_notifiers, s);
1637     block_cleanup_parameters(s);
1638 }
1639 
1640 static void migrate_fd_cleanup_schedule(MigrationState *s)
1641 {
1642     /*
1643      * Ref the state for bh, because it may be called when
1644      * there're already no other refs
1645      */
1646     object_ref(OBJECT(s));
1647     qemu_bh_schedule(s->cleanup_bh);
1648 }
1649 
1650 static void migrate_fd_cleanup_bh(void *opaque)
1651 {
1652     MigrationState *s = opaque;
1653     migrate_fd_cleanup(s);
1654     object_unref(OBJECT(s));
1655 }
1656 
1657 void migrate_set_error(MigrationState *s, const Error *error)
1658 {
1659     QEMU_LOCK_GUARD(&s->error_mutex);
1660     if (!s->error) {
1661         s->error = error_copy(error);
1662     }
1663 }
1664 
1665 void migrate_fd_error(MigrationState *s, const Error *error)
1666 {
1667     trace_migrate_fd_error(error_get_pretty(error));
1668     assert(s->to_dst_file == NULL);
1669     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1670                       MIGRATION_STATUS_FAILED);
1671     migrate_set_error(s, error);
1672 }
1673 
1674 static void migrate_fd_cancel(MigrationState *s)
1675 {
1676     int old_state ;
1677     QEMUFile *f = migrate_get_current()->to_dst_file;
1678     trace_migrate_fd_cancel();
1679 
1680     if (s->rp_state.from_dst_file) {
1681         /* shutdown the rp socket, so causing the rp thread to shutdown */
1682         qemu_file_shutdown(s->rp_state.from_dst_file);
1683     }
1684 
1685     do {
1686         old_state = s->state;
1687         if (!migration_is_running(old_state)) {
1688             break;
1689         }
1690         /* If the migration is paused, kick it out of the pause */
1691         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1692             qemu_sem_post(&s->pause_sem);
1693         }
1694         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1695     } while (s->state != MIGRATION_STATUS_CANCELLING);
1696 
1697     /*
1698      * If we're unlucky the migration code might be stuck somewhere in a
1699      * send/write while the network has failed and is waiting to timeout;
1700      * if we've got shutdown(2) available then we can force it to quit.
1701      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1702      * called in a bh, so there is no race against this cancel.
1703      */
1704     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1705         qemu_file_shutdown(f);
1706     }
1707     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1708         Error *local_err = NULL;
1709 
1710         bdrv_invalidate_cache_all(&local_err);
1711         if (local_err) {
1712             error_report_err(local_err);
1713         } else {
1714             s->block_inactive = false;
1715         }
1716     }
1717 }
1718 
1719 void add_migration_state_change_notifier(Notifier *notify)
1720 {
1721     notifier_list_add(&migration_state_notifiers, notify);
1722 }
1723 
1724 void remove_migration_state_change_notifier(Notifier *notify)
1725 {
1726     notifier_remove(notify);
1727 }
1728 
1729 bool migration_in_setup(MigrationState *s)
1730 {
1731     return s->state == MIGRATION_STATUS_SETUP;
1732 }
1733 
1734 bool migration_has_finished(MigrationState *s)
1735 {
1736     return s->state == MIGRATION_STATUS_COMPLETED;
1737 }
1738 
1739 bool migration_has_failed(MigrationState *s)
1740 {
1741     return (s->state == MIGRATION_STATUS_CANCELLED ||
1742             s->state == MIGRATION_STATUS_FAILED);
1743 }
1744 
1745 bool migration_in_postcopy(void)
1746 {
1747     MigrationState *s = migrate_get_current();
1748 
1749     switch (s->state) {
1750     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1751     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1752     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1753         return true;
1754     default:
1755         return false;
1756     }
1757 }
1758 
1759 bool migration_in_postcopy_after_devices(MigrationState *s)
1760 {
1761     return migration_in_postcopy() && s->postcopy_after_devices;
1762 }
1763 
1764 bool migration_is_idle(void)
1765 {
1766     MigrationState *s = current_migration;
1767 
1768     if (!s) {
1769         return true;
1770     }
1771 
1772     switch (s->state) {
1773     case MIGRATION_STATUS_NONE:
1774     case MIGRATION_STATUS_CANCELLED:
1775     case MIGRATION_STATUS_COMPLETED:
1776     case MIGRATION_STATUS_FAILED:
1777         return true;
1778     case MIGRATION_STATUS_SETUP:
1779     case MIGRATION_STATUS_CANCELLING:
1780     case MIGRATION_STATUS_ACTIVE:
1781     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1782     case MIGRATION_STATUS_COLO:
1783     case MIGRATION_STATUS_PRE_SWITCHOVER:
1784     case MIGRATION_STATUS_DEVICE:
1785     case MIGRATION_STATUS_WAIT_UNPLUG:
1786         return false;
1787     case MIGRATION_STATUS__MAX:
1788         g_assert_not_reached();
1789     }
1790 
1791     return false;
1792 }
1793 
1794 bool migration_is_active(MigrationState *s)
1795 {
1796     return (s->state == MIGRATION_STATUS_ACTIVE ||
1797             s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1798 }
1799 
1800 void migrate_init(MigrationState *s)
1801 {
1802     /*
1803      * Reinitialise all migration state, except
1804      * parameters/capabilities that the user set, and
1805      * locks.
1806      */
1807     s->cleanup_bh = 0;
1808     s->to_dst_file = NULL;
1809     s->state = MIGRATION_STATUS_NONE;
1810     s->rp_state.from_dst_file = NULL;
1811     s->rp_state.error = false;
1812     s->mbps = 0.0;
1813     s->pages_per_second = 0.0;
1814     s->downtime = 0;
1815     s->expected_downtime = 0;
1816     s->setup_time = 0;
1817     s->start_postcopy = false;
1818     s->postcopy_after_devices = false;
1819     s->migration_thread_running = false;
1820     error_free(s->error);
1821     s->error = NULL;
1822 
1823     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1824 
1825     s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1826     s->total_time = 0;
1827     s->vm_was_running = false;
1828     s->iteration_initial_bytes = 0;
1829     s->threshold_size = 0;
1830 }
1831 
1832 static GSList *migration_blockers;
1833 
1834 int migrate_add_blocker(Error *reason, Error **errp)
1835 {
1836     if (only_migratable) {
1837         error_propagate_prepend(errp, error_copy(reason),
1838                                 "disallowing migration blocker "
1839                                 "(--only-migratable) for: ");
1840         return -EACCES;
1841     }
1842 
1843     if (migration_is_idle()) {
1844         migration_blockers = g_slist_prepend(migration_blockers, reason);
1845         return 0;
1846     }
1847 
1848     error_propagate_prepend(errp, error_copy(reason),
1849                             "disallowing migration blocker "
1850                             "(migration in progress) for: ");
1851     return -EBUSY;
1852 }
1853 
1854 void migrate_del_blocker(Error *reason)
1855 {
1856     migration_blockers = g_slist_remove(migration_blockers, reason);
1857 }
1858 
1859 void qmp_migrate_incoming(const char *uri, Error **errp)
1860 {
1861     Error *local_err = NULL;
1862     static bool once = true;
1863 
1864     if (!deferred_incoming) {
1865         error_setg(errp, "For use with '-incoming defer'");
1866         return;
1867     }
1868     if (!once) {
1869         error_setg(errp, "The incoming migration has already been started");
1870         return;
1871     }
1872 
1873     qemu_start_incoming_migration(uri, &local_err);
1874 
1875     if (local_err) {
1876         error_propagate(errp, local_err);
1877         return;
1878     }
1879 
1880     once = false;
1881 }
1882 
1883 void qmp_migrate_recover(const char *uri, Error **errp)
1884 {
1885     MigrationIncomingState *mis = migration_incoming_get_current();
1886 
1887     if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1888         error_setg(errp, "Migrate recover can only be run "
1889                    "when postcopy is paused.");
1890         return;
1891     }
1892 
1893     if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1894                        false, true) == true) {
1895         error_setg(errp, "Migrate recovery is triggered already");
1896         return;
1897     }
1898 
1899     /*
1900      * Note that this call will never start a real migration; it will
1901      * only re-setup the migration stream and poke existing migration
1902      * to continue using that newly established channel.
1903      */
1904     qemu_start_incoming_migration(uri, errp);
1905 }
1906 
1907 void qmp_migrate_pause(Error **errp)
1908 {
1909     MigrationState *ms = migrate_get_current();
1910     MigrationIncomingState *mis = migration_incoming_get_current();
1911     int ret;
1912 
1913     if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1914         /* Source side, during postcopy */
1915         qemu_mutex_lock(&ms->qemu_file_lock);
1916         ret = qemu_file_shutdown(ms->to_dst_file);
1917         qemu_mutex_unlock(&ms->qemu_file_lock);
1918         if (ret) {
1919             error_setg(errp, "Failed to pause source migration");
1920         }
1921         return;
1922     }
1923 
1924     if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1925         ret = qemu_file_shutdown(mis->from_src_file);
1926         if (ret) {
1927             error_setg(errp, "Failed to pause destination migration");
1928         }
1929         return;
1930     }
1931 
1932     error_setg(errp, "migrate-pause is currently only supported "
1933                "during postcopy-active state");
1934 }
1935 
1936 bool migration_is_blocked(Error **errp)
1937 {
1938     if (qemu_savevm_state_blocked(errp)) {
1939         return true;
1940     }
1941 
1942     if (migration_blockers) {
1943         error_propagate(errp, error_copy(migration_blockers->data));
1944         return true;
1945     }
1946 
1947     return false;
1948 }
1949 
1950 /* Returns true if continue to migrate, or false if error detected */
1951 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1952                             bool resume, Error **errp)
1953 {
1954     Error *local_err = NULL;
1955 
1956     if (resume) {
1957         if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1958             error_setg(errp, "Cannot resume if there is no "
1959                        "paused migration");
1960             return false;
1961         }
1962 
1963         /*
1964          * Postcopy recovery won't work well with release-ram
1965          * capability since release-ram will drop the page buffer as
1966          * long as the page is put into the send buffer.  So if there
1967          * is a network failure happened, any page buffers that have
1968          * not yet reached the destination VM but have already been
1969          * sent from the source VM will be lost forever.  Let's refuse
1970          * the client from resuming such a postcopy migration.
1971          * Luckily release-ram was designed to only be used when src
1972          * and destination VMs are on the same host, so it should be
1973          * fine.
1974          */
1975         if (migrate_release_ram()) {
1976             error_setg(errp, "Postcopy recovery cannot work "
1977                        "when release-ram capability is set");
1978             return false;
1979         }
1980 
1981         /* This is a resume, skip init status */
1982         return true;
1983     }
1984 
1985     if (migration_is_running(s->state)) {
1986         error_setg(errp, QERR_MIGRATION_ACTIVE);
1987         return false;
1988     }
1989 
1990     if (runstate_check(RUN_STATE_INMIGRATE)) {
1991         error_setg(errp, "Guest is waiting for an incoming migration");
1992         return false;
1993     }
1994 
1995     if (migration_is_blocked(errp)) {
1996         return false;
1997     }
1998 
1999     if (blk || blk_inc) {
2000         if (migrate_use_block() || migrate_use_block_incremental()) {
2001             error_setg(errp, "Command options are incompatible with "
2002                        "current migration capabilities");
2003             return false;
2004         }
2005         migrate_set_block_enabled(true, &local_err);
2006         if (local_err) {
2007             error_propagate(errp, local_err);
2008             return false;
2009         }
2010         s->must_remove_block_options = true;
2011     }
2012 
2013     if (blk_inc) {
2014         migrate_set_block_incremental(s, true);
2015     }
2016 
2017     migrate_init(s);
2018     /*
2019      * set ram_counters memory to zero for a
2020      * new migration
2021      */
2022     memset(&ram_counters, 0, sizeof(ram_counters));
2023 
2024     return true;
2025 }
2026 
2027 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2028                  bool has_inc, bool inc, bool has_detach, bool detach,
2029                  bool has_resume, bool resume, Error **errp)
2030 {
2031     Error *local_err = NULL;
2032     MigrationState *s = migrate_get_current();
2033     const char *p;
2034 
2035     if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2036                          has_resume && resume, errp)) {
2037         /* Error detected, put into errp */
2038         return;
2039     }
2040 
2041     if (strstart(uri, "tcp:", &p)) {
2042         tcp_start_outgoing_migration(s, p, &local_err);
2043 #ifdef CONFIG_RDMA
2044     } else if (strstart(uri, "rdma:", &p)) {
2045         rdma_start_outgoing_migration(s, p, &local_err);
2046 #endif
2047     } else if (strstart(uri, "exec:", &p)) {
2048         exec_start_outgoing_migration(s, p, &local_err);
2049     } else if (strstart(uri, "unix:", &p)) {
2050         unix_start_outgoing_migration(s, p, &local_err);
2051     } else if (strstart(uri, "fd:", &p)) {
2052         fd_start_outgoing_migration(s, p, &local_err);
2053     } else {
2054         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2055                    "a valid migration protocol");
2056         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2057                           MIGRATION_STATUS_FAILED);
2058         block_cleanup_parameters(s);
2059         return;
2060     }
2061 
2062     if (local_err) {
2063         migrate_fd_error(s, local_err);
2064         error_propagate(errp, local_err);
2065         return;
2066     }
2067 }
2068 
2069 void qmp_migrate_cancel(Error **errp)
2070 {
2071     migrate_fd_cancel(migrate_get_current());
2072 }
2073 
2074 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2075 {
2076     MigrationState *s = migrate_get_current();
2077     if (s->state != state) {
2078         error_setg(errp,  "Migration not in expected state: %s",
2079                    MigrationStatus_str(s->state));
2080         return;
2081     }
2082     qemu_sem_post(&s->pause_sem);
2083 }
2084 
2085 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
2086 {
2087     MigrateSetParameters p = {
2088         .has_xbzrle_cache_size = true,
2089         .xbzrle_cache_size = value,
2090     };
2091 
2092     qmp_migrate_set_parameters(&p, errp);
2093 }
2094 
2095 int64_t qmp_query_migrate_cache_size(Error **errp)
2096 {
2097     return migrate_xbzrle_cache_size();
2098 }
2099 
2100 void qmp_migrate_set_speed(int64_t value, Error **errp)
2101 {
2102     MigrateSetParameters p = {
2103         .has_max_bandwidth = true,
2104         .max_bandwidth = value,
2105     };
2106 
2107     qmp_migrate_set_parameters(&p, errp);
2108 }
2109 
2110 void qmp_migrate_set_downtime(double value, Error **errp)
2111 {
2112     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2113         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
2114                    "downtime_limit",
2115                    "an integer in the range of 0 to "
2116                     stringify(MAX_MIGRATE_DOWNTIME_SECONDS)" seconds");
2117         return;
2118     }
2119 
2120     value *= 1000; /* Convert to milliseconds */
2121 
2122     MigrateSetParameters p = {
2123         .has_downtime_limit = true,
2124         .downtime_limit = (int64_t)value,
2125     };
2126 
2127     qmp_migrate_set_parameters(&p, errp);
2128 }
2129 
2130 bool migrate_release_ram(void)
2131 {
2132     MigrationState *s;
2133 
2134     s = migrate_get_current();
2135 
2136     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2137 }
2138 
2139 bool migrate_postcopy_ram(void)
2140 {
2141     MigrationState *s;
2142 
2143     s = migrate_get_current();
2144 
2145     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2146 }
2147 
2148 bool migrate_postcopy(void)
2149 {
2150     return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2151 }
2152 
2153 bool migrate_auto_converge(void)
2154 {
2155     MigrationState *s;
2156 
2157     s = migrate_get_current();
2158 
2159     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2160 }
2161 
2162 bool migrate_zero_blocks(void)
2163 {
2164     MigrationState *s;
2165 
2166     s = migrate_get_current();
2167 
2168     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2169 }
2170 
2171 bool migrate_postcopy_blocktime(void)
2172 {
2173     MigrationState *s;
2174 
2175     s = migrate_get_current();
2176 
2177     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2178 }
2179 
2180 bool migrate_use_compression(void)
2181 {
2182     MigrationState *s;
2183 
2184     s = migrate_get_current();
2185 
2186     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2187 }
2188 
2189 int migrate_compress_level(void)
2190 {
2191     MigrationState *s;
2192 
2193     s = migrate_get_current();
2194 
2195     return s->parameters.compress_level;
2196 }
2197 
2198 int migrate_compress_threads(void)
2199 {
2200     MigrationState *s;
2201 
2202     s = migrate_get_current();
2203 
2204     return s->parameters.compress_threads;
2205 }
2206 
2207 int migrate_compress_wait_thread(void)
2208 {
2209     MigrationState *s;
2210 
2211     s = migrate_get_current();
2212 
2213     return s->parameters.compress_wait_thread;
2214 }
2215 
2216 int migrate_decompress_threads(void)
2217 {
2218     MigrationState *s;
2219 
2220     s = migrate_get_current();
2221 
2222     return s->parameters.decompress_threads;
2223 }
2224 
2225 bool migrate_dirty_bitmaps(void)
2226 {
2227     MigrationState *s;
2228 
2229     s = migrate_get_current();
2230 
2231     return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2232 }
2233 
2234 bool migrate_ignore_shared(void)
2235 {
2236     MigrationState *s;
2237 
2238     s = migrate_get_current();
2239 
2240     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2241 }
2242 
2243 bool migrate_validate_uuid(void)
2244 {
2245     MigrationState *s;
2246 
2247     s = migrate_get_current();
2248 
2249     return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2250 }
2251 
2252 bool migrate_use_events(void)
2253 {
2254     MigrationState *s;
2255 
2256     s = migrate_get_current();
2257 
2258     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2259 }
2260 
2261 bool migrate_use_multifd(void)
2262 {
2263     MigrationState *s;
2264 
2265     s = migrate_get_current();
2266 
2267     return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2268 }
2269 
2270 bool migrate_pause_before_switchover(void)
2271 {
2272     MigrationState *s;
2273 
2274     s = migrate_get_current();
2275 
2276     return s->enabled_capabilities[
2277         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2278 }
2279 
2280 int migrate_multifd_channels(void)
2281 {
2282     MigrationState *s;
2283 
2284     s = migrate_get_current();
2285 
2286     return s->parameters.multifd_channels;
2287 }
2288 
2289 MultiFDCompression migrate_multifd_compression(void)
2290 {
2291     MigrationState *s;
2292 
2293     s = migrate_get_current();
2294 
2295     return s->parameters.multifd_compression;
2296 }
2297 
2298 int migrate_multifd_zlib_level(void)
2299 {
2300     MigrationState *s;
2301 
2302     s = migrate_get_current();
2303 
2304     return s->parameters.multifd_zlib_level;
2305 }
2306 
2307 int migrate_multifd_zstd_level(void)
2308 {
2309     MigrationState *s;
2310 
2311     s = migrate_get_current();
2312 
2313     return s->parameters.multifd_zstd_level;
2314 }
2315 
2316 int migrate_use_xbzrle(void)
2317 {
2318     MigrationState *s;
2319 
2320     s = migrate_get_current();
2321 
2322     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2323 }
2324 
2325 int64_t migrate_xbzrle_cache_size(void)
2326 {
2327     MigrationState *s;
2328 
2329     s = migrate_get_current();
2330 
2331     return s->parameters.xbzrle_cache_size;
2332 }
2333 
2334 static int64_t migrate_max_postcopy_bandwidth(void)
2335 {
2336     MigrationState *s;
2337 
2338     s = migrate_get_current();
2339 
2340     return s->parameters.max_postcopy_bandwidth;
2341 }
2342 
2343 bool migrate_use_block(void)
2344 {
2345     MigrationState *s;
2346 
2347     s = migrate_get_current();
2348 
2349     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2350 }
2351 
2352 bool migrate_use_return_path(void)
2353 {
2354     MigrationState *s;
2355 
2356     s = migrate_get_current();
2357 
2358     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2359 }
2360 
2361 bool migrate_use_block_incremental(void)
2362 {
2363     MigrationState *s;
2364 
2365     s = migrate_get_current();
2366 
2367     return s->parameters.block_incremental;
2368 }
2369 
2370 /* migration thread support */
2371 /*
2372  * Something bad happened to the RP stream, mark an error
2373  * The caller shall print or trace something to indicate why
2374  */
2375 static void mark_source_rp_bad(MigrationState *s)
2376 {
2377     s->rp_state.error = true;
2378 }
2379 
2380 static struct rp_cmd_args {
2381     ssize_t     len; /* -1 = variable */
2382     const char *name;
2383 } rp_cmd_args[] = {
2384     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
2385     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
2386     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
2387     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
2388     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
2389     [MIG_RP_MSG_RECV_BITMAP]    = { .len = -1, .name = "RECV_BITMAP" },
2390     [MIG_RP_MSG_RESUME_ACK]     = { .len =  4, .name = "RESUME_ACK" },
2391     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
2392 };
2393 
2394 /*
2395  * Process a request for pages received on the return path,
2396  * We're allowed to send more than requested (e.g. to round to our page size)
2397  * and we don't need to send pages that have already been sent.
2398  */
2399 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2400                                        ram_addr_t start, size_t len)
2401 {
2402     long our_host_ps = qemu_real_host_page_size;
2403 
2404     trace_migrate_handle_rp_req_pages(rbname, start, len);
2405 
2406     /*
2407      * Since we currently insist on matching page sizes, just sanity check
2408      * we're being asked for whole host pages.
2409      */
2410     if (start & (our_host_ps-1) ||
2411        (len & (our_host_ps-1))) {
2412         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2413                      " len: %zd", __func__, start, len);
2414         mark_source_rp_bad(ms);
2415         return;
2416     }
2417 
2418     if (ram_save_queue_pages(rbname, start, len)) {
2419         mark_source_rp_bad(ms);
2420     }
2421 }
2422 
2423 /* Return true to retry, false to quit */
2424 static bool postcopy_pause_return_path_thread(MigrationState *s)
2425 {
2426     trace_postcopy_pause_return_path();
2427 
2428     qemu_sem_wait(&s->postcopy_pause_rp_sem);
2429 
2430     trace_postcopy_pause_return_path_continued();
2431 
2432     return true;
2433 }
2434 
2435 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2436 {
2437     RAMBlock *block = qemu_ram_block_by_name(block_name);
2438 
2439     if (!block) {
2440         error_report("%s: invalid block name '%s'", __func__, block_name);
2441         return -EINVAL;
2442     }
2443 
2444     /* Fetch the received bitmap and refresh the dirty bitmap */
2445     return ram_dirty_bitmap_reload(s, block);
2446 }
2447 
2448 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2449 {
2450     trace_source_return_path_thread_resume_ack(value);
2451 
2452     if (value != MIGRATION_RESUME_ACK_VALUE) {
2453         error_report("%s: illegal resume_ack value %"PRIu32,
2454                      __func__, value);
2455         return -1;
2456     }
2457 
2458     /* Now both sides are active. */
2459     migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2460                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2461 
2462     /* Notify send thread that time to continue send pages */
2463     qemu_sem_post(&s->rp_state.rp_sem);
2464 
2465     return 0;
2466 }
2467 
2468 /*
2469  * Handles messages sent on the return path towards the source VM
2470  *
2471  */
2472 static void *source_return_path_thread(void *opaque)
2473 {
2474     MigrationState *ms = opaque;
2475     QEMUFile *rp = ms->rp_state.from_dst_file;
2476     uint16_t header_len, header_type;
2477     uint8_t buf[512];
2478     uint32_t tmp32, sibling_error;
2479     ram_addr_t start = 0; /* =0 to silence warning */
2480     size_t  len = 0, expected_len;
2481     int res;
2482 
2483     trace_source_return_path_thread_entry();
2484     rcu_register_thread();
2485 
2486 retry:
2487     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2488            migration_is_setup_or_active(ms->state)) {
2489         trace_source_return_path_thread_loop_top();
2490         header_type = qemu_get_be16(rp);
2491         header_len = qemu_get_be16(rp);
2492 
2493         if (qemu_file_get_error(rp)) {
2494             mark_source_rp_bad(ms);
2495             goto out;
2496         }
2497 
2498         if (header_type >= MIG_RP_MSG_MAX ||
2499             header_type == MIG_RP_MSG_INVALID) {
2500             error_report("RP: Received invalid message 0x%04x length 0x%04x",
2501                          header_type, header_len);
2502             mark_source_rp_bad(ms);
2503             goto out;
2504         }
2505 
2506         if ((rp_cmd_args[header_type].len != -1 &&
2507             header_len != rp_cmd_args[header_type].len) ||
2508             header_len > sizeof(buf)) {
2509             error_report("RP: Received '%s' message (0x%04x) with"
2510                          "incorrect length %d expecting %zu",
2511                          rp_cmd_args[header_type].name, header_type, header_len,
2512                          (size_t)rp_cmd_args[header_type].len);
2513             mark_source_rp_bad(ms);
2514             goto out;
2515         }
2516 
2517         /* We know we've got a valid header by this point */
2518         res = qemu_get_buffer(rp, buf, header_len);
2519         if (res != header_len) {
2520             error_report("RP: Failed reading data for message 0x%04x"
2521                          " read %d expected %d",
2522                          header_type, res, header_len);
2523             mark_source_rp_bad(ms);
2524             goto out;
2525         }
2526 
2527         /* OK, we have the message and the data */
2528         switch (header_type) {
2529         case MIG_RP_MSG_SHUT:
2530             sibling_error = ldl_be_p(buf);
2531             trace_source_return_path_thread_shut(sibling_error);
2532             if (sibling_error) {
2533                 error_report("RP: Sibling indicated error %d", sibling_error);
2534                 mark_source_rp_bad(ms);
2535             }
2536             /*
2537              * We'll let the main thread deal with closing the RP
2538              * we could do a shutdown(2) on it, but we're the only user
2539              * anyway, so there's nothing gained.
2540              */
2541             goto out;
2542 
2543         case MIG_RP_MSG_PONG:
2544             tmp32 = ldl_be_p(buf);
2545             trace_source_return_path_thread_pong(tmp32);
2546             break;
2547 
2548         case MIG_RP_MSG_REQ_PAGES:
2549             start = ldq_be_p(buf);
2550             len = ldl_be_p(buf + 8);
2551             migrate_handle_rp_req_pages(ms, NULL, start, len);
2552             break;
2553 
2554         case MIG_RP_MSG_REQ_PAGES_ID:
2555             expected_len = 12 + 1; /* header + termination */
2556 
2557             if (header_len >= expected_len) {
2558                 start = ldq_be_p(buf);
2559                 len = ldl_be_p(buf + 8);
2560                 /* Now we expect an idstr */
2561                 tmp32 = buf[12]; /* Length of the following idstr */
2562                 buf[13 + tmp32] = '\0';
2563                 expected_len += tmp32;
2564             }
2565             if (header_len != expected_len) {
2566                 error_report("RP: Req_Page_id with length %d expecting %zd",
2567                              header_len, expected_len);
2568                 mark_source_rp_bad(ms);
2569                 goto out;
2570             }
2571             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2572             break;
2573 
2574         case MIG_RP_MSG_RECV_BITMAP:
2575             if (header_len < 1) {
2576                 error_report("%s: missing block name", __func__);
2577                 mark_source_rp_bad(ms);
2578                 goto out;
2579             }
2580             /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2581             buf[buf[0] + 1] = '\0';
2582             if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2583                 mark_source_rp_bad(ms);
2584                 goto out;
2585             }
2586             break;
2587 
2588         case MIG_RP_MSG_RESUME_ACK:
2589             tmp32 = ldl_be_p(buf);
2590             if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2591                 mark_source_rp_bad(ms);
2592                 goto out;
2593             }
2594             break;
2595 
2596         default:
2597             break;
2598         }
2599     }
2600 
2601 out:
2602     res = qemu_file_get_error(rp);
2603     if (res) {
2604         if (res == -EIO && migration_in_postcopy()) {
2605             /*
2606              * Maybe there is something we can do: it looks like a
2607              * network down issue, and we pause for a recovery.
2608              */
2609             if (postcopy_pause_return_path_thread(ms)) {
2610                 /* Reload rp, reset the rest */
2611                 if (rp != ms->rp_state.from_dst_file) {
2612                     qemu_fclose(rp);
2613                     rp = ms->rp_state.from_dst_file;
2614                 }
2615                 ms->rp_state.error = false;
2616                 goto retry;
2617             }
2618         }
2619 
2620         trace_source_return_path_thread_bad_end();
2621         mark_source_rp_bad(ms);
2622     }
2623 
2624     trace_source_return_path_thread_end();
2625     ms->rp_state.from_dst_file = NULL;
2626     qemu_fclose(rp);
2627     rcu_unregister_thread();
2628     return NULL;
2629 }
2630 
2631 static int open_return_path_on_source(MigrationState *ms,
2632                                       bool create_thread)
2633 {
2634 
2635     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2636     if (!ms->rp_state.from_dst_file) {
2637         return -1;
2638     }
2639 
2640     trace_open_return_path_on_source();
2641 
2642     if (!create_thread) {
2643         /* We're done */
2644         return 0;
2645     }
2646 
2647     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2648                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2649 
2650     trace_open_return_path_on_source_continue();
2651 
2652     return 0;
2653 }
2654 
2655 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2656 static int await_return_path_close_on_source(MigrationState *ms)
2657 {
2658     /*
2659      * If this is a normal exit then the destination will send a SHUT and the
2660      * rp_thread will exit, however if there's an error we need to cause
2661      * it to exit.
2662      */
2663     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2664         /*
2665          * shutdown(2), if we have it, will cause it to unblock if it's stuck
2666          * waiting for the destination.
2667          */
2668         qemu_file_shutdown(ms->rp_state.from_dst_file);
2669         mark_source_rp_bad(ms);
2670     }
2671     trace_await_return_path_close_on_source_joining();
2672     qemu_thread_join(&ms->rp_state.rp_thread);
2673     trace_await_return_path_close_on_source_close();
2674     return ms->rp_state.error;
2675 }
2676 
2677 /*
2678  * Switch from normal iteration to postcopy
2679  * Returns non-0 on error
2680  */
2681 static int postcopy_start(MigrationState *ms)
2682 {
2683     int ret;
2684     QIOChannelBuffer *bioc;
2685     QEMUFile *fb;
2686     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2687     int64_t bandwidth = migrate_max_postcopy_bandwidth();
2688     bool restart_block = false;
2689     int cur_state = MIGRATION_STATUS_ACTIVE;
2690     if (!migrate_pause_before_switchover()) {
2691         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2692                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
2693     }
2694 
2695     trace_postcopy_start();
2696     qemu_mutex_lock_iothread();
2697     trace_postcopy_start_set_run();
2698 
2699     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2700     global_state_store();
2701     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2702     if (ret < 0) {
2703         goto fail;
2704     }
2705 
2706     ret = migration_maybe_pause(ms, &cur_state,
2707                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2708     if (ret < 0) {
2709         goto fail;
2710     }
2711 
2712     ret = bdrv_inactivate_all();
2713     if (ret < 0) {
2714         goto fail;
2715     }
2716     restart_block = true;
2717 
2718     /*
2719      * Cause any non-postcopiable, but iterative devices to
2720      * send out their final data.
2721      */
2722     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2723 
2724     /*
2725      * in Finish migrate and with the io-lock held everything should
2726      * be quiet, but we've potentially still got dirty pages and we
2727      * need to tell the destination to throw any pages it's already received
2728      * that are dirty
2729      */
2730     if (migrate_postcopy_ram()) {
2731         if (ram_postcopy_send_discard_bitmap(ms)) {
2732             error_report("postcopy send discard bitmap failed");
2733             goto fail;
2734         }
2735     }
2736 
2737     /*
2738      * send rest of state - note things that are doing postcopy
2739      * will notice we're in POSTCOPY_ACTIVE and not actually
2740      * wrap their state up here
2741      */
2742     /* 0 max-postcopy-bandwidth means unlimited */
2743     if (!bandwidth) {
2744         qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2745     } else {
2746         qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2747     }
2748     if (migrate_postcopy_ram()) {
2749         /* Ping just for debugging, helps line traces up */
2750         qemu_savevm_send_ping(ms->to_dst_file, 2);
2751     }
2752 
2753     /*
2754      * While loading the device state we may trigger page transfer
2755      * requests and the fd must be free to process those, and thus
2756      * the destination must read the whole device state off the fd before
2757      * it starts processing it.  Unfortunately the ad-hoc migration format
2758      * doesn't allow the destination to know the size to read without fully
2759      * parsing it through each devices load-state code (especially the open
2760      * coded devices that use get/put).
2761      * So we wrap the device state up in a package with a length at the start;
2762      * to do this we use a qemu_buf to hold the whole of the device state.
2763      */
2764     bioc = qio_channel_buffer_new(4096);
2765     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2766     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2767     object_unref(OBJECT(bioc));
2768 
2769     /*
2770      * Make sure the receiver can get incoming pages before we send the rest
2771      * of the state
2772      */
2773     qemu_savevm_send_postcopy_listen(fb);
2774 
2775     qemu_savevm_state_complete_precopy(fb, false, false);
2776     if (migrate_postcopy_ram()) {
2777         qemu_savevm_send_ping(fb, 3);
2778     }
2779 
2780     qemu_savevm_send_postcopy_run(fb);
2781 
2782     /* <><> end of stuff going into the package */
2783 
2784     /* Last point of recovery; as soon as we send the package the destination
2785      * can open devices and potentially start running.
2786      * Lets just check again we've not got any errors.
2787      */
2788     ret = qemu_file_get_error(ms->to_dst_file);
2789     if (ret) {
2790         error_report("postcopy_start: Migration stream errored (pre package)");
2791         goto fail_closefb;
2792     }
2793 
2794     restart_block = false;
2795 
2796     /* Now send that blob */
2797     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2798         goto fail_closefb;
2799     }
2800     qemu_fclose(fb);
2801 
2802     /* Send a notify to give a chance for anything that needs to happen
2803      * at the transition to postcopy and after the device state; in particular
2804      * spice needs to trigger a transition now
2805      */
2806     ms->postcopy_after_devices = true;
2807     notifier_list_notify(&migration_state_notifiers, ms);
2808 
2809     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2810 
2811     qemu_mutex_unlock_iothread();
2812 
2813     if (migrate_postcopy_ram()) {
2814         /*
2815          * Although this ping is just for debug, it could potentially be
2816          * used for getting a better measurement of downtime at the source.
2817          */
2818         qemu_savevm_send_ping(ms->to_dst_file, 4);
2819     }
2820 
2821     if (migrate_release_ram()) {
2822         ram_postcopy_migrated_memory_release(ms);
2823     }
2824 
2825     ret = qemu_file_get_error(ms->to_dst_file);
2826     if (ret) {
2827         error_report("postcopy_start: Migration stream errored");
2828         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2829                               MIGRATION_STATUS_FAILED);
2830     }
2831 
2832     return ret;
2833 
2834 fail_closefb:
2835     qemu_fclose(fb);
2836 fail:
2837     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2838                           MIGRATION_STATUS_FAILED);
2839     if (restart_block) {
2840         /* A failure happened early enough that we know the destination hasn't
2841          * accessed block devices, so we're safe to recover.
2842          */
2843         Error *local_err = NULL;
2844 
2845         bdrv_invalidate_cache_all(&local_err);
2846         if (local_err) {
2847             error_report_err(local_err);
2848         }
2849     }
2850     qemu_mutex_unlock_iothread();
2851     return -1;
2852 }
2853 
2854 /**
2855  * migration_maybe_pause: Pause if required to by
2856  * migrate_pause_before_switchover called with the iothread locked
2857  * Returns: 0 on success
2858  */
2859 static int migration_maybe_pause(MigrationState *s,
2860                                  int *current_active_state,
2861                                  int new_state)
2862 {
2863     if (!migrate_pause_before_switchover()) {
2864         return 0;
2865     }
2866 
2867     /* Since leaving this state is not atomic with posting the semaphore
2868      * it's possible that someone could have issued multiple migrate_continue
2869      * and the semaphore is incorrectly positive at this point;
2870      * the docs say it's undefined to reinit a semaphore that's already
2871      * init'd, so use timedwait to eat up any existing posts.
2872      */
2873     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2874         /* This block intentionally left blank */
2875     }
2876 
2877     /*
2878      * If the migration is cancelled when it is in the completion phase,
2879      * the migration state is set to MIGRATION_STATUS_CANCELLING.
2880      * So we don't need to wait a semaphore, otherwise we would always
2881      * wait for the 'pause_sem' semaphore.
2882      */
2883     if (s->state != MIGRATION_STATUS_CANCELLING) {
2884         qemu_mutex_unlock_iothread();
2885         migrate_set_state(&s->state, *current_active_state,
2886                           MIGRATION_STATUS_PRE_SWITCHOVER);
2887         qemu_sem_wait(&s->pause_sem);
2888         migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2889                           new_state);
2890         *current_active_state = new_state;
2891         qemu_mutex_lock_iothread();
2892     }
2893 
2894     return s->state == new_state ? 0 : -EINVAL;
2895 }
2896 
2897 /**
2898  * migration_completion: Used by migration_thread when there's not much left.
2899  *   The caller 'breaks' the loop when this returns.
2900  *
2901  * @s: Current migration state
2902  */
2903 static void migration_completion(MigrationState *s)
2904 {
2905     int ret;
2906     int current_active_state = s->state;
2907 
2908     if (s->state == MIGRATION_STATUS_ACTIVE) {
2909         qemu_mutex_lock_iothread();
2910         s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2911         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2912         s->vm_was_running = runstate_is_running();
2913         ret = global_state_store();
2914 
2915         if (!ret) {
2916             bool inactivate = !migrate_colo_enabled();
2917             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2918             if (ret >= 0) {
2919                 ret = migration_maybe_pause(s, &current_active_state,
2920                                             MIGRATION_STATUS_DEVICE);
2921             }
2922             if (ret >= 0) {
2923                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2924                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2925                                                          inactivate);
2926             }
2927             if (inactivate && ret >= 0) {
2928                 s->block_inactive = true;
2929             }
2930         }
2931         qemu_mutex_unlock_iothread();
2932 
2933         if (ret < 0) {
2934             goto fail;
2935         }
2936     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2937         trace_migration_completion_postcopy_end();
2938 
2939         qemu_savevm_state_complete_postcopy(s->to_dst_file);
2940         trace_migration_completion_postcopy_end_after_complete();
2941     }
2942 
2943     /*
2944      * If rp was opened we must clean up the thread before
2945      * cleaning everything else up (since if there are no failures
2946      * it will wait for the destination to send it's status in
2947      * a SHUT command).
2948      */
2949     if (s->rp_state.from_dst_file) {
2950         int rp_error;
2951         trace_migration_return_path_end_before();
2952         rp_error = await_return_path_close_on_source(s);
2953         trace_migration_return_path_end_after(rp_error);
2954         if (rp_error) {
2955             goto fail_invalidate;
2956         }
2957     }
2958 
2959     if (qemu_file_get_error(s->to_dst_file)) {
2960         trace_migration_completion_file_err();
2961         goto fail_invalidate;
2962     }
2963 
2964     if (!migrate_colo_enabled()) {
2965         migrate_set_state(&s->state, current_active_state,
2966                           MIGRATION_STATUS_COMPLETED);
2967     }
2968 
2969     return;
2970 
2971 fail_invalidate:
2972     /* If not doing postcopy, vm_start() will be called: let's regain
2973      * control on images.
2974      */
2975     if (s->state == MIGRATION_STATUS_ACTIVE ||
2976         s->state == MIGRATION_STATUS_DEVICE) {
2977         Error *local_err = NULL;
2978 
2979         qemu_mutex_lock_iothread();
2980         bdrv_invalidate_cache_all(&local_err);
2981         if (local_err) {
2982             error_report_err(local_err);
2983         } else {
2984             s->block_inactive = false;
2985         }
2986         qemu_mutex_unlock_iothread();
2987     }
2988 
2989 fail:
2990     migrate_set_state(&s->state, current_active_state,
2991                       MIGRATION_STATUS_FAILED);
2992 }
2993 
2994 bool migrate_colo_enabled(void)
2995 {
2996     MigrationState *s = migrate_get_current();
2997     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2998 }
2999 
3000 typedef enum MigThrError {
3001     /* No error detected */
3002     MIG_THR_ERR_NONE = 0,
3003     /* Detected error, but resumed successfully */
3004     MIG_THR_ERR_RECOVERED = 1,
3005     /* Detected fatal error, need to exit */
3006     MIG_THR_ERR_FATAL = 2,
3007 } MigThrError;
3008 
3009 static int postcopy_resume_handshake(MigrationState *s)
3010 {
3011     qemu_savevm_send_postcopy_resume(s->to_dst_file);
3012 
3013     while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3014         qemu_sem_wait(&s->rp_state.rp_sem);
3015     }
3016 
3017     if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3018         return 0;
3019     }
3020 
3021     return -1;
3022 }
3023 
3024 /* Return zero if success, or <0 for error */
3025 static int postcopy_do_resume(MigrationState *s)
3026 {
3027     int ret;
3028 
3029     /*
3030      * Call all the resume_prepare() hooks, so that modules can be
3031      * ready for the migration resume.
3032      */
3033     ret = qemu_savevm_state_resume_prepare(s);
3034     if (ret) {
3035         error_report("%s: resume_prepare() failure detected: %d",
3036                      __func__, ret);
3037         return ret;
3038     }
3039 
3040     /*
3041      * Last handshake with destination on the resume (destination will
3042      * switch to postcopy-active afterwards)
3043      */
3044     ret = postcopy_resume_handshake(s);
3045     if (ret) {
3046         error_report("%s: handshake failed: %d", __func__, ret);
3047         return ret;
3048     }
3049 
3050     return 0;
3051 }
3052 
3053 /*
3054  * We don't return until we are in a safe state to continue current
3055  * postcopy migration.  Returns MIG_THR_ERR_RECOVERED if recovered, or
3056  * MIG_THR_ERR_FATAL if unrecovery failure happened.
3057  */
3058 static MigThrError postcopy_pause(MigrationState *s)
3059 {
3060     assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3061 
3062     while (true) {
3063         QEMUFile *file;
3064 
3065         migrate_set_state(&s->state, s->state,
3066                           MIGRATION_STATUS_POSTCOPY_PAUSED);
3067 
3068         /* Current channel is possibly broken. Release it. */
3069         assert(s->to_dst_file);
3070         qemu_mutex_lock(&s->qemu_file_lock);
3071         file = s->to_dst_file;
3072         s->to_dst_file = NULL;
3073         qemu_mutex_unlock(&s->qemu_file_lock);
3074 
3075         qemu_file_shutdown(file);
3076         qemu_fclose(file);
3077 
3078         error_report("Detected IO failure for postcopy. "
3079                      "Migration paused.");
3080 
3081         /*
3082          * We wait until things fixed up. Then someone will setup the
3083          * status back for us.
3084          */
3085         while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3086             qemu_sem_wait(&s->postcopy_pause_sem);
3087         }
3088 
3089         if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3090             /* Woken up by a recover procedure. Give it a shot */
3091 
3092             /*
3093              * Firstly, let's wake up the return path now, with a new
3094              * return path channel.
3095              */
3096             qemu_sem_post(&s->postcopy_pause_rp_sem);
3097 
3098             /* Do the resume logic */
3099             if (postcopy_do_resume(s) == 0) {
3100                 /* Let's continue! */
3101                 trace_postcopy_pause_continued();
3102                 return MIG_THR_ERR_RECOVERED;
3103             } else {
3104                 /*
3105                  * Something wrong happened during the recovery, let's
3106                  * pause again. Pause is always better than throwing
3107                  * data away.
3108                  */
3109                 continue;
3110             }
3111         } else {
3112             /* This is not right... Time to quit. */
3113             return MIG_THR_ERR_FATAL;
3114         }
3115     }
3116 }
3117 
3118 static MigThrError migration_detect_error(MigrationState *s)
3119 {
3120     int ret;
3121     int state = s->state;
3122     Error *local_error = NULL;
3123 
3124     if (state == MIGRATION_STATUS_CANCELLING ||
3125         state == MIGRATION_STATUS_CANCELLED) {
3126         /* End the migration, but don't set the state to failed */
3127         return MIG_THR_ERR_FATAL;
3128     }
3129 
3130     /* Try to detect any file errors */
3131     ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3132     if (!ret) {
3133         /* Everything is fine */
3134         assert(!local_error);
3135         return MIG_THR_ERR_NONE;
3136     }
3137 
3138     if (local_error) {
3139         migrate_set_error(s, local_error);
3140         error_free(local_error);
3141     }
3142 
3143     if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3144         /*
3145          * For postcopy, we allow the network to be down for a
3146          * while. After that, it can be continued by a
3147          * recovery phase.
3148          */
3149         return postcopy_pause(s);
3150     } else {
3151         /*
3152          * For precopy (or postcopy with error outside IO), we fail
3153          * with no time.
3154          */
3155         migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3156         trace_migration_thread_file_err();
3157 
3158         /* Time to stop the migration, now. */
3159         return MIG_THR_ERR_FATAL;
3160     }
3161 }
3162 
3163 /* How many bytes have we transferred since the beginning of the migration */
3164 static uint64_t migration_total_bytes(MigrationState *s)
3165 {
3166     return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3167 }
3168 
3169 static void migration_calculate_complete(MigrationState *s)
3170 {
3171     uint64_t bytes = migration_total_bytes(s);
3172     int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3173     int64_t transfer_time;
3174 
3175     s->total_time = end_time - s->start_time;
3176     if (!s->downtime) {
3177         /*
3178          * It's still not set, so we are precopy migration.  For
3179          * postcopy, downtime is calculated during postcopy_start().
3180          */
3181         s->downtime = end_time - s->downtime_start;
3182     }
3183 
3184     transfer_time = s->total_time - s->setup_time;
3185     if (transfer_time) {
3186         s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3187     }
3188 }
3189 
3190 static void update_iteration_initial_status(MigrationState *s)
3191 {
3192     /*
3193      * Update these three fields at the same time to avoid mismatch info lead
3194      * wrong speed calculation.
3195      */
3196     s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3197     s->iteration_initial_bytes = migration_total_bytes(s);
3198     s->iteration_initial_pages = ram_get_total_transferred_pages();
3199 }
3200 
3201 static void migration_update_counters(MigrationState *s,
3202                                       int64_t current_time)
3203 {
3204     uint64_t transferred, transferred_pages, time_spent;
3205     uint64_t current_bytes; /* bytes transferred since the beginning */
3206     double bandwidth;
3207 
3208     if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3209         return;
3210     }
3211 
3212     current_bytes = migration_total_bytes(s);
3213     transferred = current_bytes - s->iteration_initial_bytes;
3214     time_spent = current_time - s->iteration_start_time;
3215     bandwidth = (double)transferred / time_spent;
3216     s->threshold_size = bandwidth * s->parameters.downtime_limit;
3217 
3218     s->mbps = (((double) transferred * 8.0) /
3219                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3220 
3221     transferred_pages = ram_get_total_transferred_pages() -
3222                             s->iteration_initial_pages;
3223     s->pages_per_second = (double) transferred_pages /
3224                              (((double) time_spent / 1000.0));
3225 
3226     /*
3227      * if we haven't sent anything, we don't want to
3228      * recalculate. 10000 is a small enough number for our purposes
3229      */
3230     if (ram_counters.dirty_pages_rate && transferred > 10000) {
3231         s->expected_downtime = ram_counters.remaining / bandwidth;
3232     }
3233 
3234     qemu_file_reset_rate_limit(s->to_dst_file);
3235 
3236     update_iteration_initial_status(s);
3237 
3238     trace_migrate_transferred(transferred, time_spent,
3239                               bandwidth, s->threshold_size);
3240 }
3241 
3242 /* Migration thread iteration status */
3243 typedef enum {
3244     MIG_ITERATE_RESUME,         /* Resume current iteration */
3245     MIG_ITERATE_SKIP,           /* Skip current iteration */
3246     MIG_ITERATE_BREAK,          /* Break the loop */
3247 } MigIterateState;
3248 
3249 /*
3250  * Return true if continue to the next iteration directly, false
3251  * otherwise.
3252  */
3253 static MigIterateState migration_iteration_run(MigrationState *s)
3254 {
3255     uint64_t pending_size, pend_pre, pend_compat, pend_post;
3256     bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3257 
3258     qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3259                               &pend_compat, &pend_post);
3260     pending_size = pend_pre + pend_compat + pend_post;
3261 
3262     trace_migrate_pending(pending_size, s->threshold_size,
3263                           pend_pre, pend_compat, pend_post);
3264 
3265     if (pending_size && pending_size >= s->threshold_size) {
3266         /* Still a significant amount to transfer */
3267         if (!in_postcopy && pend_pre <= s->threshold_size &&
3268             atomic_read(&s->start_postcopy)) {
3269             if (postcopy_start(s)) {
3270                 error_report("%s: postcopy failed to start", __func__);
3271             }
3272             return MIG_ITERATE_SKIP;
3273         }
3274         /* Just another iteration step */
3275         qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3276     } else {
3277         trace_migration_thread_low_pending(pending_size);
3278         migration_completion(s);
3279         return MIG_ITERATE_BREAK;
3280     }
3281 
3282     return MIG_ITERATE_RESUME;
3283 }
3284 
3285 static void migration_iteration_finish(MigrationState *s)
3286 {
3287     /* If we enabled cpu throttling for auto-converge, turn it off. */
3288     cpu_throttle_stop();
3289 
3290     qemu_mutex_lock_iothread();
3291     switch (s->state) {
3292     case MIGRATION_STATUS_COMPLETED:
3293         migration_calculate_complete(s);
3294         runstate_set(RUN_STATE_POSTMIGRATE);
3295         break;
3296 
3297     case MIGRATION_STATUS_ACTIVE:
3298         /*
3299          * We should really assert here, but since it's during
3300          * migration, let's try to reduce the usage of assertions.
3301          */
3302         if (!migrate_colo_enabled()) {
3303             error_report("%s: critical error: calling COLO code without "
3304                          "COLO enabled", __func__);
3305         }
3306         migrate_start_colo_process(s);
3307         /*
3308          * Fixme: we will run VM in COLO no matter its old running state.
3309          * After exited COLO, we will keep running.
3310          */
3311         s->vm_was_running = true;
3312         /* Fallthrough */
3313     case MIGRATION_STATUS_FAILED:
3314     case MIGRATION_STATUS_CANCELLED:
3315     case MIGRATION_STATUS_CANCELLING:
3316         if (s->vm_was_running) {
3317             vm_start();
3318         } else {
3319             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3320                 runstate_set(RUN_STATE_POSTMIGRATE);
3321             }
3322         }
3323         break;
3324 
3325     default:
3326         /* Should not reach here, but if so, forgive the VM. */
3327         error_report("%s: Unknown ending state %d", __func__, s->state);
3328         break;
3329     }
3330     migrate_fd_cleanup_schedule(s);
3331     qemu_mutex_unlock_iothread();
3332 }
3333 
3334 void migration_make_urgent_request(void)
3335 {
3336     qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3337 }
3338 
3339 void migration_consume_urgent_request(void)
3340 {
3341     qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3342 }
3343 
3344 /* Returns true if the rate limiting was broken by an urgent request */
3345 bool migration_rate_limit(void)
3346 {
3347     int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3348     MigrationState *s = migrate_get_current();
3349 
3350     bool urgent = false;
3351     migration_update_counters(s, now);
3352     if (qemu_file_rate_limit(s->to_dst_file)) {
3353         /*
3354          * Wait for a delay to do rate limiting OR
3355          * something urgent to post the semaphore.
3356          */
3357         int ms = s->iteration_start_time + BUFFER_DELAY - now;
3358         trace_migration_rate_limit_pre(ms);
3359         if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3360             /*
3361              * We were woken by one or more urgent things but
3362              * the timedwait will have consumed one of them.
3363              * The service routine for the urgent wake will dec
3364              * the semaphore itself for each item it consumes,
3365              * so add this one we just eat back.
3366              */
3367             qemu_sem_post(&s->rate_limit_sem);
3368             urgent = true;
3369         }
3370         trace_migration_rate_limit_post(urgent);
3371     }
3372     return urgent;
3373 }
3374 
3375 /*
3376  * Master migration thread on the source VM.
3377  * It drives the migration and pumps the data down the outgoing channel.
3378  */
3379 static void *migration_thread(void *opaque)
3380 {
3381     MigrationState *s = opaque;
3382     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3383     MigThrError thr_error;
3384     bool urgent = false;
3385 
3386     rcu_register_thread();
3387 
3388     object_ref(OBJECT(s));
3389     update_iteration_initial_status(s);
3390 
3391     qemu_savevm_state_header(s->to_dst_file);
3392 
3393     /*
3394      * If we opened the return path, we need to make sure dst has it
3395      * opened as well.
3396      */
3397     if (s->rp_state.from_dst_file) {
3398         /* Now tell the dest that it should open its end so it can reply */
3399         qemu_savevm_send_open_return_path(s->to_dst_file);
3400 
3401         /* And do a ping that will make stuff easier to debug */
3402         qemu_savevm_send_ping(s->to_dst_file, 1);
3403     }
3404 
3405     if (migrate_postcopy()) {
3406         /*
3407          * Tell the destination that we *might* want to do postcopy later;
3408          * if the other end can't do postcopy it should fail now, nice and
3409          * early.
3410          */
3411         qemu_savevm_send_postcopy_advise(s->to_dst_file);
3412     }
3413 
3414     if (migrate_colo_enabled()) {
3415         /* Notify migration destination that we enable COLO */
3416         qemu_savevm_send_colo_enable(s->to_dst_file);
3417     }
3418 
3419     qemu_savevm_state_setup(s->to_dst_file);
3420 
3421     if (qemu_savevm_state_guest_unplug_pending()) {
3422         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3423                           MIGRATION_STATUS_WAIT_UNPLUG);
3424 
3425         while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3426                qemu_savevm_state_guest_unplug_pending()) {
3427             qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3428         }
3429 
3430         migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3431                 MIGRATION_STATUS_ACTIVE);
3432     }
3433 
3434     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3435     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3436                       MIGRATION_STATUS_ACTIVE);
3437 
3438     trace_migration_thread_setup_complete();
3439 
3440     while (migration_is_active(s)) {
3441         if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3442             MigIterateState iter_state = migration_iteration_run(s);
3443             if (iter_state == MIG_ITERATE_SKIP) {
3444                 continue;
3445             } else if (iter_state == MIG_ITERATE_BREAK) {
3446                 break;
3447             }
3448         }
3449 
3450         /*
3451          * Try to detect any kind of failures, and see whether we
3452          * should stop the migration now.
3453          */
3454         thr_error = migration_detect_error(s);
3455         if (thr_error == MIG_THR_ERR_FATAL) {
3456             /* Stop migration */
3457             break;
3458         } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3459             /*
3460              * Just recovered from a e.g. network failure, reset all
3461              * the local variables. This is important to avoid
3462              * breaking transferred_bytes and bandwidth calculation
3463              */
3464             update_iteration_initial_status(s);
3465         }
3466 
3467         urgent = migration_rate_limit();
3468     }
3469 
3470     trace_migration_thread_after_loop();
3471     migration_iteration_finish(s);
3472     object_unref(OBJECT(s));
3473     rcu_unregister_thread();
3474     return NULL;
3475 }
3476 
3477 void migrate_fd_connect(MigrationState *s, Error *error_in)
3478 {
3479     Error *local_err = NULL;
3480     int64_t rate_limit;
3481     bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3482 
3483     s->expected_downtime = s->parameters.downtime_limit;
3484     if (resume) {
3485         assert(s->cleanup_bh);
3486     } else {
3487         assert(!s->cleanup_bh);
3488         s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3489     }
3490     if (error_in) {
3491         migrate_fd_error(s, error_in);
3492         migrate_fd_cleanup(s);
3493         return;
3494     }
3495 
3496     if (resume) {
3497         /* This is a resumed migration */
3498         rate_limit = s->parameters.max_postcopy_bandwidth /
3499             XFER_LIMIT_RATIO;
3500     } else {
3501         /* This is a fresh new migration */
3502         rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3503 
3504         /* Notify before starting migration thread */
3505         notifier_list_notify(&migration_state_notifiers, s);
3506     }
3507 
3508     qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3509     qemu_file_set_blocking(s->to_dst_file, true);
3510 
3511     /*
3512      * Open the return path. For postcopy, it is used exclusively. For
3513      * precopy, only if user specified "return-path" capability would
3514      * QEMU uses the return path.
3515      */
3516     if (migrate_postcopy_ram() || migrate_use_return_path()) {
3517         if (open_return_path_on_source(s, !resume)) {
3518             error_report("Unable to open return-path for postcopy");
3519             migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3520             migrate_fd_cleanup(s);
3521             return;
3522         }
3523     }
3524 
3525     if (resume) {
3526         /* Wakeup the main migration thread to do the recovery */
3527         migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3528                           MIGRATION_STATUS_POSTCOPY_RECOVER);
3529         qemu_sem_post(&s->postcopy_pause_sem);
3530         return;
3531     }
3532 
3533     if (multifd_save_setup(&local_err) != 0) {
3534         error_report_err(local_err);
3535         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3536                           MIGRATION_STATUS_FAILED);
3537         migrate_fd_cleanup(s);
3538         return;
3539     }
3540     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3541                        QEMU_THREAD_JOINABLE);
3542     s->migration_thread_running = true;
3543 }
3544 
3545 void migration_global_dump(Monitor *mon)
3546 {
3547     MigrationState *ms = migrate_get_current();
3548 
3549     monitor_printf(mon, "globals:\n");
3550     monitor_printf(mon, "store-global-state: %s\n",
3551                    ms->store_global_state ? "on" : "off");
3552     monitor_printf(mon, "only-migratable: %s\n",
3553                    only_migratable ? "on" : "off");
3554     monitor_printf(mon, "send-configuration: %s\n",
3555                    ms->send_configuration ? "on" : "off");
3556     monitor_printf(mon, "send-section-footer: %s\n",
3557                    ms->send_section_footer ? "on" : "off");
3558     monitor_printf(mon, "decompress-error-check: %s\n",
3559                    ms->decompress_error_check ? "on" : "off");
3560     monitor_printf(mon, "clear-bitmap-shift: %u\n",
3561                    ms->clear_bitmap_shift);
3562 }
3563 
3564 #define DEFINE_PROP_MIG_CAP(name, x)             \
3565     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3566 
3567 static Property migration_properties[] = {
3568     DEFINE_PROP_BOOL("store-global-state", MigrationState,
3569                      store_global_state, true),
3570     DEFINE_PROP_BOOL("send-configuration", MigrationState,
3571                      send_configuration, true),
3572     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3573                      send_section_footer, true),
3574     DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3575                       decompress_error_check, true),
3576     DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
3577                       clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
3578 
3579     /* Migration parameters */
3580     DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3581                       parameters.compress_level,
3582                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
3583     DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3584                       parameters.compress_threads,
3585                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3586     DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3587                       parameters.compress_wait_thread, true),
3588     DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3589                       parameters.decompress_threads,
3590                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3591     DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState,
3592                       parameters.throttle_trigger_threshold,
3593                       DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD),
3594     DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3595                       parameters.cpu_throttle_initial,
3596                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3597     DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3598                       parameters.cpu_throttle_increment,
3599                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3600     DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3601                       parameters.max_bandwidth, MAX_THROTTLE),
3602     DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3603                       parameters.downtime_limit,
3604                       DEFAULT_MIGRATE_SET_DOWNTIME),
3605     DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3606                       parameters.x_checkpoint_delay,
3607                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3608     DEFINE_PROP_UINT8("multifd-channels", MigrationState,
3609                       parameters.multifd_channels,
3610                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3611     DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState,
3612                       parameters.multifd_compression,
3613                       DEFAULT_MIGRATE_MULTIFD_COMPRESSION),
3614     DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState,
3615                       parameters.multifd_zlib_level,
3616                       DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL),
3617     DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState,
3618                       parameters.multifd_zstd_level,
3619                       DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL),
3620     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3621                       parameters.xbzrle_cache_size,
3622                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3623     DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3624                       parameters.max_postcopy_bandwidth,
3625                       DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3626     DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3627                       parameters.max_cpu_throttle,
3628                       DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3629     DEFINE_PROP_SIZE("announce-initial", MigrationState,
3630                       parameters.announce_initial,
3631                       DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
3632     DEFINE_PROP_SIZE("announce-max", MigrationState,
3633                       parameters.announce_max,
3634                       DEFAULT_MIGRATE_ANNOUNCE_MAX),
3635     DEFINE_PROP_SIZE("announce-rounds", MigrationState,
3636                       parameters.announce_rounds,
3637                       DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
3638     DEFINE_PROP_SIZE("announce-step", MigrationState,
3639                       parameters.announce_step,
3640                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
3641 
3642     /* Migration capabilities */
3643     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3644     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3645     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3646     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3647     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3648     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3649     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3650     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3651     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3652     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3653     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3654     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD),
3655 
3656     DEFINE_PROP_END_OF_LIST(),
3657 };
3658 
3659 static void migration_class_init(ObjectClass *klass, void *data)
3660 {
3661     DeviceClass *dc = DEVICE_CLASS(klass);
3662 
3663     dc->user_creatable = false;
3664     device_class_set_props(dc, migration_properties);
3665 }
3666 
3667 static void migration_instance_finalize(Object *obj)
3668 {
3669     MigrationState *ms = MIGRATION_OBJ(obj);
3670     MigrationParameters *params = &ms->parameters;
3671 
3672     qemu_mutex_destroy(&ms->error_mutex);
3673     qemu_mutex_destroy(&ms->qemu_file_lock);
3674     g_free(params->tls_hostname);
3675     g_free(params->tls_creds);
3676     qemu_sem_destroy(&ms->wait_unplug_sem);
3677     qemu_sem_destroy(&ms->rate_limit_sem);
3678     qemu_sem_destroy(&ms->pause_sem);
3679     qemu_sem_destroy(&ms->postcopy_pause_sem);
3680     qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3681     qemu_sem_destroy(&ms->rp_state.rp_sem);
3682     error_free(ms->error);
3683 }
3684 
3685 static void migration_instance_init(Object *obj)
3686 {
3687     MigrationState *ms = MIGRATION_OBJ(obj);
3688     MigrationParameters *params = &ms->parameters;
3689 
3690     ms->state = MIGRATION_STATUS_NONE;
3691     ms->mbps = -1;
3692     ms->pages_per_second = -1;
3693     qemu_sem_init(&ms->pause_sem, 0);
3694     qemu_mutex_init(&ms->error_mutex);
3695 
3696     params->tls_hostname = g_strdup("");
3697     params->tls_creds = g_strdup("");
3698 
3699     /* Set has_* up only for parameter checks */
3700     params->has_compress_level = true;
3701     params->has_compress_threads = true;
3702     params->has_decompress_threads = true;
3703     params->has_throttle_trigger_threshold = true;
3704     params->has_cpu_throttle_initial = true;
3705     params->has_cpu_throttle_increment = true;
3706     params->has_max_bandwidth = true;
3707     params->has_downtime_limit = true;
3708     params->has_x_checkpoint_delay = true;
3709     params->has_block_incremental = true;
3710     params->has_multifd_channels = true;
3711     params->has_multifd_compression = true;
3712     params->has_multifd_zlib_level = true;
3713     params->has_multifd_zstd_level = true;
3714     params->has_xbzrle_cache_size = true;
3715     params->has_max_postcopy_bandwidth = true;
3716     params->has_max_cpu_throttle = true;
3717     params->has_announce_initial = true;
3718     params->has_announce_max = true;
3719     params->has_announce_rounds = true;
3720     params->has_announce_step = true;
3721 
3722     qemu_sem_init(&ms->postcopy_pause_sem, 0);
3723     qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3724     qemu_sem_init(&ms->rp_state.rp_sem, 0);
3725     qemu_sem_init(&ms->rate_limit_sem, 0);
3726     qemu_sem_init(&ms->wait_unplug_sem, 0);
3727     qemu_mutex_init(&ms->qemu_file_lock);
3728 }
3729 
3730 /*
3731  * Return true if check pass, false otherwise. Error will be put
3732  * inside errp if provided.
3733  */
3734 static bool migration_object_check(MigrationState *ms, Error **errp)
3735 {
3736     MigrationCapabilityStatusList *head = NULL;
3737     /* Assuming all off */
3738     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3739     int i;
3740 
3741     if (!migrate_params_check(&ms->parameters, errp)) {
3742         return false;
3743     }
3744 
3745     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3746         if (ms->enabled_capabilities[i]) {
3747             head = migrate_cap_add(head, i, true);
3748         }
3749     }
3750 
3751     ret = migrate_caps_check(cap_list, head, errp);
3752 
3753     /* It works with head == NULL */
3754     qapi_free_MigrationCapabilityStatusList(head);
3755 
3756     return ret;
3757 }
3758 
3759 static const TypeInfo migration_type = {
3760     .name = TYPE_MIGRATION,
3761     /*
3762      * NOTE: TYPE_MIGRATION is not really a device, as the object is
3763      * not created using qdev_create(), it is not attached to the qdev
3764      * device tree, and it is never realized.
3765      *
3766      * TODO: Make this TYPE_OBJECT once QOM provides something like
3767      * TYPE_DEVICE's "-global" properties.
3768      */
3769     .parent = TYPE_DEVICE,
3770     .class_init = migration_class_init,
3771     .class_size = sizeof(MigrationClass),
3772     .instance_size = sizeof(MigrationState),
3773     .instance_init = migration_instance_init,
3774     .instance_finalize = migration_instance_finalize,
3775 };
3776 
3777 static void register_migration_types(void)
3778 {
3779     type_register_static(&migration_type);
3780 }
3781 
3782 type_init(register_migration_types);
3783