xref: /openbmc/qemu/migration/migration.c (revision d5938f29fea29581725426f203a74da746ca03e7)
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 "rdma.h"
25 #include "ram.h"
26 #include "migration/global_state.h"
27 #include "migration/misc.h"
28 #include "migration.h"
29 #include "savevm.h"
30 #include "qemu-file-channel.h"
31 #include "qemu-file.h"
32 #include "migration/vmstate.h"
33 #include "block/block.h"
34 #include "qapi/error.h"
35 #include "qapi/clone-visitor.h"
36 #include "qapi/qapi-visit-sockets.h"
37 #include "qapi/qapi-commands-migration.h"
38 #include "qapi/qapi-events-migration.h"
39 #include "qapi/qmp/qerror.h"
40 #include "qapi/qmp/qnull.h"
41 #include "qemu/rcu.h"
42 #include "block.h"
43 #include "postcopy-ram.h"
44 #include "qemu/thread.h"
45 #include "trace.h"
46 #include "exec/target_page.h"
47 #include "io/channel-buffer.h"
48 #include "migration/colo.h"
49 #include "hw/boards.h"
50 #include "hw/qdev-properties.h"
51 #include "monitor/monitor.h"
52 #include "net/announce.h"
53 
54 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
55 
56 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
57  * data. */
58 #define BUFFER_DELAY     100
59 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
60 
61 /* Time in milliseconds we are allowed to stop the source,
62  * for sending the last part */
63 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
64 
65 /* Maximum migrate downtime set to 2000 seconds */
66 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
67 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
68 
69 /* Default compression thread count */
70 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
71 /* Default decompression thread count, usually decompression is at
72  * least 4 times as fast as compression.*/
73 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
74 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
75 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
76 /* Define default autoconverge cpu throttle migration parameters */
77 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
78 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
79 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
80 
81 /* Migration XBZRLE default cache size */
82 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
83 
84 /* The delay time (in ms) between two COLO checkpoints */
85 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
86 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
87 
88 /* Background transfer rate for postcopy, 0 means unlimited, note
89  * that page requests can still exceed this limit.
90  */
91 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
92 
93 /*
94  * Parameters for self_announce_delay giving a stream of RARP/ARP
95  * packets after migration.
96  */
97 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL  50
98 #define DEFAULT_MIGRATE_ANNOUNCE_MAX     550
99 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS    5
100 #define DEFAULT_MIGRATE_ANNOUNCE_STEP    100
101 
102 static NotifierList migration_state_notifiers =
103     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
104 
105 static bool deferred_incoming;
106 
107 /* Messages sent on the return path from destination to source */
108 enum mig_rp_message_type {
109     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
110     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
111     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
112 
113     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
114     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
115     MIG_RP_MSG_RECV_BITMAP,  /* send recved_bitmap back to source */
116     MIG_RP_MSG_RESUME_ACK,   /* tell source that we are ready to resume */
117 
118     MIG_RP_MSG_MAX
119 };
120 
121 /* When we add fault tolerance, we could have several
122    migrations at once.  For now we don't need to add
123    dynamic creation of migration */
124 
125 static MigrationState *current_migration;
126 static MigrationIncomingState *current_incoming;
127 
128 static bool migration_object_check(MigrationState *ms, Error **errp);
129 static int migration_maybe_pause(MigrationState *s,
130                                  int *current_active_state,
131                                  int new_state);
132 static void migrate_fd_cancel(MigrationState *s);
133 
134 void migration_object_init(void)
135 {
136     MachineState *ms = MACHINE(qdev_get_machine());
137     Error *err = NULL;
138 
139     /* This can only be called once. */
140     assert(!current_migration);
141     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
142 
143     /*
144      * Init the migrate incoming object as well no matter whether
145      * we'll use it or not.
146      */
147     assert(!current_incoming);
148     current_incoming = g_new0(MigrationIncomingState, 1);
149     current_incoming->state = MIGRATION_STATUS_NONE;
150     current_incoming->postcopy_remote_fds =
151         g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
152     qemu_mutex_init(&current_incoming->rp_mutex);
153     qemu_event_init(&current_incoming->main_thread_load_event, false);
154     qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
155     qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
156 
157     init_dirty_bitmap_incoming_migration();
158 
159     if (!migration_object_check(current_migration, &err)) {
160         error_report_err(err);
161         exit(1);
162     }
163 
164     /*
165      * We cannot really do this in migration_instance_init() since at
166      * that time global properties are not yet applied, then this
167      * value will be definitely replaced by something else.
168      */
169     if (ms->enforce_config_section) {
170         current_migration->send_configuration = true;
171     }
172 }
173 
174 void migration_shutdown(void)
175 {
176     /*
177      * Cancel the current migration - that will (eventually)
178      * stop the migration using this structure
179      */
180     migrate_fd_cancel(current_migration);
181     object_unref(OBJECT(current_migration));
182 }
183 
184 /* For outgoing */
185 MigrationState *migrate_get_current(void)
186 {
187     /* This can only be called after the object created. */
188     assert(current_migration);
189     return current_migration;
190 }
191 
192 MigrationIncomingState *migration_incoming_get_current(void)
193 {
194     assert(current_incoming);
195     return current_incoming;
196 }
197 
198 void migration_incoming_state_destroy(void)
199 {
200     struct MigrationIncomingState *mis = migration_incoming_get_current();
201 
202     if (mis->to_src_file) {
203         /* Tell source that we are done */
204         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
205         qemu_fclose(mis->to_src_file);
206         mis->to_src_file = NULL;
207     }
208 
209     if (mis->from_src_file) {
210         qemu_fclose(mis->from_src_file);
211         mis->from_src_file = NULL;
212     }
213     if (mis->postcopy_remote_fds) {
214         g_array_free(mis->postcopy_remote_fds, TRUE);
215         mis->postcopy_remote_fds = NULL;
216     }
217 
218     qemu_event_reset(&mis->main_thread_load_event);
219 
220     if (mis->socket_address_list) {
221         qapi_free_SocketAddressList(mis->socket_address_list);
222         mis->socket_address_list = NULL;
223     }
224 }
225 
226 static void migrate_generate_event(int new_state)
227 {
228     if (migrate_use_events()) {
229         qapi_event_send_migration(new_state);
230     }
231 }
232 
233 static bool migrate_late_block_activate(void)
234 {
235     MigrationState *s;
236 
237     s = migrate_get_current();
238 
239     return s->enabled_capabilities[
240         MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
241 }
242 
243 /*
244  * Called on -incoming with a defer: uri.
245  * The migration can be started later after any parameters have been
246  * changed.
247  */
248 static void deferred_incoming_migration(Error **errp)
249 {
250     if (deferred_incoming) {
251         error_setg(errp, "Incoming migration already deferred");
252     }
253     deferred_incoming = true;
254 }
255 
256 /*
257  * Send a message on the return channel back to the source
258  * of the migration.
259  */
260 static int migrate_send_rp_message(MigrationIncomingState *mis,
261                                    enum mig_rp_message_type message_type,
262                                    uint16_t len, void *data)
263 {
264     int ret = 0;
265 
266     trace_migrate_send_rp_message((int)message_type, len);
267     qemu_mutex_lock(&mis->rp_mutex);
268 
269     /*
270      * It's possible that the file handle got lost due to network
271      * failures.
272      */
273     if (!mis->to_src_file) {
274         ret = -EIO;
275         goto error;
276     }
277 
278     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
279     qemu_put_be16(mis->to_src_file, len);
280     qemu_put_buffer(mis->to_src_file, data, len);
281     qemu_fflush(mis->to_src_file);
282 
283     /* It's possible that qemu file got error during sending */
284     ret = qemu_file_get_error(mis->to_src_file);
285 
286 error:
287     qemu_mutex_unlock(&mis->rp_mutex);
288     return ret;
289 }
290 
291 /* Request a range of pages from the source VM at the given
292  * start address.
293  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
294  *           as the last request (a name must have been given previously)
295  *   Start: Address offset within the RB
296  *   Len: Length in bytes required - must be a multiple of pagesize
297  */
298 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
299                               ram_addr_t start, size_t len)
300 {
301     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
302     size_t msglen = 12; /* start + len */
303     enum mig_rp_message_type msg_type;
304 
305     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
306     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
307 
308     if (rbname) {
309         int rbname_len = strlen(rbname);
310         assert(rbname_len < 256);
311 
312         bufc[msglen++] = rbname_len;
313         memcpy(bufc + msglen, rbname, rbname_len);
314         msglen += rbname_len;
315         msg_type = MIG_RP_MSG_REQ_PAGES_ID;
316     } else {
317         msg_type = MIG_RP_MSG_REQ_PAGES;
318     }
319 
320     return migrate_send_rp_message(mis, msg_type, msglen, bufc);
321 }
322 
323 static bool migration_colo_enabled;
324 bool migration_incoming_colo_enabled(void)
325 {
326     return migration_colo_enabled;
327 }
328 
329 void migration_incoming_disable_colo(void)
330 {
331     migration_colo_enabled = false;
332 }
333 
334 void migration_incoming_enable_colo(void)
335 {
336     migration_colo_enabled = true;
337 }
338 
339 void migrate_add_address(SocketAddress *address)
340 {
341     MigrationIncomingState *mis = migration_incoming_get_current();
342     SocketAddressList *addrs;
343 
344     addrs = g_new0(SocketAddressList, 1);
345     addrs->next = mis->socket_address_list;
346     mis->socket_address_list = addrs;
347     addrs->value = QAPI_CLONE(SocketAddress, address);
348 }
349 
350 void qemu_start_incoming_migration(const char *uri, Error **errp)
351 {
352     const char *p;
353 
354     qapi_event_send_migration(MIGRATION_STATUS_SETUP);
355     if (!strcmp(uri, "defer")) {
356         deferred_incoming_migration(errp);
357     } else if (strstart(uri, "tcp:", &p)) {
358         tcp_start_incoming_migration(p, errp);
359 #ifdef CONFIG_RDMA
360     } else if (strstart(uri, "rdma:", &p)) {
361         rdma_start_incoming_migration(p, errp);
362 #endif
363     } else if (strstart(uri, "exec:", &p)) {
364         exec_start_incoming_migration(p, errp);
365     } else if (strstart(uri, "unix:", &p)) {
366         unix_start_incoming_migration(p, errp);
367     } else if (strstart(uri, "fd:", &p)) {
368         fd_start_incoming_migration(p, errp);
369     } else {
370         error_setg(errp, "unknown migration protocol: %s", uri);
371     }
372 }
373 
374 static void process_incoming_migration_bh(void *opaque)
375 {
376     Error *local_err = NULL;
377     MigrationIncomingState *mis = opaque;
378 
379     /* If capability late_block_activate is set:
380      * Only fire up the block code now if we're going to restart the
381      * VM, else 'cont' will do it.
382      * This causes file locking to happen; so we don't want it to happen
383      * unless we really are starting the VM.
384      */
385     if (!migrate_late_block_activate() ||
386          (autostart && (!global_state_received() ||
387             global_state_get_runstate() == RUN_STATE_RUNNING))) {
388         /* Make sure all file formats flush their mutable metadata.
389          * If we get an error here, just don't restart the VM yet. */
390         bdrv_invalidate_cache_all(&local_err);
391         if (local_err) {
392             error_report_err(local_err);
393             local_err = NULL;
394             autostart = false;
395         }
396     }
397 
398     /*
399      * This must happen after all error conditions are dealt with and
400      * we're sure the VM is going to be running on this host.
401      */
402     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
403 
404     if (multifd_load_cleanup(&local_err) != 0) {
405         error_report_err(local_err);
406         autostart = false;
407     }
408     /* If global state section was not received or we are in running
409        state, we need to obey autostart. Any other state is set with
410        runstate_set. */
411 
412     dirty_bitmap_mig_before_vm_start();
413 
414     if (!global_state_received() ||
415         global_state_get_runstate() == RUN_STATE_RUNNING) {
416         if (autostart) {
417             vm_start();
418         } else {
419             runstate_set(RUN_STATE_PAUSED);
420         }
421     } else if (migration_incoming_colo_enabled()) {
422         migration_incoming_disable_colo();
423         vm_start();
424     } else {
425         runstate_set(global_state_get_runstate());
426     }
427     /*
428      * This must happen after any state changes since as soon as an external
429      * observer sees this event they might start to prod at the VM assuming
430      * it's ready to use.
431      */
432     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
433                       MIGRATION_STATUS_COMPLETED);
434     qemu_bh_delete(mis->bh);
435     migration_incoming_state_destroy();
436 }
437 
438 static void process_incoming_migration_co(void *opaque)
439 {
440     MigrationIncomingState *mis = migration_incoming_get_current();
441     PostcopyState ps;
442     int ret;
443     Error *local_err = NULL;
444 
445     assert(mis->from_src_file);
446     mis->migration_incoming_co = qemu_coroutine_self();
447     mis->largest_page_size = qemu_ram_pagesize_largest();
448     postcopy_state_set(POSTCOPY_INCOMING_NONE);
449     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
450                       MIGRATION_STATUS_ACTIVE);
451     ret = qemu_loadvm_state(mis->from_src_file);
452 
453     ps = postcopy_state_get();
454     trace_process_incoming_migration_co_end(ret, ps);
455     if (ps != POSTCOPY_INCOMING_NONE) {
456         if (ps == POSTCOPY_INCOMING_ADVISE) {
457             /*
458              * Where a migration had postcopy enabled (and thus went to advise)
459              * but managed to complete within the precopy period, we can use
460              * the normal exit.
461              */
462             postcopy_ram_incoming_cleanup(mis);
463         } else if (ret >= 0) {
464             /*
465              * Postcopy was started, cleanup should happen at the end of the
466              * postcopy thread.
467              */
468             trace_process_incoming_migration_co_postcopy_end_main();
469             return;
470         }
471         /* Else if something went wrong then just fall out of the normal exit */
472     }
473 
474     /* we get COLO info, and know if we are in COLO mode */
475     if (!ret && migration_incoming_colo_enabled()) {
476         /* Make sure all file formats flush their mutable metadata */
477         bdrv_invalidate_cache_all(&local_err);
478         if (local_err) {
479             error_report_err(local_err);
480             goto fail;
481         }
482 
483         if (colo_init_ram_cache() < 0) {
484             error_report("Init ram cache failed");
485             goto fail;
486         }
487 
488         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
489              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
490         mis->have_colo_incoming_thread = true;
491         qemu_coroutine_yield();
492 
493         /* Wait checkpoint incoming thread exit before free resource */
494         qemu_thread_join(&mis->colo_incoming_thread);
495         /* We hold the global iothread lock, so it is safe here */
496         colo_release_ram_cache();
497     }
498 
499     if (ret < 0) {
500         error_report("load of migration failed: %s", strerror(-ret));
501         goto fail;
502     }
503     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
504     qemu_bh_schedule(mis->bh);
505     mis->migration_incoming_co = NULL;
506     return;
507 fail:
508     local_err = NULL;
509     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
510                       MIGRATION_STATUS_FAILED);
511     qemu_fclose(mis->from_src_file);
512     if (multifd_load_cleanup(&local_err) != 0) {
513         error_report_err(local_err);
514     }
515     exit(EXIT_FAILURE);
516 }
517 
518 static void migration_incoming_setup(QEMUFile *f)
519 {
520     MigrationIncomingState *mis = migration_incoming_get_current();
521 
522     if (multifd_load_setup() != 0) {
523         /* We haven't been able to create multifd threads
524            nothing better to do */
525         exit(EXIT_FAILURE);
526     }
527 
528     if (!mis->from_src_file) {
529         mis->from_src_file = f;
530     }
531     qemu_file_set_blocking(f, false);
532 }
533 
534 void migration_incoming_process(void)
535 {
536     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
537     qemu_coroutine_enter(co);
538 }
539 
540 /* Returns true if recovered from a paused migration, otherwise false */
541 static bool postcopy_try_recover(QEMUFile *f)
542 {
543     MigrationIncomingState *mis = migration_incoming_get_current();
544 
545     if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
546         /* Resumed from a paused postcopy migration */
547 
548         mis->from_src_file = f;
549         /* Postcopy has standalone thread to do vm load */
550         qemu_file_set_blocking(f, true);
551 
552         /* Re-configure the return path */
553         mis->to_src_file = qemu_file_get_return_path(f);
554 
555         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
556                           MIGRATION_STATUS_POSTCOPY_RECOVER);
557 
558         /*
559          * Here, we only wake up the main loading thread (while the
560          * fault thread will still be waiting), so that we can receive
561          * commands from source now, and answer it if needed. The
562          * fault thread will be woken up afterwards until we are sure
563          * that source is ready to reply to page requests.
564          */
565         qemu_sem_post(&mis->postcopy_pause_sem_dst);
566         return true;
567     }
568 
569     return false;
570 }
571 
572 void migration_fd_process_incoming(QEMUFile *f)
573 {
574     if (postcopy_try_recover(f)) {
575         return;
576     }
577 
578     migration_incoming_setup(f);
579     migration_incoming_process();
580 }
581 
582 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
583 {
584     MigrationIncomingState *mis = migration_incoming_get_current();
585     bool start_migration;
586 
587     if (!mis->from_src_file) {
588         /* The first connection (multifd may have multiple) */
589         QEMUFile *f = qemu_fopen_channel_input(ioc);
590 
591         /* If it's a recovery, we're done */
592         if (postcopy_try_recover(f)) {
593             return;
594         }
595 
596         migration_incoming_setup(f);
597 
598         /*
599          * Common migration only needs one channel, so we can start
600          * right now.  Multifd needs more than one channel, we wait.
601          */
602         start_migration = !migrate_use_multifd();
603     } else {
604         Error *local_err = NULL;
605         /* Multiple connections */
606         assert(migrate_use_multifd());
607         start_migration = multifd_recv_new_channel(ioc, &local_err);
608         if (local_err) {
609             error_propagate(errp, local_err);
610             return;
611         }
612     }
613 
614     if (start_migration) {
615         migration_incoming_process();
616     }
617 }
618 
619 /**
620  * @migration_has_all_channels: We have received all channels that we need
621  *
622  * Returns true when we have got connections to all the channels that
623  * we need for migration.
624  */
625 bool migration_has_all_channels(void)
626 {
627     MigrationIncomingState *mis = migration_incoming_get_current();
628     bool all_channels;
629 
630     all_channels = multifd_recv_all_channels_created();
631 
632     return all_channels && mis->from_src_file != NULL;
633 }
634 
635 /*
636  * Send a 'SHUT' message on the return channel with the given value
637  * to indicate that we've finished with the RP.  Non-0 value indicates
638  * error.
639  */
640 void migrate_send_rp_shut(MigrationIncomingState *mis,
641                           uint32_t value)
642 {
643     uint32_t buf;
644 
645     buf = cpu_to_be32(value);
646     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
647 }
648 
649 /*
650  * Send a 'PONG' message on the return channel with the given value
651  * (normally in response to a 'PING')
652  */
653 void migrate_send_rp_pong(MigrationIncomingState *mis,
654                           uint32_t value)
655 {
656     uint32_t buf;
657 
658     buf = cpu_to_be32(value);
659     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
660 }
661 
662 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
663                                  char *block_name)
664 {
665     char buf[512];
666     int len;
667     int64_t res;
668 
669     /*
670      * First, we send the header part. It contains only the len of
671      * idstr, and the idstr itself.
672      */
673     len = strlen(block_name);
674     buf[0] = len;
675     memcpy(buf + 1, block_name, len);
676 
677     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
678         error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
679                      __func__);
680         return;
681     }
682 
683     migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
684 
685     /*
686      * Next, we dump the received bitmap to the stream.
687      *
688      * TODO: currently we are safe since we are the only one that is
689      * using the to_src_file handle (fault thread is still paused),
690      * and it's ok even not taking the mutex. However the best way is
691      * to take the lock before sending the message header, and release
692      * the lock after sending the bitmap.
693      */
694     qemu_mutex_lock(&mis->rp_mutex);
695     res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
696     qemu_mutex_unlock(&mis->rp_mutex);
697 
698     trace_migrate_send_rp_recv_bitmap(block_name, res);
699 }
700 
701 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
702 {
703     uint32_t buf;
704 
705     buf = cpu_to_be32(value);
706     migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
707 }
708 
709 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
710 {
711     MigrationCapabilityStatusList *head = NULL;
712     MigrationCapabilityStatusList *caps;
713     MigrationState *s = migrate_get_current();
714     int i;
715 
716     caps = NULL; /* silence compiler warning */
717     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
718 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
719         if (i == MIGRATION_CAPABILITY_BLOCK) {
720             continue;
721         }
722 #endif
723         if (head == NULL) {
724             head = g_malloc0(sizeof(*caps));
725             caps = head;
726         } else {
727             caps->next = g_malloc0(sizeof(*caps));
728             caps = caps->next;
729         }
730         caps->value =
731             g_malloc(sizeof(*caps->value));
732         caps->value->capability = i;
733         caps->value->state = s->enabled_capabilities[i];
734     }
735 
736     return head;
737 }
738 
739 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
740 {
741     MigrationParameters *params;
742     MigrationState *s = migrate_get_current();
743 
744     /* TODO use QAPI_CLONE() instead of duplicating it inline */
745     params = g_malloc0(sizeof(*params));
746     params->has_compress_level = true;
747     params->compress_level = s->parameters.compress_level;
748     params->has_compress_threads = true;
749     params->compress_threads = s->parameters.compress_threads;
750     params->has_compress_wait_thread = true;
751     params->compress_wait_thread = s->parameters.compress_wait_thread;
752     params->has_decompress_threads = true;
753     params->decompress_threads = s->parameters.decompress_threads;
754     params->has_cpu_throttle_initial = true;
755     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
756     params->has_cpu_throttle_increment = true;
757     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
758     params->has_tls_creds = true;
759     params->tls_creds = g_strdup(s->parameters.tls_creds);
760     params->has_tls_hostname = true;
761     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
762     params->has_tls_authz = true;
763     params->tls_authz = g_strdup(s->parameters.tls_authz);
764     params->has_max_bandwidth = true;
765     params->max_bandwidth = s->parameters.max_bandwidth;
766     params->has_downtime_limit = true;
767     params->downtime_limit = s->parameters.downtime_limit;
768     params->has_x_checkpoint_delay = true;
769     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
770     params->has_block_incremental = true;
771     params->block_incremental = s->parameters.block_incremental;
772     params->has_multifd_channels = true;
773     params->multifd_channels = s->parameters.multifd_channels;
774     params->has_xbzrle_cache_size = true;
775     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
776     params->has_max_postcopy_bandwidth = true;
777     params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
778     params->has_max_cpu_throttle = true;
779     params->max_cpu_throttle = s->parameters.max_cpu_throttle;
780     params->has_announce_initial = true;
781     params->announce_initial = s->parameters.announce_initial;
782     params->has_announce_max = true;
783     params->announce_max = s->parameters.announce_max;
784     params->has_announce_rounds = true;
785     params->announce_rounds = s->parameters.announce_rounds;
786     params->has_announce_step = true;
787     params->announce_step = s->parameters.announce_step;
788 
789     return params;
790 }
791 
792 AnnounceParameters *migrate_announce_params(void)
793 {
794     static AnnounceParameters ap;
795 
796     MigrationState *s = migrate_get_current();
797 
798     ap.initial = s->parameters.announce_initial;
799     ap.max = s->parameters.announce_max;
800     ap.rounds = s->parameters.announce_rounds;
801     ap.step = s->parameters.announce_step;
802 
803     return &ap;
804 }
805 
806 /*
807  * Return true if we're already in the middle of a migration
808  * (i.e. any of the active or setup states)
809  */
810 bool migration_is_setup_or_active(int state)
811 {
812     switch (state) {
813     case MIGRATION_STATUS_ACTIVE:
814     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
815     case MIGRATION_STATUS_POSTCOPY_PAUSED:
816     case MIGRATION_STATUS_POSTCOPY_RECOVER:
817     case MIGRATION_STATUS_SETUP:
818     case MIGRATION_STATUS_PRE_SWITCHOVER:
819     case MIGRATION_STATUS_DEVICE:
820         return true;
821 
822     default:
823         return false;
824 
825     }
826 }
827 
828 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
829 {
830     info->has_ram = true;
831     info->ram = g_malloc0(sizeof(*info->ram));
832     info->ram->transferred = ram_counters.transferred;
833     info->ram->total = ram_bytes_total();
834     info->ram->duplicate = ram_counters.duplicate;
835     /* legacy value.  It is not used anymore */
836     info->ram->skipped = 0;
837     info->ram->normal = ram_counters.normal;
838     info->ram->normal_bytes = ram_counters.normal *
839         qemu_target_page_size();
840     info->ram->mbps = s->mbps;
841     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
842     info->ram->postcopy_requests = ram_counters.postcopy_requests;
843     info->ram->page_size = qemu_target_page_size();
844     info->ram->multifd_bytes = ram_counters.multifd_bytes;
845     info->ram->pages_per_second = s->pages_per_second;
846 
847     if (migrate_use_xbzrle()) {
848         info->has_xbzrle_cache = true;
849         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
850         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
851         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
852         info->xbzrle_cache->pages = xbzrle_counters.pages;
853         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
854         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
855         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
856     }
857 
858     if (migrate_use_compression()) {
859         info->has_compression = true;
860         info->compression = g_malloc0(sizeof(*info->compression));
861         info->compression->pages = compression_counters.pages;
862         info->compression->busy = compression_counters.busy;
863         info->compression->busy_rate = compression_counters.busy_rate;
864         info->compression->compressed_size =
865                                     compression_counters.compressed_size;
866         info->compression->compression_rate =
867                                     compression_counters.compression_rate;
868     }
869 
870     if (cpu_throttle_active()) {
871         info->has_cpu_throttle_percentage = true;
872         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
873     }
874 
875     if (s->state != MIGRATION_STATUS_COMPLETED) {
876         info->ram->remaining = ram_bytes_remaining();
877         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
878     }
879 }
880 
881 static void populate_disk_info(MigrationInfo *info)
882 {
883     if (blk_mig_active()) {
884         info->has_disk = true;
885         info->disk = g_malloc0(sizeof(*info->disk));
886         info->disk->transferred = blk_mig_bytes_transferred();
887         info->disk->remaining = blk_mig_bytes_remaining();
888         info->disk->total = blk_mig_bytes_total();
889     }
890 }
891 
892 static void fill_source_migration_info(MigrationInfo *info)
893 {
894     MigrationState *s = migrate_get_current();
895 
896     switch (s->state) {
897     case MIGRATION_STATUS_NONE:
898         /* no migration has happened ever */
899         /* do not overwrite destination migration status */
900         return;
901         break;
902     case MIGRATION_STATUS_SETUP:
903         info->has_status = true;
904         info->has_total_time = false;
905         break;
906     case MIGRATION_STATUS_ACTIVE:
907     case MIGRATION_STATUS_CANCELLING:
908     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
909     case MIGRATION_STATUS_PRE_SWITCHOVER:
910     case MIGRATION_STATUS_DEVICE:
911     case MIGRATION_STATUS_POSTCOPY_PAUSED:
912     case MIGRATION_STATUS_POSTCOPY_RECOVER:
913          /* TODO add some postcopy stats */
914         info->has_status = true;
915         info->has_total_time = true;
916         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
917             - s->start_time;
918         info->has_expected_downtime = true;
919         info->expected_downtime = s->expected_downtime;
920         info->has_setup_time = true;
921         info->setup_time = s->setup_time;
922 
923         populate_ram_info(info, s);
924         populate_disk_info(info);
925         break;
926     case MIGRATION_STATUS_COLO:
927         info->has_status = true;
928         /* TODO: display COLO specific information (checkpoint info etc.) */
929         break;
930     case MIGRATION_STATUS_COMPLETED:
931         info->has_status = true;
932         info->has_total_time = true;
933         info->total_time = s->total_time;
934         info->has_downtime = true;
935         info->downtime = s->downtime;
936         info->has_setup_time = true;
937         info->setup_time = s->setup_time;
938 
939         populate_ram_info(info, s);
940         break;
941     case MIGRATION_STATUS_FAILED:
942         info->has_status = true;
943         if (s->error) {
944             info->has_error_desc = true;
945             info->error_desc = g_strdup(error_get_pretty(s->error));
946         }
947         break;
948     case MIGRATION_STATUS_CANCELLED:
949         info->has_status = true;
950         break;
951     }
952     info->status = s->state;
953 }
954 
955 /**
956  * @migration_caps_check - check capability validity
957  *
958  * @cap_list: old capability list, array of bool
959  * @params: new capabilities to be applied soon
960  * @errp: set *errp if the check failed, with reason
961  *
962  * Returns true if check passed, otherwise false.
963  */
964 static bool migrate_caps_check(bool *cap_list,
965                                MigrationCapabilityStatusList *params,
966                                Error **errp)
967 {
968     MigrationCapabilityStatusList *cap;
969     bool old_postcopy_cap;
970     MigrationIncomingState *mis = migration_incoming_get_current();
971 
972     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
973 
974     for (cap = params; cap; cap = cap->next) {
975         cap_list[cap->value->capability] = cap->value->state;
976     }
977 
978 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
979     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
980         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
981                    "block migration");
982         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
983         return false;
984     }
985 #endif
986 
987 #ifndef CONFIG_REPLICATION
988     if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
989         error_setg(errp, "QEMU compiled without replication module"
990                    " can't enable COLO");
991         error_append_hint(errp, "Please enable replication before COLO.\n");
992         return false;
993     }
994 #endif
995 
996     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
997         if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
998             /* The decompression threads asynchronously write into RAM
999              * rather than use the atomic copies needed to avoid
1000              * userfaulting.  It should be possible to fix the decompression
1001              * threads for compatibility in future.
1002              */
1003             error_setg(errp, "Postcopy is not currently compatible "
1004                        "with compression");
1005             return false;
1006         }
1007 
1008         /* This check is reasonably expensive, so only when it's being
1009          * set the first time, also it's only the destination that needs
1010          * special support.
1011          */
1012         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1013             !postcopy_ram_supported_by_host(mis)) {
1014             /* postcopy_ram_supported_by_host will have emitted a more
1015              * detailed message
1016              */
1017             error_setg(errp, "Postcopy is not supported");
1018             return false;
1019         }
1020 
1021         if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1022             error_setg(errp, "Postcopy is not compatible with ignore-shared");
1023             return false;
1024         }
1025     }
1026 
1027     return true;
1028 }
1029 
1030 static void fill_destination_migration_info(MigrationInfo *info)
1031 {
1032     MigrationIncomingState *mis = migration_incoming_get_current();
1033 
1034     if (mis->socket_address_list) {
1035         info->has_socket_address = true;
1036         info->socket_address =
1037             QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1038     }
1039 
1040     switch (mis->state) {
1041     case MIGRATION_STATUS_NONE:
1042         return;
1043         break;
1044     case MIGRATION_STATUS_SETUP:
1045     case MIGRATION_STATUS_CANCELLING:
1046     case MIGRATION_STATUS_CANCELLED:
1047     case MIGRATION_STATUS_ACTIVE:
1048     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1049     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1050     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1051     case MIGRATION_STATUS_FAILED:
1052     case MIGRATION_STATUS_COLO:
1053         info->has_status = true;
1054         break;
1055     case MIGRATION_STATUS_COMPLETED:
1056         info->has_status = true;
1057         fill_destination_postcopy_migration_info(info);
1058         break;
1059     }
1060     info->status = mis->state;
1061 }
1062 
1063 MigrationInfo *qmp_query_migrate(Error **errp)
1064 {
1065     MigrationInfo *info = g_malloc0(sizeof(*info));
1066 
1067     fill_destination_migration_info(info);
1068     fill_source_migration_info(info);
1069 
1070     return info;
1071 }
1072 
1073 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1074                                   Error **errp)
1075 {
1076     MigrationState *s = migrate_get_current();
1077     MigrationCapabilityStatusList *cap;
1078     bool cap_list[MIGRATION_CAPABILITY__MAX];
1079 
1080     if (migration_is_setup_or_active(s->state)) {
1081         error_setg(errp, QERR_MIGRATION_ACTIVE);
1082         return;
1083     }
1084 
1085     memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1086     if (!migrate_caps_check(cap_list, params, errp)) {
1087         return;
1088     }
1089 
1090     for (cap = params; cap; cap = cap->next) {
1091         s->enabled_capabilities[cap->value->capability] = cap->value->state;
1092     }
1093 }
1094 
1095 /*
1096  * Check whether the parameters are valid. Error will be put into errp
1097  * (if provided). Return true if valid, otherwise false.
1098  */
1099 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1100 {
1101     if (params->has_compress_level &&
1102         (params->compress_level > 9)) {
1103         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1104                    "is invalid, it should be in the range of 0 to 9");
1105         return false;
1106     }
1107 
1108     if (params->has_compress_threads && (params->compress_threads < 1)) {
1109         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1110                    "compress_threads",
1111                    "is invalid, it should be in the range of 1 to 255");
1112         return false;
1113     }
1114 
1115     if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1116         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1117                    "decompress_threads",
1118                    "is invalid, it should be in the range of 1 to 255");
1119         return false;
1120     }
1121 
1122     if (params->has_cpu_throttle_initial &&
1123         (params->cpu_throttle_initial < 1 ||
1124          params->cpu_throttle_initial > 99)) {
1125         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1126                    "cpu_throttle_initial",
1127                    "an integer in the range of 1 to 99");
1128         return false;
1129     }
1130 
1131     if (params->has_cpu_throttle_increment &&
1132         (params->cpu_throttle_increment < 1 ||
1133          params->cpu_throttle_increment > 99)) {
1134         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1135                    "cpu_throttle_increment",
1136                    "an integer in the range of 1 to 99");
1137         return false;
1138     }
1139 
1140     if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1141         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
1142                          " range of 0 to %zu bytes/second", SIZE_MAX);
1143         return false;
1144     }
1145 
1146     if (params->has_downtime_limit &&
1147         (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1148         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1149                          "the range of 0 to %d milliseconds",
1150                          MAX_MIGRATE_DOWNTIME);
1151         return false;
1152     }
1153 
1154     /* x_checkpoint_delay is now always positive */
1155 
1156     if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1157         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1158                    "multifd_channels",
1159                    "is invalid, it should be in the range of 1 to 255");
1160         return false;
1161     }
1162 
1163     if (params->has_xbzrle_cache_size &&
1164         (params->xbzrle_cache_size < qemu_target_page_size() ||
1165          !is_power_of_2(params->xbzrle_cache_size))) {
1166         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1167                    "xbzrle_cache_size",
1168                    "is invalid, it should be bigger than target page size"
1169                    " and a power of two");
1170         return false;
1171     }
1172 
1173     if (params->has_max_cpu_throttle &&
1174         (params->max_cpu_throttle < params->cpu_throttle_initial ||
1175          params->max_cpu_throttle > 99)) {
1176         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1177                    "max_cpu_throttle",
1178                    "an integer in the range of cpu_throttle_initial to 99");
1179         return false;
1180     }
1181 
1182     if (params->has_announce_initial &&
1183         params->announce_initial > 100000) {
1184         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1185                    "announce_initial",
1186                    "is invalid, it must be less than 100000 ms");
1187         return false;
1188     }
1189     if (params->has_announce_max &&
1190         params->announce_max > 100000) {
1191         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1192                    "announce_max",
1193                    "is invalid, it must be less than 100000 ms");
1194        return false;
1195     }
1196     if (params->has_announce_rounds &&
1197         params->announce_rounds > 1000) {
1198         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1199                    "announce_rounds",
1200                    "is invalid, it must be in the range of 0 to 1000");
1201        return false;
1202     }
1203     if (params->has_announce_step &&
1204         (params->announce_step < 1 ||
1205         params->announce_step > 10000)) {
1206         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1207                    "announce_step",
1208                    "is invalid, it must be in the range of 1 to 10000 ms");
1209        return false;
1210     }
1211     return true;
1212 }
1213 
1214 static void migrate_params_test_apply(MigrateSetParameters *params,
1215                                       MigrationParameters *dest)
1216 {
1217     *dest = migrate_get_current()->parameters;
1218 
1219     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1220 
1221     if (params->has_compress_level) {
1222         dest->compress_level = params->compress_level;
1223     }
1224 
1225     if (params->has_compress_threads) {
1226         dest->compress_threads = params->compress_threads;
1227     }
1228 
1229     if (params->has_compress_wait_thread) {
1230         dest->compress_wait_thread = params->compress_wait_thread;
1231     }
1232 
1233     if (params->has_decompress_threads) {
1234         dest->decompress_threads = params->decompress_threads;
1235     }
1236 
1237     if (params->has_cpu_throttle_initial) {
1238         dest->cpu_throttle_initial = params->cpu_throttle_initial;
1239     }
1240 
1241     if (params->has_cpu_throttle_increment) {
1242         dest->cpu_throttle_increment = params->cpu_throttle_increment;
1243     }
1244 
1245     if (params->has_tls_creds) {
1246         assert(params->tls_creds->type == QTYPE_QSTRING);
1247         dest->tls_creds = g_strdup(params->tls_creds->u.s);
1248     }
1249 
1250     if (params->has_tls_hostname) {
1251         assert(params->tls_hostname->type == QTYPE_QSTRING);
1252         dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1253     }
1254 
1255     if (params->has_max_bandwidth) {
1256         dest->max_bandwidth = params->max_bandwidth;
1257     }
1258 
1259     if (params->has_downtime_limit) {
1260         dest->downtime_limit = params->downtime_limit;
1261     }
1262 
1263     if (params->has_x_checkpoint_delay) {
1264         dest->x_checkpoint_delay = params->x_checkpoint_delay;
1265     }
1266 
1267     if (params->has_block_incremental) {
1268         dest->block_incremental = params->block_incremental;
1269     }
1270     if (params->has_multifd_channels) {
1271         dest->multifd_channels = params->multifd_channels;
1272     }
1273     if (params->has_xbzrle_cache_size) {
1274         dest->xbzrle_cache_size = params->xbzrle_cache_size;
1275     }
1276     if (params->has_max_postcopy_bandwidth) {
1277         dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1278     }
1279     if (params->has_max_cpu_throttle) {
1280         dest->max_cpu_throttle = params->max_cpu_throttle;
1281     }
1282     if (params->has_announce_initial) {
1283         dest->announce_initial = params->announce_initial;
1284     }
1285     if (params->has_announce_max) {
1286         dest->announce_max = params->announce_max;
1287     }
1288     if (params->has_announce_rounds) {
1289         dest->announce_rounds = params->announce_rounds;
1290     }
1291     if (params->has_announce_step) {
1292         dest->announce_step = params->announce_step;
1293     }
1294 }
1295 
1296 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1297 {
1298     MigrationState *s = migrate_get_current();
1299 
1300     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1301 
1302     if (params->has_compress_level) {
1303         s->parameters.compress_level = params->compress_level;
1304     }
1305 
1306     if (params->has_compress_threads) {
1307         s->parameters.compress_threads = params->compress_threads;
1308     }
1309 
1310     if (params->has_compress_wait_thread) {
1311         s->parameters.compress_wait_thread = params->compress_wait_thread;
1312     }
1313 
1314     if (params->has_decompress_threads) {
1315         s->parameters.decompress_threads = params->decompress_threads;
1316     }
1317 
1318     if (params->has_cpu_throttle_initial) {
1319         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1320     }
1321 
1322     if (params->has_cpu_throttle_increment) {
1323         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1324     }
1325 
1326     if (params->has_tls_creds) {
1327         g_free(s->parameters.tls_creds);
1328         assert(params->tls_creds->type == QTYPE_QSTRING);
1329         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1330     }
1331 
1332     if (params->has_tls_hostname) {
1333         g_free(s->parameters.tls_hostname);
1334         assert(params->tls_hostname->type == QTYPE_QSTRING);
1335         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1336     }
1337 
1338     if (params->has_tls_authz) {
1339         g_free(s->parameters.tls_authz);
1340         assert(params->tls_authz->type == QTYPE_QSTRING);
1341         s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1342     }
1343 
1344     if (params->has_max_bandwidth) {
1345         s->parameters.max_bandwidth = params->max_bandwidth;
1346         if (s->to_dst_file && !migration_in_postcopy()) {
1347             qemu_file_set_rate_limit(s->to_dst_file,
1348                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1349         }
1350     }
1351 
1352     if (params->has_downtime_limit) {
1353         s->parameters.downtime_limit = params->downtime_limit;
1354     }
1355 
1356     if (params->has_x_checkpoint_delay) {
1357         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1358         if (migration_in_colo_state()) {
1359             colo_checkpoint_notify(s);
1360         }
1361     }
1362 
1363     if (params->has_block_incremental) {
1364         s->parameters.block_incremental = params->block_incremental;
1365     }
1366     if (params->has_multifd_channels) {
1367         s->parameters.multifd_channels = params->multifd_channels;
1368     }
1369     if (params->has_xbzrle_cache_size) {
1370         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1371         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1372     }
1373     if (params->has_max_postcopy_bandwidth) {
1374         s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1375         if (s->to_dst_file && migration_in_postcopy()) {
1376             qemu_file_set_rate_limit(s->to_dst_file,
1377                     s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1378         }
1379     }
1380     if (params->has_max_cpu_throttle) {
1381         s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1382     }
1383     if (params->has_announce_initial) {
1384         s->parameters.announce_initial = params->announce_initial;
1385     }
1386     if (params->has_announce_max) {
1387         s->parameters.announce_max = params->announce_max;
1388     }
1389     if (params->has_announce_rounds) {
1390         s->parameters.announce_rounds = params->announce_rounds;
1391     }
1392     if (params->has_announce_step) {
1393         s->parameters.announce_step = params->announce_step;
1394     }
1395 }
1396 
1397 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1398 {
1399     MigrationParameters tmp;
1400 
1401     /* TODO Rewrite "" to null instead */
1402     if (params->has_tls_creds
1403         && params->tls_creds->type == QTYPE_QNULL) {
1404         qobject_unref(params->tls_creds->u.n);
1405         params->tls_creds->type = QTYPE_QSTRING;
1406         params->tls_creds->u.s = strdup("");
1407     }
1408     /* TODO Rewrite "" to null instead */
1409     if (params->has_tls_hostname
1410         && params->tls_hostname->type == QTYPE_QNULL) {
1411         qobject_unref(params->tls_hostname->u.n);
1412         params->tls_hostname->type = QTYPE_QSTRING;
1413         params->tls_hostname->u.s = strdup("");
1414     }
1415 
1416     migrate_params_test_apply(params, &tmp);
1417 
1418     if (!migrate_params_check(&tmp, errp)) {
1419         /* Invalid parameter */
1420         return;
1421     }
1422 
1423     migrate_params_apply(params, errp);
1424 }
1425 
1426 
1427 void qmp_migrate_start_postcopy(Error **errp)
1428 {
1429     MigrationState *s = migrate_get_current();
1430 
1431     if (!migrate_postcopy()) {
1432         error_setg(errp, "Enable postcopy with migrate_set_capability before"
1433                          " the start of migration");
1434         return;
1435     }
1436 
1437     if (s->state == MIGRATION_STATUS_NONE) {
1438         error_setg(errp, "Postcopy must be started after migration has been"
1439                          " started");
1440         return;
1441     }
1442     /*
1443      * we don't error if migration has finished since that would be racy
1444      * with issuing this command.
1445      */
1446     atomic_set(&s->start_postcopy, true);
1447 }
1448 
1449 /* shared migration helpers */
1450 
1451 void migrate_set_state(int *state, int old_state, int new_state)
1452 {
1453     assert(new_state < MIGRATION_STATUS__MAX);
1454     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1455         trace_migrate_set_state(MigrationStatus_str(new_state));
1456         migrate_generate_event(new_state);
1457     }
1458 }
1459 
1460 static MigrationCapabilityStatusList *migrate_cap_add(
1461     MigrationCapabilityStatusList *list,
1462     MigrationCapability index,
1463     bool state)
1464 {
1465     MigrationCapabilityStatusList *cap;
1466 
1467     cap = g_new0(MigrationCapabilityStatusList, 1);
1468     cap->value = g_new0(MigrationCapabilityStatus, 1);
1469     cap->value->capability = index;
1470     cap->value->state = state;
1471     cap->next = list;
1472 
1473     return cap;
1474 }
1475 
1476 void migrate_set_block_enabled(bool value, Error **errp)
1477 {
1478     MigrationCapabilityStatusList *cap;
1479 
1480     cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1481     qmp_migrate_set_capabilities(cap, errp);
1482     qapi_free_MigrationCapabilityStatusList(cap);
1483 }
1484 
1485 static void migrate_set_block_incremental(MigrationState *s, bool value)
1486 {
1487     s->parameters.block_incremental = value;
1488 }
1489 
1490 static void block_cleanup_parameters(MigrationState *s)
1491 {
1492     if (s->must_remove_block_options) {
1493         /* setting to false can never fail */
1494         migrate_set_block_enabled(false, &error_abort);
1495         migrate_set_block_incremental(s, false);
1496         s->must_remove_block_options = false;
1497     }
1498 }
1499 
1500 static void migrate_fd_cleanup(MigrationState *s)
1501 {
1502     qemu_bh_delete(s->cleanup_bh);
1503     s->cleanup_bh = NULL;
1504 
1505     qemu_savevm_state_cleanup();
1506 
1507     if (s->to_dst_file) {
1508         QEMUFile *tmp;
1509 
1510         trace_migrate_fd_cleanup();
1511         qemu_mutex_unlock_iothread();
1512         if (s->migration_thread_running) {
1513             qemu_thread_join(&s->thread);
1514             s->migration_thread_running = false;
1515         }
1516         qemu_mutex_lock_iothread();
1517 
1518         multifd_save_cleanup();
1519         qemu_mutex_lock(&s->qemu_file_lock);
1520         tmp = s->to_dst_file;
1521         s->to_dst_file = NULL;
1522         qemu_mutex_unlock(&s->qemu_file_lock);
1523         /*
1524          * Close the file handle without the lock to make sure the
1525          * critical section won't block for long.
1526          */
1527         qemu_fclose(tmp);
1528     }
1529 
1530     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1531            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1532 
1533     if (s->state == MIGRATION_STATUS_CANCELLING) {
1534         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1535                           MIGRATION_STATUS_CANCELLED);
1536     }
1537 
1538     if (s->error) {
1539         /* It is used on info migrate.  We can't free it */
1540         error_report_err(error_copy(s->error));
1541     }
1542     notifier_list_notify(&migration_state_notifiers, s);
1543     block_cleanup_parameters(s);
1544 }
1545 
1546 static void migrate_fd_cleanup_schedule(MigrationState *s)
1547 {
1548     /*
1549      * Ref the state for bh, because it may be called when
1550      * there're already no other refs
1551      */
1552     object_ref(OBJECT(s));
1553     qemu_bh_schedule(s->cleanup_bh);
1554 }
1555 
1556 static void migrate_fd_cleanup_bh(void *opaque)
1557 {
1558     MigrationState *s = opaque;
1559     migrate_fd_cleanup(s);
1560     object_unref(OBJECT(s));
1561 }
1562 
1563 void migrate_set_error(MigrationState *s, const Error *error)
1564 {
1565     qemu_mutex_lock(&s->error_mutex);
1566     if (!s->error) {
1567         s->error = error_copy(error);
1568     }
1569     qemu_mutex_unlock(&s->error_mutex);
1570 }
1571 
1572 void migrate_fd_error(MigrationState *s, const Error *error)
1573 {
1574     trace_migrate_fd_error(error_get_pretty(error));
1575     assert(s->to_dst_file == NULL);
1576     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1577                       MIGRATION_STATUS_FAILED);
1578     migrate_set_error(s, error);
1579 }
1580 
1581 static void migrate_fd_cancel(MigrationState *s)
1582 {
1583     int old_state ;
1584     QEMUFile *f = migrate_get_current()->to_dst_file;
1585     trace_migrate_fd_cancel();
1586 
1587     if (s->rp_state.from_dst_file) {
1588         /* shutdown the rp socket, so causing the rp thread to shutdown */
1589         qemu_file_shutdown(s->rp_state.from_dst_file);
1590     }
1591 
1592     do {
1593         old_state = s->state;
1594         if (!migration_is_setup_or_active(old_state)) {
1595             break;
1596         }
1597         /* If the migration is paused, kick it out of the pause */
1598         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1599             qemu_sem_post(&s->pause_sem);
1600         }
1601         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1602     } while (s->state != MIGRATION_STATUS_CANCELLING);
1603 
1604     /*
1605      * If we're unlucky the migration code might be stuck somewhere in a
1606      * send/write while the network has failed and is waiting to timeout;
1607      * if we've got shutdown(2) available then we can force it to quit.
1608      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1609      * called in a bh, so there is no race against this cancel.
1610      */
1611     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1612         qemu_file_shutdown(f);
1613     }
1614     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1615         Error *local_err = NULL;
1616 
1617         bdrv_invalidate_cache_all(&local_err);
1618         if (local_err) {
1619             error_report_err(local_err);
1620         } else {
1621             s->block_inactive = false;
1622         }
1623     }
1624 }
1625 
1626 void add_migration_state_change_notifier(Notifier *notify)
1627 {
1628     notifier_list_add(&migration_state_notifiers, notify);
1629 }
1630 
1631 void remove_migration_state_change_notifier(Notifier *notify)
1632 {
1633     notifier_remove(notify);
1634 }
1635 
1636 bool migration_in_setup(MigrationState *s)
1637 {
1638     return s->state == MIGRATION_STATUS_SETUP;
1639 }
1640 
1641 bool migration_has_finished(MigrationState *s)
1642 {
1643     return s->state == MIGRATION_STATUS_COMPLETED;
1644 }
1645 
1646 bool migration_has_failed(MigrationState *s)
1647 {
1648     return (s->state == MIGRATION_STATUS_CANCELLED ||
1649             s->state == MIGRATION_STATUS_FAILED);
1650 }
1651 
1652 bool migration_in_postcopy(void)
1653 {
1654     MigrationState *s = migrate_get_current();
1655 
1656     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1657 }
1658 
1659 bool migration_in_postcopy_after_devices(MigrationState *s)
1660 {
1661     return migration_in_postcopy() && s->postcopy_after_devices;
1662 }
1663 
1664 bool migration_is_idle(void)
1665 {
1666     MigrationState *s = current_migration;
1667 
1668     if (!s) {
1669         return true;
1670     }
1671 
1672     switch (s->state) {
1673     case MIGRATION_STATUS_NONE:
1674     case MIGRATION_STATUS_CANCELLED:
1675     case MIGRATION_STATUS_COMPLETED:
1676     case MIGRATION_STATUS_FAILED:
1677         return true;
1678     case MIGRATION_STATUS_SETUP:
1679     case MIGRATION_STATUS_CANCELLING:
1680     case MIGRATION_STATUS_ACTIVE:
1681     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1682     case MIGRATION_STATUS_COLO:
1683     case MIGRATION_STATUS_PRE_SWITCHOVER:
1684     case MIGRATION_STATUS_DEVICE:
1685         return false;
1686     case MIGRATION_STATUS__MAX:
1687         g_assert_not_reached();
1688     }
1689 
1690     return false;
1691 }
1692 
1693 void migrate_init(MigrationState *s)
1694 {
1695     /*
1696      * Reinitialise all migration state, except
1697      * parameters/capabilities that the user set, and
1698      * locks.
1699      */
1700     s->bytes_xfer = 0;
1701     s->cleanup_bh = 0;
1702     s->to_dst_file = NULL;
1703     s->state = MIGRATION_STATUS_NONE;
1704     s->rp_state.from_dst_file = NULL;
1705     s->rp_state.error = false;
1706     s->mbps = 0.0;
1707     s->pages_per_second = 0.0;
1708     s->downtime = 0;
1709     s->expected_downtime = 0;
1710     s->setup_time = 0;
1711     s->start_postcopy = false;
1712     s->postcopy_after_devices = false;
1713     s->migration_thread_running = false;
1714     error_free(s->error);
1715     s->error = NULL;
1716 
1717     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1718 
1719     s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1720     s->total_time = 0;
1721     s->vm_was_running = false;
1722     s->iteration_initial_bytes = 0;
1723     s->threshold_size = 0;
1724 }
1725 
1726 static GSList *migration_blockers;
1727 
1728 int migrate_add_blocker(Error *reason, Error **errp)
1729 {
1730     if (only_migratable) {
1731         error_propagate_prepend(errp, error_copy(reason),
1732                                 "disallowing migration blocker "
1733                                 "(--only-migratable) for: ");
1734         return -EACCES;
1735     }
1736 
1737     if (migration_is_idle()) {
1738         migration_blockers = g_slist_prepend(migration_blockers, reason);
1739         return 0;
1740     }
1741 
1742     error_propagate_prepend(errp, error_copy(reason),
1743                             "disallowing migration blocker "
1744                             "(migration in progress) for: ");
1745     return -EBUSY;
1746 }
1747 
1748 void migrate_del_blocker(Error *reason)
1749 {
1750     migration_blockers = g_slist_remove(migration_blockers, reason);
1751 }
1752 
1753 void qmp_migrate_incoming(const char *uri, Error **errp)
1754 {
1755     Error *local_err = NULL;
1756     static bool once = true;
1757 
1758     if (!deferred_incoming) {
1759         error_setg(errp, "For use with '-incoming defer'");
1760         return;
1761     }
1762     if (!once) {
1763         error_setg(errp, "The incoming migration has already been started");
1764     }
1765 
1766     qemu_start_incoming_migration(uri, &local_err);
1767 
1768     if (local_err) {
1769         error_propagate(errp, local_err);
1770         return;
1771     }
1772 
1773     once = false;
1774 }
1775 
1776 void qmp_migrate_recover(const char *uri, Error **errp)
1777 {
1778     MigrationIncomingState *mis = migration_incoming_get_current();
1779 
1780     if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1781         error_setg(errp, "Migrate recover can only be run "
1782                    "when postcopy is paused.");
1783         return;
1784     }
1785 
1786     if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1787                        false, true) == true) {
1788         error_setg(errp, "Migrate recovery is triggered already");
1789         return;
1790     }
1791 
1792     /*
1793      * Note that this call will never start a real migration; it will
1794      * only re-setup the migration stream and poke existing migration
1795      * to continue using that newly established channel.
1796      */
1797     qemu_start_incoming_migration(uri, errp);
1798 }
1799 
1800 void qmp_migrate_pause(Error **errp)
1801 {
1802     MigrationState *ms = migrate_get_current();
1803     MigrationIncomingState *mis = migration_incoming_get_current();
1804     int ret;
1805 
1806     if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1807         /* Source side, during postcopy */
1808         qemu_mutex_lock(&ms->qemu_file_lock);
1809         ret = qemu_file_shutdown(ms->to_dst_file);
1810         qemu_mutex_unlock(&ms->qemu_file_lock);
1811         if (ret) {
1812             error_setg(errp, "Failed to pause source migration");
1813         }
1814         return;
1815     }
1816 
1817     if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1818         ret = qemu_file_shutdown(mis->from_src_file);
1819         if (ret) {
1820             error_setg(errp, "Failed to pause destination migration");
1821         }
1822         return;
1823     }
1824 
1825     error_setg(errp, "migrate-pause is currently only supported "
1826                "during postcopy-active state");
1827 }
1828 
1829 bool migration_is_blocked(Error **errp)
1830 {
1831     if (qemu_savevm_state_blocked(errp)) {
1832         return true;
1833     }
1834 
1835     if (migration_blockers) {
1836         error_propagate(errp, error_copy(migration_blockers->data));
1837         return true;
1838     }
1839 
1840     return false;
1841 }
1842 
1843 /* Returns true if continue to migrate, or false if error detected */
1844 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1845                             bool resume, Error **errp)
1846 {
1847     Error *local_err = NULL;
1848 
1849     if (resume) {
1850         if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1851             error_setg(errp, "Cannot resume if there is no "
1852                        "paused migration");
1853             return false;
1854         }
1855 
1856         /*
1857          * Postcopy recovery won't work well with release-ram
1858          * capability since release-ram will drop the page buffer as
1859          * long as the page is put into the send buffer.  So if there
1860          * is a network failure happened, any page buffers that have
1861          * not yet reached the destination VM but have already been
1862          * sent from the source VM will be lost forever.  Let's refuse
1863          * the client from resuming such a postcopy migration.
1864          * Luckily release-ram was designed to only be used when src
1865          * and destination VMs are on the same host, so it should be
1866          * fine.
1867          */
1868         if (migrate_release_ram()) {
1869             error_setg(errp, "Postcopy recovery cannot work "
1870                        "when release-ram capability is set");
1871             return false;
1872         }
1873 
1874         /* This is a resume, skip init status */
1875         return true;
1876     }
1877 
1878     if (migration_is_setup_or_active(s->state) ||
1879         s->state == MIGRATION_STATUS_CANCELLING ||
1880         s->state == MIGRATION_STATUS_COLO) {
1881         error_setg(errp, QERR_MIGRATION_ACTIVE);
1882         return false;
1883     }
1884 
1885     if (runstate_check(RUN_STATE_INMIGRATE)) {
1886         error_setg(errp, "Guest is waiting for an incoming migration");
1887         return false;
1888     }
1889 
1890     if (migration_is_blocked(errp)) {
1891         return false;
1892     }
1893 
1894     if (blk || blk_inc) {
1895         if (migrate_use_block() || migrate_use_block_incremental()) {
1896             error_setg(errp, "Command options are incompatible with "
1897                        "current migration capabilities");
1898             return false;
1899         }
1900         migrate_set_block_enabled(true, &local_err);
1901         if (local_err) {
1902             error_propagate(errp, local_err);
1903             return false;
1904         }
1905         s->must_remove_block_options = true;
1906     }
1907 
1908     if (blk_inc) {
1909         migrate_set_block_incremental(s, true);
1910     }
1911 
1912     migrate_init(s);
1913 
1914     return true;
1915 }
1916 
1917 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1918                  bool has_inc, bool inc, bool has_detach, bool detach,
1919                  bool has_resume, bool resume, Error **errp)
1920 {
1921     Error *local_err = NULL;
1922     MigrationState *s = migrate_get_current();
1923     const char *p;
1924 
1925     if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1926                          has_resume && resume, errp)) {
1927         /* Error detected, put into errp */
1928         return;
1929     }
1930 
1931     if (strstart(uri, "tcp:", &p)) {
1932         tcp_start_outgoing_migration(s, p, &local_err);
1933 #ifdef CONFIG_RDMA
1934     } else if (strstart(uri, "rdma:", &p)) {
1935         rdma_start_outgoing_migration(s, p, &local_err);
1936 #endif
1937     } else if (strstart(uri, "exec:", &p)) {
1938         exec_start_outgoing_migration(s, p, &local_err);
1939     } else if (strstart(uri, "unix:", &p)) {
1940         unix_start_outgoing_migration(s, p, &local_err);
1941     } else if (strstart(uri, "fd:", &p)) {
1942         fd_start_outgoing_migration(s, p, &local_err);
1943     } else {
1944         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1945                    "a valid migration protocol");
1946         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1947                           MIGRATION_STATUS_FAILED);
1948         block_cleanup_parameters(s);
1949         return;
1950     }
1951 
1952     if (local_err) {
1953         migrate_fd_error(s, local_err);
1954         error_propagate(errp, local_err);
1955         return;
1956     }
1957 }
1958 
1959 void qmp_migrate_cancel(Error **errp)
1960 {
1961     migrate_fd_cancel(migrate_get_current());
1962 }
1963 
1964 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1965 {
1966     MigrationState *s = migrate_get_current();
1967     if (s->state != state) {
1968         error_setg(errp,  "Migration not in expected state: %s",
1969                    MigrationStatus_str(s->state));
1970         return;
1971     }
1972     qemu_sem_post(&s->pause_sem);
1973 }
1974 
1975 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1976 {
1977     MigrateSetParameters p = {
1978         .has_xbzrle_cache_size = true,
1979         .xbzrle_cache_size = value,
1980     };
1981 
1982     qmp_migrate_set_parameters(&p, errp);
1983 }
1984 
1985 int64_t qmp_query_migrate_cache_size(Error **errp)
1986 {
1987     return migrate_xbzrle_cache_size();
1988 }
1989 
1990 void qmp_migrate_set_speed(int64_t value, Error **errp)
1991 {
1992     MigrateSetParameters p = {
1993         .has_max_bandwidth = true,
1994         .max_bandwidth = value,
1995     };
1996 
1997     qmp_migrate_set_parameters(&p, errp);
1998 }
1999 
2000 void qmp_migrate_set_downtime(double value, Error **errp)
2001 {
2002     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2003         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
2004                          "the range of 0 to %d seconds",
2005                          MAX_MIGRATE_DOWNTIME_SECONDS);
2006         return;
2007     }
2008 
2009     value *= 1000; /* Convert to milliseconds */
2010     value = MAX(0, MIN(INT64_MAX, value));
2011 
2012     MigrateSetParameters p = {
2013         .has_downtime_limit = true,
2014         .downtime_limit = value,
2015     };
2016 
2017     qmp_migrate_set_parameters(&p, errp);
2018 }
2019 
2020 bool migrate_release_ram(void)
2021 {
2022     MigrationState *s;
2023 
2024     s = migrate_get_current();
2025 
2026     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2027 }
2028 
2029 bool migrate_postcopy_ram(void)
2030 {
2031     MigrationState *s;
2032 
2033     s = migrate_get_current();
2034 
2035     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2036 }
2037 
2038 bool migrate_postcopy(void)
2039 {
2040     return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2041 }
2042 
2043 bool migrate_auto_converge(void)
2044 {
2045     MigrationState *s;
2046 
2047     s = migrate_get_current();
2048 
2049     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2050 }
2051 
2052 bool migrate_zero_blocks(void)
2053 {
2054     MigrationState *s;
2055 
2056     s = migrate_get_current();
2057 
2058     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2059 }
2060 
2061 bool migrate_postcopy_blocktime(void)
2062 {
2063     MigrationState *s;
2064 
2065     s = migrate_get_current();
2066 
2067     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2068 }
2069 
2070 bool migrate_use_compression(void)
2071 {
2072     MigrationState *s;
2073 
2074     s = migrate_get_current();
2075 
2076     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2077 }
2078 
2079 int migrate_compress_level(void)
2080 {
2081     MigrationState *s;
2082 
2083     s = migrate_get_current();
2084 
2085     return s->parameters.compress_level;
2086 }
2087 
2088 int migrate_compress_threads(void)
2089 {
2090     MigrationState *s;
2091 
2092     s = migrate_get_current();
2093 
2094     return s->parameters.compress_threads;
2095 }
2096 
2097 int migrate_compress_wait_thread(void)
2098 {
2099     MigrationState *s;
2100 
2101     s = migrate_get_current();
2102 
2103     return s->parameters.compress_wait_thread;
2104 }
2105 
2106 int migrate_decompress_threads(void)
2107 {
2108     MigrationState *s;
2109 
2110     s = migrate_get_current();
2111 
2112     return s->parameters.decompress_threads;
2113 }
2114 
2115 bool migrate_dirty_bitmaps(void)
2116 {
2117     MigrationState *s;
2118 
2119     s = migrate_get_current();
2120 
2121     return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2122 }
2123 
2124 bool migrate_ignore_shared(void)
2125 {
2126     MigrationState *s;
2127 
2128     s = migrate_get_current();
2129 
2130     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2131 }
2132 
2133 bool migrate_use_events(void)
2134 {
2135     MigrationState *s;
2136 
2137     s = migrate_get_current();
2138 
2139     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2140 }
2141 
2142 bool migrate_use_multifd(void)
2143 {
2144     MigrationState *s;
2145 
2146     s = migrate_get_current();
2147 
2148     return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2149 }
2150 
2151 bool migrate_pause_before_switchover(void)
2152 {
2153     MigrationState *s;
2154 
2155     s = migrate_get_current();
2156 
2157     return s->enabled_capabilities[
2158         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2159 }
2160 
2161 int migrate_multifd_channels(void)
2162 {
2163     MigrationState *s;
2164 
2165     s = migrate_get_current();
2166 
2167     return s->parameters.multifd_channels;
2168 }
2169 
2170 int migrate_use_xbzrle(void)
2171 {
2172     MigrationState *s;
2173 
2174     s = migrate_get_current();
2175 
2176     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2177 }
2178 
2179 int64_t migrate_xbzrle_cache_size(void)
2180 {
2181     MigrationState *s;
2182 
2183     s = migrate_get_current();
2184 
2185     return s->parameters.xbzrle_cache_size;
2186 }
2187 
2188 static int64_t migrate_max_postcopy_bandwidth(void)
2189 {
2190     MigrationState *s;
2191 
2192     s = migrate_get_current();
2193 
2194     return s->parameters.max_postcopy_bandwidth;
2195 }
2196 
2197 bool migrate_use_block(void)
2198 {
2199     MigrationState *s;
2200 
2201     s = migrate_get_current();
2202 
2203     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2204 }
2205 
2206 bool migrate_use_return_path(void)
2207 {
2208     MigrationState *s;
2209 
2210     s = migrate_get_current();
2211 
2212     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2213 }
2214 
2215 bool migrate_use_block_incremental(void)
2216 {
2217     MigrationState *s;
2218 
2219     s = migrate_get_current();
2220 
2221     return s->parameters.block_incremental;
2222 }
2223 
2224 /* migration thread support */
2225 /*
2226  * Something bad happened to the RP stream, mark an error
2227  * The caller shall print or trace something to indicate why
2228  */
2229 static void mark_source_rp_bad(MigrationState *s)
2230 {
2231     s->rp_state.error = true;
2232 }
2233 
2234 static struct rp_cmd_args {
2235     ssize_t     len; /* -1 = variable */
2236     const char *name;
2237 } rp_cmd_args[] = {
2238     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
2239     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
2240     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
2241     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
2242     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
2243     [MIG_RP_MSG_RECV_BITMAP]    = { .len = -1, .name = "RECV_BITMAP" },
2244     [MIG_RP_MSG_RESUME_ACK]     = { .len =  4, .name = "RESUME_ACK" },
2245     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
2246 };
2247 
2248 /*
2249  * Process a request for pages received on the return path,
2250  * We're allowed to send more than requested (e.g. to round to our page size)
2251  * and we don't need to send pages that have already been sent.
2252  */
2253 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2254                                        ram_addr_t start, size_t len)
2255 {
2256     long our_host_ps = getpagesize();
2257 
2258     trace_migrate_handle_rp_req_pages(rbname, start, len);
2259 
2260     /*
2261      * Since we currently insist on matching page sizes, just sanity check
2262      * we're being asked for whole host pages.
2263      */
2264     if (start & (our_host_ps-1) ||
2265        (len & (our_host_ps-1))) {
2266         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2267                      " len: %zd", __func__, start, len);
2268         mark_source_rp_bad(ms);
2269         return;
2270     }
2271 
2272     if (ram_save_queue_pages(rbname, start, len)) {
2273         mark_source_rp_bad(ms);
2274     }
2275 }
2276 
2277 /* Return true to retry, false to quit */
2278 static bool postcopy_pause_return_path_thread(MigrationState *s)
2279 {
2280     trace_postcopy_pause_return_path();
2281 
2282     qemu_sem_wait(&s->postcopy_pause_rp_sem);
2283 
2284     trace_postcopy_pause_return_path_continued();
2285 
2286     return true;
2287 }
2288 
2289 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2290 {
2291     RAMBlock *block = qemu_ram_block_by_name(block_name);
2292 
2293     if (!block) {
2294         error_report("%s: invalid block name '%s'", __func__, block_name);
2295         return -EINVAL;
2296     }
2297 
2298     /* Fetch the received bitmap and refresh the dirty bitmap */
2299     return ram_dirty_bitmap_reload(s, block);
2300 }
2301 
2302 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2303 {
2304     trace_source_return_path_thread_resume_ack(value);
2305 
2306     if (value != MIGRATION_RESUME_ACK_VALUE) {
2307         error_report("%s: illegal resume_ack value %"PRIu32,
2308                      __func__, value);
2309         return -1;
2310     }
2311 
2312     /* Now both sides are active. */
2313     migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2314                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2315 
2316     /* Notify send thread that time to continue send pages */
2317     qemu_sem_post(&s->rp_state.rp_sem);
2318 
2319     return 0;
2320 }
2321 
2322 /*
2323  * Handles messages sent on the return path towards the source VM
2324  *
2325  */
2326 static void *source_return_path_thread(void *opaque)
2327 {
2328     MigrationState *ms = opaque;
2329     QEMUFile *rp = ms->rp_state.from_dst_file;
2330     uint16_t header_len, header_type;
2331     uint8_t buf[512];
2332     uint32_t tmp32, sibling_error;
2333     ram_addr_t start = 0; /* =0 to silence warning */
2334     size_t  len = 0, expected_len;
2335     int res;
2336 
2337     trace_source_return_path_thread_entry();
2338     rcu_register_thread();
2339 
2340 retry:
2341     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2342            migration_is_setup_or_active(ms->state)) {
2343         trace_source_return_path_thread_loop_top();
2344         header_type = qemu_get_be16(rp);
2345         header_len = qemu_get_be16(rp);
2346 
2347         if (qemu_file_get_error(rp)) {
2348             mark_source_rp_bad(ms);
2349             goto out;
2350         }
2351 
2352         if (header_type >= MIG_RP_MSG_MAX ||
2353             header_type == MIG_RP_MSG_INVALID) {
2354             error_report("RP: Received invalid message 0x%04x length 0x%04x",
2355                     header_type, header_len);
2356             mark_source_rp_bad(ms);
2357             goto out;
2358         }
2359 
2360         if ((rp_cmd_args[header_type].len != -1 &&
2361             header_len != rp_cmd_args[header_type].len) ||
2362             header_len > sizeof(buf)) {
2363             error_report("RP: Received '%s' message (0x%04x) with"
2364                     "incorrect length %d expecting %zu",
2365                     rp_cmd_args[header_type].name, header_type, header_len,
2366                     (size_t)rp_cmd_args[header_type].len);
2367             mark_source_rp_bad(ms);
2368             goto out;
2369         }
2370 
2371         /* We know we've got a valid header by this point */
2372         res = qemu_get_buffer(rp, buf, header_len);
2373         if (res != header_len) {
2374             error_report("RP: Failed reading data for message 0x%04x"
2375                          " read %d expected %d",
2376                          header_type, res, header_len);
2377             mark_source_rp_bad(ms);
2378             goto out;
2379         }
2380 
2381         /* OK, we have the message and the data */
2382         switch (header_type) {
2383         case MIG_RP_MSG_SHUT:
2384             sibling_error = ldl_be_p(buf);
2385             trace_source_return_path_thread_shut(sibling_error);
2386             if (sibling_error) {
2387                 error_report("RP: Sibling indicated error %d", sibling_error);
2388                 mark_source_rp_bad(ms);
2389             }
2390             /*
2391              * We'll let the main thread deal with closing the RP
2392              * we could do a shutdown(2) on it, but we're the only user
2393              * anyway, so there's nothing gained.
2394              */
2395             goto out;
2396 
2397         case MIG_RP_MSG_PONG:
2398             tmp32 = ldl_be_p(buf);
2399             trace_source_return_path_thread_pong(tmp32);
2400             break;
2401 
2402         case MIG_RP_MSG_REQ_PAGES:
2403             start = ldq_be_p(buf);
2404             len = ldl_be_p(buf + 8);
2405             migrate_handle_rp_req_pages(ms, NULL, start, len);
2406             break;
2407 
2408         case MIG_RP_MSG_REQ_PAGES_ID:
2409             expected_len = 12 + 1; /* header + termination */
2410 
2411             if (header_len >= expected_len) {
2412                 start = ldq_be_p(buf);
2413                 len = ldl_be_p(buf + 8);
2414                 /* Now we expect an idstr */
2415                 tmp32 = buf[12]; /* Length of the following idstr */
2416                 buf[13 + tmp32] = '\0';
2417                 expected_len += tmp32;
2418             }
2419             if (header_len != expected_len) {
2420                 error_report("RP: Req_Page_id with length %d expecting %zd",
2421                         header_len, expected_len);
2422                 mark_source_rp_bad(ms);
2423                 goto out;
2424             }
2425             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2426             break;
2427 
2428         case MIG_RP_MSG_RECV_BITMAP:
2429             if (header_len < 1) {
2430                 error_report("%s: missing block name", __func__);
2431                 mark_source_rp_bad(ms);
2432                 goto out;
2433             }
2434             /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2435             buf[buf[0] + 1] = '\0';
2436             if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2437                 mark_source_rp_bad(ms);
2438                 goto out;
2439             }
2440             break;
2441 
2442         case MIG_RP_MSG_RESUME_ACK:
2443             tmp32 = ldl_be_p(buf);
2444             if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2445                 mark_source_rp_bad(ms);
2446                 goto out;
2447             }
2448             break;
2449 
2450         default:
2451             break;
2452         }
2453     }
2454 
2455 out:
2456     res = qemu_file_get_error(rp);
2457     if (res) {
2458         if (res == -EIO) {
2459             /*
2460              * Maybe there is something we can do: it looks like a
2461              * network down issue, and we pause for a recovery.
2462              */
2463             if (postcopy_pause_return_path_thread(ms)) {
2464                 /* Reload rp, reset the rest */
2465                 if (rp != ms->rp_state.from_dst_file) {
2466                     qemu_fclose(rp);
2467                     rp = ms->rp_state.from_dst_file;
2468                 }
2469                 ms->rp_state.error = false;
2470                 goto retry;
2471             }
2472         }
2473 
2474         trace_source_return_path_thread_bad_end();
2475         mark_source_rp_bad(ms);
2476     }
2477 
2478     trace_source_return_path_thread_end();
2479     ms->rp_state.from_dst_file = NULL;
2480     qemu_fclose(rp);
2481     rcu_unregister_thread();
2482     return NULL;
2483 }
2484 
2485 static int open_return_path_on_source(MigrationState *ms,
2486                                       bool create_thread)
2487 {
2488 
2489     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2490     if (!ms->rp_state.from_dst_file) {
2491         return -1;
2492     }
2493 
2494     trace_open_return_path_on_source();
2495 
2496     if (!create_thread) {
2497         /* We're done */
2498         return 0;
2499     }
2500 
2501     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2502                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2503 
2504     trace_open_return_path_on_source_continue();
2505 
2506     return 0;
2507 }
2508 
2509 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2510 static int await_return_path_close_on_source(MigrationState *ms)
2511 {
2512     /*
2513      * If this is a normal exit then the destination will send a SHUT and the
2514      * rp_thread will exit, however if there's an error we need to cause
2515      * it to exit.
2516      */
2517     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2518         /*
2519          * shutdown(2), if we have it, will cause it to unblock if it's stuck
2520          * waiting for the destination.
2521          */
2522         qemu_file_shutdown(ms->rp_state.from_dst_file);
2523         mark_source_rp_bad(ms);
2524     }
2525     trace_await_return_path_close_on_source_joining();
2526     qemu_thread_join(&ms->rp_state.rp_thread);
2527     trace_await_return_path_close_on_source_close();
2528     return ms->rp_state.error;
2529 }
2530 
2531 /*
2532  * Switch from normal iteration to postcopy
2533  * Returns non-0 on error
2534  */
2535 static int postcopy_start(MigrationState *ms)
2536 {
2537     int ret;
2538     QIOChannelBuffer *bioc;
2539     QEMUFile *fb;
2540     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2541     int64_t bandwidth = migrate_max_postcopy_bandwidth();
2542     bool restart_block = false;
2543     int cur_state = MIGRATION_STATUS_ACTIVE;
2544     if (!migrate_pause_before_switchover()) {
2545         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2546                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
2547     }
2548 
2549     trace_postcopy_start();
2550     qemu_mutex_lock_iothread();
2551     trace_postcopy_start_set_run();
2552 
2553     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2554     global_state_store();
2555     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2556     if (ret < 0) {
2557         goto fail;
2558     }
2559 
2560     ret = migration_maybe_pause(ms, &cur_state,
2561                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2562     if (ret < 0) {
2563         goto fail;
2564     }
2565 
2566     ret = bdrv_inactivate_all();
2567     if (ret < 0) {
2568         goto fail;
2569     }
2570     restart_block = true;
2571 
2572     /*
2573      * Cause any non-postcopiable, but iterative devices to
2574      * send out their final data.
2575      */
2576     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2577 
2578     /*
2579      * in Finish migrate and with the io-lock held everything should
2580      * be quiet, but we've potentially still got dirty pages and we
2581      * need to tell the destination to throw any pages it's already received
2582      * that are dirty
2583      */
2584     if (migrate_postcopy_ram()) {
2585         if (ram_postcopy_send_discard_bitmap(ms)) {
2586             error_report("postcopy send discard bitmap failed");
2587             goto fail;
2588         }
2589     }
2590 
2591     /*
2592      * send rest of state - note things that are doing postcopy
2593      * will notice we're in POSTCOPY_ACTIVE and not actually
2594      * wrap their state up here
2595      */
2596     /* 0 max-postcopy-bandwidth means unlimited */
2597     if (!bandwidth) {
2598         qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2599     } else {
2600         qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2601     }
2602     if (migrate_postcopy_ram()) {
2603         /* Ping just for debugging, helps line traces up */
2604         qemu_savevm_send_ping(ms->to_dst_file, 2);
2605     }
2606 
2607     /*
2608      * While loading the device state we may trigger page transfer
2609      * requests and the fd must be free to process those, and thus
2610      * the destination must read the whole device state off the fd before
2611      * it starts processing it.  Unfortunately the ad-hoc migration format
2612      * doesn't allow the destination to know the size to read without fully
2613      * parsing it through each devices load-state code (especially the open
2614      * coded devices that use get/put).
2615      * So we wrap the device state up in a package with a length at the start;
2616      * to do this we use a qemu_buf to hold the whole of the device state.
2617      */
2618     bioc = qio_channel_buffer_new(4096);
2619     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2620     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2621     object_unref(OBJECT(bioc));
2622 
2623     /*
2624      * Make sure the receiver can get incoming pages before we send the rest
2625      * of the state
2626      */
2627     qemu_savevm_send_postcopy_listen(fb);
2628 
2629     qemu_savevm_state_complete_precopy(fb, false, false);
2630     if (migrate_postcopy_ram()) {
2631         qemu_savevm_send_ping(fb, 3);
2632     }
2633 
2634     qemu_savevm_send_postcopy_run(fb);
2635 
2636     /* <><> end of stuff going into the package */
2637 
2638     /* Last point of recovery; as soon as we send the package the destination
2639      * can open devices and potentially start running.
2640      * Lets just check again we've not got any errors.
2641      */
2642     ret = qemu_file_get_error(ms->to_dst_file);
2643     if (ret) {
2644         error_report("postcopy_start: Migration stream errored (pre package)");
2645         goto fail_closefb;
2646     }
2647 
2648     restart_block = false;
2649 
2650     /* Now send that blob */
2651     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2652         goto fail_closefb;
2653     }
2654     qemu_fclose(fb);
2655 
2656     /* Send a notify to give a chance for anything that needs to happen
2657      * at the transition to postcopy and after the device state; in particular
2658      * spice needs to trigger a transition now
2659      */
2660     ms->postcopy_after_devices = true;
2661     notifier_list_notify(&migration_state_notifiers, ms);
2662 
2663     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2664 
2665     qemu_mutex_unlock_iothread();
2666 
2667     if (migrate_postcopy_ram()) {
2668         /*
2669          * Although this ping is just for debug, it could potentially be
2670          * used for getting a better measurement of downtime at the source.
2671          */
2672         qemu_savevm_send_ping(ms->to_dst_file, 4);
2673     }
2674 
2675     if (migrate_release_ram()) {
2676         ram_postcopy_migrated_memory_release(ms);
2677     }
2678 
2679     ret = qemu_file_get_error(ms->to_dst_file);
2680     if (ret) {
2681         error_report("postcopy_start: Migration stream errored");
2682         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2683                               MIGRATION_STATUS_FAILED);
2684     }
2685 
2686     return ret;
2687 
2688 fail_closefb:
2689     qemu_fclose(fb);
2690 fail:
2691     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2692                           MIGRATION_STATUS_FAILED);
2693     if (restart_block) {
2694         /* A failure happened early enough that we know the destination hasn't
2695          * accessed block devices, so we're safe to recover.
2696          */
2697         Error *local_err = NULL;
2698 
2699         bdrv_invalidate_cache_all(&local_err);
2700         if (local_err) {
2701             error_report_err(local_err);
2702         }
2703     }
2704     qemu_mutex_unlock_iothread();
2705     return -1;
2706 }
2707 
2708 /**
2709  * migration_maybe_pause: Pause if required to by
2710  * migrate_pause_before_switchover called with the iothread locked
2711  * Returns: 0 on success
2712  */
2713 static int migration_maybe_pause(MigrationState *s,
2714                                  int *current_active_state,
2715                                  int new_state)
2716 {
2717     if (!migrate_pause_before_switchover()) {
2718         return 0;
2719     }
2720 
2721     /* Since leaving this state is not atomic with posting the semaphore
2722      * it's possible that someone could have issued multiple migrate_continue
2723      * and the semaphore is incorrectly positive at this point;
2724      * the docs say it's undefined to reinit a semaphore that's already
2725      * init'd, so use timedwait to eat up any existing posts.
2726      */
2727     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2728         /* This block intentionally left blank */
2729     }
2730 
2731     qemu_mutex_unlock_iothread();
2732     migrate_set_state(&s->state, *current_active_state,
2733                       MIGRATION_STATUS_PRE_SWITCHOVER);
2734     qemu_sem_wait(&s->pause_sem);
2735     migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2736                       new_state);
2737     *current_active_state = new_state;
2738     qemu_mutex_lock_iothread();
2739 
2740     return s->state == new_state ? 0 : -EINVAL;
2741 }
2742 
2743 /**
2744  * migration_completion: Used by migration_thread when there's not much left.
2745  *   The caller 'breaks' the loop when this returns.
2746  *
2747  * @s: Current migration state
2748  */
2749 static void migration_completion(MigrationState *s)
2750 {
2751     int ret;
2752     int current_active_state = s->state;
2753 
2754     if (s->state == MIGRATION_STATUS_ACTIVE) {
2755         qemu_mutex_lock_iothread();
2756         s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2757         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2758         s->vm_was_running = runstate_is_running();
2759         ret = global_state_store();
2760 
2761         if (!ret) {
2762             bool inactivate = !migrate_colo_enabled();
2763             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2764             if (ret >= 0) {
2765                 ret = migration_maybe_pause(s, &current_active_state,
2766                                             MIGRATION_STATUS_DEVICE);
2767             }
2768             if (ret >= 0) {
2769                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2770                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2771                                                          inactivate);
2772             }
2773             if (inactivate && ret >= 0) {
2774                 s->block_inactive = true;
2775             }
2776         }
2777         qemu_mutex_unlock_iothread();
2778 
2779         if (ret < 0) {
2780             goto fail;
2781         }
2782     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2783         trace_migration_completion_postcopy_end();
2784 
2785         qemu_savevm_state_complete_postcopy(s->to_dst_file);
2786         trace_migration_completion_postcopy_end_after_complete();
2787     }
2788 
2789     /*
2790      * If rp was opened we must clean up the thread before
2791      * cleaning everything else up (since if there are no failures
2792      * it will wait for the destination to send it's status in
2793      * a SHUT command).
2794      */
2795     if (s->rp_state.from_dst_file) {
2796         int rp_error;
2797         trace_migration_return_path_end_before();
2798         rp_error = await_return_path_close_on_source(s);
2799         trace_migration_return_path_end_after(rp_error);
2800         if (rp_error) {
2801             goto fail_invalidate;
2802         }
2803     }
2804 
2805     if (qemu_file_get_error(s->to_dst_file)) {
2806         trace_migration_completion_file_err();
2807         goto fail_invalidate;
2808     }
2809 
2810     if (!migrate_colo_enabled()) {
2811         migrate_set_state(&s->state, current_active_state,
2812                           MIGRATION_STATUS_COMPLETED);
2813     }
2814 
2815     return;
2816 
2817 fail_invalidate:
2818     /* If not doing postcopy, vm_start() will be called: let's regain
2819      * control on images.
2820      */
2821     if (s->state == MIGRATION_STATUS_ACTIVE ||
2822         s->state == MIGRATION_STATUS_DEVICE) {
2823         Error *local_err = NULL;
2824 
2825         qemu_mutex_lock_iothread();
2826         bdrv_invalidate_cache_all(&local_err);
2827         if (local_err) {
2828             error_report_err(local_err);
2829         } else {
2830             s->block_inactive = false;
2831         }
2832         qemu_mutex_unlock_iothread();
2833     }
2834 
2835 fail:
2836     migrate_set_state(&s->state, current_active_state,
2837                       MIGRATION_STATUS_FAILED);
2838 }
2839 
2840 bool migrate_colo_enabled(void)
2841 {
2842     MigrationState *s = migrate_get_current();
2843     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2844 }
2845 
2846 typedef enum MigThrError {
2847     /* No error detected */
2848     MIG_THR_ERR_NONE = 0,
2849     /* Detected error, but resumed successfully */
2850     MIG_THR_ERR_RECOVERED = 1,
2851     /* Detected fatal error, need to exit */
2852     MIG_THR_ERR_FATAL = 2,
2853 } MigThrError;
2854 
2855 static int postcopy_resume_handshake(MigrationState *s)
2856 {
2857     qemu_savevm_send_postcopy_resume(s->to_dst_file);
2858 
2859     while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2860         qemu_sem_wait(&s->rp_state.rp_sem);
2861     }
2862 
2863     if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2864         return 0;
2865     }
2866 
2867     return -1;
2868 }
2869 
2870 /* Return zero if success, or <0 for error */
2871 static int postcopy_do_resume(MigrationState *s)
2872 {
2873     int ret;
2874 
2875     /*
2876      * Call all the resume_prepare() hooks, so that modules can be
2877      * ready for the migration resume.
2878      */
2879     ret = qemu_savevm_state_resume_prepare(s);
2880     if (ret) {
2881         error_report("%s: resume_prepare() failure detected: %d",
2882                      __func__, ret);
2883         return ret;
2884     }
2885 
2886     /*
2887      * Last handshake with destination on the resume (destination will
2888      * switch to postcopy-active afterwards)
2889      */
2890     ret = postcopy_resume_handshake(s);
2891     if (ret) {
2892         error_report("%s: handshake failed: %d", __func__, ret);
2893         return ret;
2894     }
2895 
2896     return 0;
2897 }
2898 
2899 /*
2900  * We don't return until we are in a safe state to continue current
2901  * postcopy migration.  Returns MIG_THR_ERR_RECOVERED if recovered, or
2902  * MIG_THR_ERR_FATAL if unrecovery failure happened.
2903  */
2904 static MigThrError postcopy_pause(MigrationState *s)
2905 {
2906     assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2907 
2908     while (true) {
2909         QEMUFile *file;
2910 
2911         migrate_set_state(&s->state, s->state,
2912                           MIGRATION_STATUS_POSTCOPY_PAUSED);
2913 
2914         /* Current channel is possibly broken. Release it. */
2915         assert(s->to_dst_file);
2916         qemu_mutex_lock(&s->qemu_file_lock);
2917         file = s->to_dst_file;
2918         s->to_dst_file = NULL;
2919         qemu_mutex_unlock(&s->qemu_file_lock);
2920 
2921         qemu_file_shutdown(file);
2922         qemu_fclose(file);
2923 
2924         error_report("Detected IO failure for postcopy. "
2925                      "Migration paused.");
2926 
2927         /*
2928          * We wait until things fixed up. Then someone will setup the
2929          * status back for us.
2930          */
2931         while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2932             qemu_sem_wait(&s->postcopy_pause_sem);
2933         }
2934 
2935         if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2936             /* Woken up by a recover procedure. Give it a shot */
2937 
2938             /*
2939              * Firstly, let's wake up the return path now, with a new
2940              * return path channel.
2941              */
2942             qemu_sem_post(&s->postcopy_pause_rp_sem);
2943 
2944             /* Do the resume logic */
2945             if (postcopy_do_resume(s) == 0) {
2946                 /* Let's continue! */
2947                 trace_postcopy_pause_continued();
2948                 return MIG_THR_ERR_RECOVERED;
2949             } else {
2950                 /*
2951                  * Something wrong happened during the recovery, let's
2952                  * pause again. Pause is always better than throwing
2953                  * data away.
2954                  */
2955                 continue;
2956             }
2957         } else {
2958             /* This is not right... Time to quit. */
2959             return MIG_THR_ERR_FATAL;
2960         }
2961     }
2962 }
2963 
2964 static MigThrError migration_detect_error(MigrationState *s)
2965 {
2966     int ret;
2967     int state = s->state;
2968 
2969     if (state == MIGRATION_STATUS_CANCELLING ||
2970         state == MIGRATION_STATUS_CANCELLED) {
2971         /* End the migration, but don't set the state to failed */
2972         return MIG_THR_ERR_FATAL;
2973     }
2974 
2975     /* Try to detect any file errors */
2976     ret = qemu_file_get_error(s->to_dst_file);
2977 
2978     if (!ret) {
2979         /* Everything is fine */
2980         return MIG_THR_ERR_NONE;
2981     }
2982 
2983     if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
2984         /*
2985          * For postcopy, we allow the network to be down for a
2986          * while. After that, it can be continued by a
2987          * recovery phase.
2988          */
2989         return postcopy_pause(s);
2990     } else {
2991         /*
2992          * For precopy (or postcopy with error outside IO), we fail
2993          * with no time.
2994          */
2995         migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
2996         trace_migration_thread_file_err();
2997 
2998         /* Time to stop the migration, now. */
2999         return MIG_THR_ERR_FATAL;
3000     }
3001 }
3002 
3003 /* How many bytes have we transferred since the beggining of the migration */
3004 static uint64_t migration_total_bytes(MigrationState *s)
3005 {
3006     return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3007 }
3008 
3009 static void migration_calculate_complete(MigrationState *s)
3010 {
3011     uint64_t bytes = migration_total_bytes(s);
3012     int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3013     int64_t transfer_time;
3014 
3015     s->total_time = end_time - s->start_time;
3016     if (!s->downtime) {
3017         /*
3018          * It's still not set, so we are precopy migration.  For
3019          * postcopy, downtime is calculated during postcopy_start().
3020          */
3021         s->downtime = end_time - s->downtime_start;
3022     }
3023 
3024     transfer_time = s->total_time - s->setup_time;
3025     if (transfer_time) {
3026         s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3027     }
3028 }
3029 
3030 static void migration_update_counters(MigrationState *s,
3031                                       int64_t current_time)
3032 {
3033     uint64_t transferred, transferred_pages, time_spent;
3034     uint64_t current_bytes; /* bytes transferred since the beginning */
3035     double bandwidth;
3036 
3037     if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3038         return;
3039     }
3040 
3041     current_bytes = migration_total_bytes(s);
3042     transferred = current_bytes - s->iteration_initial_bytes;
3043     time_spent = current_time - s->iteration_start_time;
3044     bandwidth = (double)transferred / time_spent;
3045     s->threshold_size = bandwidth * s->parameters.downtime_limit;
3046 
3047     s->mbps = (((double) transferred * 8.0) /
3048                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3049 
3050     transferred_pages = ram_get_total_transferred_pages() -
3051                             s->iteration_initial_pages;
3052     s->pages_per_second = (double) transferred_pages /
3053                              (((double) time_spent / 1000.0));
3054 
3055     /*
3056      * if we haven't sent anything, we don't want to
3057      * recalculate. 10000 is a small enough number for our purposes
3058      */
3059     if (ram_counters.dirty_pages_rate && transferred > 10000) {
3060         s->expected_downtime = ram_counters.remaining / bandwidth;
3061     }
3062 
3063     qemu_file_reset_rate_limit(s->to_dst_file);
3064 
3065     s->iteration_start_time = current_time;
3066     s->iteration_initial_bytes = current_bytes;
3067     s->iteration_initial_pages = ram_get_total_transferred_pages();
3068 
3069     trace_migrate_transferred(transferred, time_spent,
3070                               bandwidth, s->threshold_size);
3071 }
3072 
3073 /* Migration thread iteration status */
3074 typedef enum {
3075     MIG_ITERATE_RESUME,         /* Resume current iteration */
3076     MIG_ITERATE_SKIP,           /* Skip current iteration */
3077     MIG_ITERATE_BREAK,          /* Break the loop */
3078 } MigIterateState;
3079 
3080 /*
3081  * Return true if continue to the next iteration directly, false
3082  * otherwise.
3083  */
3084 static MigIterateState migration_iteration_run(MigrationState *s)
3085 {
3086     uint64_t pending_size, pend_pre, pend_compat, pend_post;
3087     bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3088 
3089     qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3090                               &pend_compat, &pend_post);
3091     pending_size = pend_pre + pend_compat + pend_post;
3092 
3093     trace_migrate_pending(pending_size, s->threshold_size,
3094                           pend_pre, pend_compat, pend_post);
3095 
3096     if (pending_size && pending_size >= s->threshold_size) {
3097         /* Still a significant amount to transfer */
3098         if (migrate_postcopy() && !in_postcopy &&
3099             pend_pre <= s->threshold_size &&
3100             atomic_read(&s->start_postcopy)) {
3101             if (postcopy_start(s)) {
3102                 error_report("%s: postcopy failed to start", __func__);
3103             }
3104             return MIG_ITERATE_SKIP;
3105         }
3106         /* Just another iteration step */
3107         qemu_savevm_state_iterate(s->to_dst_file,
3108             s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3109     } else {
3110         trace_migration_thread_low_pending(pending_size);
3111         migration_completion(s);
3112         return MIG_ITERATE_BREAK;
3113     }
3114 
3115     return MIG_ITERATE_RESUME;
3116 }
3117 
3118 static void migration_iteration_finish(MigrationState *s)
3119 {
3120     /* If we enabled cpu throttling for auto-converge, turn it off. */
3121     cpu_throttle_stop();
3122 
3123     qemu_mutex_lock_iothread();
3124     switch (s->state) {
3125     case MIGRATION_STATUS_COMPLETED:
3126         migration_calculate_complete(s);
3127         runstate_set(RUN_STATE_POSTMIGRATE);
3128         break;
3129 
3130     case MIGRATION_STATUS_ACTIVE:
3131         /*
3132          * We should really assert here, but since it's during
3133          * migration, let's try to reduce the usage of assertions.
3134          */
3135         if (!migrate_colo_enabled()) {
3136             error_report("%s: critical error: calling COLO code without "
3137                          "COLO enabled", __func__);
3138         }
3139         migrate_start_colo_process(s);
3140         /*
3141          * Fixme: we will run VM in COLO no matter its old running state.
3142          * After exited COLO, we will keep running.
3143          */
3144         s->vm_was_running = true;
3145         /* Fallthrough */
3146     case MIGRATION_STATUS_FAILED:
3147     case MIGRATION_STATUS_CANCELLED:
3148     case MIGRATION_STATUS_CANCELLING:
3149         if (s->vm_was_running) {
3150             vm_start();
3151         } else {
3152             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3153                 runstate_set(RUN_STATE_POSTMIGRATE);
3154             }
3155         }
3156         break;
3157 
3158     default:
3159         /* Should not reach here, but if so, forgive the VM. */
3160         error_report("%s: Unknown ending state %d", __func__, s->state);
3161         break;
3162     }
3163     migrate_fd_cleanup_schedule(s);
3164     qemu_mutex_unlock_iothread();
3165 }
3166 
3167 void migration_make_urgent_request(void)
3168 {
3169     qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3170 }
3171 
3172 void migration_consume_urgent_request(void)
3173 {
3174     qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3175 }
3176 
3177 /*
3178  * Master migration thread on the source VM.
3179  * It drives the migration and pumps the data down the outgoing channel.
3180  */
3181 static void *migration_thread(void *opaque)
3182 {
3183     MigrationState *s = opaque;
3184     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3185     MigThrError thr_error;
3186     bool urgent = false;
3187 
3188     rcu_register_thread();
3189 
3190     object_ref(OBJECT(s));
3191     s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3192 
3193     qemu_savevm_state_header(s->to_dst_file);
3194 
3195     /*
3196      * If we opened the return path, we need to make sure dst has it
3197      * opened as well.
3198      */
3199     if (s->rp_state.from_dst_file) {
3200         /* Now tell the dest that it should open its end so it can reply */
3201         qemu_savevm_send_open_return_path(s->to_dst_file);
3202 
3203         /* And do a ping that will make stuff easier to debug */
3204         qemu_savevm_send_ping(s->to_dst_file, 1);
3205     }
3206 
3207     if (migrate_postcopy()) {
3208         /*
3209          * Tell the destination that we *might* want to do postcopy later;
3210          * if the other end can't do postcopy it should fail now, nice and
3211          * early.
3212          */
3213         qemu_savevm_send_postcopy_advise(s->to_dst_file);
3214     }
3215 
3216     if (migrate_colo_enabled()) {
3217         /* Notify migration destination that we enable COLO */
3218         qemu_savevm_send_colo_enable(s->to_dst_file);
3219     }
3220 
3221     qemu_savevm_state_setup(s->to_dst_file);
3222 
3223     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3224     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3225                       MIGRATION_STATUS_ACTIVE);
3226 
3227     trace_migration_thread_setup_complete();
3228 
3229     while (s->state == MIGRATION_STATUS_ACTIVE ||
3230            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3231         int64_t current_time;
3232 
3233         if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3234             MigIterateState iter_state = migration_iteration_run(s);
3235             if (iter_state == MIG_ITERATE_SKIP) {
3236                 continue;
3237             } else if (iter_state == MIG_ITERATE_BREAK) {
3238                 break;
3239             }
3240         }
3241 
3242         /*
3243          * Try to detect any kind of failures, and see whether we
3244          * should stop the migration now.
3245          */
3246         thr_error = migration_detect_error(s);
3247         if (thr_error == MIG_THR_ERR_FATAL) {
3248             /* Stop migration */
3249             break;
3250         } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3251             /*
3252              * Just recovered from a e.g. network failure, reset all
3253              * the local variables. This is important to avoid
3254              * breaking transferred_bytes and bandwidth calculation
3255              */
3256             s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3257             s->iteration_initial_bytes = 0;
3258         }
3259 
3260         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3261 
3262         migration_update_counters(s, current_time);
3263 
3264         urgent = false;
3265         if (qemu_file_rate_limit(s->to_dst_file)) {
3266             /* Wait for a delay to do rate limiting OR
3267              * something urgent to post the semaphore.
3268              */
3269             int ms = s->iteration_start_time + BUFFER_DELAY - current_time;
3270             trace_migration_thread_ratelimit_pre(ms);
3271             if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3272                 /* We were worken by one or more urgent things but
3273                  * the timedwait will have consumed one of them.
3274                  * The service routine for the urgent wake will dec
3275                  * the semaphore itself for each item it consumes,
3276                  * so add this one we just eat back.
3277                  */
3278                 qemu_sem_post(&s->rate_limit_sem);
3279                 urgent = true;
3280             }
3281             trace_migration_thread_ratelimit_post(urgent);
3282         }
3283     }
3284 
3285     trace_migration_thread_after_loop();
3286     migration_iteration_finish(s);
3287     object_unref(OBJECT(s));
3288     rcu_unregister_thread();
3289     return NULL;
3290 }
3291 
3292 void migrate_fd_connect(MigrationState *s, Error *error_in)
3293 {
3294     int64_t rate_limit;
3295     bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3296 
3297     s->expected_downtime = s->parameters.downtime_limit;
3298     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3299     if (error_in) {
3300         migrate_fd_error(s, error_in);
3301         migrate_fd_cleanup(s);
3302         return;
3303     }
3304 
3305     if (resume) {
3306         /* This is a resumed migration */
3307         rate_limit = INT64_MAX;
3308     } else {
3309         /* This is a fresh new migration */
3310         rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3311 
3312         /* Notify before starting migration thread */
3313         notifier_list_notify(&migration_state_notifiers, s);
3314     }
3315 
3316     qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3317     qemu_file_set_blocking(s->to_dst_file, true);
3318 
3319     /*
3320      * Open the return path. For postcopy, it is used exclusively. For
3321      * precopy, only if user specified "return-path" capability would
3322      * QEMU uses the return path.
3323      */
3324     if (migrate_postcopy_ram() || migrate_use_return_path()) {
3325         if (open_return_path_on_source(s, !resume)) {
3326             error_report("Unable to open return-path for postcopy");
3327             migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3328             migrate_fd_cleanup(s);
3329             return;
3330         }
3331     }
3332 
3333     if (resume) {
3334         /* Wakeup the main migration thread to do the recovery */
3335         migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3336                           MIGRATION_STATUS_POSTCOPY_RECOVER);
3337         qemu_sem_post(&s->postcopy_pause_sem);
3338         return;
3339     }
3340 
3341     if (multifd_save_setup() != 0) {
3342         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3343                           MIGRATION_STATUS_FAILED);
3344         migrate_fd_cleanup(s);
3345         return;
3346     }
3347     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3348                        QEMU_THREAD_JOINABLE);
3349     s->migration_thread_running = true;
3350 }
3351 
3352 void migration_global_dump(Monitor *mon)
3353 {
3354     MigrationState *ms = migrate_get_current();
3355 
3356     monitor_printf(mon, "globals:\n");
3357     monitor_printf(mon, "store-global-state: %s\n",
3358                    ms->store_global_state ? "on" : "off");
3359     monitor_printf(mon, "only-migratable: %s\n",
3360                    only_migratable ? "on" : "off");
3361     monitor_printf(mon, "send-configuration: %s\n",
3362                    ms->send_configuration ? "on" : "off");
3363     monitor_printf(mon, "send-section-footer: %s\n",
3364                    ms->send_section_footer ? "on" : "off");
3365     monitor_printf(mon, "decompress-error-check: %s\n",
3366                    ms->decompress_error_check ? "on" : "off");
3367     monitor_printf(mon, "clear-bitmap-shift: %u\n",
3368                    ms->clear_bitmap_shift);
3369 }
3370 
3371 #define DEFINE_PROP_MIG_CAP(name, x)             \
3372     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3373 
3374 static Property migration_properties[] = {
3375     DEFINE_PROP_BOOL("store-global-state", MigrationState,
3376                      store_global_state, true),
3377     DEFINE_PROP_BOOL("send-configuration", MigrationState,
3378                      send_configuration, true),
3379     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3380                      send_section_footer, true),
3381     DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3382                       decompress_error_check, true),
3383     DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
3384                       clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
3385 
3386     /* Migration parameters */
3387     DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3388                       parameters.compress_level,
3389                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
3390     DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3391                       parameters.compress_threads,
3392                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3393     DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3394                       parameters.compress_wait_thread, true),
3395     DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3396                       parameters.decompress_threads,
3397                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3398     DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3399                       parameters.cpu_throttle_initial,
3400                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3401     DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3402                       parameters.cpu_throttle_increment,
3403                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3404     DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3405                       parameters.max_bandwidth, MAX_THROTTLE),
3406     DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3407                       parameters.downtime_limit,
3408                       DEFAULT_MIGRATE_SET_DOWNTIME),
3409     DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3410                       parameters.x_checkpoint_delay,
3411                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3412     DEFINE_PROP_UINT8("multifd-channels", MigrationState,
3413                       parameters.multifd_channels,
3414                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3415     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3416                       parameters.xbzrle_cache_size,
3417                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3418     DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3419                       parameters.max_postcopy_bandwidth,
3420                       DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3421     DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3422                       parameters.max_cpu_throttle,
3423                       DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3424     DEFINE_PROP_SIZE("announce-initial", MigrationState,
3425                       parameters.announce_initial,
3426                       DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
3427     DEFINE_PROP_SIZE("announce-max", MigrationState,
3428                       parameters.announce_max,
3429                       DEFAULT_MIGRATE_ANNOUNCE_MAX),
3430     DEFINE_PROP_SIZE("announce-rounds", MigrationState,
3431                       parameters.announce_rounds,
3432                       DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
3433     DEFINE_PROP_SIZE("announce-step", MigrationState,
3434                       parameters.announce_step,
3435                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
3436 
3437     /* Migration capabilities */
3438     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3439     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3440     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3441     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3442     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3443     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3444     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3445     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3446     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3447     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3448     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3449     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD),
3450 
3451     DEFINE_PROP_END_OF_LIST(),
3452 };
3453 
3454 static void migration_class_init(ObjectClass *klass, void *data)
3455 {
3456     DeviceClass *dc = DEVICE_CLASS(klass);
3457 
3458     dc->user_creatable = false;
3459     dc->props = migration_properties;
3460 }
3461 
3462 static void migration_instance_finalize(Object *obj)
3463 {
3464     MigrationState *ms = MIGRATION_OBJ(obj);
3465     MigrationParameters *params = &ms->parameters;
3466 
3467     qemu_mutex_destroy(&ms->error_mutex);
3468     qemu_mutex_destroy(&ms->qemu_file_lock);
3469     g_free(params->tls_hostname);
3470     g_free(params->tls_creds);
3471     qemu_sem_destroy(&ms->rate_limit_sem);
3472     qemu_sem_destroy(&ms->pause_sem);
3473     qemu_sem_destroy(&ms->postcopy_pause_sem);
3474     qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3475     qemu_sem_destroy(&ms->rp_state.rp_sem);
3476     error_free(ms->error);
3477 }
3478 
3479 static void migration_instance_init(Object *obj)
3480 {
3481     MigrationState *ms = MIGRATION_OBJ(obj);
3482     MigrationParameters *params = &ms->parameters;
3483 
3484     ms->state = MIGRATION_STATUS_NONE;
3485     ms->mbps = -1;
3486     ms->pages_per_second = -1;
3487     qemu_sem_init(&ms->pause_sem, 0);
3488     qemu_mutex_init(&ms->error_mutex);
3489 
3490     params->tls_hostname = g_strdup("");
3491     params->tls_creds = g_strdup("");
3492 
3493     /* Set has_* up only for parameter checks */
3494     params->has_compress_level = true;
3495     params->has_compress_threads = true;
3496     params->has_decompress_threads = true;
3497     params->has_cpu_throttle_initial = true;
3498     params->has_cpu_throttle_increment = true;
3499     params->has_max_bandwidth = true;
3500     params->has_downtime_limit = true;
3501     params->has_x_checkpoint_delay = true;
3502     params->has_block_incremental = true;
3503     params->has_multifd_channels = true;
3504     params->has_xbzrle_cache_size = true;
3505     params->has_max_postcopy_bandwidth = true;
3506     params->has_max_cpu_throttle = true;
3507     params->has_announce_initial = true;
3508     params->has_announce_max = true;
3509     params->has_announce_rounds = true;
3510     params->has_announce_step = true;
3511 
3512     qemu_sem_init(&ms->postcopy_pause_sem, 0);
3513     qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3514     qemu_sem_init(&ms->rp_state.rp_sem, 0);
3515     qemu_sem_init(&ms->rate_limit_sem, 0);
3516     qemu_mutex_init(&ms->qemu_file_lock);
3517 }
3518 
3519 /*
3520  * Return true if check pass, false otherwise. Error will be put
3521  * inside errp if provided.
3522  */
3523 static bool migration_object_check(MigrationState *ms, Error **errp)
3524 {
3525     MigrationCapabilityStatusList *head = NULL;
3526     /* Assuming all off */
3527     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3528     int i;
3529 
3530     if (!migrate_params_check(&ms->parameters, errp)) {
3531         return false;
3532     }
3533 
3534     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3535         if (ms->enabled_capabilities[i]) {
3536             head = migrate_cap_add(head, i, true);
3537         }
3538     }
3539 
3540     ret = migrate_caps_check(cap_list, head, errp);
3541 
3542     /* It works with head == NULL */
3543     qapi_free_MigrationCapabilityStatusList(head);
3544 
3545     return ret;
3546 }
3547 
3548 static const TypeInfo migration_type = {
3549     .name = TYPE_MIGRATION,
3550     /*
3551      * NOTE: TYPE_MIGRATION is not really a device, as the object is
3552      * not created using qdev_create(), it is not attached to the qdev
3553      * device tree, and it is never realized.
3554      *
3555      * TODO: Make this TYPE_OBJECT once QOM provides something like
3556      * TYPE_DEVICE's "-global" properties.
3557      */
3558     .parent = TYPE_DEVICE,
3559     .class_init = migration_class_init,
3560     .class_size = sizeof(MigrationClass),
3561     .instance_size = sizeof(MigrationState),
3562     .instance_init = migration_instance_init,
3563     .instance_finalize = migration_instance_finalize,
3564 };
3565 
3566 static void register_migration_types(void)
3567 {
3568     type_register_static(&migration_type);
3569 }
3570 
3571 type_init(register_migration_types);
3572