Merge remote-tracking branch 'remotes/alistair23/tags/pull-riscv-to-apply-20211022...
[qemu.git] / migration / migration.c
1 /*
2 * QEMU live migration
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
14 */
15
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
21 #include "exec.h"
22 #include "fd.h"
23 #include "socket.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
27 #include "rdma.h"
28 #include "ram.h"
29 #include "migration/global_state.h"
30 #include "migration/misc.h"
31 #include "migration.h"
32 #include "savevm.h"
33 #include "qemu-file-channel.h"
34 #include "qemu-file.h"
35 #include "migration/vmstate.h"
36 #include "block/block.h"
37 #include "qapi/error.h"
38 #include "qapi/clone-visitor.h"
39 #include "qapi/qapi-visit-migration.h"
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qapi-commands-migration.h"
42 #include "qapi/qapi-events-migration.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qapi/qmp/qnull.h"
45 #include "qemu/rcu.h"
46 #include "block.h"
47 #include "postcopy-ram.h"
48 #include "qemu/thread.h"
49 #include "trace.h"
50 #include "exec/target_page.h"
51 #include "io/channel-buffer.h"
52 #include "migration/colo.h"
53 #include "hw/boards.h"
54 #include "hw/qdev-properties.h"
55 #include "hw/qdev-properties-system.h"
56 #include "monitor/monitor.h"
57 #include "net/announce.h"
58 #include "qemu/queue.h"
59 #include "multifd.h"
60 #include "qemu/yank.h"
61 #include "sysemu/cpus.h"
62
63 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
64
65 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
66 * data. */
67 #define BUFFER_DELAY 100
68 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
69
70 /* Time in milliseconds we are allowed to stop the source,
71 * for sending the last part */
72 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
73
74 /* Maximum migrate downtime set to 2000 seconds */
75 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
76 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
77
78 /* Default compression thread count */
79 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
80 /* Default decompression thread count, usually decompression is at
81 * least 4 times as fast as compression.*/
82 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
83 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
84 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
85 /* Define default autoconverge cpu throttle migration parameters */
86 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
87 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
88 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
89 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
90
91 /* Migration XBZRLE default cache size */
92 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
93
94 /* The delay time (in ms) between two COLO checkpoints */
95 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
96 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
97 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
98 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
99 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
100 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
101 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
102
103 /* Background transfer rate for postcopy, 0 means unlimited, note
104 * that page requests can still exceed this limit.
105 */
106 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
107
108 /*
109 * Parameters for self_announce_delay giving a stream of RARP/ARP
110 * packets after migration.
111 */
112 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
113 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
114 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
115 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
116
117 static NotifierList migration_state_notifiers =
118 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
119
120 /* Messages sent on the return path from destination to source */
121 enum mig_rp_message_type {
122 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
123 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
124 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
125
126 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
127 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
128 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
129 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
130
131 MIG_RP_MSG_MAX
132 };
133
134 /* Migration capabilities set */
135 struct MigrateCapsSet {
136 int size; /* Capability set size */
137 MigrationCapability caps[]; /* Variadic array of capabilities */
138 };
139 typedef struct MigrateCapsSet MigrateCapsSet;
140
141 /* Define and initialize MigrateCapsSet */
142 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \
143 MigrateCapsSet _name = { \
144 .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
145 .caps = { __VA_ARGS__ } \
146 }
147
148 /* Background-snapshot compatibility check list */
149 static const
150 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot,
151 MIGRATION_CAPABILITY_POSTCOPY_RAM,
152 MIGRATION_CAPABILITY_DIRTY_BITMAPS,
153 MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME,
154 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE,
155 MIGRATION_CAPABILITY_RETURN_PATH,
156 MIGRATION_CAPABILITY_MULTIFD,
157 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER,
158 MIGRATION_CAPABILITY_AUTO_CONVERGE,
159 MIGRATION_CAPABILITY_RELEASE_RAM,
160 MIGRATION_CAPABILITY_RDMA_PIN_ALL,
161 MIGRATION_CAPABILITY_COMPRESS,
162 MIGRATION_CAPABILITY_XBZRLE,
163 MIGRATION_CAPABILITY_X_COLO,
164 MIGRATION_CAPABILITY_VALIDATE_UUID);
165
166 /* When we add fault tolerance, we could have several
167 migrations at once. For now we don't need to add
168 dynamic creation of migration */
169
170 static MigrationState *current_migration;
171 static MigrationIncomingState *current_incoming;
172
173 static GSList *migration_blockers;
174
175 static bool migration_object_check(MigrationState *ms, Error **errp);
176 static int migration_maybe_pause(MigrationState *s,
177 int *current_active_state,
178 int new_state);
179 static void migrate_fd_cancel(MigrationState *s);
180
181 static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp)
182 {
183 uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp;
184
185 return (a > b) - (a < b);
186 }
187
188 void migration_object_init(void)
189 {
190 Error *err = NULL;
191
192 /* This can only be called once. */
193 assert(!current_migration);
194 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
195
196 /*
197 * Init the migrate incoming object as well no matter whether
198 * we'll use it or not.
199 */
200 assert(!current_incoming);
201 current_incoming = g_new0(MigrationIncomingState, 1);
202 current_incoming->state = MIGRATION_STATUS_NONE;
203 current_incoming->postcopy_remote_fds =
204 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
205 qemu_mutex_init(&current_incoming->rp_mutex);
206 qemu_event_init(&current_incoming->main_thread_load_event, false);
207 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
208 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
209 qemu_mutex_init(&current_incoming->page_request_mutex);
210 current_incoming->page_requested = g_tree_new(page_request_addr_cmp);
211
212 if (!migration_object_check(current_migration, &err)) {
213 error_report_err(err);
214 exit(1);
215 }
216
217 blk_mig_init();
218 ram_mig_init();
219 dirty_bitmap_mig_init();
220 }
221
222 void migration_cancel(void)
223 {
224 migrate_fd_cancel(current_migration);
225 }
226
227 void migration_shutdown(void)
228 {
229 /*
230 * Cancel the current migration - that will (eventually)
231 * stop the migration using this structure
232 */
233 migration_cancel();
234 object_unref(OBJECT(current_migration));
235
236 /*
237 * Cancel outgoing migration of dirty bitmaps. It should
238 * at least unref used block nodes.
239 */
240 dirty_bitmap_mig_cancel_outgoing();
241
242 /*
243 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
244 * are non-critical data, and their loss never considered as
245 * something serious.
246 */
247 dirty_bitmap_mig_cancel_incoming();
248 }
249
250 /* For outgoing */
251 MigrationState *migrate_get_current(void)
252 {
253 /* This can only be called after the object created. */
254 assert(current_migration);
255 return current_migration;
256 }
257
258 MigrationIncomingState *migration_incoming_get_current(void)
259 {
260 assert(current_incoming);
261 return current_incoming;
262 }
263
264 void migration_incoming_state_destroy(void)
265 {
266 struct MigrationIncomingState *mis = migration_incoming_get_current();
267
268 if (mis->to_src_file) {
269 /* Tell source that we are done */
270 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
271 qemu_fclose(mis->to_src_file);
272 mis->to_src_file = NULL;
273 }
274
275 if (mis->from_src_file) {
276 qemu_fclose(mis->from_src_file);
277 mis->from_src_file = NULL;
278 }
279 if (mis->postcopy_remote_fds) {
280 g_array_free(mis->postcopy_remote_fds, TRUE);
281 mis->postcopy_remote_fds = NULL;
282 }
283 if (mis->transport_cleanup) {
284 mis->transport_cleanup(mis->transport_data);
285 }
286
287 qemu_event_reset(&mis->main_thread_load_event);
288
289 if (mis->page_requested) {
290 g_tree_destroy(mis->page_requested);
291 mis->page_requested = NULL;
292 }
293
294 if (mis->socket_address_list) {
295 qapi_free_SocketAddressList(mis->socket_address_list);
296 mis->socket_address_list = NULL;
297 }
298
299 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
300 }
301
302 static void migrate_generate_event(int new_state)
303 {
304 if (migrate_use_events()) {
305 qapi_event_send_migration(new_state);
306 }
307 }
308
309 static bool migrate_late_block_activate(void)
310 {
311 MigrationState *s;
312
313 s = migrate_get_current();
314
315 return s->enabled_capabilities[
316 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
317 }
318
319 /*
320 * Send a message on the return channel back to the source
321 * of the migration.
322 */
323 static int migrate_send_rp_message(MigrationIncomingState *mis,
324 enum mig_rp_message_type message_type,
325 uint16_t len, void *data)
326 {
327 int ret = 0;
328
329 trace_migrate_send_rp_message((int)message_type, len);
330 QEMU_LOCK_GUARD(&mis->rp_mutex);
331
332 /*
333 * It's possible that the file handle got lost due to network
334 * failures.
335 */
336 if (!mis->to_src_file) {
337 ret = -EIO;
338 return ret;
339 }
340
341 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
342 qemu_put_be16(mis->to_src_file, len);
343 qemu_put_buffer(mis->to_src_file, data, len);
344 qemu_fflush(mis->to_src_file);
345
346 /* It's possible that qemu file got error during sending */
347 ret = qemu_file_get_error(mis->to_src_file);
348
349 return ret;
350 }
351
352 /* Request one page from the source VM at the given start address.
353 * rb: the RAMBlock to request the page in
354 * Start: Address offset within the RB
355 * Len: Length in bytes required - must be a multiple of pagesize
356 */
357 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
358 RAMBlock *rb, ram_addr_t start)
359 {
360 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
361 size_t msglen = 12; /* start + len */
362 size_t len = qemu_ram_pagesize(rb);
363 enum mig_rp_message_type msg_type;
364 const char *rbname;
365 int rbname_len;
366
367 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
368 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
369
370 /*
371 * We maintain the last ramblock that we requested for page. Note that we
372 * don't need locking because this function will only be called within the
373 * postcopy ram fault thread.
374 */
375 if (rb != mis->last_rb) {
376 mis->last_rb = rb;
377
378 rbname = qemu_ram_get_idstr(rb);
379 rbname_len = strlen(rbname);
380
381 assert(rbname_len < 256);
382
383 bufc[msglen++] = rbname_len;
384 memcpy(bufc + msglen, rbname, rbname_len);
385 msglen += rbname_len;
386 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
387 } else {
388 msg_type = MIG_RP_MSG_REQ_PAGES;
389 }
390
391 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
392 }
393
394 int migrate_send_rp_req_pages(MigrationIncomingState *mis,
395 RAMBlock *rb, ram_addr_t start, uint64_t haddr)
396 {
397 void *aligned = (void *)(uintptr_t)(haddr & (-qemu_ram_pagesize(rb)));
398 bool received = false;
399
400 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
401 received = ramblock_recv_bitmap_test_byte_offset(rb, start);
402 if (!received && !g_tree_lookup(mis->page_requested, aligned)) {
403 /*
404 * The page has not been received, and it's not yet in the page
405 * request list. Queue it. Set the value of element to 1, so that
406 * things like g_tree_lookup() will return TRUE (1) when found.
407 */
408 g_tree_insert(mis->page_requested, aligned, (gpointer)1);
409 mis->page_requested_count++;
410 trace_postcopy_page_req_add(aligned, mis->page_requested_count);
411 }
412 }
413
414 /*
415 * If the page is there, skip sending the message. We don't even need the
416 * lock because as long as the page arrived, it'll be there forever.
417 */
418 if (received) {
419 return 0;
420 }
421
422 return migrate_send_rp_message_req_pages(mis, rb, start);
423 }
424
425 static bool migration_colo_enabled;
426 bool migration_incoming_colo_enabled(void)
427 {
428 return migration_colo_enabled;
429 }
430
431 void migration_incoming_disable_colo(void)
432 {
433 ram_block_discard_disable(false);
434 migration_colo_enabled = false;
435 }
436
437 int migration_incoming_enable_colo(void)
438 {
439 if (ram_block_discard_disable(true)) {
440 error_report("COLO: cannot disable RAM discard");
441 return -EBUSY;
442 }
443 migration_colo_enabled = true;
444 return 0;
445 }
446
447 void migrate_add_address(SocketAddress *address)
448 {
449 MigrationIncomingState *mis = migration_incoming_get_current();
450
451 QAPI_LIST_PREPEND(mis->socket_address_list,
452 QAPI_CLONE(SocketAddress, address));
453 }
454
455 static void qemu_start_incoming_migration(const char *uri, Error **errp)
456 {
457 const char *p = NULL;
458
459 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
460 return;
461 }
462
463 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
464 if (strstart(uri, "tcp:", &p) ||
465 strstart(uri, "unix:", NULL) ||
466 strstart(uri, "vsock:", NULL)) {
467 socket_start_incoming_migration(p ? p : uri, errp);
468 #ifdef CONFIG_RDMA
469 } else if (strstart(uri, "rdma:", &p)) {
470 rdma_start_incoming_migration(p, errp);
471 #endif
472 } else if (strstart(uri, "exec:", &p)) {
473 exec_start_incoming_migration(p, errp);
474 } else if (strstart(uri, "fd:", &p)) {
475 fd_start_incoming_migration(p, errp);
476 } else {
477 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
478 error_setg(errp, "unknown migration protocol: %s", uri);
479 }
480 }
481
482 static void process_incoming_migration_bh(void *opaque)
483 {
484 Error *local_err = NULL;
485 MigrationIncomingState *mis = opaque;
486
487 /* If capability late_block_activate is set:
488 * Only fire up the block code now if we're going to restart the
489 * VM, else 'cont' will do it.
490 * This causes file locking to happen; so we don't want it to happen
491 * unless we really are starting the VM.
492 */
493 if (!migrate_late_block_activate() ||
494 (autostart && (!global_state_received() ||
495 global_state_get_runstate() == RUN_STATE_RUNNING))) {
496 /* Make sure all file formats flush their mutable metadata.
497 * If we get an error here, just don't restart the VM yet. */
498 bdrv_invalidate_cache_all(&local_err);
499 if (local_err) {
500 error_report_err(local_err);
501 local_err = NULL;
502 autostart = false;
503 }
504 }
505
506 /*
507 * This must happen after all error conditions are dealt with and
508 * we're sure the VM is going to be running on this host.
509 */
510 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
511
512 if (multifd_load_cleanup(&local_err) != 0) {
513 error_report_err(local_err);
514 autostart = false;
515 }
516 /* If global state section was not received or we are in running
517 state, we need to obey autostart. Any other state is set with
518 runstate_set. */
519
520 dirty_bitmap_mig_before_vm_start();
521
522 if (!global_state_received() ||
523 global_state_get_runstate() == RUN_STATE_RUNNING) {
524 if (autostart) {
525 vm_start();
526 } else {
527 runstate_set(RUN_STATE_PAUSED);
528 }
529 } else if (migration_incoming_colo_enabled()) {
530 migration_incoming_disable_colo();
531 vm_start();
532 } else {
533 runstate_set(global_state_get_runstate());
534 }
535 /*
536 * This must happen after any state changes since as soon as an external
537 * observer sees this event they might start to prod at the VM assuming
538 * it's ready to use.
539 */
540 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
541 MIGRATION_STATUS_COMPLETED);
542 qemu_bh_delete(mis->bh);
543 migration_incoming_state_destroy();
544 }
545
546 static void process_incoming_migration_co(void *opaque)
547 {
548 MigrationIncomingState *mis = migration_incoming_get_current();
549 PostcopyState ps;
550 int ret;
551 Error *local_err = NULL;
552
553 assert(mis->from_src_file);
554 mis->migration_incoming_co = qemu_coroutine_self();
555 mis->largest_page_size = qemu_ram_pagesize_largest();
556 postcopy_state_set(POSTCOPY_INCOMING_NONE);
557 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
558 MIGRATION_STATUS_ACTIVE);
559 ret = qemu_loadvm_state(mis->from_src_file);
560
561 ps = postcopy_state_get();
562 trace_process_incoming_migration_co_end(ret, ps);
563 if (ps != POSTCOPY_INCOMING_NONE) {
564 if (ps == POSTCOPY_INCOMING_ADVISE) {
565 /*
566 * Where a migration had postcopy enabled (and thus went to advise)
567 * but managed to complete within the precopy period, we can use
568 * the normal exit.
569 */
570 postcopy_ram_incoming_cleanup(mis);
571 } else if (ret >= 0) {
572 /*
573 * Postcopy was started, cleanup should happen at the end of the
574 * postcopy thread.
575 */
576 trace_process_incoming_migration_co_postcopy_end_main();
577 return;
578 }
579 /* Else if something went wrong then just fall out of the normal exit */
580 }
581
582 /* we get COLO info, and know if we are in COLO mode */
583 if (!ret && migration_incoming_colo_enabled()) {
584 /* Make sure all file formats flush their mutable metadata */
585 bdrv_invalidate_cache_all(&local_err);
586 if (local_err) {
587 error_report_err(local_err);
588 goto fail;
589 }
590
591 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
592 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
593 mis->have_colo_incoming_thread = true;
594 qemu_coroutine_yield();
595
596 /* Wait checkpoint incoming thread exit before free resource */
597 qemu_thread_join(&mis->colo_incoming_thread);
598 /* We hold the global iothread lock, so it is safe here */
599 colo_release_ram_cache();
600 }
601
602 if (ret < 0) {
603 error_report("load of migration failed: %s", strerror(-ret));
604 goto fail;
605 }
606 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
607 qemu_bh_schedule(mis->bh);
608 mis->migration_incoming_co = NULL;
609 return;
610 fail:
611 local_err = NULL;
612 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
613 MIGRATION_STATUS_FAILED);
614 qemu_fclose(mis->from_src_file);
615 if (multifd_load_cleanup(&local_err) != 0) {
616 error_report_err(local_err);
617 }
618 exit(EXIT_FAILURE);
619 }
620
621 /**
622 * @migration_incoming_setup: Setup incoming migration
623 *
624 * Returns 0 for no error or 1 for error
625 *
626 * @f: file for main migration channel
627 * @errp: where to put errors
628 */
629 static int migration_incoming_setup(QEMUFile *f, Error **errp)
630 {
631 MigrationIncomingState *mis = migration_incoming_get_current();
632 Error *local_err = NULL;
633
634 if (multifd_load_setup(&local_err) != 0) {
635 /* We haven't been able to create multifd threads
636 nothing better to do */
637 error_report_err(local_err);
638 exit(EXIT_FAILURE);
639 }
640
641 if (!mis->from_src_file) {
642 mis->from_src_file = f;
643 }
644 qemu_file_set_blocking(f, false);
645 return 0;
646 }
647
648 void migration_incoming_process(void)
649 {
650 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
651 qemu_coroutine_enter(co);
652 }
653
654 /* Returns true if recovered from a paused migration, otherwise false */
655 static bool postcopy_try_recover(QEMUFile *f)
656 {
657 MigrationIncomingState *mis = migration_incoming_get_current();
658
659 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
660 /* Resumed from a paused postcopy migration */
661
662 mis->from_src_file = f;
663 /* Postcopy has standalone thread to do vm load */
664 qemu_file_set_blocking(f, true);
665
666 /* Re-configure the return path */
667 mis->to_src_file = qemu_file_get_return_path(f);
668
669 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
670 MIGRATION_STATUS_POSTCOPY_RECOVER);
671
672 /*
673 * Here, we only wake up the main loading thread (while the
674 * fault thread will still be waiting), so that we can receive
675 * commands from source now, and answer it if needed. The
676 * fault thread will be woken up afterwards until we are sure
677 * that source is ready to reply to page requests.
678 */
679 qemu_sem_post(&mis->postcopy_pause_sem_dst);
680 return true;
681 }
682
683 return false;
684 }
685
686 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
687 {
688 Error *local_err = NULL;
689
690 if (postcopy_try_recover(f)) {
691 return;
692 }
693
694 if (migration_incoming_setup(f, &local_err)) {
695 error_propagate(errp, local_err);
696 return;
697 }
698 migration_incoming_process();
699 }
700
701 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
702 {
703 MigrationIncomingState *mis = migration_incoming_get_current();
704 Error *local_err = NULL;
705 bool start_migration;
706
707 if (!mis->from_src_file) {
708 /* The first connection (multifd may have multiple) */
709 QEMUFile *f = qemu_fopen_channel_input(ioc);
710
711 /* If it's a recovery, we're done */
712 if (postcopy_try_recover(f)) {
713 return;
714 }
715
716 if (migration_incoming_setup(f, &local_err)) {
717 error_propagate(errp, local_err);
718 return;
719 }
720
721 /*
722 * Common migration only needs one channel, so we can start
723 * right now. Multifd needs more than one channel, we wait.
724 */
725 start_migration = !migrate_use_multifd();
726 } else {
727 /* Multiple connections */
728 assert(migrate_use_multifd());
729 start_migration = multifd_recv_new_channel(ioc, &local_err);
730 if (local_err) {
731 error_propagate(errp, local_err);
732 return;
733 }
734 }
735
736 if (start_migration) {
737 migration_incoming_process();
738 }
739 }
740
741 /**
742 * @migration_has_all_channels: We have received all channels that we need
743 *
744 * Returns true when we have got connections to all the channels that
745 * we need for migration.
746 */
747 bool migration_has_all_channels(void)
748 {
749 MigrationIncomingState *mis = migration_incoming_get_current();
750 bool all_channels;
751
752 all_channels = multifd_recv_all_channels_created();
753
754 return all_channels && mis->from_src_file != NULL;
755 }
756
757 /*
758 * Send a 'SHUT' message on the return channel with the given value
759 * to indicate that we've finished with the RP. Non-0 value indicates
760 * error.
761 */
762 void migrate_send_rp_shut(MigrationIncomingState *mis,
763 uint32_t value)
764 {
765 uint32_t buf;
766
767 buf = cpu_to_be32(value);
768 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
769 }
770
771 /*
772 * Send a 'PONG' message on the return channel with the given value
773 * (normally in response to a 'PING')
774 */
775 void migrate_send_rp_pong(MigrationIncomingState *mis,
776 uint32_t value)
777 {
778 uint32_t buf;
779
780 buf = cpu_to_be32(value);
781 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
782 }
783
784 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
785 char *block_name)
786 {
787 char buf[512];
788 int len;
789 int64_t res;
790
791 /*
792 * First, we send the header part. It contains only the len of
793 * idstr, and the idstr itself.
794 */
795 len = strlen(block_name);
796 buf[0] = len;
797 memcpy(buf + 1, block_name, len);
798
799 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
800 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
801 __func__);
802 return;
803 }
804
805 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
806
807 /*
808 * Next, we dump the received bitmap to the stream.
809 *
810 * TODO: currently we are safe since we are the only one that is
811 * using the to_src_file handle (fault thread is still paused),
812 * and it's ok even not taking the mutex. However the best way is
813 * to take the lock before sending the message header, and release
814 * the lock after sending the bitmap.
815 */
816 qemu_mutex_lock(&mis->rp_mutex);
817 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
818 qemu_mutex_unlock(&mis->rp_mutex);
819
820 trace_migrate_send_rp_recv_bitmap(block_name, res);
821 }
822
823 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
824 {
825 uint32_t buf;
826
827 buf = cpu_to_be32(value);
828 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
829 }
830
831 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
832 {
833 MigrationCapabilityStatusList *head = NULL, **tail = &head;
834 MigrationCapabilityStatus *caps;
835 MigrationState *s = migrate_get_current();
836 int i;
837
838 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
839 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
840 if (i == MIGRATION_CAPABILITY_BLOCK) {
841 continue;
842 }
843 #endif
844 caps = g_malloc0(sizeof(*caps));
845 caps->capability = i;
846 caps->state = s->enabled_capabilities[i];
847 QAPI_LIST_APPEND(tail, caps);
848 }
849
850 return head;
851 }
852
853 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
854 {
855 MigrationParameters *params;
856 MigrationState *s = migrate_get_current();
857
858 /* TODO use QAPI_CLONE() instead of duplicating it inline */
859 params = g_malloc0(sizeof(*params));
860 params->has_compress_level = true;
861 params->compress_level = s->parameters.compress_level;
862 params->has_compress_threads = true;
863 params->compress_threads = s->parameters.compress_threads;
864 params->has_compress_wait_thread = true;
865 params->compress_wait_thread = s->parameters.compress_wait_thread;
866 params->has_decompress_threads = true;
867 params->decompress_threads = s->parameters.decompress_threads;
868 params->has_throttle_trigger_threshold = true;
869 params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
870 params->has_cpu_throttle_initial = true;
871 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
872 params->has_cpu_throttle_increment = true;
873 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
874 params->has_cpu_throttle_tailslow = true;
875 params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
876 params->has_tls_creds = true;
877 params->tls_creds = g_strdup(s->parameters.tls_creds);
878 params->has_tls_hostname = true;
879 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
880 params->has_tls_authz = true;
881 params->tls_authz = g_strdup(s->parameters.tls_authz ?
882 s->parameters.tls_authz : "");
883 params->has_max_bandwidth = true;
884 params->max_bandwidth = s->parameters.max_bandwidth;
885 params->has_downtime_limit = true;
886 params->downtime_limit = s->parameters.downtime_limit;
887 params->has_x_checkpoint_delay = true;
888 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
889 params->has_block_incremental = true;
890 params->block_incremental = s->parameters.block_incremental;
891 params->has_multifd_channels = true;
892 params->multifd_channels = s->parameters.multifd_channels;
893 params->has_multifd_compression = true;
894 params->multifd_compression = s->parameters.multifd_compression;
895 params->has_multifd_zlib_level = true;
896 params->multifd_zlib_level = s->parameters.multifd_zlib_level;
897 params->has_multifd_zstd_level = true;
898 params->multifd_zstd_level = s->parameters.multifd_zstd_level;
899 params->has_xbzrle_cache_size = true;
900 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
901 params->has_max_postcopy_bandwidth = true;
902 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
903 params->has_max_cpu_throttle = true;
904 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
905 params->has_announce_initial = true;
906 params->announce_initial = s->parameters.announce_initial;
907 params->has_announce_max = true;
908 params->announce_max = s->parameters.announce_max;
909 params->has_announce_rounds = true;
910 params->announce_rounds = s->parameters.announce_rounds;
911 params->has_announce_step = true;
912 params->announce_step = s->parameters.announce_step;
913
914 if (s->parameters.has_block_bitmap_mapping) {
915 params->has_block_bitmap_mapping = true;
916 params->block_bitmap_mapping =
917 QAPI_CLONE(BitmapMigrationNodeAliasList,
918 s->parameters.block_bitmap_mapping);
919 }
920
921 return params;
922 }
923
924 AnnounceParameters *migrate_announce_params(void)
925 {
926 static AnnounceParameters ap;
927
928 MigrationState *s = migrate_get_current();
929
930 ap.initial = s->parameters.announce_initial;
931 ap.max = s->parameters.announce_max;
932 ap.rounds = s->parameters.announce_rounds;
933 ap.step = s->parameters.announce_step;
934
935 return &ap;
936 }
937
938 /*
939 * Return true if we're already in the middle of a migration
940 * (i.e. any of the active or setup states)
941 */
942 bool migration_is_setup_or_active(int state)
943 {
944 switch (state) {
945 case MIGRATION_STATUS_ACTIVE:
946 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
947 case MIGRATION_STATUS_POSTCOPY_PAUSED:
948 case MIGRATION_STATUS_POSTCOPY_RECOVER:
949 case MIGRATION_STATUS_SETUP:
950 case MIGRATION_STATUS_PRE_SWITCHOVER:
951 case MIGRATION_STATUS_DEVICE:
952 case MIGRATION_STATUS_WAIT_UNPLUG:
953 case MIGRATION_STATUS_COLO:
954 return true;
955
956 default:
957 return false;
958
959 }
960 }
961
962 bool migration_is_running(int state)
963 {
964 switch (state) {
965 case MIGRATION_STATUS_ACTIVE:
966 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
967 case MIGRATION_STATUS_POSTCOPY_PAUSED:
968 case MIGRATION_STATUS_POSTCOPY_RECOVER:
969 case MIGRATION_STATUS_SETUP:
970 case MIGRATION_STATUS_PRE_SWITCHOVER:
971 case MIGRATION_STATUS_DEVICE:
972 case MIGRATION_STATUS_WAIT_UNPLUG:
973 case MIGRATION_STATUS_CANCELLING:
974 return true;
975
976 default:
977 return false;
978
979 }
980 }
981
982 static void populate_time_info(MigrationInfo *info, MigrationState *s)
983 {
984 info->has_status = true;
985 info->has_setup_time = true;
986 info->setup_time = s->setup_time;
987 if (s->state == MIGRATION_STATUS_COMPLETED) {
988 info->has_total_time = true;
989 info->total_time = s->total_time;
990 info->has_downtime = true;
991 info->downtime = s->downtime;
992 } else {
993 info->has_total_time = true;
994 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
995 s->start_time;
996 info->has_expected_downtime = true;
997 info->expected_downtime = s->expected_downtime;
998 }
999 }
1000
1001 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
1002 {
1003 info->has_ram = true;
1004 info->ram = g_malloc0(sizeof(*info->ram));
1005 info->ram->transferred = ram_counters.transferred;
1006 info->ram->total = ram_bytes_total();
1007 info->ram->duplicate = ram_counters.duplicate;
1008 /* legacy value. It is not used anymore */
1009 info->ram->skipped = 0;
1010 info->ram->normal = ram_counters.normal;
1011 info->ram->normal_bytes = ram_counters.normal *
1012 qemu_target_page_size();
1013 info->ram->mbps = s->mbps;
1014 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
1015 info->ram->postcopy_requests = ram_counters.postcopy_requests;
1016 info->ram->page_size = qemu_target_page_size();
1017 info->ram->multifd_bytes = ram_counters.multifd_bytes;
1018 info->ram->pages_per_second = s->pages_per_second;
1019
1020 if (migrate_use_xbzrle()) {
1021 info->has_xbzrle_cache = true;
1022 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
1023 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
1024 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
1025 info->xbzrle_cache->pages = xbzrle_counters.pages;
1026 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
1027 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
1028 info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
1029 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
1030 }
1031
1032 if (migrate_use_compression()) {
1033 info->has_compression = true;
1034 info->compression = g_malloc0(sizeof(*info->compression));
1035 info->compression->pages = compression_counters.pages;
1036 info->compression->busy = compression_counters.busy;
1037 info->compression->busy_rate = compression_counters.busy_rate;
1038 info->compression->compressed_size =
1039 compression_counters.compressed_size;
1040 info->compression->compression_rate =
1041 compression_counters.compression_rate;
1042 }
1043
1044 if (cpu_throttle_active()) {
1045 info->has_cpu_throttle_percentage = true;
1046 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
1047 }
1048
1049 if (s->state != MIGRATION_STATUS_COMPLETED) {
1050 info->ram->remaining = ram_bytes_remaining();
1051 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
1052 }
1053 }
1054
1055 static void populate_disk_info(MigrationInfo *info)
1056 {
1057 if (blk_mig_active()) {
1058 info->has_disk = true;
1059 info->disk = g_malloc0(sizeof(*info->disk));
1060 info->disk->transferred = blk_mig_bytes_transferred();
1061 info->disk->remaining = blk_mig_bytes_remaining();
1062 info->disk->total = blk_mig_bytes_total();
1063 }
1064 }
1065
1066 static void fill_source_migration_info(MigrationInfo *info)
1067 {
1068 MigrationState *s = migrate_get_current();
1069 GSList *cur_blocker = migration_blockers;
1070
1071 info->blocked_reasons = NULL;
1072
1073 /*
1074 * There are two types of reasons a migration might be blocked;
1075 * a) devices marked in VMState as non-migratable, and
1076 * b) Explicit migration blockers
1077 * We need to add both of them here.
1078 */
1079 qemu_savevm_non_migratable_list(&info->blocked_reasons);
1080
1081 while (cur_blocker) {
1082 QAPI_LIST_PREPEND(info->blocked_reasons,
1083 g_strdup(error_get_pretty(cur_blocker->data)));
1084 cur_blocker = g_slist_next(cur_blocker);
1085 }
1086 info->has_blocked_reasons = info->blocked_reasons != NULL;
1087
1088 switch (s->state) {
1089 case MIGRATION_STATUS_NONE:
1090 /* no migration has happened ever */
1091 /* do not overwrite destination migration status */
1092 return;
1093 case MIGRATION_STATUS_SETUP:
1094 info->has_status = true;
1095 info->has_total_time = false;
1096 break;
1097 case MIGRATION_STATUS_ACTIVE:
1098 case MIGRATION_STATUS_CANCELLING:
1099 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1100 case MIGRATION_STATUS_PRE_SWITCHOVER:
1101 case MIGRATION_STATUS_DEVICE:
1102 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1103 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1104 /* TODO add some postcopy stats */
1105 populate_time_info(info, s);
1106 populate_ram_info(info, s);
1107 populate_disk_info(info);
1108 populate_vfio_info(info);
1109 break;
1110 case MIGRATION_STATUS_COLO:
1111 info->has_status = true;
1112 /* TODO: display COLO specific information (checkpoint info etc.) */
1113 break;
1114 case MIGRATION_STATUS_COMPLETED:
1115 populate_time_info(info, s);
1116 populate_ram_info(info, s);
1117 populate_vfio_info(info);
1118 break;
1119 case MIGRATION_STATUS_FAILED:
1120 info->has_status = true;
1121 if (s->error) {
1122 info->has_error_desc = true;
1123 info->error_desc = g_strdup(error_get_pretty(s->error));
1124 }
1125 break;
1126 case MIGRATION_STATUS_CANCELLED:
1127 info->has_status = true;
1128 break;
1129 case MIGRATION_STATUS_WAIT_UNPLUG:
1130 info->has_status = true;
1131 break;
1132 }
1133 info->status = s->state;
1134 }
1135
1136 typedef enum WriteTrackingSupport {
1137 WT_SUPPORT_UNKNOWN = 0,
1138 WT_SUPPORT_ABSENT,
1139 WT_SUPPORT_AVAILABLE,
1140 WT_SUPPORT_COMPATIBLE
1141 } WriteTrackingSupport;
1142
1143 static
1144 WriteTrackingSupport migrate_query_write_tracking(void)
1145 {
1146 /* Check if kernel supports required UFFD features */
1147 if (!ram_write_tracking_available()) {
1148 return WT_SUPPORT_ABSENT;
1149 }
1150 /*
1151 * Check if current memory configuration is
1152 * compatible with required UFFD features.
1153 */
1154 if (!ram_write_tracking_compatible()) {
1155 return WT_SUPPORT_AVAILABLE;
1156 }
1157
1158 return WT_SUPPORT_COMPATIBLE;
1159 }
1160
1161 /**
1162 * @migration_caps_check - check capability validity
1163 *
1164 * @cap_list: old capability list, array of bool
1165 * @params: new capabilities to be applied soon
1166 * @errp: set *errp if the check failed, with reason
1167 *
1168 * Returns true if check passed, otherwise false.
1169 */
1170 static bool migrate_caps_check(bool *cap_list,
1171 MigrationCapabilityStatusList *params,
1172 Error **errp)
1173 {
1174 MigrationCapabilityStatusList *cap;
1175 bool old_postcopy_cap;
1176 MigrationIncomingState *mis = migration_incoming_get_current();
1177
1178 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1179
1180 for (cap = params; cap; cap = cap->next) {
1181 cap_list[cap->value->capability] = cap->value->state;
1182 }
1183
1184 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1185 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1186 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1187 "block migration");
1188 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1189 return false;
1190 }
1191 #endif
1192
1193 #ifndef CONFIG_REPLICATION
1194 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1195 error_setg(errp, "QEMU compiled without replication module"
1196 " can't enable COLO");
1197 error_append_hint(errp, "Please enable replication before COLO.\n");
1198 return false;
1199 }
1200 #endif
1201
1202 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1203 /* This check is reasonably expensive, so only when it's being
1204 * set the first time, also it's only the destination that needs
1205 * special support.
1206 */
1207 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1208 !postcopy_ram_supported_by_host(mis)) {
1209 /* postcopy_ram_supported_by_host will have emitted a more
1210 * detailed message
1211 */
1212 error_setg(errp, "Postcopy is not supported");
1213 return false;
1214 }
1215
1216 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1217 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1218 return false;
1219 }
1220 }
1221
1222 if (cap_list[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]) {
1223 WriteTrackingSupport wt_support;
1224 int idx;
1225 /*
1226 * Check if 'background-snapshot' capability is supported by
1227 * host kernel and compatible with guest memory configuration.
1228 */
1229 wt_support = migrate_query_write_tracking();
1230 if (wt_support < WT_SUPPORT_AVAILABLE) {
1231 error_setg(errp, "Background-snapshot is not supported by host kernel");
1232 return false;
1233 }
1234 if (wt_support < WT_SUPPORT_COMPATIBLE) {
1235 error_setg(errp, "Background-snapshot is not compatible "
1236 "with guest memory configuration");
1237 return false;
1238 }
1239
1240 /*
1241 * Check if there are any migration capabilities
1242 * incompatible with 'background-snapshot'.
1243 */
1244 for (idx = 0; idx < check_caps_background_snapshot.size; idx++) {
1245 int incomp_cap = check_caps_background_snapshot.caps[idx];
1246 if (cap_list[incomp_cap]) {
1247 error_setg(errp,
1248 "Background-snapshot is not compatible with %s",
1249 MigrationCapability_str(incomp_cap));
1250 return false;
1251 }
1252 }
1253 }
1254
1255 return true;
1256 }
1257
1258 static void fill_destination_migration_info(MigrationInfo *info)
1259 {
1260 MigrationIncomingState *mis = migration_incoming_get_current();
1261
1262 if (mis->socket_address_list) {
1263 info->has_socket_address = true;
1264 info->socket_address =
1265 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1266 }
1267
1268 switch (mis->state) {
1269 case MIGRATION_STATUS_NONE:
1270 return;
1271 case MIGRATION_STATUS_SETUP:
1272 case MIGRATION_STATUS_CANCELLING:
1273 case MIGRATION_STATUS_CANCELLED:
1274 case MIGRATION_STATUS_ACTIVE:
1275 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1276 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1277 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1278 case MIGRATION_STATUS_FAILED:
1279 case MIGRATION_STATUS_COLO:
1280 info->has_status = true;
1281 break;
1282 case MIGRATION_STATUS_COMPLETED:
1283 info->has_status = true;
1284 fill_destination_postcopy_migration_info(info);
1285 break;
1286 }
1287 info->status = mis->state;
1288 }
1289
1290 MigrationInfo *qmp_query_migrate(Error **errp)
1291 {
1292 MigrationInfo *info = g_malloc0(sizeof(*info));
1293
1294 fill_destination_migration_info(info);
1295 fill_source_migration_info(info);
1296
1297 return info;
1298 }
1299
1300 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1301 Error **errp)
1302 {
1303 MigrationState *s = migrate_get_current();
1304 MigrationCapabilityStatusList *cap;
1305 bool cap_list[MIGRATION_CAPABILITY__MAX];
1306
1307 if (migration_is_running(s->state)) {
1308 error_setg(errp, QERR_MIGRATION_ACTIVE);
1309 return;
1310 }
1311
1312 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1313 if (!migrate_caps_check(cap_list, params, errp)) {
1314 return;
1315 }
1316
1317 for (cap = params; cap; cap = cap->next) {
1318 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1319 }
1320 }
1321
1322 /*
1323 * Check whether the parameters are valid. Error will be put into errp
1324 * (if provided). Return true if valid, otherwise false.
1325 */
1326 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1327 {
1328 if (params->has_compress_level &&
1329 (params->compress_level > 9)) {
1330 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1331 "a value between 0 and 9");
1332 return false;
1333 }
1334
1335 if (params->has_compress_threads && (params->compress_threads < 1)) {
1336 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1337 "compress_threads",
1338 "a value between 1 and 255");
1339 return false;
1340 }
1341
1342 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1343 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1344 "decompress_threads",
1345 "a value between 1 and 255");
1346 return false;
1347 }
1348
1349 if (params->has_throttle_trigger_threshold &&
1350 (params->throttle_trigger_threshold < 1 ||
1351 params->throttle_trigger_threshold > 100)) {
1352 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1353 "throttle_trigger_threshold",
1354 "an integer in the range of 1 to 100");
1355 return false;
1356 }
1357
1358 if (params->has_cpu_throttle_initial &&
1359 (params->cpu_throttle_initial < 1 ||
1360 params->cpu_throttle_initial > 99)) {
1361 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1362 "cpu_throttle_initial",
1363 "an integer in the range of 1 to 99");
1364 return false;
1365 }
1366
1367 if (params->has_cpu_throttle_increment &&
1368 (params->cpu_throttle_increment < 1 ||
1369 params->cpu_throttle_increment > 99)) {
1370 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1371 "cpu_throttle_increment",
1372 "an integer in the range of 1 to 99");
1373 return false;
1374 }
1375
1376 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1377 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1378 "max_bandwidth",
1379 "an integer in the range of 0 to "stringify(SIZE_MAX)
1380 " bytes/second");
1381 return false;
1382 }
1383
1384 if (params->has_downtime_limit &&
1385 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1386 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1387 "downtime_limit",
1388 "an integer in the range of 0 to "
1389 stringify(MAX_MIGRATE_DOWNTIME)" ms");
1390 return false;
1391 }
1392
1393 /* x_checkpoint_delay is now always positive */
1394
1395 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1396 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1397 "multifd_channels",
1398 "a value between 1 and 255");
1399 return false;
1400 }
1401
1402 if (params->has_multifd_zlib_level &&
1403 (params->multifd_zlib_level > 9)) {
1404 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1405 "a value between 0 and 9");
1406 return false;
1407 }
1408
1409 if (params->has_multifd_zstd_level &&
1410 (params->multifd_zstd_level > 20)) {
1411 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1412 "a value between 0 and 20");
1413 return false;
1414 }
1415
1416 if (params->has_xbzrle_cache_size &&
1417 (params->xbzrle_cache_size < qemu_target_page_size() ||
1418 !is_power_of_2(params->xbzrle_cache_size))) {
1419 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1420 "xbzrle_cache_size",
1421 "a power of two no less than the target page size");
1422 return false;
1423 }
1424
1425 if (params->has_max_cpu_throttle &&
1426 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1427 params->max_cpu_throttle > 99)) {
1428 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1429 "max_cpu_throttle",
1430 "an integer in the range of cpu_throttle_initial to 99");
1431 return false;
1432 }
1433
1434 if (params->has_announce_initial &&
1435 params->announce_initial > 100000) {
1436 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1437 "announce_initial",
1438 "a value between 0 and 100000");
1439 return false;
1440 }
1441 if (params->has_announce_max &&
1442 params->announce_max > 100000) {
1443 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1444 "announce_max",
1445 "a value between 0 and 100000");
1446 return false;
1447 }
1448 if (params->has_announce_rounds &&
1449 params->announce_rounds > 1000) {
1450 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1451 "announce_rounds",
1452 "a value between 0 and 1000");
1453 return false;
1454 }
1455 if (params->has_announce_step &&
1456 (params->announce_step < 1 ||
1457 params->announce_step > 10000)) {
1458 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1459 "announce_step",
1460 "a value between 0 and 10000");
1461 return false;
1462 }
1463
1464 if (params->has_block_bitmap_mapping &&
1465 !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) {
1466 error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: ");
1467 return false;
1468 }
1469
1470 return true;
1471 }
1472
1473 static void migrate_params_test_apply(MigrateSetParameters *params,
1474 MigrationParameters *dest)
1475 {
1476 *dest = migrate_get_current()->parameters;
1477
1478 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1479
1480 if (params->has_compress_level) {
1481 dest->compress_level = params->compress_level;
1482 }
1483
1484 if (params->has_compress_threads) {
1485 dest->compress_threads = params->compress_threads;
1486 }
1487
1488 if (params->has_compress_wait_thread) {
1489 dest->compress_wait_thread = params->compress_wait_thread;
1490 }
1491
1492 if (params->has_decompress_threads) {
1493 dest->decompress_threads = params->decompress_threads;
1494 }
1495
1496 if (params->has_throttle_trigger_threshold) {
1497 dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
1498 }
1499
1500 if (params->has_cpu_throttle_initial) {
1501 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1502 }
1503
1504 if (params->has_cpu_throttle_increment) {
1505 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1506 }
1507
1508 if (params->has_cpu_throttle_tailslow) {
1509 dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1510 }
1511
1512 if (params->has_tls_creds) {
1513 assert(params->tls_creds->type == QTYPE_QSTRING);
1514 dest->tls_creds = params->tls_creds->u.s;
1515 }
1516
1517 if (params->has_tls_hostname) {
1518 assert(params->tls_hostname->type == QTYPE_QSTRING);
1519 dest->tls_hostname = params->tls_hostname->u.s;
1520 }
1521
1522 if (params->has_max_bandwidth) {
1523 dest->max_bandwidth = params->max_bandwidth;
1524 }
1525
1526 if (params->has_downtime_limit) {
1527 dest->downtime_limit = params->downtime_limit;
1528 }
1529
1530 if (params->has_x_checkpoint_delay) {
1531 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1532 }
1533
1534 if (params->has_block_incremental) {
1535 dest->block_incremental = params->block_incremental;
1536 }
1537 if (params->has_multifd_channels) {
1538 dest->multifd_channels = params->multifd_channels;
1539 }
1540 if (params->has_multifd_compression) {
1541 dest->multifd_compression = params->multifd_compression;
1542 }
1543 if (params->has_xbzrle_cache_size) {
1544 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1545 }
1546 if (params->has_max_postcopy_bandwidth) {
1547 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1548 }
1549 if (params->has_max_cpu_throttle) {
1550 dest->max_cpu_throttle = params->max_cpu_throttle;
1551 }
1552 if (params->has_announce_initial) {
1553 dest->announce_initial = params->announce_initial;
1554 }
1555 if (params->has_announce_max) {
1556 dest->announce_max = params->announce_max;
1557 }
1558 if (params->has_announce_rounds) {
1559 dest->announce_rounds = params->announce_rounds;
1560 }
1561 if (params->has_announce_step) {
1562 dest->announce_step = params->announce_step;
1563 }
1564
1565 if (params->has_block_bitmap_mapping) {
1566 dest->has_block_bitmap_mapping = true;
1567 dest->block_bitmap_mapping = params->block_bitmap_mapping;
1568 }
1569 }
1570
1571 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1572 {
1573 MigrationState *s = migrate_get_current();
1574
1575 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1576
1577 if (params->has_compress_level) {
1578 s->parameters.compress_level = params->compress_level;
1579 }
1580
1581 if (params->has_compress_threads) {
1582 s->parameters.compress_threads = params->compress_threads;
1583 }
1584
1585 if (params->has_compress_wait_thread) {
1586 s->parameters.compress_wait_thread = params->compress_wait_thread;
1587 }
1588
1589 if (params->has_decompress_threads) {
1590 s->parameters.decompress_threads = params->decompress_threads;
1591 }
1592
1593 if (params->has_throttle_trigger_threshold) {
1594 s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
1595 }
1596
1597 if (params->has_cpu_throttle_initial) {
1598 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1599 }
1600
1601 if (params->has_cpu_throttle_increment) {
1602 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1603 }
1604
1605 if (params->has_cpu_throttle_tailslow) {
1606 s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1607 }
1608
1609 if (params->has_tls_creds) {
1610 g_free(s->parameters.tls_creds);
1611 assert(params->tls_creds->type == QTYPE_QSTRING);
1612 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1613 }
1614
1615 if (params->has_tls_hostname) {
1616 g_free(s->parameters.tls_hostname);
1617 assert(params->tls_hostname->type == QTYPE_QSTRING);
1618 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1619 }
1620
1621 if (params->has_tls_authz) {
1622 g_free(s->parameters.tls_authz);
1623 assert(params->tls_authz->type == QTYPE_QSTRING);
1624 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1625 }
1626
1627 if (params->has_max_bandwidth) {
1628 s->parameters.max_bandwidth = params->max_bandwidth;
1629 if (s->to_dst_file && !migration_in_postcopy()) {
1630 qemu_file_set_rate_limit(s->to_dst_file,
1631 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1632 }
1633 }
1634
1635 if (params->has_downtime_limit) {
1636 s->parameters.downtime_limit = params->downtime_limit;
1637 }
1638
1639 if (params->has_x_checkpoint_delay) {
1640 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1641 if (migration_in_colo_state()) {
1642 colo_checkpoint_notify(s);
1643 }
1644 }
1645
1646 if (params->has_block_incremental) {
1647 s->parameters.block_incremental = params->block_incremental;
1648 }
1649 if (params->has_multifd_channels) {
1650 s->parameters.multifd_channels = params->multifd_channels;
1651 }
1652 if (params->has_multifd_compression) {
1653 s->parameters.multifd_compression = params->multifd_compression;
1654 }
1655 if (params->has_xbzrle_cache_size) {
1656 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1657 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1658 }
1659 if (params->has_max_postcopy_bandwidth) {
1660 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1661 if (s->to_dst_file && migration_in_postcopy()) {
1662 qemu_file_set_rate_limit(s->to_dst_file,
1663 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1664 }
1665 }
1666 if (params->has_max_cpu_throttle) {
1667 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1668 }
1669 if (params->has_announce_initial) {
1670 s->parameters.announce_initial = params->announce_initial;
1671 }
1672 if (params->has_announce_max) {
1673 s->parameters.announce_max = params->announce_max;
1674 }
1675 if (params->has_announce_rounds) {
1676 s->parameters.announce_rounds = params->announce_rounds;
1677 }
1678 if (params->has_announce_step) {
1679 s->parameters.announce_step = params->announce_step;
1680 }
1681
1682 if (params->has_block_bitmap_mapping) {
1683 qapi_free_BitmapMigrationNodeAliasList(
1684 s->parameters.block_bitmap_mapping);
1685
1686 s->parameters.has_block_bitmap_mapping = true;
1687 s->parameters.block_bitmap_mapping =
1688 QAPI_CLONE(BitmapMigrationNodeAliasList,
1689 params->block_bitmap_mapping);
1690 }
1691 }
1692
1693 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1694 {
1695 MigrationParameters tmp;
1696
1697 /* TODO Rewrite "" to null instead */
1698 if (params->has_tls_creds
1699 && params->tls_creds->type == QTYPE_QNULL) {
1700 qobject_unref(params->tls_creds->u.n);
1701 params->tls_creds->type = QTYPE_QSTRING;
1702 params->tls_creds->u.s = strdup("");
1703 }
1704 /* TODO Rewrite "" to null instead */
1705 if (params->has_tls_hostname
1706 && params->tls_hostname->type == QTYPE_QNULL) {
1707 qobject_unref(params->tls_hostname->u.n);
1708 params->tls_hostname->type = QTYPE_QSTRING;
1709 params->tls_hostname->u.s = strdup("");
1710 }
1711
1712 migrate_params_test_apply(params, &tmp);
1713
1714 if (!migrate_params_check(&tmp, errp)) {
1715 /* Invalid parameter */
1716 return;
1717 }
1718
1719 migrate_params_apply(params, errp);
1720 }
1721
1722
1723 void qmp_migrate_start_postcopy(Error **errp)
1724 {
1725 MigrationState *s = migrate_get_current();
1726
1727 if (!migrate_postcopy()) {
1728 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1729 " the start of migration");
1730 return;
1731 }
1732
1733 if (s->state == MIGRATION_STATUS_NONE) {
1734 error_setg(errp, "Postcopy must be started after migration has been"
1735 " started");
1736 return;
1737 }
1738 /*
1739 * we don't error if migration has finished since that would be racy
1740 * with issuing this command.
1741 */
1742 qatomic_set(&s->start_postcopy, true);
1743 }
1744
1745 /* shared migration helpers */
1746
1747 void migrate_set_state(int *state, int old_state, int new_state)
1748 {
1749 assert(new_state < MIGRATION_STATUS__MAX);
1750 if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
1751 trace_migrate_set_state(MigrationStatus_str(new_state));
1752 migrate_generate_event(new_state);
1753 }
1754 }
1755
1756 static MigrationCapabilityStatus *migrate_cap_add(MigrationCapability index,
1757 bool state)
1758 {
1759 MigrationCapabilityStatus *cap;
1760
1761 cap = g_new0(MigrationCapabilityStatus, 1);
1762 cap->capability = index;
1763 cap->state = state;
1764
1765 return cap;
1766 }
1767
1768 void migrate_set_block_enabled(bool value, Error **errp)
1769 {
1770 MigrationCapabilityStatusList *cap = NULL;
1771
1772 QAPI_LIST_PREPEND(cap, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK, value));
1773 qmp_migrate_set_capabilities(cap, errp);
1774 qapi_free_MigrationCapabilityStatusList(cap);
1775 }
1776
1777 static void migrate_set_block_incremental(MigrationState *s, bool value)
1778 {
1779 s->parameters.block_incremental = value;
1780 }
1781
1782 static void block_cleanup_parameters(MigrationState *s)
1783 {
1784 if (s->must_remove_block_options) {
1785 /* setting to false can never fail */
1786 migrate_set_block_enabled(false, &error_abort);
1787 migrate_set_block_incremental(s, false);
1788 s->must_remove_block_options = false;
1789 }
1790 }
1791
1792 static void migrate_fd_cleanup(MigrationState *s)
1793 {
1794 qemu_bh_delete(s->cleanup_bh);
1795 s->cleanup_bh = NULL;
1796
1797 qemu_savevm_state_cleanup();
1798
1799 if (s->to_dst_file) {
1800 QEMUFile *tmp;
1801
1802 trace_migrate_fd_cleanup();
1803 qemu_mutex_unlock_iothread();
1804 if (s->migration_thread_running) {
1805 qemu_thread_join(&s->thread);
1806 s->migration_thread_running = false;
1807 }
1808 qemu_mutex_lock_iothread();
1809
1810 multifd_save_cleanup();
1811 qemu_mutex_lock(&s->qemu_file_lock);
1812 tmp = s->to_dst_file;
1813 s->to_dst_file = NULL;
1814 qemu_mutex_unlock(&s->qemu_file_lock);
1815 /*
1816 * Close the file handle without the lock to make sure the
1817 * critical section won't block for long.
1818 */
1819 qemu_fclose(tmp);
1820 }
1821
1822 assert(!migration_is_active(s));
1823
1824 if (s->state == MIGRATION_STATUS_CANCELLING) {
1825 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1826 MIGRATION_STATUS_CANCELLED);
1827 }
1828
1829 if (s->error) {
1830 /* It is used on info migrate. We can't free it */
1831 error_report_err(error_copy(s->error));
1832 }
1833 notifier_list_notify(&migration_state_notifiers, s);
1834 block_cleanup_parameters(s);
1835 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
1836 }
1837
1838 static void migrate_fd_cleanup_schedule(MigrationState *s)
1839 {
1840 /*
1841 * Ref the state for bh, because it may be called when
1842 * there're already no other refs
1843 */
1844 object_ref(OBJECT(s));
1845 qemu_bh_schedule(s->cleanup_bh);
1846 }
1847
1848 static void migrate_fd_cleanup_bh(void *opaque)
1849 {
1850 MigrationState *s = opaque;
1851 migrate_fd_cleanup(s);
1852 object_unref(OBJECT(s));
1853 }
1854
1855 void migrate_set_error(MigrationState *s, const Error *error)
1856 {
1857 QEMU_LOCK_GUARD(&s->error_mutex);
1858 if (!s->error) {
1859 s->error = error_copy(error);
1860 }
1861 }
1862
1863 void migrate_fd_error(MigrationState *s, const Error *error)
1864 {
1865 trace_migrate_fd_error(error_get_pretty(error));
1866 assert(s->to_dst_file == NULL);
1867 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1868 MIGRATION_STATUS_FAILED);
1869 migrate_set_error(s, error);
1870 }
1871
1872 static void migrate_fd_cancel(MigrationState *s)
1873 {
1874 int old_state ;
1875 QEMUFile *f = migrate_get_current()->to_dst_file;
1876 trace_migrate_fd_cancel();
1877
1878 if (s->rp_state.from_dst_file) {
1879 /* shutdown the rp socket, so causing the rp thread to shutdown */
1880 qemu_file_shutdown(s->rp_state.from_dst_file);
1881 }
1882
1883 do {
1884 old_state = s->state;
1885 if (!migration_is_running(old_state)) {
1886 break;
1887 }
1888 /* If the migration is paused, kick it out of the pause */
1889 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1890 qemu_sem_post(&s->pause_sem);
1891 }
1892 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1893 } while (s->state != MIGRATION_STATUS_CANCELLING);
1894
1895 /*
1896 * If we're unlucky the migration code might be stuck somewhere in a
1897 * send/write while the network has failed and is waiting to timeout;
1898 * if we've got shutdown(2) available then we can force it to quit.
1899 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1900 * called in a bh, so there is no race against this cancel.
1901 */
1902 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1903 qemu_file_shutdown(f);
1904 }
1905 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1906 Error *local_err = NULL;
1907
1908 bdrv_invalidate_cache_all(&local_err);
1909 if (local_err) {
1910 error_report_err(local_err);
1911 } else {
1912 s->block_inactive = false;
1913 }
1914 }
1915 }
1916
1917 void add_migration_state_change_notifier(Notifier *notify)
1918 {
1919 notifier_list_add(&migration_state_notifiers, notify);
1920 }
1921
1922 void remove_migration_state_change_notifier(Notifier *notify)
1923 {
1924 notifier_remove(notify);
1925 }
1926
1927 bool migration_in_setup(MigrationState *s)
1928 {
1929 return s->state == MIGRATION_STATUS_SETUP;
1930 }
1931
1932 bool migration_has_finished(MigrationState *s)
1933 {
1934 return s->state == MIGRATION_STATUS_COMPLETED;
1935 }
1936
1937 bool migration_has_failed(MigrationState *s)
1938 {
1939 return (s->state == MIGRATION_STATUS_CANCELLED ||
1940 s->state == MIGRATION_STATUS_FAILED);
1941 }
1942
1943 bool migration_in_postcopy(void)
1944 {
1945 MigrationState *s = migrate_get_current();
1946
1947 switch (s->state) {
1948 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1949 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1950 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1951 return true;
1952 default:
1953 return false;
1954 }
1955 }
1956
1957 bool migration_in_postcopy_after_devices(MigrationState *s)
1958 {
1959 return migration_in_postcopy() && s->postcopy_after_devices;
1960 }
1961
1962 bool migration_in_incoming_postcopy(void)
1963 {
1964 PostcopyState ps = postcopy_state_get();
1965
1966 return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
1967 }
1968
1969 bool migration_in_bg_snapshot(void)
1970 {
1971 MigrationState *s = migrate_get_current();
1972
1973 return migrate_background_snapshot() &&
1974 migration_is_setup_or_active(s->state);
1975 }
1976
1977 bool migration_is_idle(void)
1978 {
1979 MigrationState *s = current_migration;
1980
1981 if (!s) {
1982 return true;
1983 }
1984
1985 switch (s->state) {
1986 case MIGRATION_STATUS_NONE:
1987 case MIGRATION_STATUS_CANCELLED:
1988 case MIGRATION_STATUS_COMPLETED:
1989 case MIGRATION_STATUS_FAILED:
1990 return true;
1991 case MIGRATION_STATUS_SETUP:
1992 case MIGRATION_STATUS_CANCELLING:
1993 case MIGRATION_STATUS_ACTIVE:
1994 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1995 case MIGRATION_STATUS_COLO:
1996 case MIGRATION_STATUS_PRE_SWITCHOVER:
1997 case MIGRATION_STATUS_DEVICE:
1998 case MIGRATION_STATUS_WAIT_UNPLUG:
1999 return false;
2000 case MIGRATION_STATUS__MAX:
2001 g_assert_not_reached();
2002 }
2003
2004 return false;
2005 }
2006
2007 bool migration_is_active(MigrationState *s)
2008 {
2009 return (s->state == MIGRATION_STATUS_ACTIVE ||
2010 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2011 }
2012
2013 void migrate_init(MigrationState *s)
2014 {
2015 /*
2016 * Reinitialise all migration state, except
2017 * parameters/capabilities that the user set, and
2018 * locks.
2019 */
2020 s->cleanup_bh = 0;
2021 s->vm_start_bh = 0;
2022 s->to_dst_file = NULL;
2023 s->state = MIGRATION_STATUS_NONE;
2024 s->rp_state.from_dst_file = NULL;
2025 s->rp_state.error = false;
2026 s->mbps = 0.0;
2027 s->pages_per_second = 0.0;
2028 s->downtime = 0;
2029 s->expected_downtime = 0;
2030 s->setup_time = 0;
2031 s->start_postcopy = false;
2032 s->postcopy_after_devices = false;
2033 s->migration_thread_running = false;
2034 error_free(s->error);
2035 s->error = NULL;
2036 s->hostname = NULL;
2037
2038 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
2039
2040 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2041 s->total_time = 0;
2042 s->vm_was_running = false;
2043 s->iteration_initial_bytes = 0;
2044 s->threshold_size = 0;
2045 }
2046
2047 int migrate_add_blocker(Error *reason, Error **errp)
2048 {
2049 if (only_migratable) {
2050 error_propagate_prepend(errp, error_copy(reason),
2051 "disallowing migration blocker "
2052 "(--only-migratable) for: ");
2053 return -EACCES;
2054 }
2055
2056 if (migration_is_idle()) {
2057 migration_blockers = g_slist_prepend(migration_blockers, reason);
2058 return 0;
2059 }
2060
2061 error_propagate_prepend(errp, error_copy(reason),
2062 "disallowing migration blocker "
2063 "(migration in progress) for: ");
2064 return -EBUSY;
2065 }
2066
2067 void migrate_del_blocker(Error *reason)
2068 {
2069 migration_blockers = g_slist_remove(migration_blockers, reason);
2070 }
2071
2072 void qmp_migrate_incoming(const char *uri, Error **errp)
2073 {
2074 Error *local_err = NULL;
2075 static bool once = true;
2076
2077 if (!once) {
2078 error_setg(errp, "The incoming migration has already been started");
2079 return;
2080 }
2081 if (!runstate_check(RUN_STATE_INMIGRATE)) {
2082 error_setg(errp, "'-incoming' was not specified on the command line");
2083 return;
2084 }
2085
2086 qemu_start_incoming_migration(uri, &local_err);
2087
2088 if (local_err) {
2089 error_propagate(errp, local_err);
2090 return;
2091 }
2092
2093 once = false;
2094 }
2095
2096 void qmp_migrate_recover(const char *uri, Error **errp)
2097 {
2098 MigrationIncomingState *mis = migration_incoming_get_current();
2099
2100 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2101 error_setg(errp, "Migrate recover can only be run "
2102 "when postcopy is paused.");
2103 return;
2104 }
2105
2106 if (qatomic_cmpxchg(&mis->postcopy_recover_triggered,
2107 false, true) == true) {
2108 error_setg(errp, "Migrate recovery is triggered already");
2109 return;
2110 }
2111
2112 /*
2113 * Note that this call will never start a real migration; it will
2114 * only re-setup the migration stream and poke existing migration
2115 * to continue using that newly established channel.
2116 */
2117 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2118 qemu_start_incoming_migration(uri, errp);
2119 }
2120
2121 void qmp_migrate_pause(Error **errp)
2122 {
2123 MigrationState *ms = migrate_get_current();
2124 MigrationIncomingState *mis = migration_incoming_get_current();
2125 int ret;
2126
2127 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2128 /* Source side, during postcopy */
2129 qemu_mutex_lock(&ms->qemu_file_lock);
2130 ret = qemu_file_shutdown(ms->to_dst_file);
2131 qemu_mutex_unlock(&ms->qemu_file_lock);
2132 if (ret) {
2133 error_setg(errp, "Failed to pause source migration");
2134 }
2135 return;
2136 }
2137
2138 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2139 ret = qemu_file_shutdown(mis->from_src_file);
2140 if (ret) {
2141 error_setg(errp, "Failed to pause destination migration");
2142 }
2143 return;
2144 }
2145
2146 error_setg(errp, "migrate-pause is currently only supported "
2147 "during postcopy-active state");
2148 }
2149
2150 bool migration_is_blocked(Error **errp)
2151 {
2152 if (qemu_savevm_state_blocked(errp)) {
2153 return true;
2154 }
2155
2156 if (migration_blockers) {
2157 error_propagate(errp, error_copy(migration_blockers->data));
2158 return true;
2159 }
2160
2161 return false;
2162 }
2163
2164 /* Returns true if continue to migrate, or false if error detected */
2165 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
2166 bool resume, Error **errp)
2167 {
2168 Error *local_err = NULL;
2169
2170 if (resume) {
2171 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2172 error_setg(errp, "Cannot resume if there is no "
2173 "paused migration");
2174 return false;
2175 }
2176
2177 /*
2178 * Postcopy recovery won't work well with release-ram
2179 * capability since release-ram will drop the page buffer as
2180 * long as the page is put into the send buffer. So if there
2181 * is a network failure happened, any page buffers that have
2182 * not yet reached the destination VM but have already been
2183 * sent from the source VM will be lost forever. Let's refuse
2184 * the client from resuming such a postcopy migration.
2185 * Luckily release-ram was designed to only be used when src
2186 * and destination VMs are on the same host, so it should be
2187 * fine.
2188 */
2189 if (migrate_release_ram()) {
2190 error_setg(errp, "Postcopy recovery cannot work "
2191 "when release-ram capability is set");
2192 return false;
2193 }
2194
2195 /* This is a resume, skip init status */
2196 return true;
2197 }
2198
2199 if (migration_is_running(s->state)) {
2200 error_setg(errp, QERR_MIGRATION_ACTIVE);
2201 return false;
2202 }
2203
2204 if (runstate_check(RUN_STATE_INMIGRATE)) {
2205 error_setg(errp, "Guest is waiting for an incoming migration");
2206 return false;
2207 }
2208
2209 if (runstate_check(RUN_STATE_POSTMIGRATE)) {
2210 error_setg(errp, "Can't migrate the vm that was paused due to "
2211 "previous migration");
2212 return false;
2213 }
2214
2215 if (migration_is_blocked(errp)) {
2216 return false;
2217 }
2218
2219 if (blk || blk_inc) {
2220 if (migrate_colo_enabled()) {
2221 error_setg(errp, "No disk migration is required in COLO mode");
2222 return false;
2223 }
2224 if (migrate_use_block() || migrate_use_block_incremental()) {
2225 error_setg(errp, "Command options are incompatible with "
2226 "current migration capabilities");
2227 return false;
2228 }
2229 migrate_set_block_enabled(true, &local_err);
2230 if (local_err) {
2231 error_propagate(errp, local_err);
2232 return false;
2233 }
2234 s->must_remove_block_options = true;
2235 }
2236
2237 if (blk_inc) {
2238 migrate_set_block_incremental(s, true);
2239 }
2240
2241 migrate_init(s);
2242 /*
2243 * set ram_counters memory to zero for a
2244 * new migration
2245 */
2246 memset(&ram_counters, 0, sizeof(ram_counters));
2247
2248 return true;
2249 }
2250
2251 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2252 bool has_inc, bool inc, bool has_detach, bool detach,
2253 bool has_resume, bool resume, Error **errp)
2254 {
2255 Error *local_err = NULL;
2256 MigrationState *s = migrate_get_current();
2257 const char *p = NULL;
2258
2259 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2260 has_resume && resume, errp)) {
2261 /* Error detected, put into errp */
2262 return;
2263 }
2264
2265 if (!(has_resume && resume)) {
2266 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
2267 return;
2268 }
2269 }
2270
2271 if (strstart(uri, "tcp:", &p) ||
2272 strstart(uri, "unix:", NULL) ||
2273 strstart(uri, "vsock:", NULL)) {
2274 socket_start_outgoing_migration(s, p ? p : uri, &local_err);
2275 #ifdef CONFIG_RDMA
2276 } else if (strstart(uri, "rdma:", &p)) {
2277 rdma_start_outgoing_migration(s, p, &local_err);
2278 #endif
2279 } else if (strstart(uri, "exec:", &p)) {
2280 exec_start_outgoing_migration(s, p, &local_err);
2281 } else if (strstart(uri, "fd:", &p)) {
2282 fd_start_outgoing_migration(s, p, &local_err);
2283 } else {
2284 if (!(has_resume && resume)) {
2285 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2286 }
2287 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2288 "a valid migration protocol");
2289 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2290 MIGRATION_STATUS_FAILED);
2291 block_cleanup_parameters(s);
2292 return;
2293 }
2294
2295 if (local_err) {
2296 if (!(has_resume && resume)) {
2297 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2298 }
2299 migrate_fd_error(s, local_err);
2300 error_propagate(errp, local_err);
2301 return;
2302 }
2303 }
2304
2305 void qmp_migrate_cancel(Error **errp)
2306 {
2307 migration_cancel();
2308 }
2309
2310 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2311 {
2312 MigrationState *s = migrate_get_current();
2313 if (s->state != state) {
2314 error_setg(errp, "Migration not in expected state: %s",
2315 MigrationStatus_str(s->state));
2316 return;
2317 }
2318 qemu_sem_post(&s->pause_sem);
2319 }
2320
2321 bool migrate_release_ram(void)
2322 {
2323 MigrationState *s;
2324
2325 s = migrate_get_current();
2326
2327 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2328 }
2329
2330 bool migrate_postcopy_ram(void)
2331 {
2332 MigrationState *s;
2333
2334 s = migrate_get_current();
2335
2336 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2337 }
2338
2339 bool migrate_postcopy(void)
2340 {
2341 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2342 }
2343
2344 bool migrate_auto_converge(void)
2345 {
2346 MigrationState *s;
2347
2348 s = migrate_get_current();
2349
2350 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2351 }
2352
2353 bool migrate_zero_blocks(void)
2354 {
2355 MigrationState *s;
2356
2357 s = migrate_get_current();
2358
2359 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2360 }
2361
2362 bool migrate_postcopy_blocktime(void)
2363 {
2364 MigrationState *s;
2365
2366 s = migrate_get_current();
2367
2368 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2369 }
2370
2371 bool migrate_use_compression(void)
2372 {
2373 MigrationState *s;
2374
2375 s = migrate_get_current();
2376
2377 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2378 }
2379
2380 int migrate_compress_level(void)
2381 {
2382 MigrationState *s;
2383
2384 s = migrate_get_current();
2385
2386 return s->parameters.compress_level;
2387 }
2388
2389 int migrate_compress_threads(void)
2390 {
2391 MigrationState *s;
2392
2393 s = migrate_get_current();
2394
2395 return s->parameters.compress_threads;
2396 }
2397
2398 int migrate_compress_wait_thread(void)
2399 {
2400 MigrationState *s;
2401
2402 s = migrate_get_current();
2403
2404 return s->parameters.compress_wait_thread;
2405 }
2406
2407 int migrate_decompress_threads(void)
2408 {
2409 MigrationState *s;
2410
2411 s = migrate_get_current();
2412
2413 return s->parameters.decompress_threads;
2414 }
2415
2416 bool migrate_dirty_bitmaps(void)
2417 {
2418 MigrationState *s;
2419
2420 s = migrate_get_current();
2421
2422 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2423 }
2424
2425 bool migrate_ignore_shared(void)
2426 {
2427 MigrationState *s;
2428
2429 s = migrate_get_current();
2430
2431 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2432 }
2433
2434 bool migrate_validate_uuid(void)
2435 {
2436 MigrationState *s;
2437
2438 s = migrate_get_current();
2439
2440 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2441 }
2442
2443 bool migrate_use_events(void)
2444 {
2445 MigrationState *s;
2446
2447 s = migrate_get_current();
2448
2449 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2450 }
2451
2452 bool migrate_use_multifd(void)
2453 {
2454 MigrationState *s;
2455
2456 s = migrate_get_current();
2457
2458 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2459 }
2460
2461 bool migrate_pause_before_switchover(void)
2462 {
2463 MigrationState *s;
2464
2465 s = migrate_get_current();
2466
2467 return s->enabled_capabilities[
2468 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2469 }
2470
2471 int migrate_multifd_channels(void)
2472 {
2473 MigrationState *s;
2474
2475 s = migrate_get_current();
2476
2477 return s->parameters.multifd_channels;
2478 }
2479
2480 MultiFDCompression migrate_multifd_compression(void)
2481 {
2482 MigrationState *s;
2483
2484 s = migrate_get_current();
2485
2486 return s->parameters.multifd_compression;
2487 }
2488
2489 int migrate_multifd_zlib_level(void)
2490 {
2491 MigrationState *s;
2492
2493 s = migrate_get_current();
2494
2495 return s->parameters.multifd_zlib_level;
2496 }
2497
2498 int migrate_multifd_zstd_level(void)
2499 {
2500 MigrationState *s;
2501
2502 s = migrate_get_current();
2503
2504 return s->parameters.multifd_zstd_level;
2505 }
2506
2507 int migrate_use_xbzrle(void)
2508 {
2509 MigrationState *s;
2510
2511 s = migrate_get_current();
2512
2513 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2514 }
2515
2516 uint64_t migrate_xbzrle_cache_size(void)
2517 {
2518 MigrationState *s;
2519
2520 s = migrate_get_current();
2521
2522 return s->parameters.xbzrle_cache_size;
2523 }
2524
2525 static int64_t migrate_max_postcopy_bandwidth(void)
2526 {
2527 MigrationState *s;
2528
2529 s = migrate_get_current();
2530
2531 return s->parameters.max_postcopy_bandwidth;
2532 }
2533
2534 bool migrate_use_block(void)
2535 {
2536 MigrationState *s;
2537
2538 s = migrate_get_current();
2539
2540 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2541 }
2542
2543 bool migrate_use_return_path(void)
2544 {
2545 MigrationState *s;
2546
2547 s = migrate_get_current();
2548
2549 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2550 }
2551
2552 bool migrate_use_block_incremental(void)
2553 {
2554 MigrationState *s;
2555
2556 s = migrate_get_current();
2557
2558 return s->parameters.block_incremental;
2559 }
2560
2561 bool migrate_background_snapshot(void)
2562 {
2563 MigrationState *s;
2564
2565 s = migrate_get_current();
2566
2567 return s->enabled_capabilities[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT];
2568 }
2569
2570 /* migration thread support */
2571 /*
2572 * Something bad happened to the RP stream, mark an error
2573 * The caller shall print or trace something to indicate why
2574 */
2575 static void mark_source_rp_bad(MigrationState *s)
2576 {
2577 s->rp_state.error = true;
2578 }
2579
2580 static struct rp_cmd_args {
2581 ssize_t len; /* -1 = variable */
2582 const char *name;
2583 } rp_cmd_args[] = {
2584 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2585 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2586 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2587 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2588 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2589 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2590 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2591 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2592 };
2593
2594 /*
2595 * Process a request for pages received on the return path,
2596 * We're allowed to send more than requested (e.g. to round to our page size)
2597 * and we don't need to send pages that have already been sent.
2598 */
2599 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2600 ram_addr_t start, size_t len)
2601 {
2602 long our_host_ps = qemu_real_host_page_size;
2603
2604 trace_migrate_handle_rp_req_pages(rbname, start, len);
2605
2606 /*
2607 * Since we currently insist on matching page sizes, just sanity check
2608 * we're being asked for whole host pages.
2609 */
2610 if (start & (our_host_ps - 1) ||
2611 (len & (our_host_ps - 1))) {
2612 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2613 " len: %zd", __func__, start, len);
2614 mark_source_rp_bad(ms);
2615 return;
2616 }
2617
2618 if (ram_save_queue_pages(rbname, start, len)) {
2619 mark_source_rp_bad(ms);
2620 }
2621 }
2622
2623 /* Return true to retry, false to quit */
2624 static bool postcopy_pause_return_path_thread(MigrationState *s)
2625 {
2626 trace_postcopy_pause_return_path();
2627
2628 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2629
2630 trace_postcopy_pause_return_path_continued();
2631
2632 return true;
2633 }
2634
2635 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2636 {
2637 RAMBlock *block = qemu_ram_block_by_name(block_name);
2638
2639 if (!block) {
2640 error_report("%s: invalid block name '%s'", __func__, block_name);
2641 return -EINVAL;
2642 }
2643
2644 /* Fetch the received bitmap and refresh the dirty bitmap */
2645 return ram_dirty_bitmap_reload(s, block);
2646 }
2647
2648 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2649 {
2650 trace_source_return_path_thread_resume_ack(value);
2651
2652 if (value != MIGRATION_RESUME_ACK_VALUE) {
2653 error_report("%s: illegal resume_ack value %"PRIu32,
2654 __func__, value);
2655 return -1;
2656 }
2657
2658 /* Now both sides are active. */
2659 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2660 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2661
2662 /* Notify send thread that time to continue send pages */
2663 qemu_sem_post(&s->rp_state.rp_sem);
2664
2665 return 0;
2666 }
2667
2668 /*
2669 * Handles messages sent on the return path towards the source VM
2670 *
2671 */
2672 static void *source_return_path_thread(void *opaque)
2673 {
2674 MigrationState *ms = opaque;
2675 QEMUFile *rp = ms->rp_state.from_dst_file;
2676 uint16_t header_len, header_type;
2677 uint8_t buf[512];
2678 uint32_t tmp32, sibling_error;
2679 ram_addr_t start = 0; /* =0 to silence warning */
2680 size_t len = 0, expected_len;
2681 int res;
2682
2683 trace_source_return_path_thread_entry();
2684 rcu_register_thread();
2685
2686 retry:
2687 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2688 migration_is_setup_or_active(ms->state)) {
2689 trace_source_return_path_thread_loop_top();
2690 header_type = qemu_get_be16(rp);
2691 header_len = qemu_get_be16(rp);
2692
2693 if (qemu_file_get_error(rp)) {
2694 mark_source_rp_bad(ms);
2695 goto out;
2696 }
2697
2698 if (header_type >= MIG_RP_MSG_MAX ||
2699 header_type == MIG_RP_MSG_INVALID) {
2700 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2701 header_type, header_len);
2702 mark_source_rp_bad(ms);
2703 goto out;
2704 }
2705
2706 if ((rp_cmd_args[header_type].len != -1 &&
2707 header_len != rp_cmd_args[header_type].len) ||
2708 header_len > sizeof(buf)) {
2709 error_report("RP: Received '%s' message (0x%04x) with"
2710 "incorrect length %d expecting %zu",
2711 rp_cmd_args[header_type].name, header_type, header_len,
2712 (size_t)rp_cmd_args[header_type].len);
2713 mark_source_rp_bad(ms);
2714 goto out;
2715 }
2716
2717 /* We know we've got a valid header by this point */
2718 res = qemu_get_buffer(rp, buf, header_len);
2719 if (res != header_len) {
2720 error_report("RP: Failed reading data for message 0x%04x"
2721 " read %d expected %d",
2722 header_type, res, header_len);
2723 mark_source_rp_bad(ms);
2724 goto out;
2725 }
2726
2727 /* OK, we have the message and the data */
2728 switch (header_type) {
2729 case MIG_RP_MSG_SHUT:
2730 sibling_error = ldl_be_p(buf);
2731 trace_source_return_path_thread_shut(sibling_error);
2732 if (sibling_error) {
2733 error_report("RP: Sibling indicated error %d", sibling_error);
2734 mark_source_rp_bad(ms);
2735 }
2736 /*
2737 * We'll let the main thread deal with closing the RP
2738 * we could do a shutdown(2) on it, but we're the only user
2739 * anyway, so there's nothing gained.
2740 */
2741 goto out;
2742
2743 case MIG_RP_MSG_PONG:
2744 tmp32 = ldl_be_p(buf);
2745 trace_source_return_path_thread_pong(tmp32);
2746 break;
2747
2748 case MIG_RP_MSG_REQ_PAGES:
2749 start = ldq_be_p(buf);
2750 len = ldl_be_p(buf + 8);
2751 migrate_handle_rp_req_pages(ms, NULL, start, len);
2752 break;
2753
2754 case MIG_RP_MSG_REQ_PAGES_ID:
2755 expected_len = 12 + 1; /* header + termination */
2756
2757 if (header_len >= expected_len) {
2758 start = ldq_be_p(buf);
2759 len = ldl_be_p(buf + 8);
2760 /* Now we expect an idstr */
2761 tmp32 = buf[12]; /* Length of the following idstr */
2762 buf[13 + tmp32] = '\0';
2763 expected_len += tmp32;
2764 }
2765 if (header_len != expected_len) {
2766 error_report("RP: Req_Page_id with length %d expecting %zd",
2767 header_len, expected_len);
2768 mark_source_rp_bad(ms);
2769 goto out;
2770 }
2771 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2772 break;
2773
2774 case MIG_RP_MSG_RECV_BITMAP:
2775 if (header_len < 1) {
2776 error_report("%s: missing block name", __func__);
2777 mark_source_rp_bad(ms);
2778 goto out;
2779 }
2780 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2781 buf[buf[0] + 1] = '\0';
2782 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2783 mark_source_rp_bad(ms);
2784 goto out;
2785 }
2786 break;
2787
2788 case MIG_RP_MSG_RESUME_ACK:
2789 tmp32 = ldl_be_p(buf);
2790 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2791 mark_source_rp_bad(ms);
2792 goto out;
2793 }
2794 break;
2795
2796 default:
2797 break;
2798 }
2799 }
2800
2801 out:
2802 res = qemu_file_get_error(rp);
2803 if (res) {
2804 if (res == -EIO && migration_in_postcopy()) {
2805 /*
2806 * Maybe there is something we can do: it looks like a
2807 * network down issue, and we pause for a recovery.
2808 */
2809 if (postcopy_pause_return_path_thread(ms)) {
2810 /* Reload rp, reset the rest */
2811 if (rp != ms->rp_state.from_dst_file) {
2812 qemu_fclose(rp);
2813 rp = ms->rp_state.from_dst_file;
2814 }
2815 ms->rp_state.error = false;
2816 goto retry;
2817 }
2818 }
2819
2820 trace_source_return_path_thread_bad_end();
2821 mark_source_rp_bad(ms);
2822 }
2823
2824 trace_source_return_path_thread_end();
2825 ms->rp_state.from_dst_file = NULL;
2826 qemu_fclose(rp);
2827 rcu_unregister_thread();
2828 return NULL;
2829 }
2830
2831 static int open_return_path_on_source(MigrationState *ms,
2832 bool create_thread)
2833 {
2834
2835 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2836 if (!ms->rp_state.from_dst_file) {
2837 return -1;
2838 }
2839
2840 trace_open_return_path_on_source();
2841
2842 if (!create_thread) {
2843 /* We're done */
2844 return 0;
2845 }
2846
2847 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2848 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2849
2850 trace_open_return_path_on_source_continue();
2851
2852 return 0;
2853 }
2854
2855 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2856 static int await_return_path_close_on_source(MigrationState *ms)
2857 {
2858 /*
2859 * If this is a normal exit then the destination will send a SHUT and the
2860 * rp_thread will exit, however if there's an error we need to cause
2861 * it to exit.
2862 */
2863 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2864 /*
2865 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2866 * waiting for the destination.
2867 */
2868 qemu_file_shutdown(ms->rp_state.from_dst_file);
2869 mark_source_rp_bad(ms);
2870 }
2871 trace_await_return_path_close_on_source_joining();
2872 qemu_thread_join(&ms->rp_state.rp_thread);
2873 trace_await_return_path_close_on_source_close();
2874 return ms->rp_state.error;
2875 }
2876
2877 /*
2878 * Switch from normal iteration to postcopy
2879 * Returns non-0 on error
2880 */
2881 static int postcopy_start(MigrationState *ms)
2882 {
2883 int ret;
2884 QIOChannelBuffer *bioc;
2885 QEMUFile *fb;
2886 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2887 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2888 bool restart_block = false;
2889 int cur_state = MIGRATION_STATUS_ACTIVE;
2890 if (!migrate_pause_before_switchover()) {
2891 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2892 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2893 }
2894
2895 trace_postcopy_start();
2896 qemu_mutex_lock_iothread();
2897 trace_postcopy_start_set_run();
2898
2899 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2900 global_state_store();
2901 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2902 if (ret < 0) {
2903 goto fail;
2904 }
2905
2906 ret = migration_maybe_pause(ms, &cur_state,
2907 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2908 if (ret < 0) {
2909 goto fail;
2910 }
2911
2912 ret = bdrv_inactivate_all();
2913 if (ret < 0) {
2914 goto fail;
2915 }
2916 restart_block = true;
2917
2918 /*
2919 * Cause any non-postcopiable, but iterative devices to
2920 * send out their final data.
2921 */
2922 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2923
2924 /*
2925 * in Finish migrate and with the io-lock held everything should
2926 * be quiet, but we've potentially still got dirty pages and we
2927 * need to tell the destination to throw any pages it's already received
2928 * that are dirty
2929 */
2930 if (migrate_postcopy_ram()) {
2931 if (ram_postcopy_send_discard_bitmap(ms)) {
2932 error_report("postcopy send discard bitmap failed");
2933 goto fail;
2934 }
2935 }
2936
2937 /*
2938 * send rest of state - note things that are doing postcopy
2939 * will notice we're in POSTCOPY_ACTIVE and not actually
2940 * wrap their state up here
2941 */
2942 /* 0 max-postcopy-bandwidth means unlimited */
2943 if (!bandwidth) {
2944 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2945 } else {
2946 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2947 }
2948 if (migrate_postcopy_ram()) {
2949 /* Ping just for debugging, helps line traces up */
2950 qemu_savevm_send_ping(ms->to_dst_file, 2);
2951 }
2952
2953 /*
2954 * While loading the device state we may trigger page transfer
2955 * requests and the fd must be free to process those, and thus
2956 * the destination must read the whole device state off the fd before
2957 * it starts processing it. Unfortunately the ad-hoc migration format
2958 * doesn't allow the destination to know the size to read without fully
2959 * parsing it through each devices load-state code (especially the open
2960 * coded devices that use get/put).
2961 * So we wrap the device state up in a package with a length at the start;
2962 * to do this we use a qemu_buf to hold the whole of the device state.
2963 */
2964 bioc = qio_channel_buffer_new(4096);
2965 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2966 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2967 object_unref(OBJECT(bioc));
2968
2969 /*
2970 * Make sure the receiver can get incoming pages before we send the rest
2971 * of the state
2972 */
2973 qemu_savevm_send_postcopy_listen(fb);
2974
2975 qemu_savevm_state_complete_precopy(fb, false, false);
2976 if (migrate_postcopy_ram()) {
2977 qemu_savevm_send_ping(fb, 3);
2978 }
2979
2980 qemu_savevm_send_postcopy_run(fb);
2981
2982 /* <><> end of stuff going into the package */
2983
2984 /* Last point of recovery; as soon as we send the package the destination
2985 * can open devices and potentially start running.
2986 * Lets just check again we've not got any errors.
2987 */
2988 ret = qemu_file_get_error(ms->to_dst_file);
2989 if (ret) {
2990 error_report("postcopy_start: Migration stream errored (pre package)");
2991 goto fail_closefb;
2992 }
2993
2994 restart_block = false;
2995
2996 /* Now send that blob */
2997 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2998 goto fail_closefb;
2999 }
3000 qemu_fclose(fb);
3001
3002 /* Send a notify to give a chance for anything that needs to happen
3003 * at the transition to postcopy and after the device state; in particular
3004 * spice needs to trigger a transition now
3005 */
3006 ms->postcopy_after_devices = true;
3007 notifier_list_notify(&migration_state_notifiers, ms);
3008
3009 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
3010
3011 qemu_mutex_unlock_iothread();
3012
3013 if (migrate_postcopy_ram()) {
3014 /*
3015 * Although this ping is just for debug, it could potentially be
3016 * used for getting a better measurement of downtime at the source.
3017 */
3018 qemu_savevm_send_ping(ms->to_dst_file, 4);
3019 }
3020
3021 if (migrate_release_ram()) {
3022 ram_postcopy_migrated_memory_release(ms);
3023 }
3024
3025 ret = qemu_file_get_error(ms->to_dst_file);
3026 if (ret) {
3027 error_report("postcopy_start: Migration stream errored");
3028 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
3029 MIGRATION_STATUS_FAILED);
3030 }
3031
3032 return ret;
3033
3034 fail_closefb:
3035 qemu_fclose(fb);
3036 fail:
3037 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
3038 MIGRATION_STATUS_FAILED);
3039 if (restart_block) {
3040 /* A failure happened early enough that we know the destination hasn't
3041 * accessed block devices, so we're safe to recover.
3042 */
3043 Error *local_err = NULL;
3044
3045 bdrv_invalidate_cache_all(&local_err);
3046 if (local_err) {
3047 error_report_err(local_err);
3048 }
3049 }
3050 qemu_mutex_unlock_iothread();
3051 return -1;
3052 }
3053
3054 /**
3055 * migration_maybe_pause: Pause if required to by
3056 * migrate_pause_before_switchover called with the iothread locked
3057 * Returns: 0 on success
3058 */
3059 static int migration_maybe_pause(MigrationState *s,
3060 int *current_active_state,
3061 int new_state)
3062 {
3063 if (!migrate_pause_before_switchover()) {
3064 return 0;
3065 }
3066
3067 /* Since leaving this state is not atomic with posting the semaphore
3068 * it's possible that someone could have issued multiple migrate_continue
3069 * and the semaphore is incorrectly positive at this point;
3070 * the docs say it's undefined to reinit a semaphore that's already
3071 * init'd, so use timedwait to eat up any existing posts.
3072 */
3073 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
3074 /* This block intentionally left blank */
3075 }
3076
3077 /*
3078 * If the migration is cancelled when it is in the completion phase,
3079 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3080 * So we don't need to wait a semaphore, otherwise we would always
3081 * wait for the 'pause_sem' semaphore.
3082 */
3083 if (s->state != MIGRATION_STATUS_CANCELLING) {
3084 qemu_mutex_unlock_iothread();
3085 migrate_set_state(&s->state, *current_active_state,
3086 MIGRATION_STATUS_PRE_SWITCHOVER);
3087 qemu_sem_wait(&s->pause_sem);
3088 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
3089 new_state);
3090 *current_active_state = new_state;
3091 qemu_mutex_lock_iothread();
3092 }
3093
3094 return s->state == new_state ? 0 : -EINVAL;
3095 }
3096
3097 /**
3098 * migration_completion: Used by migration_thread when there's not much left.
3099 * The caller 'breaks' the loop when this returns.
3100 *
3101 * @s: Current migration state
3102 */
3103 static void migration_completion(MigrationState *s)
3104 {
3105 int ret;
3106 int current_active_state = s->state;
3107
3108 if (s->state == MIGRATION_STATUS_ACTIVE) {
3109 qemu_mutex_lock_iothread();
3110 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3111 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3112 s->vm_was_running = runstate_is_running();
3113 ret = global_state_store();
3114
3115 if (!ret) {
3116 bool inactivate = !migrate_colo_enabled();
3117 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
3118 trace_migration_completion_vm_stop(ret);
3119 if (ret >= 0) {
3120 ret = migration_maybe_pause(s, &current_active_state,
3121 MIGRATION_STATUS_DEVICE);
3122 }
3123 if (ret >= 0) {
3124 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3125 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
3126 inactivate);
3127 }
3128 if (inactivate && ret >= 0) {
3129 s->block_inactive = true;
3130 }
3131 }
3132 qemu_mutex_unlock_iothread();
3133
3134 if (ret < 0) {
3135 goto fail;
3136 }
3137 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3138 trace_migration_completion_postcopy_end();
3139
3140 qemu_savevm_state_complete_postcopy(s->to_dst_file);
3141 trace_migration_completion_postcopy_end_after_complete();
3142 } else if (s->state == MIGRATION_STATUS_CANCELLING) {
3143 goto fail;
3144 }
3145
3146 /*
3147 * If rp was opened we must clean up the thread before
3148 * cleaning everything else up (since if there are no failures
3149 * it will wait for the destination to send it's status in
3150 * a SHUT command).
3151 */
3152 if (s->rp_state.from_dst_file) {
3153 int rp_error;
3154 trace_migration_return_path_end_before();
3155 rp_error = await_return_path_close_on_source(s);
3156 trace_migration_return_path_end_after(rp_error);
3157 if (rp_error) {
3158 goto fail_invalidate;
3159 }
3160 }
3161
3162 if (qemu_file_get_error(s->to_dst_file)) {
3163 trace_migration_completion_file_err();
3164 goto fail_invalidate;
3165 }
3166
3167 if (!migrate_colo_enabled()) {
3168 migrate_set_state(&s->state, current_active_state,
3169 MIGRATION_STATUS_COMPLETED);
3170 }
3171
3172 return;
3173
3174 fail_invalidate:
3175 /* If not doing postcopy, vm_start() will be called: let's regain
3176 * control on images.
3177 */
3178 if (s->state == MIGRATION_STATUS_ACTIVE ||
3179 s->state == MIGRATION_STATUS_DEVICE) {
3180 Error *local_err = NULL;
3181
3182 qemu_mutex_lock_iothread();
3183 bdrv_invalidate_cache_all(&local_err);
3184 if (local_err) {
3185 error_report_err(local_err);
3186 } else {
3187 s->block_inactive = false;
3188 }
3189 qemu_mutex_unlock_iothread();
3190 }
3191
3192 fail:
3193 migrate_set_state(&s->state, current_active_state,
3194 MIGRATION_STATUS_FAILED);
3195 }
3196
3197 /**
3198 * bg_migration_completion: Used by bg_migration_thread when after all the
3199 * RAM has been saved. The caller 'breaks' the loop when this returns.
3200 *
3201 * @s: Current migration state
3202 */
3203 static void bg_migration_completion(MigrationState *s)
3204 {
3205 int current_active_state = s->state;
3206
3207 /*
3208 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3209 * memory ranges, flush kernel wait queues and wake up threads
3210 * waiting for write fault to be resolved.
3211 */
3212 ram_write_tracking_stop();
3213
3214 if (s->state == MIGRATION_STATUS_ACTIVE) {
3215 /*
3216 * By this moment we have RAM content saved into the migration stream.
3217 * The next step is to flush the non-RAM content (device state)
3218 * right after the ram content. The device state has been stored into
3219 * the temporary buffer before RAM saving started.
3220 */
3221 qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage);
3222 qemu_fflush(s->to_dst_file);
3223 } else if (s->state == MIGRATION_STATUS_CANCELLING) {
3224 goto fail;
3225 }
3226
3227 if (qemu_file_get_error(s->to_dst_file)) {
3228 trace_migration_completion_file_err();
3229 goto fail;
3230 }
3231
3232 migrate_set_state(&s->state, current_active_state,
3233 MIGRATION_STATUS_COMPLETED);
3234 return;
3235
3236 fail:
3237 migrate_set_state(&s->state, current_active_state,
3238 MIGRATION_STATUS_FAILED);
3239 }
3240
3241 bool migrate_colo_enabled(void)
3242 {
3243 MigrationState *s = migrate_get_current();
3244 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
3245 }
3246
3247 typedef enum MigThrError {
3248 /* No error detected */
3249 MIG_THR_ERR_NONE = 0,
3250 /* Detected error, but resumed successfully */
3251 MIG_THR_ERR_RECOVERED = 1,
3252 /* Detected fatal error, need to exit */
3253 MIG_THR_ERR_FATAL = 2,
3254 } MigThrError;
3255
3256 static int postcopy_resume_handshake(MigrationState *s)
3257 {
3258 qemu_savevm_send_postcopy_resume(s->to_dst_file);
3259
3260 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3261 qemu_sem_wait(&s->rp_state.rp_sem);
3262 }
3263
3264 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3265 return 0;
3266 }
3267
3268 return -1;
3269 }
3270
3271 /* Return zero if success, or <0 for error */
3272 static int postcopy_do_resume(MigrationState *s)
3273 {
3274 int ret;
3275
3276 /*
3277 * Call all the resume_prepare() hooks, so that modules can be
3278 * ready for the migration resume.
3279 */
3280 ret = qemu_savevm_state_resume_prepare(s);
3281 if (ret) {
3282 error_report("%s: resume_prepare() failure detected: %d",
3283 __func__, ret);
3284 return ret;
3285 }
3286
3287 /*
3288 * Last handshake with destination on the resume (destination will
3289 * switch to postcopy-active afterwards)
3290 */
3291 ret = postcopy_resume_handshake(s);
3292 if (ret) {
3293 error_report("%s: handshake failed: %d", __func__, ret);
3294 return ret;
3295 }
3296
3297 return 0;
3298 }
3299
3300 /*
3301 * We don't return until we are in a safe state to continue current
3302 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3303 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3304 */
3305 static MigThrError postcopy_pause(MigrationState *s)
3306 {
3307 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3308
3309 while (true) {
3310 QEMUFile *file;
3311
3312 /* Current channel is possibly broken. Release it. */
3313 assert(s->to_dst_file);
3314 qemu_mutex_lock(&s->qemu_file_lock);
3315 file = s->to_dst_file;
3316 s->to_dst_file = NULL;
3317 qemu_mutex_unlock(&s->qemu_file_lock);
3318
3319 qemu_file_shutdown(file);
3320 qemu_fclose(file);
3321
3322 migrate_set_state(&s->state, s->state,
3323 MIGRATION_STATUS_POSTCOPY_PAUSED);
3324
3325 error_report("Detected IO failure for postcopy. "
3326 "Migration paused.");
3327
3328 /*
3329 * We wait until things fixed up. Then someone will setup the
3330 * status back for us.
3331 */
3332 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3333 qemu_sem_wait(&s->postcopy_pause_sem);
3334 }
3335
3336 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3337 /* Woken up by a recover procedure. Give it a shot */
3338
3339 /*
3340 * Firstly, let's wake up the return path now, with a new
3341 * return path channel.
3342 */
3343 qemu_sem_post(&s->postcopy_pause_rp_sem);
3344
3345 /* Do the resume logic */
3346 if (postcopy_do_resume(s) == 0) {
3347 /* Let's continue! */
3348 trace_postcopy_pause_continued();
3349 return MIG_THR_ERR_RECOVERED;
3350 } else {
3351 /*
3352 * Something wrong happened during the recovery, let's
3353 * pause again. Pause is always better than throwing
3354 * data away.
3355 */
3356 continue;
3357 }
3358 } else {
3359 /* This is not right... Time to quit. */
3360 return MIG_THR_ERR_FATAL;
3361 }
3362 }
3363 }
3364
3365 static MigThrError migration_detect_error(MigrationState *s)
3366 {
3367 int ret;
3368 int state = s->state;
3369 Error *local_error = NULL;
3370
3371 if (state == MIGRATION_STATUS_CANCELLING ||
3372 state == MIGRATION_STATUS_CANCELLED) {
3373 /* End the migration, but don't set the state to failed */
3374 return MIG_THR_ERR_FATAL;
3375 }
3376
3377 /* Try to detect any file errors */
3378 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3379 if (!ret) {
3380 /* Everything is fine */
3381 assert(!local_error);
3382 return MIG_THR_ERR_NONE;
3383 }
3384
3385 if (local_error) {
3386 migrate_set_error(s, local_error);
3387 error_free(local_error);
3388 }
3389
3390 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3391 /*
3392 * For postcopy, we allow the network to be down for a
3393 * while. After that, it can be continued by a
3394 * recovery phase.
3395 */
3396 return postcopy_pause(s);
3397 } else {
3398 /*
3399 * For precopy (or postcopy with error outside IO), we fail
3400 * with no time.
3401 */
3402 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3403 trace_migration_thread_file_err();
3404
3405 /* Time to stop the migration, now. */
3406 return MIG_THR_ERR_FATAL;
3407 }
3408 }
3409
3410 /* How many bytes have we transferred since the beginning of the migration */
3411 static uint64_t migration_total_bytes(MigrationState *s)
3412 {
3413 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3414 }
3415
3416 static void migration_calculate_complete(MigrationState *s)
3417 {
3418 uint64_t bytes = migration_total_bytes(s);
3419 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3420 int64_t transfer_time;
3421
3422 s->total_time = end_time - s->start_time;
3423 if (!s->downtime) {
3424 /*
3425 * It's still not set, so we are precopy migration. For
3426 * postcopy, downtime is calculated during postcopy_start().
3427 */
3428 s->downtime = end_time - s->downtime_start;
3429 }
3430
3431 transfer_time = s->total_time - s->setup_time;
3432 if (transfer_time) {
3433 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3434 }
3435 }
3436
3437 static void update_iteration_initial_status(MigrationState *s)
3438 {
3439 /*
3440 * Update these three fields at the same time to avoid mismatch info lead
3441 * wrong speed calculation.
3442 */
3443 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3444 s->iteration_initial_bytes = migration_total_bytes(s);
3445 s->iteration_initial_pages = ram_get_total_transferred_pages();
3446 }
3447
3448 static void migration_update_counters(MigrationState *s,
3449 int64_t current_time)
3450 {
3451 uint64_t transferred, transferred_pages, time_spent;
3452 uint64_t current_bytes; /* bytes transferred since the beginning */
3453 double bandwidth;
3454
3455 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3456 return;
3457 }
3458
3459 current_bytes = migration_total_bytes(s);
3460 transferred = current_bytes - s->iteration_initial_bytes;
3461 time_spent = current_time - s->iteration_start_time;
3462 bandwidth = (double)transferred / time_spent;
3463 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3464
3465 s->mbps = (((double) transferred * 8.0) /
3466 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3467
3468 transferred_pages = ram_get_total_transferred_pages() -
3469 s->iteration_initial_pages;
3470 s->pages_per_second = (double) transferred_pages /
3471 (((double) time_spent / 1000.0));
3472
3473 /*
3474 * if we haven't sent anything, we don't want to
3475 * recalculate. 10000 is a small enough number for our purposes
3476 */
3477 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3478 s->expected_downtime = ram_counters.remaining / bandwidth;
3479 }
3480
3481 qemu_file_reset_rate_limit(s->to_dst_file);
3482
3483 update_iteration_initial_status(s);
3484
3485 trace_migrate_transferred(transferred, time_spent,
3486 bandwidth, s->threshold_size);
3487 }
3488
3489 /* Migration thread iteration status */
3490 typedef enum {
3491 MIG_ITERATE_RESUME, /* Resume current iteration */
3492 MIG_ITERATE_SKIP, /* Skip current iteration */
3493 MIG_ITERATE_BREAK, /* Break the loop */
3494 } MigIterateState;
3495
3496 /*
3497 * Return true if continue to the next iteration directly, false
3498 * otherwise.
3499 */
3500 static MigIterateState migration_iteration_run(MigrationState *s)
3501 {
3502 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3503 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3504
3505 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3506 &pend_compat, &pend_post);
3507 pending_size = pend_pre + pend_compat + pend_post;
3508
3509 trace_migrate_pending(pending_size, s->threshold_size,
3510 pend_pre, pend_compat, pend_post);
3511
3512 if (pending_size && pending_size >= s->threshold_size) {
3513 /* Still a significant amount to transfer */
3514 if (!in_postcopy && pend_pre <= s->threshold_size &&
3515 qatomic_read(&s->start_postcopy)) {
3516 if (postcopy_start(s)) {
3517 error_report("%s: postcopy failed to start", __func__);
3518 }
3519 return MIG_ITERATE_SKIP;
3520 }
3521 /* Just another iteration step */
3522 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3523 } else {
3524 trace_migration_thread_low_pending(pending_size);
3525 migration_completion(s);
3526 return MIG_ITERATE_BREAK;
3527 }
3528
3529 return MIG_ITERATE_RESUME;
3530 }
3531
3532 static void migration_iteration_finish(MigrationState *s)
3533 {
3534 /* If we enabled cpu throttling for auto-converge, turn it off. */
3535 cpu_throttle_stop();
3536
3537 qemu_mutex_lock_iothread();
3538 switch (s->state) {
3539 case MIGRATION_STATUS_COMPLETED:
3540 migration_calculate_complete(s);
3541 runstate_set(RUN_STATE_POSTMIGRATE);
3542 break;
3543
3544 case MIGRATION_STATUS_ACTIVE:
3545 /*
3546 * We should really assert here, but since it's during
3547 * migration, let's try to reduce the usage of assertions.
3548 */
3549 if (!migrate_colo_enabled()) {
3550 error_report("%s: critical error: calling COLO code without "
3551 "COLO enabled", __func__);
3552 }
3553 migrate_start_colo_process(s);
3554 /*
3555 * Fixme: we will run VM in COLO no matter its old running state.
3556 * After exited COLO, we will keep running.
3557 */
3558 s->vm_was_running = true;
3559 /* Fallthrough */
3560 case MIGRATION_STATUS_FAILED:
3561 case MIGRATION_STATUS_CANCELLED:
3562 case MIGRATION_STATUS_CANCELLING:
3563 if (s->vm_was_running) {
3564 vm_start();
3565 } else {
3566 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3567 runstate_set(RUN_STATE_POSTMIGRATE);
3568 }
3569 }
3570 break;
3571
3572 default:
3573 /* Should not reach here, but if so, forgive the VM. */
3574 error_report("%s: Unknown ending state %d", __func__, s->state);
3575 break;
3576 }
3577 migrate_fd_cleanup_schedule(s);
3578 qemu_mutex_unlock_iothread();
3579 }
3580
3581 static void bg_migration_iteration_finish(MigrationState *s)
3582 {
3583 qemu_mutex_lock_iothread();
3584 switch (s->state) {
3585 case MIGRATION_STATUS_COMPLETED:
3586 migration_calculate_complete(s);
3587 break;
3588
3589 case MIGRATION_STATUS_ACTIVE:
3590 case MIGRATION_STATUS_FAILED:
3591 case MIGRATION_STATUS_CANCELLED:
3592 case MIGRATION_STATUS_CANCELLING:
3593 break;
3594
3595 default:
3596 /* Should not reach here, but if so, forgive the VM. */
3597 error_report("%s: Unknown ending state %d", __func__, s->state);
3598 break;
3599 }
3600
3601 migrate_fd_cleanup_schedule(s);
3602 qemu_mutex_unlock_iothread();
3603 }
3604
3605 /*
3606 * Return true if continue to the next iteration directly, false
3607 * otherwise.
3608 */
3609 static MigIterateState bg_migration_iteration_run(MigrationState *s)
3610 {
3611 int res;
3612
3613 res = qemu_savevm_state_iterate(s->to_dst_file, false);
3614 if (res > 0) {
3615 bg_migration_completion(s);
3616 return MIG_ITERATE_BREAK;
3617 }
3618
3619 return MIG_ITERATE_RESUME;
3620 }
3621
3622 void migration_make_urgent_request(void)
3623 {
3624 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3625 }
3626
3627 void migration_consume_urgent_request(void)
3628 {
3629 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3630 }
3631
3632 /* Returns true if the rate limiting was broken by an urgent request */
3633 bool migration_rate_limit(void)
3634 {
3635 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3636 MigrationState *s = migrate_get_current();
3637
3638 bool urgent = false;
3639 migration_update_counters(s, now);
3640 if (qemu_file_rate_limit(s->to_dst_file)) {
3641
3642 if (qemu_file_get_error(s->to_dst_file)) {
3643 return false;
3644 }
3645 /*
3646 * Wait for a delay to do rate limiting OR
3647 * something urgent to post the semaphore.
3648 */
3649 int ms = s->iteration_start_time + BUFFER_DELAY - now;
3650 trace_migration_rate_limit_pre(ms);
3651 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3652 /*
3653 * We were woken by one or more urgent things but
3654 * the timedwait will have consumed one of them.
3655 * The service routine for the urgent wake will dec
3656 * the semaphore itself for each item it consumes,
3657 * so add this one we just eat back.
3658 */
3659 qemu_sem_post(&s->rate_limit_sem);
3660 urgent = true;
3661 }
3662 trace_migration_rate_limit_post(urgent);
3663 }
3664 return urgent;
3665 }
3666
3667 /*
3668 * Master migration thread on the source VM.
3669 * It drives the migration and pumps the data down the outgoing channel.
3670 */
3671 static void *migration_thread(void *opaque)
3672 {
3673 MigrationState *s = opaque;
3674 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3675 MigThrError thr_error;
3676 bool urgent = false;
3677
3678 rcu_register_thread();
3679
3680 object_ref(OBJECT(s));
3681 update_iteration_initial_status(s);
3682
3683 qemu_savevm_state_header(s->to_dst_file);
3684
3685 /*
3686 * If we opened the return path, we need to make sure dst has it
3687 * opened as well.
3688 */
3689 if (s->rp_state.from_dst_file) {
3690 /* Now tell the dest that it should open its end so it can reply */
3691 qemu_savevm_send_open_return_path(s->to_dst_file);
3692
3693 /* And do a ping that will make stuff easier to debug */
3694 qemu_savevm_send_ping(s->to_dst_file, 1);
3695 }
3696
3697 if (migrate_postcopy()) {
3698 /*
3699 * Tell the destination that we *might* want to do postcopy later;
3700 * if the other end can't do postcopy it should fail now, nice and
3701 * early.
3702 */
3703 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3704 }
3705
3706 if (migrate_colo_enabled()) {
3707 /* Notify migration destination that we enable COLO */
3708 qemu_savevm_send_colo_enable(s->to_dst_file);
3709 }
3710
3711 qemu_savevm_state_setup(s->to_dst_file);
3712
3713 if (qemu_savevm_state_guest_unplug_pending()) {
3714 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3715 MIGRATION_STATUS_WAIT_UNPLUG);
3716
3717 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3718 qemu_savevm_state_guest_unplug_pending()) {
3719 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3720 }
3721
3722 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3723 MIGRATION_STATUS_ACTIVE);
3724 }
3725
3726 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3727 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3728 MIGRATION_STATUS_ACTIVE);
3729
3730 trace_migration_thread_setup_complete();
3731
3732 while (migration_is_active(s)) {
3733 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3734 MigIterateState iter_state = migration_iteration_run(s);
3735 if (iter_state == MIG_ITERATE_SKIP) {
3736 continue;
3737 } else if (iter_state == MIG_ITERATE_BREAK) {
3738 break;
3739 }
3740 }
3741
3742 /*
3743 * Try to detect any kind of failures, and see whether we
3744 * should stop the migration now.
3745 */
3746 thr_error = migration_detect_error(s);
3747 if (thr_error == MIG_THR_ERR_FATAL) {
3748 /* Stop migration */
3749 break;
3750 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3751 /*
3752 * Just recovered from a e.g. network failure, reset all
3753 * the local variables. This is important to avoid
3754 * breaking transferred_bytes and bandwidth calculation
3755 */
3756 update_iteration_initial_status(s);
3757 }
3758
3759 urgent = migration_rate_limit();
3760 }
3761
3762 trace_migration_thread_after_loop();
3763 migration_iteration_finish(s);
3764 object_unref(OBJECT(s));
3765 rcu_unregister_thread();
3766 return NULL;
3767 }
3768
3769 static void bg_migration_vm_start_bh(void *opaque)
3770 {
3771 MigrationState *s = opaque;
3772
3773 qemu_bh_delete(s->vm_start_bh);
3774 s->vm_start_bh = NULL;
3775
3776 vm_start();
3777 s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start;
3778 }
3779
3780 /**
3781 * Background snapshot thread, based on live migration code.
3782 * This is an alternative implementation of live migration mechanism
3783 * introduced specifically to support background snapshots.
3784 *
3785 * It takes advantage of userfault_fd write protection mechanism introduced
3786 * in v5.7 kernel. Compared to existing dirty page logging migration much
3787 * lesser stream traffic is produced resulting in smaller snapshot images,
3788 * simply cause of no page duplicates can get into the stream.
3789 *
3790 * Another key point is that generated vmstate stream reflects machine state
3791 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3792 * mechanism, which effectively results in that saved snapshot is the state of VM
3793 * at the end of the process.
3794 */
3795 static void *bg_migration_thread(void *opaque)
3796 {
3797 MigrationState *s = opaque;
3798 int64_t setup_start;
3799 MigThrError thr_error;
3800 QEMUFile *fb;
3801 bool early_fail = true;
3802
3803 rcu_register_thread();
3804 object_ref(OBJECT(s));
3805
3806 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3807
3808 setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3809 /*
3810 * We want to save vmstate for the moment when migration has been
3811 * initiated but also we want to save RAM content while VM is running.
3812 * The RAM content should appear first in the vmstate. So, we first
3813 * stash the non-RAM part of the vmstate to the temporary buffer,
3814 * then write RAM part of the vmstate to the migration stream
3815 * with vCPUs running and, finally, write stashed non-RAM part of
3816 * the vmstate from the buffer to the migration stream.
3817 */
3818 s->bioc = qio_channel_buffer_new(512 * 1024);
3819 qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer");
3820 fb = qemu_fopen_channel_output(QIO_CHANNEL(s->bioc));
3821 object_unref(OBJECT(s->bioc));
3822
3823 update_iteration_initial_status(s);
3824
3825 /*
3826 * Prepare for tracking memory writes with UFFD-WP - populate
3827 * RAM pages before protecting.
3828 */
3829 #ifdef __linux__
3830 ram_write_tracking_prepare();
3831 #endif
3832
3833 qemu_savevm_state_header(s->to_dst_file);
3834 qemu_savevm_state_setup(s->to_dst_file);
3835
3836 if (qemu_savevm_state_guest_unplug_pending()) {
3837 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3838 MIGRATION_STATUS_WAIT_UNPLUG);
3839
3840 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3841 qemu_savevm_state_guest_unplug_pending()) {
3842 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3843 }
3844
3845 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3846 MIGRATION_STATUS_ACTIVE);
3847 } else {
3848 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3849 MIGRATION_STATUS_ACTIVE);
3850 }
3851 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3852
3853 trace_migration_thread_setup_complete();
3854 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3855
3856 qemu_mutex_lock_iothread();
3857
3858 /*
3859 * If VM is currently in suspended state, then, to make a valid runstate
3860 * transition in vm_stop_force_state() we need to wakeup it up.
3861 */
3862 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3863 s->vm_was_running = runstate_is_running();
3864
3865 if (global_state_store()) {
3866 goto fail;
3867 }
3868 /* Forcibly stop VM before saving state of vCPUs and devices */
3869 if (vm_stop_force_state(RUN_STATE_PAUSED)) {
3870 goto fail;
3871 }
3872 /*
3873 * Put vCPUs in sync with shadow context structures, then
3874 * save their state to channel-buffer along with devices.
3875 */
3876 cpu_synchronize_all_states();
3877 if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) {
3878 goto fail;
3879 }
3880 /*
3881 * Since we are going to get non-iterable state data directly
3882 * from s->bioc->data, explicit flush is needed here.
3883 */
3884 qemu_fflush(fb);
3885
3886 /* Now initialize UFFD context and start tracking RAM writes */
3887 if (ram_write_tracking_start()) {
3888 goto fail;
3889 }
3890 early_fail = false;
3891
3892 /*
3893 * Start VM from BH handler to avoid write-fault lock here.
3894 * UFFD-WP protection for the whole RAM is already enabled so
3895 * calling VM state change notifiers from vm_start() would initiate
3896 * writes to virtio VQs memory which is in write-protected region.
3897 */
3898 s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s);
3899 qemu_bh_schedule(s->vm_start_bh);
3900
3901 qemu_mutex_unlock_iothread();
3902
3903 while (migration_is_active(s)) {
3904 MigIterateState iter_state = bg_migration_iteration_run(s);
3905 if (iter_state == MIG_ITERATE_SKIP) {
3906 continue;
3907 } else if (iter_state == MIG_ITERATE_BREAK) {
3908 break;
3909 }
3910
3911 /*
3912 * Try to detect any kind of failures, and see whether we
3913 * should stop the migration now.
3914 */
3915 thr_error = migration_detect_error(s);
3916 if (thr_error == MIG_THR_ERR_FATAL) {
3917 /* Stop migration */
3918 break;
3919 }
3920
3921 migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
3922 }
3923
3924 trace_migration_thread_after_loop();