Merge remote-tracking branch 'remotes/mcayland/tags/qemu-sparc-20210412' into staging
[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 #ifdef CONFIG_VFIO
64 #include "hw/vfio/vfio-common.h"
65 #endif
66
67 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
68
69 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
70 * data. */
71 #define BUFFER_DELAY 100
72 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
73
74 /* Time in milliseconds we are allowed to stop the source,
75 * for sending the last part */
76 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
77
78 /* Maximum migrate downtime set to 2000 seconds */
79 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
80 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
81
82 /* Default compression thread count */
83 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
84 /* Default decompression thread count, usually decompression is at
85 * least 4 times as fast as compression.*/
86 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
87 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
88 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
89 /* Define default autoconverge cpu throttle migration parameters */
90 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
91 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
92 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
93 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
94
95 /* Migration XBZRLE default cache size */
96 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
97
98 /* The delay time (in ms) between two COLO checkpoints */
99 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
100 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
101 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
102 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
103 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
104 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
105 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
106
107 /* Background transfer rate for postcopy, 0 means unlimited, note
108 * that page requests can still exceed this limit.
109 */
110 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
111
112 /*
113 * Parameters for self_announce_delay giving a stream of RARP/ARP
114 * packets after migration.
115 */
116 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
117 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
118 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
119 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
120
121 static NotifierList migration_state_notifiers =
122 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
123
124 /* Messages sent on the return path from destination to source */
125 enum mig_rp_message_type {
126 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
127 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
128 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
129
130 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
131 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
132 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
133 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
134
135 MIG_RP_MSG_MAX
136 };
137
138 /* Migration capabilities set */
139 struct MigrateCapsSet {
140 int size; /* Capability set size */
141 MigrationCapability caps[]; /* Variadic array of capabilities */
142 };
143 typedef struct MigrateCapsSet MigrateCapsSet;
144
145 /* Define and initialize MigrateCapsSet */
146 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \
147 MigrateCapsSet _name = { \
148 .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
149 .caps = { __VA_ARGS__ } \
150 }
151
152 /* Background-snapshot compatibility check list */
153 static const
154 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot,
155 MIGRATION_CAPABILITY_POSTCOPY_RAM,
156 MIGRATION_CAPABILITY_DIRTY_BITMAPS,
157 MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME,
158 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE,
159 MIGRATION_CAPABILITY_RETURN_PATH,
160 MIGRATION_CAPABILITY_MULTIFD,
161 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER,
162 MIGRATION_CAPABILITY_AUTO_CONVERGE,
163 MIGRATION_CAPABILITY_RELEASE_RAM,
164 MIGRATION_CAPABILITY_RDMA_PIN_ALL,
165 MIGRATION_CAPABILITY_COMPRESS,
166 MIGRATION_CAPABILITY_XBZRLE,
167 MIGRATION_CAPABILITY_X_COLO,
168 MIGRATION_CAPABILITY_VALIDATE_UUID);
169
170 /* When we add fault tolerance, we could have several
171 migrations at once. For now we don't need to add
172 dynamic creation of migration */
173
174 static MigrationState *current_migration;
175 static MigrationIncomingState *current_incoming;
176
177 static GSList *migration_blockers;
178
179 static bool migration_object_check(MigrationState *ms, Error **errp);
180 static int migration_maybe_pause(MigrationState *s,
181 int *current_active_state,
182 int new_state);
183 static void migrate_fd_cancel(MigrationState *s);
184
185 static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp)
186 {
187 uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp;
188
189 return (a > b) - (a < b);
190 }
191
192 void migration_object_init(void)
193 {
194 Error *err = NULL;
195
196 /* This can only be called once. */
197 assert(!current_migration);
198 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
199
200 /*
201 * Init the migrate incoming object as well no matter whether
202 * we'll use it or not.
203 */
204 assert(!current_incoming);
205 current_incoming = g_new0(MigrationIncomingState, 1);
206 current_incoming->state = MIGRATION_STATUS_NONE;
207 current_incoming->postcopy_remote_fds =
208 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
209 qemu_mutex_init(&current_incoming->rp_mutex);
210 qemu_event_init(&current_incoming->main_thread_load_event, false);
211 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
212 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
213 qemu_mutex_init(&current_incoming->page_request_mutex);
214 current_incoming->page_requested = g_tree_new(page_request_addr_cmp);
215
216 if (!migration_object_check(current_migration, &err)) {
217 error_report_err(err);
218 exit(1);
219 }
220
221 blk_mig_init();
222 ram_mig_init();
223 dirty_bitmap_mig_init();
224 }
225
226 void migration_shutdown(void)
227 {
228 /*
229 * Cancel the current migration - that will (eventually)
230 * stop the migration using this structure
231 */
232 migrate_fd_cancel(current_migration);
233 object_unref(OBJECT(current_migration));
234
235 /*
236 * Cancel outgoing migration of dirty bitmaps. It should
237 * at least unref used block nodes.
238 */
239 dirty_bitmap_mig_cancel_outgoing();
240
241 /*
242 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
243 * are non-critical data, and their loss never considered as
244 * something serious.
245 */
246 dirty_bitmap_mig_cancel_incoming();
247 }
248
249 /* For outgoing */
250 MigrationState *migrate_get_current(void)
251 {
252 /* This can only be called after the object created. */
253 assert(current_migration);
254 return current_migration;
255 }
256
257 MigrationIncomingState *migration_incoming_get_current(void)
258 {
259 assert(current_incoming);
260 return current_incoming;
261 }
262
263 void migration_incoming_state_destroy(void)
264 {
265 struct MigrationIncomingState *mis = migration_incoming_get_current();
266
267 if (mis->to_src_file) {
268 /* Tell source that we are done */
269 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
270 qemu_fclose(mis->to_src_file);
271 mis->to_src_file = NULL;
272 }
273
274 if (mis->from_src_file) {
275 qemu_fclose(mis->from_src_file);
276 mis->from_src_file = NULL;
277 }
278 if (mis->postcopy_remote_fds) {
279 g_array_free(mis->postcopy_remote_fds, TRUE);
280 mis->postcopy_remote_fds = NULL;
281 }
282
283 qemu_event_reset(&mis->main_thread_load_event);
284
285 if (mis->page_requested) {
286 g_tree_destroy(mis->page_requested);
287 mis->page_requested = NULL;
288 }
289
290 if (mis->socket_address_list) {
291 qapi_free_SocketAddressList(mis->socket_address_list);
292 mis->socket_address_list = NULL;
293 }
294
295 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
296 }
297
298 static void migrate_generate_event(int new_state)
299 {
300 if (migrate_use_events()) {
301 qapi_event_send_migration(new_state);
302 }
303 }
304
305 static bool migrate_late_block_activate(void)
306 {
307 MigrationState *s;
308
309 s = migrate_get_current();
310
311 return s->enabled_capabilities[
312 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
313 }
314
315 /*
316 * Send a message on the return channel back to the source
317 * of the migration.
318 */
319 static int migrate_send_rp_message(MigrationIncomingState *mis,
320 enum mig_rp_message_type message_type,
321 uint16_t len, void *data)
322 {
323 int ret = 0;
324
325 trace_migrate_send_rp_message((int)message_type, len);
326 qemu_mutex_lock(&mis->rp_mutex);
327
328 /*
329 * It's possible that the file handle got lost due to network
330 * failures.
331 */
332 if (!mis->to_src_file) {
333 ret = -EIO;
334 goto error;
335 }
336
337 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
338 qemu_put_be16(mis->to_src_file, len);
339 qemu_put_buffer(mis->to_src_file, data, len);
340 qemu_fflush(mis->to_src_file);
341
342 /* It's possible that qemu file got error during sending */
343 ret = qemu_file_get_error(mis->to_src_file);
344
345 error:
346 qemu_mutex_unlock(&mis->rp_mutex);
347 return ret;
348 }
349
350 /* Request one page from the source VM at the given start address.
351 * rb: the RAMBlock to request the page in
352 * Start: Address offset within the RB
353 * Len: Length in bytes required - must be a multiple of pagesize
354 */
355 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
356 RAMBlock *rb, ram_addr_t start)
357 {
358 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
359 size_t msglen = 12; /* start + len */
360 size_t len = qemu_ram_pagesize(rb);
361 enum mig_rp_message_type msg_type;
362 const char *rbname;
363 int rbname_len;
364
365 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
366 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
367
368 /*
369 * We maintain the last ramblock that we requested for page. Note that we
370 * don't need locking because this function will only be called within the
371 * postcopy ram fault thread.
372 */
373 if (rb != mis->last_rb) {
374 mis->last_rb = rb;
375
376 rbname = qemu_ram_get_idstr(rb);
377 rbname_len = strlen(rbname);
378
379 assert(rbname_len < 256);
380
381 bufc[msglen++] = rbname_len;
382 memcpy(bufc + msglen, rbname, rbname_len);
383 msglen += rbname_len;
384 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
385 } else {
386 msg_type = MIG_RP_MSG_REQ_PAGES;
387 }
388
389 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
390 }
391
392 int migrate_send_rp_req_pages(MigrationIncomingState *mis,
393 RAMBlock *rb, ram_addr_t start, uint64_t haddr)
394 {
395 void *aligned = (void *)(uintptr_t)(haddr & (-qemu_ram_pagesize(rb)));
396 bool received = false;
397
398 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
399 received = ramblock_recv_bitmap_test_byte_offset(rb, start);
400 if (!received && !g_tree_lookup(mis->page_requested, aligned)) {
401 /*
402 * The page has not been received, and it's not yet in the page
403 * request list. Queue it. Set the value of element to 1, so that
404 * things like g_tree_lookup() will return TRUE (1) when found.
405 */
406 g_tree_insert(mis->page_requested, aligned, (gpointer)1);
407 mis->page_requested_count++;
408 trace_postcopy_page_req_add(aligned, mis->page_requested_count);
409 }
410 }
411
412 /*
413 * If the page is there, skip sending the message. We don't even need the
414 * lock because as long as the page arrived, it'll be there forever.
415 */
416 if (received) {
417 return 0;
418 }
419
420 return migrate_send_rp_message_req_pages(mis, rb, start);
421 }
422
423 static bool migration_colo_enabled;
424 bool migration_incoming_colo_enabled(void)
425 {
426 return migration_colo_enabled;
427 }
428
429 void migration_incoming_disable_colo(void)
430 {
431 ram_block_discard_disable(false);
432 migration_colo_enabled = false;
433 }
434
435 int migration_incoming_enable_colo(void)
436 {
437 if (ram_block_discard_disable(true)) {
438 error_report("COLO: cannot disable RAM discard");
439 return -EBUSY;
440 }
441 migration_colo_enabled = true;
442 return 0;
443 }
444
445 void migrate_add_address(SocketAddress *address)
446 {
447 MigrationIncomingState *mis = migration_incoming_get_current();
448
449 QAPI_LIST_PREPEND(mis->socket_address_list,
450 QAPI_CLONE(SocketAddress, address));
451 }
452
453 static void qemu_start_incoming_migration(const char *uri, Error **errp)
454 {
455 const char *p = NULL;
456
457 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
458 return;
459 }
460
461 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
462 if (strstart(uri, "tcp:", &p) ||
463 strstart(uri, "unix:", NULL) ||
464 strstart(uri, "vsock:", NULL)) {
465 socket_start_incoming_migration(p ? p : uri, errp);
466 #ifdef CONFIG_RDMA
467 } else if (strstart(uri, "rdma:", &p)) {
468 rdma_start_incoming_migration(p, errp);
469 #endif
470 } else if (strstart(uri, "exec:", &p)) {
471 exec_start_incoming_migration(p, errp);
472 } else if (strstart(uri, "fd:", &p)) {
473 fd_start_incoming_migration(p, errp);
474 } else {
475 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
476 error_setg(errp, "unknown migration protocol: %s", uri);
477 }
478 }
479
480 static void process_incoming_migration_bh(void *opaque)
481 {
482 Error *local_err = NULL;
483 MigrationIncomingState *mis = opaque;
484
485 /* If capability late_block_activate is set:
486 * Only fire up the block code now if we're going to restart the
487 * VM, else 'cont' will do it.
488 * This causes file locking to happen; so we don't want it to happen
489 * unless we really are starting the VM.
490 */
491 if (!migrate_late_block_activate() ||
492 (autostart && (!global_state_received() ||
493 global_state_get_runstate() == RUN_STATE_RUNNING))) {
494 /* Make sure all file formats flush their mutable metadata.
495 * If we get an error here, just don't restart the VM yet. */
496 bdrv_invalidate_cache_all(&local_err);
497 if (local_err) {
498 error_report_err(local_err);
499 local_err = NULL;
500 autostart = false;
501 }
502 }
503
504 /*
505 * This must happen after all error conditions are dealt with and
506 * we're sure the VM is going to be running on this host.
507 */
508 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
509
510 if (multifd_load_cleanup(&local_err) != 0) {
511 error_report_err(local_err);
512 autostart = false;
513 }
514 /* If global state section was not received or we are in running
515 state, we need to obey autostart. Any other state is set with
516 runstate_set. */
517
518 dirty_bitmap_mig_before_vm_start();
519
520 if (!global_state_received() ||
521 global_state_get_runstate() == RUN_STATE_RUNNING) {
522 if (autostart) {
523 vm_start();
524 } else {
525 runstate_set(RUN_STATE_PAUSED);
526 }
527 } else if (migration_incoming_colo_enabled()) {
528 migration_incoming_disable_colo();
529 vm_start();
530 } else {
531 runstate_set(global_state_get_runstate());
532 }
533 /*
534 * This must happen after any state changes since as soon as an external
535 * observer sees this event they might start to prod at the VM assuming
536 * it's ready to use.
537 */
538 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
539 MIGRATION_STATUS_COMPLETED);
540 qemu_bh_delete(mis->bh);
541 migration_incoming_state_destroy();
542 }
543
544 static void process_incoming_migration_co(void *opaque)
545 {
546 MigrationIncomingState *mis = migration_incoming_get_current();
547 PostcopyState ps;
548 int ret;
549 Error *local_err = NULL;
550
551 assert(mis->from_src_file);
552 mis->migration_incoming_co = qemu_coroutine_self();
553 mis->largest_page_size = qemu_ram_pagesize_largest();
554 postcopy_state_set(POSTCOPY_INCOMING_NONE);
555 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
556 MIGRATION_STATUS_ACTIVE);
557 ret = qemu_loadvm_state(mis->from_src_file);
558
559 ps = postcopy_state_get();
560 trace_process_incoming_migration_co_end(ret, ps);
561 if (ps != POSTCOPY_INCOMING_NONE) {
562 if (ps == POSTCOPY_INCOMING_ADVISE) {
563 /*
564 * Where a migration had postcopy enabled (and thus went to advise)
565 * but managed to complete within the precopy period, we can use
566 * the normal exit.
567 */
568 postcopy_ram_incoming_cleanup(mis);
569 } else if (ret >= 0) {
570 /*
571 * Postcopy was started, cleanup should happen at the end of the
572 * postcopy thread.
573 */
574 trace_process_incoming_migration_co_postcopy_end_main();
575 return;
576 }
577 /* Else if something went wrong then just fall out of the normal exit */
578 }
579
580 /* we get COLO info, and know if we are in COLO mode */
581 if (!ret && migration_incoming_colo_enabled()) {
582 /* Make sure all file formats flush their mutable metadata */
583 bdrv_invalidate_cache_all(&local_err);
584 if (local_err) {
585 error_report_err(local_err);
586 goto fail;
587 }
588
589 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
590 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
591 mis->have_colo_incoming_thread = true;
592 qemu_coroutine_yield();
593
594 /* Wait checkpoint incoming thread exit before free resource */
595 qemu_thread_join(&mis->colo_incoming_thread);
596 /* We hold the global iothread lock, so it is safe here */
597 colo_release_ram_cache();
598 }
599
600 if (ret < 0) {
601 error_report("load of migration failed: %s", strerror(-ret));
602 goto fail;
603 }
604 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
605 qemu_bh_schedule(mis->bh);
606 mis->migration_incoming_co = NULL;
607 return;
608 fail:
609 local_err = NULL;
610 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
611 MIGRATION_STATUS_FAILED);
612 qemu_fclose(mis->from_src_file);
613 if (multifd_load_cleanup(&local_err) != 0) {
614 error_report_err(local_err);
615 }
616 exit(EXIT_FAILURE);
617 }
618
619 /**
620 * @migration_incoming_setup: Setup incoming migration
621 *
622 * Returns 0 for no error or 1 for error
623 *
624 * @f: file for main migration channel
625 * @errp: where to put errors
626 */
627 static int migration_incoming_setup(QEMUFile *f, Error **errp)
628 {
629 MigrationIncomingState *mis = migration_incoming_get_current();
630 Error *local_err = NULL;
631
632 if (multifd_load_setup(&local_err) != 0) {
633 /* We haven't been able to create multifd threads
634 nothing better to do */
635 error_report_err(local_err);
636 exit(EXIT_FAILURE);
637 }
638
639 if (!mis->from_src_file) {
640 mis->from_src_file = f;
641 }
642 qemu_file_set_blocking(f, false);
643 return 0;
644 }
645
646 void migration_incoming_process(void)
647 {
648 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
649 qemu_coroutine_enter(co);
650 }
651
652 /* Returns true if recovered from a paused migration, otherwise false */
653 static bool postcopy_try_recover(QEMUFile *f)
654 {
655 MigrationIncomingState *mis = migration_incoming_get_current();
656
657 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
658 /* Resumed from a paused postcopy migration */
659
660 mis->from_src_file = f;
661 /* Postcopy has standalone thread to do vm load */
662 qemu_file_set_blocking(f, true);
663
664 /* Re-configure the return path */
665 mis->to_src_file = qemu_file_get_return_path(f);
666
667 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
668 MIGRATION_STATUS_POSTCOPY_RECOVER);
669
670 /*
671 * Here, we only wake up the main loading thread (while the
672 * fault thread will still be waiting), so that we can receive
673 * commands from source now, and answer it if needed. The
674 * fault thread will be woken up afterwards until we are sure
675 * that source is ready to reply to page requests.
676 */
677 qemu_sem_post(&mis->postcopy_pause_sem_dst);
678 return true;
679 }
680
681 return false;
682 }
683
684 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
685 {
686 Error *local_err = NULL;
687
688 if (postcopy_try_recover(f)) {
689 return;
690 }
691
692 if (migration_incoming_setup(f, &local_err)) {
693 error_propagate(errp, local_err);
694 return;
695 }
696 migration_incoming_process();
697 }
698
699 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
700 {
701 MigrationIncomingState *mis = migration_incoming_get_current();
702 Error *local_err = NULL;
703 bool start_migration;
704
705 if (!mis->from_src_file) {
706 /* The first connection (multifd may have multiple) */
707 QEMUFile *f = qemu_fopen_channel_input(ioc);
708
709 /* If it's a recovery, we're done */
710 if (postcopy_try_recover(f)) {
711 return;
712 }
713
714 if (migration_incoming_setup(f, &local_err)) {
715 error_propagate(errp, local_err);
716 return;
717 }
718
719 /*
720 * Common migration only needs one channel, so we can start
721 * right now. Multifd needs more than one channel, we wait.
722 */
723 start_migration = !migrate_use_multifd();
724 } else {
725 /* Multiple connections */
726 assert(migrate_use_multifd());
727 start_migration = multifd_recv_new_channel(ioc, &local_err);
728 if (local_err) {
729 error_propagate(errp, local_err);
730 return;
731 }
732 }
733
734 if (start_migration) {
735 migration_incoming_process();
736 }
737 }
738
739 /**
740 * @migration_has_all_channels: We have received all channels that we need
741 *
742 * Returns true when we have got connections to all the channels that
743 * we need for migration.
744 */
745 bool migration_has_all_channels(void)
746 {
747 MigrationIncomingState *mis = migration_incoming_get_current();
748 bool all_channels;
749
750 all_channels = multifd_recv_all_channels_created();
751
752 return all_channels && mis->from_src_file != NULL;
753 }
754
755 /*
756 * Send a 'SHUT' message on the return channel with the given value
757 * to indicate that we've finished with the RP. Non-0 value indicates
758 * error.
759 */
760 void migrate_send_rp_shut(MigrationIncomingState *mis,
761 uint32_t value)
762 {
763 uint32_t buf;
764
765 buf = cpu_to_be32(value);
766 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
767 }
768
769 /*
770 * Send a 'PONG' message on the return channel with the given value
771 * (normally in response to a 'PING')
772 */
773 void migrate_send_rp_pong(MigrationIncomingState *mis,
774 uint32_t value)
775 {
776 uint32_t buf;
777
778 buf = cpu_to_be32(value);
779 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
780 }
781
782 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
783 char *block_name)
784 {
785 char buf[512];
786 int len;
787 int64_t res;
788
789 /*
790 * First, we send the header part. It contains only the len of
791 * idstr, and the idstr itself.
792 */
793 len = strlen(block_name);
794 buf[0] = len;
795 memcpy(buf + 1, block_name, len);
796
797 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
798 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
799 __func__);
800 return;
801 }
802
803 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
804
805 /*
806 * Next, we dump the received bitmap to the stream.
807 *
808 * TODO: currently we are safe since we are the only one that is
809 * using the to_src_file handle (fault thread is still paused),
810 * and it's ok even not taking the mutex. However the best way is
811 * to take the lock before sending the message header, and release
812 * the lock after sending the bitmap.
813 */
814 qemu_mutex_lock(&mis->rp_mutex);
815 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
816 qemu_mutex_unlock(&mis->rp_mutex);
817
818 trace_migrate_send_rp_recv_bitmap(block_name, res);
819 }
820
821 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
822 {
823 uint32_t buf;
824
825 buf = cpu_to_be32(value);
826 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
827 }
828
829 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
830 {
831 MigrationCapabilityStatusList *head = NULL, **tail = &head;
832 MigrationCapabilityStatus *caps;
833 MigrationState *s = migrate_get_current();
834 int i;
835
836 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
837 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
838 if (i == MIGRATION_CAPABILITY_BLOCK) {
839 continue;
840 }
841 #endif
842 caps = g_malloc0(sizeof(*caps));
843 caps->capability = i;
844 caps->state = s->enabled_capabilities[i];
845 QAPI_LIST_APPEND(tail, caps);
846 }
847
848 return head;
849 }
850
851 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
852 {
853 MigrationParameters *params;
854 MigrationState *s = migrate_get_current();
855
856 /* TODO use QAPI_CLONE() instead of duplicating it inline */
857 params = g_malloc0(sizeof(*params));
858 params->has_compress_level = true;
859 params->compress_level = s->parameters.compress_level;
860 params->has_compress_threads = true;
861 params->compress_threads = s->parameters.compress_threads;
862 params->has_compress_wait_thread = true;
863 params->compress_wait_thread = s->parameters.compress_wait_thread;
864 params->has_decompress_threads = true;
865 params->decompress_threads = s->parameters.decompress_threads;
866 params->has_throttle_trigger_threshold = true;
867 params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
868 params->has_cpu_throttle_initial = true;
869 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
870 params->has_cpu_throttle_increment = true;
871 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
872 params->has_cpu_throttle_tailslow = true;
873 params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
874 params->has_tls_creds = true;
875 params->tls_creds = g_strdup(s->parameters.tls_creds);
876 params->has_tls_hostname = true;
877 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
878 params->has_tls_authz = true;
879 params->tls_authz = g_strdup(s->parameters.tls_authz ?
880 s->parameters.tls_authz : "");
881 params->has_max_bandwidth = true;
882 params->max_bandwidth = s->parameters.max_bandwidth;
883 params->has_downtime_limit = true;
884 params->downtime_limit = s->parameters.downtime_limit;
885 params->has_x_checkpoint_delay = true;
886 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
887 params->has_block_incremental = true;
888 params->block_incremental = s->parameters.block_incremental;
889 params->has_multifd_channels = true;
890 params->multifd_channels = s->parameters.multifd_channels;
891 params->has_multifd_compression = true;
892 params->multifd_compression = s->parameters.multifd_compression;
893 params->has_multifd_zlib_level = true;
894 params->multifd_zlib_level = s->parameters.multifd_zlib_level;
895 params->has_multifd_zstd_level = true;
896 params->multifd_zstd_level = s->parameters.multifd_zstd_level;
897 params->has_xbzrle_cache_size = true;
898 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
899 params->has_max_postcopy_bandwidth = true;
900 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
901 params->has_max_cpu_throttle = true;
902 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
903 params->has_announce_initial = true;
904 params->announce_initial = s->parameters.announce_initial;
905 params->has_announce_max = true;
906 params->announce_max = s->parameters.announce_max;
907 params->has_announce_rounds = true;
908 params->announce_rounds = s->parameters.announce_rounds;
909 params->has_announce_step = true;
910 params->announce_step = s->parameters.announce_step;
911
912 if (s->parameters.has_block_bitmap_mapping) {
913 params->has_block_bitmap_mapping = true;
914 params->block_bitmap_mapping =
915 QAPI_CLONE(BitmapMigrationNodeAliasList,
916 s->parameters.block_bitmap_mapping);
917 }
918
919 return params;
920 }
921
922 AnnounceParameters *migrate_announce_params(void)
923 {
924 static AnnounceParameters ap;
925
926 MigrationState *s = migrate_get_current();
927
928 ap.initial = s->parameters.announce_initial;
929 ap.max = s->parameters.announce_max;
930 ap.rounds = s->parameters.announce_rounds;
931 ap.step = s->parameters.announce_step;
932
933 return &ap;
934 }
935
936 /*
937 * Return true if we're already in the middle of a migration
938 * (i.e. any of the active or setup states)
939 */
940 bool migration_is_setup_or_active(int state)
941 {
942 switch (state) {
943 case MIGRATION_STATUS_ACTIVE:
944 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
945 case MIGRATION_STATUS_POSTCOPY_PAUSED:
946 case MIGRATION_STATUS_POSTCOPY_RECOVER:
947 case MIGRATION_STATUS_SETUP:
948 case MIGRATION_STATUS_PRE_SWITCHOVER:
949 case MIGRATION_STATUS_DEVICE:
950 case MIGRATION_STATUS_WAIT_UNPLUG:
951 case MIGRATION_STATUS_COLO:
952 return true;
953
954 default:
955 return false;
956
957 }
958 }
959
960 bool migration_is_running(int state)
961 {
962 switch (state) {
963 case MIGRATION_STATUS_ACTIVE:
964 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
965 case MIGRATION_STATUS_POSTCOPY_PAUSED:
966 case MIGRATION_STATUS_POSTCOPY_RECOVER:
967 case MIGRATION_STATUS_SETUP:
968 case MIGRATION_STATUS_PRE_SWITCHOVER:
969 case MIGRATION_STATUS_DEVICE:
970 case MIGRATION_STATUS_WAIT_UNPLUG:
971 case MIGRATION_STATUS_CANCELLING:
972 return true;
973
974 default:
975 return false;
976
977 }
978 }
979
980 static void populate_time_info(MigrationInfo *info, MigrationState *s)
981 {
982 info->has_status = true;
983 info->has_setup_time = true;
984 info->setup_time = s->setup_time;
985 if (s->state == MIGRATION_STATUS_COMPLETED) {
986 info->has_total_time = true;
987 info->total_time = s->total_time;
988 info->has_downtime = true;
989 info->downtime = s->downtime;
990 } else {
991 info->has_total_time = true;
992 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
993 s->start_time;
994 info->has_expected_downtime = true;
995 info->expected_downtime = s->expected_downtime;
996 }
997 }
998
999 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
1000 {
1001 info->has_ram = true;
1002 info->ram = g_malloc0(sizeof(*info->ram));
1003 info->ram->transferred = ram_counters.transferred;
1004 info->ram->total = ram_bytes_total();
1005 info->ram->duplicate = ram_counters.duplicate;
1006 /* legacy value. It is not used anymore */
1007 info->ram->skipped = 0;
1008 info->ram->normal = ram_counters.normal;
1009 info->ram->normal_bytes = ram_counters.normal *
1010 qemu_target_page_size();
1011 info->ram->mbps = s->mbps;
1012 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
1013 info->ram->postcopy_requests = ram_counters.postcopy_requests;
1014 info->ram->page_size = qemu_target_page_size();
1015 info->ram->multifd_bytes = ram_counters.multifd_bytes;
1016 info->ram->pages_per_second = s->pages_per_second;
1017
1018 if (migrate_use_xbzrle()) {
1019 info->has_xbzrle_cache = true;
1020 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
1021 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
1022 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
1023 info->xbzrle_cache->pages = xbzrle_counters.pages;
1024 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
1025 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
1026 info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
1027 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
1028 }
1029
1030 if (migrate_use_compression()) {
1031 info->has_compression = true;
1032 info->compression = g_malloc0(sizeof(*info->compression));
1033 info->compression->pages = compression_counters.pages;
1034 info->compression->busy = compression_counters.busy;
1035 info->compression->busy_rate = compression_counters.busy_rate;
1036 info->compression->compressed_size =
1037 compression_counters.compressed_size;
1038 info->compression->compression_rate =
1039 compression_counters.compression_rate;
1040 }
1041
1042 if (cpu_throttle_active()) {
1043 info->has_cpu_throttle_percentage = true;
1044 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
1045 }
1046
1047 if (s->state != MIGRATION_STATUS_COMPLETED) {
1048 info->ram->remaining = ram_bytes_remaining();
1049 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
1050 }
1051 }
1052
1053 static void populate_disk_info(MigrationInfo *info)
1054 {
1055 if (blk_mig_active()) {
1056 info->has_disk = true;
1057 info->disk = g_malloc0(sizeof(*info->disk));
1058 info->disk->transferred = blk_mig_bytes_transferred();
1059 info->disk->remaining = blk_mig_bytes_remaining();
1060 info->disk->total = blk_mig_bytes_total();
1061 }
1062 }
1063
1064 static void populate_vfio_info(MigrationInfo *info)
1065 {
1066 #ifdef CONFIG_VFIO
1067 if (vfio_mig_active()) {
1068 info->has_vfio = true;
1069 info->vfio = g_malloc0(sizeof(*info->vfio));
1070 info->vfio->transferred = vfio_mig_bytes_transferred();
1071 }
1072 #endif
1073 }
1074
1075 static void fill_source_migration_info(MigrationInfo *info)
1076 {
1077 MigrationState *s = migrate_get_current();
1078
1079 info->blocked = migration_is_blocked(NULL);
1080 info->has_blocked_reasons = info->blocked;
1081 info->blocked_reasons = NULL;
1082 if (info->blocked) {
1083 GSList *cur_blocker = migration_blockers;
1084
1085 /*
1086 * There are two types of reasons a migration might be blocked;
1087 * a) devices marked in VMState as non-migratable, and
1088 * b) Explicit migration blockers
1089 * We need to add both of them here.
1090 */
1091 qemu_savevm_non_migratable_list(&info->blocked_reasons);
1092
1093 while (cur_blocker) {
1094 QAPI_LIST_PREPEND(info->blocked_reasons,
1095 g_strdup(error_get_pretty(cur_blocker->data)));
1096 cur_blocker = g_slist_next(cur_blocker);
1097 }
1098 }
1099
1100 switch (s->state) {
1101 case MIGRATION_STATUS_NONE:
1102 /* no migration has happened ever */
1103 /* do not overwrite destination migration status */
1104 return;
1105 case MIGRATION_STATUS_SETUP:
1106 info->has_status = true;
1107 info->has_total_time = false;
1108 break;
1109 case MIGRATION_STATUS_ACTIVE:
1110 case MIGRATION_STATUS_CANCELLING:
1111 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1112 case MIGRATION_STATUS_PRE_SWITCHOVER:
1113 case MIGRATION_STATUS_DEVICE:
1114 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1115 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1116 /* TODO add some postcopy stats */
1117 populate_time_info(info, s);
1118 populate_ram_info(info, s);
1119 populate_disk_info(info);
1120 populate_vfio_info(info);
1121 break;
1122 case MIGRATION_STATUS_COLO:
1123 info->has_status = true;
1124 /* TODO: display COLO specific information (checkpoint info etc.) */
1125 break;
1126 case MIGRATION_STATUS_COMPLETED:
1127 populate_time_info(info, s);
1128 populate_ram_info(info, s);
1129 populate_vfio_info(info);
1130 break;
1131 case MIGRATION_STATUS_FAILED:
1132 info->has_status = true;
1133 if (s->error) {
1134 info->has_error_desc = true;
1135 info->error_desc = g_strdup(error_get_pretty(s->error));
1136 }
1137 break;
1138 case MIGRATION_STATUS_CANCELLED:
1139 info->has_status = true;
1140 break;
1141 case MIGRATION_STATUS_WAIT_UNPLUG:
1142 info->has_status = true;
1143 break;
1144 }
1145 info->status = s->state;
1146 }
1147
1148 typedef enum WriteTrackingSupport {
1149 WT_SUPPORT_UNKNOWN = 0,
1150 WT_SUPPORT_ABSENT,
1151 WT_SUPPORT_AVAILABLE,
1152 WT_SUPPORT_COMPATIBLE
1153 } WriteTrackingSupport;
1154
1155 static
1156 WriteTrackingSupport migrate_query_write_tracking(void)
1157 {
1158 /* Check if kernel supports required UFFD features */
1159 if (!ram_write_tracking_available()) {
1160 return WT_SUPPORT_ABSENT;
1161 }
1162 /*
1163 * Check if current memory configuration is
1164 * compatible with required UFFD features.
1165 */
1166 if (!ram_write_tracking_compatible()) {
1167 return WT_SUPPORT_AVAILABLE;
1168 }
1169
1170 return WT_SUPPORT_COMPATIBLE;
1171 }
1172
1173 /**
1174 * @migration_caps_check - check capability validity
1175 *
1176 * @cap_list: old capability list, array of bool
1177 * @params: new capabilities to be applied soon
1178 * @errp: set *errp if the check failed, with reason
1179 *
1180 * Returns true if check passed, otherwise false.
1181 */
1182 static bool migrate_caps_check(bool *cap_list,
1183 MigrationCapabilityStatusList *params,
1184 Error **errp)
1185 {
1186 MigrationCapabilityStatusList *cap;
1187 bool old_postcopy_cap;
1188 MigrationIncomingState *mis = migration_incoming_get_current();
1189
1190 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1191
1192 for (cap = params; cap; cap = cap->next) {
1193 cap_list[cap->value->capability] = cap->value->state;
1194 }
1195
1196 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1197 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1198 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1199 "block migration");
1200 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1201 return false;
1202 }
1203 #endif
1204
1205 #ifndef CONFIG_REPLICATION
1206 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1207 error_setg(errp, "QEMU compiled without replication module"
1208 " can't enable COLO");
1209 error_append_hint(errp, "Please enable replication before COLO.\n");
1210 return false;
1211 }
1212 #endif
1213
1214 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1215 /* This check is reasonably expensive, so only when it's being
1216 * set the first time, also it's only the destination that needs
1217 * special support.
1218 */
1219 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1220 !postcopy_ram_supported_by_host(mis)) {
1221 /* postcopy_ram_supported_by_host will have emitted a more
1222 * detailed message
1223 */
1224 error_setg(errp, "Postcopy is not supported");
1225 return false;
1226 }
1227
1228 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1229 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1230 return false;
1231 }
1232 }
1233
1234 if (cap_list[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]) {
1235 WriteTrackingSupport wt_support;
1236 int idx;
1237 /*
1238 * Check if 'background-snapshot' capability is supported by
1239 * host kernel and compatible with guest memory configuration.
1240 */
1241 wt_support = migrate_query_write_tracking();
1242 if (wt_support < WT_SUPPORT_AVAILABLE) {
1243 error_setg(errp, "Background-snapshot is not supported by host kernel");
1244 return false;
1245 }
1246 if (wt_support < WT_SUPPORT_COMPATIBLE) {
1247 error_setg(errp, "Background-snapshot is not compatible "
1248 "with guest memory configuration");
1249 return false;
1250 }
1251
1252 /*
1253 * Check if there are any migration capabilities
1254 * incompatible with 'background-snapshot'.
1255 */
1256 for (idx = 0; idx < check_caps_background_snapshot.size; idx++) {
1257 int incomp_cap = check_caps_background_snapshot.caps[idx];
1258 if (cap_list[incomp_cap]) {
1259 error_setg(errp,
1260 "Background-snapshot is not compatible with %s",
1261 MigrationCapability_str(incomp_cap));
1262 return false;
1263 }
1264 }
1265 }
1266
1267 return true;
1268 }
1269
1270 static void fill_destination_migration_info(MigrationInfo *info)
1271 {
1272 MigrationIncomingState *mis = migration_incoming_get_current();
1273
1274 if (mis->socket_address_list) {
1275 info->has_socket_address = true;
1276 info->socket_address =
1277 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1278 }
1279
1280 switch (mis->state) {
1281 case MIGRATION_STATUS_NONE:
1282 return;
1283 case MIGRATION_STATUS_SETUP:
1284 case MIGRATION_STATUS_CANCELLING:
1285 case MIGRATION_STATUS_CANCELLED:
1286 case MIGRATION_STATUS_ACTIVE:
1287 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1288 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1289 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1290 case MIGRATION_STATUS_FAILED:
1291 case MIGRATION_STATUS_COLO:
1292 info->has_status = true;
1293 break;
1294 case MIGRATION_STATUS_COMPLETED:
1295 info->has_status = true;
1296 fill_destination_postcopy_migration_info(info);
1297 break;
1298 }
1299 info->status = mis->state;
1300 }
1301
1302 MigrationInfo *qmp_query_migrate(Error **errp)
1303 {
1304 MigrationInfo *info = g_malloc0(sizeof(*info));
1305
1306 fill_destination_migration_info(info);
1307 fill_source_migration_info(info);
1308
1309 return info;
1310 }
1311
1312 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1313 Error **errp)
1314 {
1315 MigrationState *s = migrate_get_current();
1316 MigrationCapabilityStatusList *cap;
1317 bool cap_list[MIGRATION_CAPABILITY__MAX];
1318
1319 if (migration_is_running(s->state)) {
1320 error_setg(errp, QERR_MIGRATION_ACTIVE);
1321 return;
1322 }
1323
1324 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1325 if (!migrate_caps_check(cap_list, params, errp)) {
1326 return;
1327 }
1328
1329 for (cap = params; cap; cap = cap->next) {
1330 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1331 }
1332 }
1333
1334 /*
1335 * Check whether the parameters are valid. Error will be put into errp
1336 * (if provided). Return true if valid, otherwise false.
1337 */
1338 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1339 {
1340 if (params->has_compress_level &&
1341 (params->compress_level > 9)) {
1342 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1343 "a value between 0 and 9");
1344 return false;
1345 }
1346
1347 if (params->has_compress_threads && (params->compress_threads < 1)) {
1348 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1349 "compress_threads",
1350 "a value between 1 and 255");
1351 return false;
1352 }
1353
1354 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1355 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1356 "decompress_threads",
1357 "a value between 1 and 255");
1358 return false;
1359 }
1360
1361 if (params->has_throttle_trigger_threshold &&
1362 (params->throttle_trigger_threshold < 1 ||
1363 params->throttle_trigger_threshold > 100)) {
1364 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1365 "throttle_trigger_threshold",
1366 "an integer in the range of 1 to 100");
1367 return false;
1368 }
1369
1370 if (params->has_cpu_throttle_initial &&
1371 (params->cpu_throttle_initial < 1 ||
1372 params->cpu_throttle_initial > 99)) {
1373 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1374 "cpu_throttle_initial",
1375 "an integer in the range of 1 to 99");
1376 return false;
1377 }
1378
1379 if (params->has_cpu_throttle_increment &&
1380 (params->cpu_throttle_increment < 1 ||
1381 params->cpu_throttle_increment > 99)) {
1382 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1383 "cpu_throttle_increment",
1384 "an integer in the range of 1 to 99");
1385 return false;
1386 }
1387
1388 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1389 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1390 "max_bandwidth",
1391 "an integer in the range of 0 to "stringify(SIZE_MAX)
1392 " bytes/second");
1393 return false;
1394 }
1395
1396 if (params->has_downtime_limit &&
1397 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1398 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1399 "downtime_limit",
1400 "an integer in the range of 0 to "
1401 stringify(MAX_MIGRATE_DOWNTIME)" ms");
1402 return false;
1403 }
1404
1405 /* x_checkpoint_delay is now always positive */
1406
1407 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1408 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1409 "multifd_channels",
1410 "a value between 1 and 255");
1411 return false;
1412 }
1413
1414 if (params->has_multifd_zlib_level &&
1415 (params->multifd_zlib_level > 9)) {
1416 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1417 "a value between 0 and 9");
1418 return false;
1419 }
1420
1421 if (params->has_multifd_zstd_level &&
1422 (params->multifd_zstd_level > 20)) {
1423 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1424 "a value between 0 and 20");
1425 return false;
1426 }
1427
1428 if (params->has_xbzrle_cache_size &&
1429 (params->xbzrle_cache_size < qemu_target_page_size() ||
1430 !is_power_of_2(params->xbzrle_cache_size))) {
1431 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1432 "xbzrle_cache_size",
1433 "a power of two no less than the target page size");
1434 return false;
1435 }
1436
1437 if (params->has_max_cpu_throttle &&
1438 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1439 params->max_cpu_throttle > 99)) {
1440 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1441 "max_cpu_throttle",
1442 "an integer in the range of cpu_throttle_initial to 99");
1443 return false;
1444 }
1445
1446 if (params->has_announce_initial &&
1447 params->announce_initial > 100000) {
1448 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1449 "announce_initial",
1450 "a value between 0 and 100000");
1451 return false;
1452 }
1453 if (params->has_announce_max &&
1454 params->announce_max > 100000) {
1455 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1456 "announce_max",
1457 "a value between 0 and 100000");
1458 return false;
1459 }
1460 if (params->has_announce_rounds &&
1461 params->announce_rounds > 1000) {
1462 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1463 "announce_rounds",
1464 "a value between 0 and 1000");
1465 return false;
1466 }
1467 if (params->has_announce_step &&
1468 (params->announce_step < 1 ||
1469 params->announce_step > 10000)) {
1470 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1471 "announce_step",
1472 "a value between 0 and 10000");
1473 return false;
1474 }
1475
1476 if (params->has_block_bitmap_mapping &&
1477 !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) {
1478 error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: ");
1479 return false;
1480 }
1481
1482 return true;
1483 }
1484
1485 static void migrate_params_test_apply(MigrateSetParameters *params,
1486 MigrationParameters *dest)
1487 {
1488 *dest = migrate_get_current()->parameters;
1489
1490 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1491
1492 if (params->has_compress_level) {
1493 dest->compress_level = params->compress_level;
1494 }
1495
1496 if (params->has_compress_threads) {
1497 dest->compress_threads = params->compress_threads;
1498 }
1499
1500 if (params->has_compress_wait_thread) {
1501 dest->compress_wait_thread = params->compress_wait_thread;
1502 }
1503
1504 if (params->has_decompress_threads) {
1505 dest->decompress_threads = params->decompress_threads;
1506 }
1507
1508 if (params->has_throttle_trigger_threshold) {
1509 dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
1510 }
1511
1512 if (params->has_cpu_throttle_initial) {
1513 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1514 }
1515
1516 if (params->has_cpu_throttle_increment) {
1517 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1518 }
1519
1520 if (params->has_cpu_throttle_tailslow) {
1521 dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1522 }
1523
1524 if (params->has_tls_creds) {
1525 assert(params->tls_creds->type == QTYPE_QSTRING);
1526 dest->tls_creds = params->tls_creds->u.s;
1527 }
1528
1529 if (params->has_tls_hostname) {
1530 assert(params->tls_hostname->type == QTYPE_QSTRING);
1531 dest->tls_hostname = params->tls_hostname->u.s;
1532 }
1533
1534 if (params->has_max_bandwidth) {
1535 dest->max_bandwidth = params->max_bandwidth;
1536 }
1537
1538 if (params->has_downtime_limit) {
1539 dest->downtime_limit = params->downtime_limit;
1540 }
1541
1542 if (params->has_x_checkpoint_delay) {
1543 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1544 }
1545
1546 if (params->has_block_incremental) {
1547 dest->block_incremental = params->block_incremental;
1548 }
1549 if (params->has_multifd_channels) {
1550 dest->multifd_channels = params->multifd_channels;
1551 }
1552 if (params->has_multifd_compression) {
1553 dest->multifd_compression = params->multifd_compression;
1554 }
1555 if (params->has_xbzrle_cache_size) {
1556 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1557 }
1558 if (params->has_max_postcopy_bandwidth) {
1559 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1560 }
1561 if (params->has_max_cpu_throttle) {
1562 dest->max_cpu_throttle = params->max_cpu_throttle;
1563 }
1564 if (params->has_announce_initial) {
1565 dest->announce_initial = params->announce_initial;
1566 }
1567 if (params->has_announce_max) {
1568 dest->announce_max = params->announce_max;
1569 }
1570 if (params->has_announce_rounds) {
1571 dest->announce_rounds = params->announce_rounds;
1572 }
1573 if (params->has_announce_step) {
1574 dest->announce_step = params->announce_step;
1575 }
1576
1577 if (params->has_block_bitmap_mapping) {
1578 dest->has_block_bitmap_mapping = true;
1579 dest->block_bitmap_mapping = params->block_bitmap_mapping;
1580 }
1581 }
1582
1583 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1584 {
1585 MigrationState *s = migrate_get_current();
1586
1587 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1588
1589 if (params->has_compress_level) {
1590 s->parameters.compress_level = params->compress_level;
1591 }
1592
1593 if (params->has_compress_threads) {
1594 s->parameters.compress_threads = params->compress_threads;
1595 }
1596
1597 if (params->has_compress_wait_thread) {
1598 s->parameters.compress_wait_thread = params->compress_wait_thread;
1599 }
1600
1601 if (params->has_decompress_threads) {
1602 s->parameters.decompress_threads = params->decompress_threads;
1603 }
1604
1605 if (params->has_throttle_trigger_threshold) {
1606 s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
1607 }
1608
1609 if (params->has_cpu_throttle_initial) {
1610 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1611 }
1612
1613 if (params->has_cpu_throttle_increment) {
1614 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1615 }
1616
1617 if (params->has_cpu_throttle_tailslow) {
1618 s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1619 }
1620
1621 if (params->has_tls_creds) {
1622 g_free(s->parameters.tls_creds);
1623 assert(params->tls_creds->type == QTYPE_QSTRING);
1624 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1625 }
1626
1627 if (params->has_tls_hostname) {
1628 g_free(s->parameters.tls_hostname);
1629 assert(params->tls_hostname->type == QTYPE_QSTRING);
1630 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1631 }
1632
1633 if (params->has_tls_authz) {
1634 g_free(s->parameters.tls_authz);
1635 assert(params->tls_authz->type == QTYPE_QSTRING);
1636 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1637 }
1638
1639 if (params->has_max_bandwidth) {
1640 s->parameters.max_bandwidth = params->max_bandwidth;
1641 if (s->to_dst_file && !migration_in_postcopy()) {
1642 qemu_file_set_rate_limit(s->to_dst_file,
1643 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1644 }
1645 }
1646
1647 if (params->has_downtime_limit) {
1648 s->parameters.downtime_limit = params->downtime_limit;
1649 }
1650
1651 if (params->has_x_checkpoint_delay) {
1652 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1653 if (migration_in_colo_state()) {
1654 colo_checkpoint_notify(s);
1655 }
1656 }
1657
1658 if (params->has_block_incremental) {
1659 s->parameters.block_incremental = params->block_incremental;
1660 }
1661 if (params->has_multifd_channels) {
1662 s->parameters.multifd_channels = params->multifd_channels;
1663 }
1664 if (params->has_multifd_compression) {
1665 s->parameters.multifd_compression = params->multifd_compression;
1666 }
1667 if (params->has_xbzrle_cache_size) {
1668 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1669 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1670 }
1671 if (params->has_max_postcopy_bandwidth) {
1672 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1673 if (s->to_dst_file && migration_in_postcopy()) {
1674 qemu_file_set_rate_limit(s->to_dst_file,
1675 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1676 }
1677 }
1678 if (params->has_max_cpu_throttle) {
1679 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1680 }
1681 if (params->has_announce_initial) {
1682 s->parameters.announce_initial = params->announce_initial;
1683 }
1684 if (params->has_announce_max) {
1685 s->parameters.announce_max = params->announce_max;
1686 }
1687 if (params->has_announce_rounds) {
1688 s->parameters.announce_rounds = params->announce_rounds;
1689 }
1690 if (params->has_announce_step) {
1691 s->parameters.announce_step = params->announce_step;
1692 }
1693
1694 if (params->has_block_bitmap_mapping) {
1695 qapi_free_BitmapMigrationNodeAliasList(
1696 s->parameters.block_bitmap_mapping);
1697
1698 s->parameters.has_block_bitmap_mapping = true;
1699 s->parameters.block_bitmap_mapping =
1700 QAPI_CLONE(BitmapMigrationNodeAliasList,
1701 params->block_bitmap_mapping);
1702 }
1703 }
1704
1705 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1706 {
1707 MigrationParameters tmp;
1708
1709 /* TODO Rewrite "" to null instead */
1710 if (params->has_tls_creds
1711 && params->tls_creds->type == QTYPE_QNULL) {
1712 qobject_unref(params->tls_creds->u.n);
1713 params->tls_creds->type = QTYPE_QSTRING;
1714 params->tls_creds->u.s = strdup("");
1715 }
1716 /* TODO Rewrite "" to null instead */
1717 if (params->has_tls_hostname
1718 && params->tls_hostname->type == QTYPE_QNULL) {
1719 qobject_unref(params->tls_hostname->u.n);
1720 params->tls_hostname->type = QTYPE_QSTRING;
1721 params->tls_hostname->u.s = strdup("");
1722 }
1723
1724 migrate_params_test_apply(params, &tmp);
1725
1726 if (!migrate_params_check(&tmp, errp)) {
1727 /* Invalid parameter */
1728 return;
1729 }
1730
1731 migrate_params_apply(params, errp);
1732 }
1733
1734
1735 void qmp_migrate_start_postcopy(Error **errp)
1736 {
1737 MigrationState *s = migrate_get_current();
1738
1739 if (!migrate_postcopy()) {
1740 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1741 " the start of migration");
1742 return;
1743 }
1744
1745 if (s->state == MIGRATION_STATUS_NONE) {
1746 error_setg(errp, "Postcopy must be started after migration has been"
1747 " started");
1748 return;
1749 }
1750 /*
1751 * we don't error if migration has finished since that would be racy
1752 * with issuing this command.
1753 */
1754 qatomic_set(&s->start_postcopy, true);
1755 }
1756
1757 /* shared migration helpers */
1758
1759 void migrate_set_state(int *state, int old_state, int new_state)
1760 {
1761 assert(new_state < MIGRATION_STATUS__MAX);
1762 if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
1763 trace_migrate_set_state(MigrationStatus_str(new_state));
1764 migrate_generate_event(new_state);
1765 }
1766 }
1767
1768 static MigrationCapabilityStatus *migrate_cap_add(MigrationCapability index,
1769 bool state)
1770 {
1771 MigrationCapabilityStatus *cap;
1772
1773 cap = g_new0(MigrationCapabilityStatus, 1);
1774 cap->capability = index;
1775 cap->state = state;
1776
1777 return cap;
1778 }
1779
1780 void migrate_set_block_enabled(bool value, Error **errp)
1781 {
1782 MigrationCapabilityStatusList *cap = NULL;
1783
1784 QAPI_LIST_PREPEND(cap, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK, value));
1785 qmp_migrate_set_capabilities(cap, errp);
1786 qapi_free_MigrationCapabilityStatusList(cap);
1787 }
1788
1789 static void migrate_set_block_incremental(MigrationState *s, bool value)
1790 {
1791 s->parameters.block_incremental = value;
1792 }
1793
1794 static void block_cleanup_parameters(MigrationState *s)
1795 {
1796 if (s->must_remove_block_options) {
1797 /* setting to false can never fail */
1798 migrate_set_block_enabled(false, &error_abort);
1799 migrate_set_block_incremental(s, false);
1800 s->must_remove_block_options = false;
1801 }
1802 }
1803
1804 static void migrate_fd_cleanup(MigrationState *s)
1805 {
1806 qemu_bh_delete(s->cleanup_bh);
1807 s->cleanup_bh = NULL;
1808
1809 qemu_savevm_state_cleanup();
1810
1811 if (s->to_dst_file) {
1812 QEMUFile *tmp;
1813
1814 trace_migrate_fd_cleanup();
1815 qemu_mutex_unlock_iothread();
1816 if (s->migration_thread_running) {
1817 qemu_thread_join(&s->thread);
1818 s->migration_thread_running = false;
1819 }
1820 qemu_mutex_lock_iothread();
1821
1822 multifd_save_cleanup();
1823 qemu_mutex_lock(&s->qemu_file_lock);
1824 tmp = s->to_dst_file;
1825 s->to_dst_file = NULL;
1826 qemu_mutex_unlock(&s->qemu_file_lock);
1827 /*
1828 * Close the file handle without the lock to make sure the
1829 * critical section won't block for long.
1830 */
1831 qemu_fclose(tmp);
1832 }
1833
1834 assert(!migration_is_active(s));
1835
1836 if (s->state == MIGRATION_STATUS_CANCELLING) {
1837 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1838 MIGRATION_STATUS_CANCELLED);
1839 }
1840
1841 if (s->error) {
1842 /* It is used on info migrate. We can't free it */
1843 error_report_err(error_copy(s->error));
1844 }
1845 notifier_list_notify(&migration_state_notifiers, s);
1846 block_cleanup_parameters(s);
1847 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
1848 }
1849
1850 static void migrate_fd_cleanup_schedule(MigrationState *s)
1851 {
1852 /*
1853 * Ref the state for bh, because it may be called when
1854 * there're already no other refs
1855 */
1856 object_ref(OBJECT(s));
1857 qemu_bh_schedule(s->cleanup_bh);
1858 }
1859
1860 static void migrate_fd_cleanup_bh(void *opaque)
1861 {
1862 MigrationState *s = opaque;
1863 migrate_fd_cleanup(s);
1864 object_unref(OBJECT(s));
1865 }
1866
1867 void migrate_set_error(MigrationState *s, const Error *error)
1868 {
1869 QEMU_LOCK_GUARD(&s->error_mutex);
1870 if (!s->error) {
1871 s->error = error_copy(error);
1872 }
1873 }
1874
1875 void migrate_fd_error(MigrationState *s, const Error *error)
1876 {
1877 trace_migrate_fd_error(error_get_pretty(error));
1878 assert(s->to_dst_file == NULL);
1879 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1880 MIGRATION_STATUS_FAILED);
1881 migrate_set_error(s, error);
1882 }
1883
1884 static void migrate_fd_cancel(MigrationState *s)
1885 {
1886 int old_state ;
1887 QEMUFile *f = migrate_get_current()->to_dst_file;
1888 trace_migrate_fd_cancel();
1889
1890 if (s->rp_state.from_dst_file) {
1891 /* shutdown the rp socket, so causing the rp thread to shutdown */
1892 qemu_file_shutdown(s->rp_state.from_dst_file);
1893 }
1894
1895 do {
1896 old_state = s->state;
1897 if (!migration_is_running(old_state)) {
1898 break;
1899 }
1900 /* If the migration is paused, kick it out of the pause */
1901 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1902 qemu_sem_post(&s->pause_sem);
1903 }
1904 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1905 } while (s->state != MIGRATION_STATUS_CANCELLING);
1906
1907 /*
1908 * If we're unlucky the migration code might be stuck somewhere in a
1909 * send/write while the network has failed and is waiting to timeout;
1910 * if we've got shutdown(2) available then we can force it to quit.
1911 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1912 * called in a bh, so there is no race against this cancel.
1913 */
1914 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1915 qemu_file_shutdown(f);
1916 }
1917 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1918 Error *local_err = NULL;
1919
1920 bdrv_invalidate_cache_all(&local_err);
1921 if (local_err) {
1922 error_report_err(local_err);
1923 } else {
1924 s->block_inactive = false;
1925 }
1926 }
1927 }
1928
1929 void add_migration_state_change_notifier(Notifier *notify)
1930 {
1931 notifier_list_add(&migration_state_notifiers, notify);
1932 }
1933
1934 void remove_migration_state_change_notifier(Notifier *notify)
1935 {
1936 notifier_remove(notify);
1937 }
1938
1939 bool migration_in_setup(MigrationState *s)
1940 {
1941 return s->state == MIGRATION_STATUS_SETUP;
1942 }
1943
1944 bool migration_has_finished(MigrationState *s)
1945 {
1946 return s->state == MIGRATION_STATUS_COMPLETED;
1947 }
1948
1949 bool migration_has_failed(MigrationState *s)
1950 {
1951 return (s->state == MIGRATION_STATUS_CANCELLED ||
1952 s->state == MIGRATION_STATUS_FAILED);
1953 }
1954
1955 bool migration_in_postcopy(void)
1956 {
1957 MigrationState *s = migrate_get_current();
1958
1959 switch (s->state) {
1960 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1961 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1962 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1963 return true;
1964 default:
1965 return false;
1966 }
1967 }
1968
1969 bool migration_in_postcopy_after_devices(MigrationState *s)
1970 {
1971 return migration_in_postcopy() && s->postcopy_after_devices;
1972 }
1973
1974 bool migration_in_incoming_postcopy(void)
1975 {
1976 PostcopyState ps = postcopy_state_get();
1977
1978 return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
1979 }
1980
1981 bool migration_is_idle(void)
1982 {
1983 MigrationState *s = current_migration;
1984
1985 if (!s) {
1986 return true;
1987 }
1988
1989 switch (s->state) {
1990 case MIGRATION_STATUS_NONE:
1991 case MIGRATION_STATUS_CANCELLED:
1992 case MIGRATION_STATUS_COMPLETED:
1993 case MIGRATION_STATUS_FAILED:
1994 return true;
1995 case MIGRATION_STATUS_SETUP:
1996 case MIGRATION_STATUS_CANCELLING:
1997 case MIGRATION_STATUS_ACTIVE:
1998 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1999 case MIGRATION_STATUS_COLO:
2000 case MIGRATION_STATUS_PRE_SWITCHOVER:
2001 case MIGRATION_STATUS_DEVICE:
2002 case MIGRATION_STATUS_WAIT_UNPLUG:
2003 return false;
2004 case MIGRATION_STATUS__MAX:
2005 g_assert_not_reached();
2006 }
2007
2008 return false;
2009 }
2010
2011 bool migration_is_active(MigrationState *s)
2012 {
2013 return (s->state == MIGRATION_STATUS_ACTIVE ||
2014 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2015 }
2016
2017 void migrate_init(MigrationState *s)
2018 {
2019 /*
2020 * Reinitialise all migration state, except
2021 * parameters/capabilities that the user set, and
2022 * locks.
2023 */
2024 s->cleanup_bh = 0;
2025 s->vm_start_bh = 0;
2026 s->to_dst_file = NULL;
2027 s->state = MIGRATION_STATUS_NONE;
2028 s->rp_state.from_dst_file = NULL;
2029 s->rp_state.error = false;
2030 s->mbps = 0.0;
2031 s->pages_per_second = 0.0;
2032 s->downtime = 0;
2033 s->expected_downtime = 0;
2034 s->setup_time = 0;
2035 s->start_postcopy = false;
2036 s->postcopy_after_devices = false;
2037 s->migration_thread_running = false;
2038 error_free(s->error);
2039 s->error = NULL;
2040 s->hostname = NULL;
2041
2042 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
2043
2044 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2045 s->total_time = 0;
2046 s->vm_was_running = false;
2047 s->iteration_initial_bytes = 0;
2048 s->threshold_size = 0;
2049 }
2050
2051 int migrate_add_blocker(Error *reason, Error **errp)
2052 {
2053 if (only_migratable) {
2054 error_propagate_prepend(errp, error_copy(reason),
2055 "disallowing migration blocker "
2056 "(--only-migratable) for: ");
2057 return -EACCES;
2058 }
2059
2060 if (migration_is_idle()) {
2061 migration_blockers = g_slist_prepend(migration_blockers, reason);
2062 return 0;
2063 }
2064
2065 error_propagate_prepend(errp, error_copy(reason),
2066 "disallowing migration blocker "
2067 "(migration in progress) for: ");
2068 return -EBUSY;
2069 }
2070
2071 void migrate_del_blocker(Error *reason)
2072 {
2073 migration_blockers = g_slist_remove(migration_blockers, reason);
2074 }
2075
2076 void qmp_migrate_incoming(const char *uri, Error **errp)
2077 {
2078 Error *local_err = NULL;
2079 static bool once = true;
2080
2081 if (!once) {
2082 error_setg(errp, "The incoming migration has already been started");
2083 return;
2084 }
2085 if (!runstate_check(RUN_STATE_INMIGRATE)) {
2086 error_setg(errp, "'-incoming' was not specified on the command line");
2087 return;
2088 }
2089
2090 qemu_start_incoming_migration(uri, &local_err);
2091
2092 if (local_err) {
2093 error_propagate(errp, local_err);
2094 return;
2095 }
2096
2097 once = false;
2098 }
2099
2100 void qmp_migrate_recover(const char *uri, Error **errp)
2101 {
2102 MigrationIncomingState *mis = migration_incoming_get_current();
2103
2104 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2105 error_setg(errp, "Migrate recover can only be run "
2106 "when postcopy is paused.");
2107 return;
2108 }
2109
2110 if (qatomic_cmpxchg(&mis->postcopy_recover_triggered,
2111 false, true) == true) {
2112 error_setg(errp, "Migrate recovery is triggered already");
2113 return;
2114 }
2115
2116 /*
2117 * Note that this call will never start a real migration; it will
2118 * only re-setup the migration stream and poke existing migration
2119 * to continue using that newly established channel.
2120 */
2121 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2122 qemu_start_incoming_migration(uri, errp);
2123 }
2124
2125 void qmp_migrate_pause(Error **errp)
2126 {
2127 MigrationState *ms = migrate_get_current();
2128 MigrationIncomingState *mis = migration_incoming_get_current();
2129 int ret;
2130
2131 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2132 /* Source side, during postcopy */
2133 qemu_mutex_lock(&ms->qemu_file_lock);
2134 ret = qemu_file_shutdown(ms->to_dst_file);
2135 qemu_mutex_unlock(&ms->qemu_file_lock);
2136 if (ret) {
2137 error_setg(errp, "Failed to pause source migration");
2138 }
2139 return;
2140 }
2141
2142 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2143 ret = qemu_file_shutdown(mis->from_src_file);
2144 if (ret) {
2145 error_setg(errp, "Failed to pause destination migration");
2146 }
2147 return;
2148 }
2149
2150 error_setg(errp, "migrate-pause is currently only supported "
2151 "during postcopy-active state");
2152 }
2153
2154 bool migration_is_blocked(Error **errp)
2155 {
2156 if (qemu_savevm_state_blocked(errp)) {
2157 return true;
2158 }
2159
2160 if (migration_blockers) {
2161 error_propagate(errp, error_copy(migration_blockers->data));
2162 return true;
2163 }
2164
2165 return false;
2166 }
2167
2168 /* Returns true if continue to migrate, or false if error detected */
2169 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
2170 bool resume, Error **errp)
2171 {
2172 Error *local_err = NULL;
2173
2174 if (resume) {
2175 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2176 error_setg(errp, "Cannot resume if there is no "
2177 "paused migration");
2178 return false;
2179 }
2180
2181 /*
2182 * Postcopy recovery won't work well with release-ram
2183 * capability since release-ram will drop the page buffer as
2184 * long as the page is put into the send buffer. So if there
2185 * is a network failure happened, any page buffers that have
2186 * not yet reached the destination VM but have already been
2187 * sent from the source VM will be lost forever. Let's refuse
2188 * the client from resuming such a postcopy migration.
2189 * Luckily release-ram was designed to only be used when src
2190 * and destination VMs are on the same host, so it should be
2191 * fine.
2192 */
2193 if (migrate_release_ram()) {
2194 error_setg(errp, "Postcopy recovery cannot work "
2195 "when release-ram capability is set");
2196 return false;
2197 }
2198
2199 /* This is a resume, skip init status */
2200 return true;
2201 }
2202
2203 if (migration_is_running(s->state)) {
2204 error_setg(errp, QERR_MIGRATION_ACTIVE);
2205 return false;
2206 }
2207
2208 if (runstate_check(RUN_STATE_INMIGRATE)) {
2209 error_setg(errp, "Guest is waiting for an incoming migration");
2210 return false;
2211 }
2212
2213 if (runstate_check(RUN_STATE_POSTMIGRATE)) {
2214 error_setg(errp, "Can't migrate the vm that was paused due to "
2215 "previous migration");
2216 return false;
2217 }
2218
2219 if (migration_is_blocked(errp)) {
2220 return false;
2221 }
2222
2223 if (blk || blk_inc) {
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 migrate_fd_cancel(migrate_get_current());
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 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
2322 {
2323 MigrateSetParameters p = {
2324 .has_xbzrle_cache_size = true,
2325 .xbzrle_cache_size = value,
2326 };
2327
2328 qmp_migrate_set_parameters(&p, errp);
2329 }
2330
2331 uint64_t qmp_query_migrate_cache_size(Error **errp)
2332 {
2333 return migrate_xbzrle_cache_size();
2334 }
2335
2336 void qmp_migrate_set_speed(int64_t value, Error **errp)
2337 {
2338 MigrateSetParameters p = {
2339 .has_max_bandwidth = true,
2340 .max_bandwidth = value,
2341 };
2342
2343 qmp_migrate_set_parameters(&p, errp);
2344 }
2345
2346 void qmp_migrate_set_downtime(double value, Error **errp)
2347 {
2348 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2349 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
2350 "downtime_limit",
2351 "an integer in the range of 0 to "
2352 stringify(MAX_MIGRATE_DOWNTIME_SECONDS)" seconds");
2353 return;
2354 }
2355
2356 value *= 1000; /* Convert to milliseconds */
2357
2358 MigrateSetParameters p = {
2359 .has_downtime_limit = true,
2360 .downtime_limit = (int64_t)value,
2361 };
2362
2363 qmp_migrate_set_parameters(&p, errp);
2364 }
2365
2366 bool migrate_release_ram(void)
2367 {
2368 MigrationState *s;
2369
2370 s = migrate_get_current();
2371
2372 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2373 }
2374
2375 bool migrate_postcopy_ram(void)
2376 {
2377 MigrationState *s;
2378
2379 s = migrate_get_current();
2380
2381 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2382 }
2383
2384 bool migrate_postcopy(void)
2385 {
2386 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2387 }
2388
2389 bool migrate_auto_converge(void)
2390 {
2391 MigrationState *s;
2392
2393 s = migrate_get_current();
2394
2395 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2396 }
2397
2398 bool migrate_zero_blocks(void)
2399 {
2400 MigrationState *s;
2401
2402 s = migrate_get_current();
2403
2404 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2405 }
2406
2407 bool migrate_postcopy_blocktime(void)
2408 {
2409 MigrationState *s;
2410
2411 s = migrate_get_current();
2412
2413 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2414 }
2415
2416 bool migrate_use_compression(void)
2417 {
2418 MigrationState *s;
2419
2420 s = migrate_get_current();
2421
2422 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2423 }
2424
2425 int migrate_compress_level(void)
2426 {
2427 MigrationState *s;
2428
2429 s = migrate_get_current();
2430
2431 return s->parameters.compress_level;
2432 }
2433
2434 int migrate_compress_threads(void)
2435 {
2436 MigrationState *s;
2437
2438 s = migrate_get_current();
2439
2440 return s->parameters.compress_threads;
2441 }
2442
2443 int migrate_compress_wait_thread(void)
2444 {
2445 MigrationState *s;
2446
2447 s = migrate_get_current();
2448
2449 return s->parameters.compress_wait_thread;
2450 }
2451
2452 int migrate_decompress_threads(void)
2453 {
2454 MigrationState *s;
2455
2456 s = migrate_get_current();
2457
2458 return s->parameters.decompress_threads;
2459 }
2460
2461 bool migrate_dirty_bitmaps(void)
2462 {
2463 MigrationState *s;
2464
2465 s = migrate_get_current();
2466
2467 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2468 }
2469
2470 bool migrate_ignore_shared(void)
2471 {
2472 MigrationState *s;
2473
2474 s = migrate_get_current();
2475
2476 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2477 }
2478
2479 bool migrate_validate_uuid(void)
2480 {
2481 MigrationState *s;
2482
2483 s = migrate_get_current();
2484
2485 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2486 }
2487
2488 bool migrate_use_events(void)
2489 {
2490 MigrationState *s;
2491
2492 s = migrate_get_current();
2493
2494 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2495 }
2496
2497 bool migrate_use_multifd(void)
2498 {
2499 MigrationState *s;
2500
2501 s = migrate_get_current();
2502
2503 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2504 }
2505
2506 bool migrate_pause_before_switchover(void)
2507 {
2508 MigrationState *s;
2509
2510 s = migrate_get_current();
2511
2512 return s->enabled_capabilities[
2513 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2514 }
2515
2516 int migrate_multifd_channels(void)
2517 {
2518 MigrationState *s;
2519
2520 s = migrate_get_current();
2521
2522 return s->parameters.multifd_channels;
2523 }
2524
2525 MultiFDCompression migrate_multifd_compression(void)
2526 {
2527 MigrationState *s;
2528
2529 s = migrate_get_current();
2530
2531 return s->parameters.multifd_compression;
2532 }
2533
2534 int migrate_multifd_zlib_level(void)
2535 {
2536 MigrationState *s;
2537
2538 s = migrate_get_current();
2539
2540 return s->parameters.multifd_zlib_level;
2541 }
2542
2543 int migrate_multifd_zstd_level(void)
2544 {
2545 MigrationState *s;
2546
2547 s = migrate_get_current();
2548
2549 return s->parameters.multifd_zstd_level;
2550 }
2551
2552 int migrate_use_xbzrle(void)
2553 {
2554 MigrationState *s;
2555
2556 s = migrate_get_current();
2557
2558 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2559 }
2560
2561 uint64_t migrate_xbzrle_cache_size(void)
2562 {
2563 MigrationState *s;
2564
2565 s = migrate_get_current();
2566
2567 return s->parameters.xbzrle_cache_size;
2568 }
2569
2570 static int64_t migrate_max_postcopy_bandwidth(void)
2571 {
2572 MigrationState *s;
2573
2574 s = migrate_get_current();
2575
2576 return s->parameters.max_postcopy_bandwidth;
2577 }
2578
2579 bool migrate_use_block(void)
2580 {
2581 MigrationState *s;
2582
2583 s = migrate_get_current();
2584
2585 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2586 }
2587
2588 bool migrate_use_return_path(void)
2589 {
2590 MigrationState *s;
2591
2592 s = migrate_get_current();
2593
2594 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2595 }
2596
2597 bool migrate_use_block_incremental(void)
2598 {
2599 MigrationState *s;
2600
2601 s = migrate_get_current();
2602
2603 return s->parameters.block_incremental;
2604 }
2605
2606 bool migrate_background_snapshot(void)
2607 {
2608 MigrationState *s;
2609
2610 s = migrate_get_current();
2611
2612 return s->enabled_capabilities[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT];
2613 }
2614
2615 /* migration thread support */
2616 /*
2617 * Something bad happened to the RP stream, mark an error
2618 * The caller shall print or trace something to indicate why
2619 */
2620 static void mark_source_rp_bad(MigrationState *s)
2621 {
2622 s->rp_state.error = true;
2623 }
2624
2625 static struct rp_cmd_args {
2626 ssize_t len; /* -1 = variable */
2627 const char *name;
2628 } rp_cmd_args[] = {
2629 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2630 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2631 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2632 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2633 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2634 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2635 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2636 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2637 };
2638
2639 /*
2640 * Process a request for pages received on the return path,
2641 * We're allowed to send more than requested (e.g. to round to our page size)
2642 * and we don't need to send pages that have already been sent.
2643 */
2644 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2645 ram_addr_t start, size_t len)
2646 {
2647 long our_host_ps = qemu_real_host_page_size;
2648
2649 trace_migrate_handle_rp_req_pages(rbname, start, len);
2650
2651 /*
2652 * Since we currently insist on matching page sizes, just sanity check
2653 * we're being asked for whole host pages.
2654 */
2655 if (start & (our_host_ps - 1) ||
2656 (len & (our_host_ps - 1))) {
2657 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2658 " len: %zd", __func__, start, len);
2659 mark_source_rp_bad(ms);
2660 return;
2661 }
2662
2663 if (ram_save_queue_pages(rbname, start, len)) {
2664 mark_source_rp_bad(ms);
2665 }
2666 }
2667
2668 /* Return true to retry, false to quit */
2669 static bool postcopy_pause_return_path_thread(MigrationState *s)
2670 {
2671 trace_postcopy_pause_return_path();
2672
2673 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2674
2675 trace_postcopy_pause_return_path_continued();
2676
2677 return true;
2678 }
2679
2680 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2681 {
2682 RAMBlock *block = qemu_ram_block_by_name(block_name);
2683
2684 if (!block) {
2685 error_report("%s: invalid block name '%s'", __func__, block_name);
2686 return -EINVAL;
2687 }
2688
2689 /* Fetch the received bitmap and refresh the dirty bitmap */
2690 return ram_dirty_bitmap_reload(s, block);
2691 }
2692
2693 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2694 {
2695 trace_source_return_path_thread_resume_ack(value);
2696
2697 if (value != MIGRATION_RESUME_ACK_VALUE) {
2698 error_report("%s: illegal resume_ack value %"PRIu32,
2699 __func__, value);
2700 return -1;
2701 }
2702
2703 /* Now both sides are active. */
2704 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2705 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2706
2707 /* Notify send thread that time to continue send pages */
2708 qemu_sem_post(&s->rp_state.rp_sem);
2709
2710 return 0;
2711 }
2712
2713 /*
2714 * Handles messages sent on the return path towards the source VM
2715 *
2716 */
2717 static void *source_return_path_thread(void *opaque)
2718 {
2719 MigrationState *ms = opaque;
2720 QEMUFile *rp = ms->rp_state.from_dst_file;
2721 uint16_t header_len, header_type;
2722 uint8_t buf[512];
2723 uint32_t tmp32, sibling_error;
2724 ram_addr_t start = 0; /* =0 to silence warning */
2725 size_t len = 0, expected_len;
2726 int res;
2727
2728 trace_source_return_path_thread_entry();
2729 rcu_register_thread();
2730
2731 retry:
2732 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2733 migration_is_setup_or_active(ms->state)) {
2734 trace_source_return_path_thread_loop_top();
2735 header_type = qemu_get_be16(rp);
2736 header_len = qemu_get_be16(rp);
2737
2738 if (qemu_file_get_error(rp)) {
2739 mark_source_rp_bad(ms);
2740 goto out;
2741 }
2742
2743 if (header_type >= MIG_RP_MSG_MAX ||
2744 header_type == MIG_RP_MSG_INVALID) {
2745 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2746 header_type, header_len);
2747 mark_source_rp_bad(ms);
2748 goto out;
2749 }
2750
2751 if ((rp_cmd_args[header_type].len != -1 &&
2752 header_len != rp_cmd_args[header_type].len) ||
2753 header_len > sizeof(buf)) {
2754 error_report("RP: Received '%s' message (0x%04x) with"
2755 "incorrect length %d expecting %zu",
2756 rp_cmd_args[header_type].name, header_type, header_len,
2757 (size_t)rp_cmd_args[header_type].len);
2758 mark_source_rp_bad(ms);
2759 goto out;
2760 }
2761
2762 /* We know we've got a valid header by this point */
2763 res = qemu_get_buffer(rp, buf, header_len);
2764 if (res != header_len) {
2765 error_report("RP: Failed reading data for message 0x%04x"
2766 " read %d expected %d",
2767 header_type, res, header_len);
2768 mark_source_rp_bad(ms);
2769 goto out;
2770 }
2771
2772 /* OK, we have the message and the data */
2773 switch (header_type) {
2774 case MIG_RP_MSG_SHUT:
2775 sibling_error = ldl_be_p(buf);
2776 trace_source_return_path_thread_shut(sibling_error);
2777 if (sibling_error) {
2778 error_report("RP: Sibling indicated error %d", sibling_error);
2779 mark_source_rp_bad(ms);
2780 }
2781 /*
2782 * We'll let the main thread deal with closing the RP
2783 * we could do a shutdown(2) on it, but we're the only user
2784 * anyway, so there's nothing gained.
2785 */
2786 goto out;
2787
2788 case MIG_RP_MSG_PONG:
2789 tmp32 = ldl_be_p(buf);
2790 trace_source_return_path_thread_pong(tmp32);
2791 break;
2792
2793 case MIG_RP_MSG_REQ_PAGES:
2794 start = ldq_be_p(buf);
2795 len = ldl_be_p(buf + 8);
2796 migrate_handle_rp_req_pages(ms, NULL, start, len);
2797 break;
2798
2799 case MIG_RP_MSG_REQ_PAGES_ID:
2800 expected_len = 12 + 1; /* header + termination */
2801
2802 if (header_len >= expected_len) {
2803 start = ldq_be_p(buf);
2804 len = ldl_be_p(buf + 8);
2805 /* Now we expect an idstr */
2806 tmp32 = buf[12]; /* Length of the following idstr */
2807 buf[13 + tmp32] = '\0';
2808 expected_len += tmp32;
2809 }
2810 if (header_len != expected_len) {
2811 error_report("RP: Req_Page_id with length %d expecting %zd",
2812 header_len, expected_len);
2813 mark_source_rp_bad(ms);
2814 goto out;
2815 }
2816 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2817 break;
2818
2819 case MIG_RP_MSG_RECV_BITMAP:
2820 if (header_len < 1) {
2821 error_report("%s: missing block name", __func__);
2822 mark_source_rp_bad(ms);
2823 goto out;
2824 }
2825 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2826 buf[buf[0] + 1] = '\0';
2827 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2828 mark_source_rp_bad(ms);
2829 goto out;
2830 }
2831 break;
2832
2833 case MIG_RP_MSG_RESUME_ACK:
2834 tmp32 = ldl_be_p(buf);
2835 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2836 mark_source_rp_bad(ms);
2837 goto out;
2838 }
2839 break;
2840
2841 default:
2842 break;
2843 }
2844 }
2845
2846 out:
2847 res = qemu_file_get_error(rp);
2848 if (res) {
2849 if (res == -EIO && migration_in_postcopy()) {
2850 /*
2851 * Maybe there is something we can do: it looks like a
2852 * network down issue, and we pause for a recovery.
2853 */
2854 if (postcopy_pause_return_path_thread(ms)) {
2855 /* Reload rp, reset the rest */
2856 if (rp != ms->rp_state.from_dst_file) {
2857 qemu_fclose(rp);
2858 rp = ms->rp_state.from_dst_file;
2859 }
2860 ms->rp_state.error = false;
2861 goto retry;
2862 }
2863 }
2864
2865 trace_source_return_path_thread_bad_end();
2866 mark_source_rp_bad(ms);
2867 }
2868
2869 trace_source_return_path_thread_end();
2870 ms->rp_state.from_dst_file = NULL;
2871 qemu_fclose(rp);
2872 rcu_unregister_thread();
2873 return NULL;
2874 }
2875
2876 static int open_return_path_on_source(MigrationState *ms,
2877 bool create_thread)
2878 {
2879
2880 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2881 if (!ms->rp_state.from_dst_file) {
2882 return -1;
2883 }
2884
2885 trace_open_return_path_on_source();
2886
2887 if (!create_thread) {
2888 /* We're done */
2889 return 0;
2890 }
2891
2892 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2893 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2894
2895 trace_open_return_path_on_source_continue();
2896
2897 return 0;
2898 }
2899
2900 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2901 static int await_return_path_close_on_source(MigrationState *ms)
2902 {
2903 /*
2904 * If this is a normal exit then the destination will send a SHUT and the
2905 * rp_thread will exit, however if there's an error we need to cause
2906 * it to exit.
2907 */
2908 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2909 /*
2910 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2911 * waiting for the destination.
2912 */
2913 qemu_file_shutdown(ms->rp_state.from_dst_file);
2914 mark_source_rp_bad(ms);
2915 }
2916 trace_await_return_path_close_on_source_joining();
2917 qemu_thread_join(&ms->rp_state.rp_thread);
2918 trace_await_return_path_close_on_source_close();
2919 return ms->rp_state.error;
2920 }
2921
2922 /*
2923 * Switch from normal iteration to postcopy
2924 * Returns non-0 on error
2925 */
2926 static int postcopy_start(MigrationState *ms)
2927 {
2928 int ret;
2929 QIOChannelBuffer *bioc;
2930 QEMUFile *fb;
2931 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2932 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2933 bool restart_block = false;
2934 int cur_state = MIGRATION_STATUS_ACTIVE;
2935 if (!migrate_pause_before_switchover()) {
2936 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2937 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2938 }
2939
2940 trace_postcopy_start();
2941 qemu_mutex_lock_iothread();
2942 trace_postcopy_start_set_run();
2943
2944 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2945 global_state_store();
2946 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2947 if (ret < 0) {
2948 goto fail;
2949 }
2950
2951 ret = migration_maybe_pause(ms, &cur_state,
2952 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2953 if (ret < 0) {
2954 goto fail;
2955 }
2956
2957 ret = bdrv_inactivate_all();
2958 if (ret < 0) {
2959 goto fail;
2960 }
2961 restart_block = true;
2962
2963 /*
2964 * Cause any non-postcopiable, but iterative devices to
2965 * send out their final data.
2966 */
2967 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2968
2969 /*
2970 * in Finish migrate and with the io-lock held everything should
2971 * be quiet, but we've potentially still got dirty pages and we
2972 * need to tell the destination to throw any pages it's already received
2973 * that are dirty
2974 */
2975 if (migrate_postcopy_ram()) {
2976 if (ram_postcopy_send_discard_bitmap(ms)) {
2977 error_report("postcopy send discard bitmap failed");
2978 goto fail;
2979 }
2980 }
2981
2982 /*
2983 * send rest of state - note things that are doing postcopy
2984 * will notice we're in POSTCOPY_ACTIVE and not actually
2985 * wrap their state up here
2986 */
2987 /* 0 max-postcopy-bandwidth means unlimited */
2988 if (!bandwidth) {
2989 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2990 } else {
2991 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2992 }
2993 if (migrate_postcopy_ram()) {
2994 /* Ping just for debugging, helps line traces up */
2995 qemu_savevm_send_ping(ms->to_dst_file, 2);
2996 }
2997
2998 /*
2999 * While loading the device state we may trigger page transfer
3000 * requests and the fd must be free to process those, and thus
3001 * the destination must read the whole device state off the fd before
3002 * it starts processing it. Unfortunately the ad-hoc migration format
3003 * doesn't allow the destination to know the size to read without fully
3004 * parsing it through each devices load-state code (especially the open
3005 * coded devices that use get/put).
3006 * So we wrap the device state up in a package with a length at the start;
3007 * to do this we use a qemu_buf to hold the whole of the device state.
3008 */
3009 bioc = qio_channel_buffer_new(4096);
3010 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
3011 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
3012 object_unref(OBJECT(bioc));
3013
3014 /*
3015 * Make sure the receiver can get incoming pages before we send the rest
3016 * of the state
3017 */
3018 qemu_savevm_send_postcopy_listen(fb);
3019
3020 qemu_savevm_state_complete_precopy(fb, false, false);
3021 if (migrate_postcopy_ram()) {
3022 qemu_savevm_send_ping(fb, 3);
3023 }
3024
3025 qemu_savevm_send_postcopy_run(fb);
3026
3027 /* <><> end of stuff going into the package */
3028
3029 /* Last point of recovery; as soon as we send the package the destination
3030 * can open devices and potentially start running.
3031 * Lets just check again we've not got any errors.
3032 */
3033 ret = qemu_file_get_error(ms->to_dst_file);
3034 if (ret) {
3035 error_report("postcopy_start: Migration stream errored (pre package)");
3036 goto fail_closefb;
3037 }
3038
3039 restart_block = false;
3040
3041 /* Now send that blob */
3042 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
3043 goto fail_closefb;
3044 }
3045 qemu_fclose(fb);
3046
3047 /* Send a notify to give a chance for anything that needs to happen
3048 * at the transition to postcopy and after the device state; in particular
3049 * spice needs to trigger a transition now
3050 */
3051 ms->postcopy_after_devices = true;
3052 notifier_list_notify(&migration_state_notifiers, ms);
3053
3054 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
3055
3056 qemu_mutex_unlock_iothread();
3057
3058 if (migrate_postcopy_ram()) {
3059 /*
3060 * Although this ping is just for debug, it could potentially be
3061 * used for getting a better measurement of downtime at the source.
3062 */
3063 qemu_savevm_send_ping(ms->to_dst_file, 4);
3064 }
3065
3066 if (migrate_release_ram()) {
3067 ram_postcopy_migrated_memory_release(ms);
3068 }
3069
3070 ret = qemu_file_get_error(ms->to_dst_file);
3071 if (ret) {
3072 error_report("postcopy_start: Migration stream errored");
3073 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
3074 MIGRATION_STATUS_FAILED);
3075 }
3076
3077 return ret;
3078
3079 fail_closefb:
3080 qemu_fclose(fb);
3081 fail:
3082 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
3083 MIGRATION_STATUS_FAILED);
3084 if (restart_block) {
3085 /* A failure happened early enough that we know the destination hasn't
3086 * accessed block devices, so we're safe to recover.
3087 */
3088 Error *local_err = NULL;
3089
3090 bdrv_invalidate_cache_all(&local_err);
3091 if (local_err) {
3092 error_report_err(local_err);
3093 }
3094 }
3095 qemu_mutex_unlock_iothread();
3096 return -1;
3097 }
3098
3099 /**
3100 * migration_maybe_pause: Pause if required to by
3101 * migrate_pause_before_switchover called with the iothread locked
3102 * Returns: 0 on success
3103 */
3104 static int migration_maybe_pause(MigrationState *s,
3105 int *current_active_state,
3106 int new_state)
3107 {
3108 if (!migrate_pause_before_switchover()) {
3109 return 0;
3110 }
3111
3112 /* Since leaving this state is not atomic with posting the semaphore
3113 * it's possible that someone could have issued multiple migrate_continue
3114 * and the semaphore is incorrectly positive at this point;
3115 * the docs say it's undefined to reinit a semaphore that's already
3116 * init'd, so use timedwait to eat up any existing posts.
3117 */
3118 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
3119 /* This block intentionally left blank */
3120 }
3121
3122 /*
3123 * If the migration is cancelled when it is in the completion phase,
3124 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3125 * So we don't need to wait a semaphore, otherwise we would always
3126 * wait for the 'pause_sem' semaphore.
3127 */
3128 if (s->state != MIGRATION_STATUS_CANCELLING) {
3129 qemu_mutex_unlock_iothread();
3130 migrate_set_state(&s->state, *current_active_state,
3131 MIGRATION_STATUS_PRE_SWITCHOVER);
3132 qemu_sem_wait(&s->pause_sem);
3133 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
3134 new_state);
3135 *current_active_state = new_state;
3136 qemu_mutex_lock_iothread();
3137 }
3138
3139 return s->state == new_state ? 0 : -EINVAL;
3140 }
3141
3142 /**
3143 * migration_completion: Used by migration_thread when there's not much left.
3144 * The caller 'breaks' the loop when this returns.
3145 *
3146 * @s: Current migration state
3147 */
3148 static void migration_completion(MigrationState *s)
3149 {
3150 int ret;
3151 int current_active_state = s->state;
3152
3153 if (s->state == MIGRATION_STATUS_ACTIVE) {
3154 qemu_mutex_lock_iothread();
3155 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3156 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3157 s->vm_was_running = runstate_is_running();
3158 ret = global_state_store();
3159
3160 if (!ret) {
3161 bool inactivate = !migrate_colo_enabled();
3162 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
3163 if (ret >= 0) {
3164 ret = migration_maybe_pause(s, &current_active_state,
3165 MIGRATION_STATUS_DEVICE);
3166 }
3167 if (ret >= 0) {
3168 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3169 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
3170 inactivate);
3171 }
3172 if (inactivate && ret >= 0) {
3173 s->block_inactive = true;
3174 }
3175 }
3176 qemu_mutex_unlock_iothread();
3177
3178 if (ret < 0) {
3179 goto fail;
3180 }
3181 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3182 trace_migration_completion_postcopy_end();
3183
3184 qemu_savevm_state_complete_postcopy(s->to_dst_file);
3185 trace_migration_completion_postcopy_end_after_complete();
3186 } else if (s->state == MIGRATION_STATUS_CANCELLING) {
3187 goto fail;
3188 }
3189
3190 /*
3191 * If rp was opened we must clean up the thread before
3192 * cleaning everything else up (since if there are no failures
3193 * it will wait for the destination to send it's status in
3194 * a SHUT command).
3195 */
3196 if (s->rp_state.from_dst_file) {
3197 int rp_error;
3198 trace_migration_return_path_end_before();
3199 rp_error = await_return_path_close_on_source(s);
3200 trace_migration_return_path_end_after(rp_error);
3201 if (rp_error) {
3202 goto fail_invalidate;
3203 }
3204 }
3205
3206 if (qemu_file_get_error(s->to_dst_file)) {
3207 trace_migration_completion_file_err();
3208 goto fail_invalidate;
3209 }
3210
3211 if (!migrate_colo_enabled()) {
3212 migrate_set_state(&s->state, current_active_state,
3213 MIGRATION_STATUS_COMPLETED);
3214 }
3215
3216 return;
3217
3218 fail_invalidate:
3219 /* If not doing postcopy, vm_start() will be called: let's regain
3220 * control on images.
3221 */
3222 if (s->state == MIGRATION_STATUS_ACTIVE ||
3223 s->state == MIGRATION_STATUS_DEVICE) {
3224 Error *local_err = NULL;
3225
3226 qemu_mutex_lock_iothread();
3227 bdrv_invalidate_cache_all(&local_err);
3228 if (local_err) {
3229 error_report_err(local_err);
3230 } else {
3231 s->block_inactive = false;
3232 }
3233 qemu_mutex_unlock_iothread();
3234 }
3235
3236 fail:
3237 migrate_set_state(&s->state, current_active_state,
3238 MIGRATION_STATUS_FAILED);
3239 }
3240
3241 /**
3242 * bg_migration_completion: Used by bg_migration_thread when after all the
3243 * RAM has been saved. The caller 'breaks' the loop when this returns.
3244 *
3245 * @s: Current migration state
3246 */
3247 static void bg_migration_completion(MigrationState *s)
3248 {
3249 int current_active_state = s->state;
3250
3251 /*
3252 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3253 * memory ranges, flush kernel wait queues and wake up threads
3254 * waiting for write fault to be resolved.
3255 */
3256 ram_write_tracking_stop();
3257
3258 if (s->state == MIGRATION_STATUS_ACTIVE) {
3259 /*
3260 * By this moment we have RAM content saved into the migration stream.
3261 * The next step is to flush the non-RAM content (device state)
3262 * right after the ram content. The device state has been stored into
3263 * the temporary buffer before RAM saving started.
3264 */
3265 qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage);
3266 qemu_fflush(s->to_dst_file);
3267 } else if (s->state == MIGRATION_STATUS_CANCELLING) {
3268 goto fail;
3269 }
3270
3271 if (qemu_file_get_error(s->to_dst_file)) {
3272 trace_migration_completion_file_err();
3273 goto fail;
3274 }
3275
3276 migrate_set_state(&s->state, current_active_state,
3277 MIGRATION_STATUS_COMPLETED);
3278 return;
3279
3280 fail:
3281 migrate_set_state(&s->state, current_active_state,
3282 MIGRATION_STATUS_FAILED);
3283 }
3284
3285 bool migrate_colo_enabled(void)
3286 {
3287 MigrationState *s = migrate_get_current();
3288 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
3289 }
3290
3291 typedef enum MigThrError {
3292 /* No error detected */
3293 MIG_THR_ERR_NONE = 0,
3294 /* Detected error, but resumed successfully */
3295 MIG_THR_ERR_RECOVERED = 1,
3296 /* Detected fatal error, need to exit */
3297 MIG_THR_ERR_FATAL = 2,
3298 } MigThrError;
3299
3300 static int postcopy_resume_handshake(MigrationState *s)
3301 {
3302 qemu_savevm_send_postcopy_resume(s->to_dst_file);
3303
3304 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3305 qemu_sem_wait(&s->rp_state.rp_sem);
3306 }
3307
3308 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3309 return 0;
3310 }
3311
3312 return -1;
3313 }
3314
3315 /* Return zero if success, or <0 for error */
3316 static int postcopy_do_resume(MigrationState *s)
3317 {
3318 int ret;
3319
3320 /*
3321 * Call all the resume_prepare() hooks, so that modules can be
3322 * ready for the migration resume.
3323 */
3324 ret = qemu_savevm_state_resume_prepare(s);
3325 if (ret) {
3326 error_report("%s: resume_prepare() failure detected: %d",
3327 __func__, ret);
3328 return ret;
3329 }
3330
3331 /*
3332 * Last handshake with destination on the resume (destination will
3333 * switch to postcopy-active afterwards)
3334 */
3335 ret = postcopy_resume_handshake(s);
3336 if (ret) {
3337 error_report("%s: handshake failed: %d", __func__, ret);
3338 return ret;
3339 }
3340
3341 return 0;
3342 }
3343
3344 /*
3345 * We don't return until we are in a safe state to continue current
3346 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3347 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3348 */
3349 static MigThrError postcopy_pause(MigrationState *s)
3350 {
3351 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3352
3353 while (true) {
3354 QEMUFile *file;
3355
3356 /* Current channel is possibly broken. Release it. */
3357 assert(s->to_dst_file);
3358 qemu_mutex_lock(&s->qemu_file_lock);
3359 file = s->to_dst_file;
3360 s->to_dst_file = NULL;
3361 qemu_mutex_unlock(&s->qemu_file_lock);
3362
3363 qemu_file_shutdown(file);
3364 qemu_fclose(file);
3365
3366 migrate_set_state(&s->state, s->state,
3367 MIGRATION_STATUS_POSTCOPY_PAUSED);
3368
3369 error_report("Detected IO failure for postcopy. "
3370 "Migration paused.");
3371
3372 /*
3373 * We wait until things fixed up. Then someone will setup the
3374 * status back for us.
3375 */
3376 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3377 qemu_sem_wait(&s->postcopy_pause_sem);
3378 }
3379
3380 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3381 /* Woken up by a recover procedure. Give it a shot */
3382
3383 /*
3384 * Firstly, let's wake up the return path now, with a new
3385 * return path channel.
3386 */
3387 qemu_sem_post(&s->postcopy_pause_rp_sem);
3388
3389 /* Do the resume logic */
3390 if (postcopy_do_resume(s) == 0) {
3391 /* Let's continue! */
3392 trace_postcopy_pause_continued();
3393 return MIG_THR_ERR_RECOVERED;
3394 } else {
3395 /*
3396 * Something wrong happened during the recovery, let's
3397 * pause again. Pause is always better than throwing
3398 * data away.
3399 */
3400 continue;
3401 }
3402 } else {
3403 /* This is not right... Time to quit. */
3404 return MIG_THR_ERR_FATAL;
3405 }
3406 }
3407 }
3408
3409 static MigThrError migration_detect_error(MigrationState *s)
3410 {
3411 int ret;
3412 int state = s->state;
3413 Error *local_error = NULL;
3414
3415 if (state == MIGRATION_STATUS_CANCELLING ||
3416 state == MIGRATION_STATUS_CANCELLED) {
3417 /* End the migration, but don't set the state to failed */
3418 return MIG_THR_ERR_FATAL;
3419 }
3420
3421 /* Try to detect any file errors */
3422 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3423 if (!ret) {
3424 /* Everything is fine */
3425 assert(!local_error);
3426 return MIG_THR_ERR_NONE;
3427 }
3428
3429 if (local_error) {
3430 migrate_set_error(s, local_error);
3431 error_free(local_error);
3432 }
3433
3434 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3435 /*
3436 * For postcopy, we allow the network to be down for a
3437 * while. After that, it can be continued by a
3438 * recovery phase.
3439 */
3440 return postcopy_pause(s);
3441 } else {
3442 /*
3443 * For precopy (or postcopy with error outside IO), we fail
3444 * with no time.
3445 */
3446 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3447 trace_migration_thread_file_err();
3448
3449 /* Time to stop the migration, now. */
3450 return MIG_THR_ERR_FATAL;
3451 }
3452 }
3453
3454 /* How many bytes have we transferred since the beginning of the migration */
3455 static uint64_t migration_total_bytes(MigrationState *s)
3456 {
3457 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3458 }
3459
3460 static void migration_calculate_complete(MigrationState *s)
3461 {
3462 uint64_t bytes = migration_total_bytes(s);
3463 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3464 int64_t transfer_time;
3465
3466 s->total_time = end_time - s->start_time;
3467 if (!s->downtime) {
3468 /*
3469 * It's still not set, so we are precopy migration. For
3470 * postcopy, downtime is calculated during postcopy_start().
3471 */
3472 s->downtime = end_time - s->downtime_start;
3473 }
3474
3475 transfer_time = s->total_time - s->setup_time;
3476 if (transfer_time) {
3477 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3478 }
3479 }
3480
3481 static void update_iteration_initial_status(MigrationState *s)
3482 {
3483 /*
3484 * Update these three fields at the same time to avoid mismatch info lead
3485 * wrong speed calculation.
3486 */
3487 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3488 s->iteration_initial_bytes = migration_total_bytes(s);
3489 s->iteration_initial_pages = ram_get_total_transferred_pages();
3490 }
3491
3492 static void migration_update_counters(MigrationState *s,
3493 int64_t current_time)
3494 {
3495 uint64_t transferred, transferred_pages, time_spent;
3496 uint64_t current_bytes; /* bytes transferred since the beginning */
3497 double bandwidth;
3498
3499 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3500 return;
3501 }
3502
3503 current_bytes = migration_total_bytes(s);
3504 transferred = current_bytes - s->iteration_initial_bytes;
3505 time_spent = current_time - s->iteration_start_time;
3506 bandwidth = (double)transferred / time_spent;
3507 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3508
3509 s->mbps = (((double) transferred * 8.0) /
3510 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3511
3512 transferred_pages = ram_get_total_transferred_pages() -
3513 s->iteration_initial_pages;
3514 s->pages_per_second = (double) transferred_pages /
3515 (((double) time_spent / 1000.0));
3516
3517 /*
3518 * if we haven't sent anything, we don't want to
3519 * recalculate. 10000 is a small enough number for our purposes
3520 */
3521 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3522 s->expected_downtime = ram_counters.remaining / bandwidth;
3523 }
3524
3525 qemu_file_reset_rate_limit(s->to_dst_file);
3526
3527 update_iteration_initial_status(s);
3528
3529 trace_migrate_transferred(transferred, time_spent,
3530 bandwidth, s->threshold_size);
3531 }
3532
3533 /* Migration thread iteration status */
3534 typedef enum {
3535 MIG_ITERATE_RESUME, /* Resume current iteration */
3536 MIG_ITERATE_SKIP, /* Skip current iteration */
3537 MIG_ITERATE_BREAK, /* Break the loop */
3538 } MigIterateState;
3539
3540 /*
3541 * Return true if continue to the next iteration directly, false
3542 * otherwise.
3543 */
3544 static MigIterateState migration_iteration_run(MigrationState *s)
3545 {
3546 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3547 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3548
3549 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3550 &pend_compat, &pend_post);
3551 pending_size = pend_pre + pend_compat + pend_post;
3552
3553 trace_migrate_pending(pending_size, s->threshold_size,
3554 pend_pre, pend_compat, pend_post);
3555
3556 if (pending_size && pending_size >= s->threshold_size) {
3557 /* Still a significant amount to transfer */
3558 if (!in_postcopy && pend_pre <= s->threshold_size &&
3559 qatomic_read(&s->start_postcopy)) {
3560 if (postcopy_start(s)) {
3561 error_report("%s: postcopy failed to start", __func__);
3562 }
3563 return MIG_ITERATE_SKIP;
3564 }
3565 /* Just another iteration step */
3566 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3567 } else {
3568 trace_migration_thread_low_pending(pending_size);
3569 migration_completion(s);
3570 return MIG_ITERATE_BREAK;
3571 }
3572
3573 return MIG_ITERATE_RESUME;
3574 }
3575
3576 static void migration_iteration_finish(MigrationState *s)
3577 {
3578 /* If we enabled cpu throttling for auto-converge, turn it off. */
3579 cpu_throttle_stop();
3580
3581 qemu_mutex_lock_iothread();
3582 switch (s->state) {
3583 case MIGRATION_STATUS_COMPLETED:
3584 migration_calculate_complete(s);
3585 runstate_set(RUN_STATE_POSTMIGRATE);
3586 break;
3587
3588 case MIGRATION_STATUS_ACTIVE:
3589 /*
3590 * We should really assert here, but since it's during
3591 * migration, let's try to reduce the usage of assertions.
3592 */
3593 if (!migrate_colo_enabled()) {
3594 error_report("%s: critical error: calling COLO code without "
3595 "COLO enabled", __func__);
3596 }
3597 migrate_start_colo_process(s);
3598 /*
3599 * Fixme: we will run VM in COLO no matter its old running state.
3600 * After exited COLO, we will keep running.
3601 */
3602 s->vm_was_running = true;
3603 /* Fallthrough */
3604 case MIGRATION_STATUS_FAILED:
3605 case MIGRATION_STATUS_CANCELLED:
3606 case MIGRATION_STATUS_CANCELLING:
3607 if (s->vm_was_running) {
3608 vm_start();
3609 } else {
3610 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3611 runstate_set(RUN_STATE_POSTMIGRATE);
3612 }
3613 }
3614 break;
3615
3616 default:
3617 /* Should not reach here, but if so, forgive the VM. */
3618 error_report("%s: Unknown ending state %d", __func__, s->state);
3619 break;
3620 }
3621 migrate_fd_cleanup_schedule(s);
3622 qemu_mutex_unlock_iothread();
3623 }
3624
3625 static void bg_migration_iteration_finish(MigrationState *s)
3626 {
3627 qemu_mutex_lock_iothread();
3628 switch (s->state) {
3629 case MIGRATION_STATUS_COMPLETED:
3630 migration_calculate_complete(s);
3631 break;
3632
3633 case MIGRATION_STATUS_ACTIVE:
3634 case MIGRATION_STATUS_FAILED:
3635 case MIGRATION_STATUS_CANCELLED:
3636 case MIGRATION_STATUS_CANCELLING:
3637 break;
3638
3639 default:
3640 /* Should not reach here, but if so, forgive the VM. */
3641 error_report("%s: Unknown ending state %d", __func__, s->state);
3642 break;
3643 }
3644
3645 migrate_fd_cleanup_schedule(s);
3646 qemu_mutex_unlock_iothread();
3647 }
3648
3649 /*
3650 * Return true if continue to the next iteration directly, false
3651 * otherwise.
3652 */
3653 static MigIterateState bg_migration_iteration_run(MigrationState *s)
3654 {
3655 int res;
3656
3657 res = qemu_savevm_state_iterate(s->to_dst_file, false);
3658 if (res > 0) {
3659 bg_migration_completion(s);
3660 return MIG_ITERATE_BREAK;
3661 }
3662
3663 return MIG_ITERATE_RESUME;
3664 }
3665
3666 void migration_make_urgent_request(void)
3667 {
3668 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3669 }
3670
3671 void migration_consume_urgent_request(void)
3672 {
3673 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3674 }
3675
3676 /* Returns true if the rate limiting was broken by an urgent request */
3677 bool migration_rate_limit(void)
3678 {
3679 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3680 MigrationState *s = migrate_get_current();
3681
3682 bool urgent = false;
3683 migration_update_counters(s, now);
3684 if (qemu_file_rate_limit(s->to_dst_file)) {
3685
3686 if (qemu_file_get_error(s->to_dst_file)) {
3687 return false;
3688 }
3689 /*
3690 * Wait for a delay to do rate limiting OR
3691 * something urgent to post the semaphore.
3692 */
3693 int ms = s->iteration_start_time + BUFFER_DELAY - now;
3694 trace_migration_rate_limit_pre(ms);
3695 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3696 /*
3697 * We were woken by one or more urgent things but
3698 * the timedwait will have consumed one of them.
3699 * The service routine for the urgent wake will dec
3700 * the semaphore itself for each item it consumes,
3701 * so add this one we just eat back.
3702 */
3703 qemu_sem_post(&s->rate_limit_sem);
3704 urgent = true;
3705 }
3706 trace_migration_rate_limit_post(urgent);
3707 }
3708 return urgent;
3709 }
3710
3711 /*
3712 * Master migration thread on the source VM.
3713 * It drives the migration and pumps the data down the outgoing channel.
3714 */
3715 static void *migration_thread(void *opaque)
3716 {
3717 MigrationState *s = opaque;
3718 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3719 MigThrError thr_error;
3720 bool urgent = false;
3721
3722 rcu_register_thread();
3723
3724 object_ref(OBJECT(s));
3725 update_iteration_initial_status(s);
3726
3727 qemu_savevm_state_header(s->to_dst_file);
3728
3729 /*
3730 * If we opened the return path, we need to make sure dst has it
3731 * opened as well.
3732 */
3733 if (s->rp_state.from_dst_file) {
3734 /* Now tell the dest that it should open its end so it can reply */
3735 qemu_savevm_send_open_return_path(s->to_dst_file);
3736
3737 /* And do a ping that will make stuff easier to debug */
3738 qemu_savevm_send_ping(s->to_dst_file, 1);
3739 }
3740
3741 if (migrate_postcopy()) {
3742 /*
3743 * Tell the destination that we *might* want to do postcopy later;
3744 * if the other end can't do postcopy it should fail now, nice and
3745 * early.
3746 */
3747 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3748 }
3749
3750 if (migrate_colo_enabled()) {
3751 /* Notify migration destination that we enable COLO */
3752 qemu_savevm_send_colo_enable(s->to_dst_file);
3753 }
3754
3755 qemu_savevm_state_setup(s->to_dst_file);
3756
3757 if (qemu_savevm_state_guest_unplug_pending()) {
3758 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3759 MIGRATION_STATUS_WAIT_UNPLUG);
3760
3761 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3762 qemu_savevm_state_guest_unplug_pending()) {
3763 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3764 }
3765
3766 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3767 MIGRATION_STATUS_ACTIVE);
3768 }
3769
3770 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3771 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3772 MIGRATION_STATUS_ACTIVE);
3773
3774 trace_migration_thread_setup_complete();
3775
3776 while (migration_is_active(s)) {
3777 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3778 MigIterateState iter_state = migration_iteration_run(s);
3779 if (iter_state == MIG_ITERATE_SKIP) {
3780 continue;
3781 } else if (iter_state == MIG_ITERATE_BREAK) {
3782 break;
3783 }
3784 }
3785
3786 /*
3787 * Try to detect any kind of failures, and see whether we
3788 * should stop the migration now.
3789 */
3790 thr_error = migration_detect_error(s);
3791 if (thr_error == MIG_THR_ERR_FATAL) {
3792 /* Stop migration */
3793 break;
3794 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3795 /*
3796 * Just recovered from a e.g. network failure, reset all
3797 * the local variables. This is important to avoid
3798 * breaking transferred_bytes and bandwidth calculation
3799 */
3800 update_iteration_initial_status(s);
3801 }
3802
3803 urgent = migration_rate_limit();
3804 }
3805
3806 trace_migration_thread_after_loop();
3807 migration_iteration_finish(s);
3808 object_unref(OBJECT(s));
3809 rcu_unregister_thread();
3810 return NULL;
3811 }
3812
3813 static void bg_migration_vm_start_bh(void *opaque)
3814 {
3815 MigrationState *s = opaque;
3816
3817 qemu_bh_delete(s->vm_start_bh);
3818 s->vm_start_bh = NULL;
3819
3820 vm_start();
3821 s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start;
3822 }
3823
3824 /**
3825 * Background snapshot thread, based on live migration code.
3826 * This is an alternative implementation of live migration mechanism
3827 * introduced specifically to support background snapshots.
3828 *
3829 * It takes advantage of userfault_fd write protection mechanism introduced
3830 * in v5.7 kernel. Compared to existing dirty page logging migration much
3831 * lesser stream traffic is produced resulting in smaller snapshot images,
3832 * simply cause of no page duplicates can get into the stream.
3833 *
3834 * Another key point is that generated vmstate stream reflects machine state
3835 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3836 * mechanism, which effectively results in that saved snapshot is the state of VM
3837 * at the end of the process.
3838 */
3839 static void *bg_migration_thread(void *opaque)
3840 {
3841 MigrationState *s = opaque;
3842 int64_t setup_start;
3843 MigThrError thr_error;
3844 QEMUFile *fb;
3845 bool early_fail = true;
3846
3847 rcu_register_thread();
3848 object_ref(OBJECT(s));
3849
3850 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3851
3852 setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3853 /*
3854 * We want to save vmstate for the moment when migration has been
3855 * initiated but also we want to save RAM content while VM is running.
3856 * The RAM content should appear first in the vmstate. So, we first
3857 * stash the non-RAM part of the vmstate to the temporary buffer,
3858 * then write RAM part of the vmstate to the migration stream
3859 * with vCPUs running and, finally, write stashed non-RAM part of
3860 * the vmstate from the buffer to the migration stream.
3861 */
3862 s->bioc = qio_channel_buffer_new(128 * 1024);
3863 qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer");
3864 fb = qemu_fopen_channel_output(QIO_CHANNEL(s->bioc));
3865 object_unref(OBJECT(s->bioc));
3866
3867 update_iteration_initial_status(s);
3868
3869 qemu_savevm_state_header(s->to_dst_file);
3870 qemu_savevm_state_setup(s->to_dst_file);
3871
3872 if (qemu_savevm_state_guest_unplug_pending()) {
3873 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3874 MIGRATION_STATUS_WAIT_UNPLUG);
3875
3876 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3877 qemu_savevm_state_guest_unplug_pending()) {
3878 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3879 }
3880
3881 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3882 MIGRATION_STATUS_ACTIVE);
3883 } else {
3884 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3885 MIGRATION_STATUS_ACTIVE);
3886 }
3887 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3888
3889 trace_migration_thread_setup_complete();
3890 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3891
3892 qemu_mutex_lock_iothread();
3893
3894 /*
3895 * If VM is currently in suspended state, then, to make a valid runstate
3896 * transition in vm_stop_force_state() we need to wakeup it up.
3897 */
3898 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3899 s->vm_was_running = runstate_is_running();
3900
3901 if (global_state_store()) {
3902 goto fail;
3903 }
3904 /* Forcibly stop VM before saving state of vCPUs and devices */
3905 if (vm_stop_force_state(RUN_STATE_PAUSED)) {
3906 goto fail;
3907 }
3908 /*
3909 * Put vCPUs in sync with shadow context structures, then
3910 * save their state to channel-buffer along with devices.
3911 */
3912 cpu_synchronize_all_states();
3913 if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) {
3914 goto fail;
3915 }
3916 /* Now initialize UFFD context and start tracking RAM writes */
3917 if (ram_write_tracking_start()) {
3918 goto fail;
3919 }
3920 early_fail = false;
3921
3922 /*