Merge remote-tracking branch 'remotes/dgilbert/tags/pull-migration-20201026a' into...
[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 "monitor/monitor.h"
56 #include "net/announce.h"
57 #include "qemu/queue.h"
58 #include "multifd.h"
59
60 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
61
62 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
63 * data. */
64 #define BUFFER_DELAY 100
65 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
66
67 /* Time in milliseconds we are allowed to stop the source,
68 * for sending the last part */
69 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
70
71 /* Maximum migrate downtime set to 2000 seconds */
72 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
73 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
74
75 /* Default compression thread count */
76 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
77 /* Default decompression thread count, usually decompression is at
78 * least 4 times as fast as compression.*/
79 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
80 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
81 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
82 /* Define default autoconverge cpu throttle migration parameters */
83 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
84 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
85 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
86 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
87
88 /* Migration XBZRLE default cache size */
89 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
90
91 /* The delay time (in ms) between two COLO checkpoints */
92 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
93 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
94 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
95 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
96 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
97 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
98 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
99
100 /* Background transfer rate for postcopy, 0 means unlimited, note
101 * that page requests can still exceed this limit.
102 */
103 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
104
105 /*
106 * Parameters for self_announce_delay giving a stream of RARP/ARP
107 * packets after migration.
108 */
109 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
110 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
111 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
112 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
113
114 static NotifierList migration_state_notifiers =
115 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
116
117 static bool deferred_incoming;
118
119 /* Messages sent on the return path from destination to source */
120 enum mig_rp_message_type {
121 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
122 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
123 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
124
125 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
126 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
127 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
128 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
129
130 MIG_RP_MSG_MAX
131 };
132
133 /* When we add fault tolerance, we could have several
134 migrations at once. For now we don't need to add
135 dynamic creation of migration */
136
137 static MigrationState *current_migration;
138 static MigrationIncomingState *current_incoming;
139
140 static bool migration_object_check(MigrationState *ms, Error **errp);
141 static int migration_maybe_pause(MigrationState *s,
142 int *current_active_state,
143 int new_state);
144 static void migrate_fd_cancel(MigrationState *s);
145
146 static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp)
147 {
148 uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp;
149
150 return (a > b) - (a < b);
151 }
152
153 void migration_object_init(void)
154 {
155 Error *err = NULL;
156
157 /* This can only be called once. */
158 assert(!current_migration);
159 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
160
161 /*
162 * Init the migrate incoming object as well no matter whether
163 * we'll use it or not.
164 */
165 assert(!current_incoming);
166 current_incoming = g_new0(MigrationIncomingState, 1);
167 current_incoming->state = MIGRATION_STATUS_NONE;
168 current_incoming->postcopy_remote_fds =
169 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
170 qemu_mutex_init(&current_incoming->rp_mutex);
171 qemu_event_init(&current_incoming->main_thread_load_event, false);
172 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
173 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
174 qemu_mutex_init(&current_incoming->page_request_mutex);
175 current_incoming->page_requested = g_tree_new(page_request_addr_cmp);
176
177 if (!migration_object_check(current_migration, &err)) {
178 error_report_err(err);
179 exit(1);
180 }
181 }
182
183 void migration_shutdown(void)
184 {
185 /*
186 * Cancel the current migration - that will (eventually)
187 * stop the migration using this structure
188 */
189 migrate_fd_cancel(current_migration);
190 object_unref(OBJECT(current_migration));
191
192 /*
193 * Cancel outgoing migration of dirty bitmaps. It should
194 * at least unref used block nodes.
195 */
196 dirty_bitmap_mig_cancel_outgoing();
197
198 /*
199 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
200 * are non-critical data, and their loss never considered as
201 * something serious.
202 */
203 dirty_bitmap_mig_cancel_incoming();
204 }
205
206 /* For outgoing */
207 MigrationState *migrate_get_current(void)
208 {
209 /* This can only be called after the object created. */
210 assert(current_migration);
211 return current_migration;
212 }
213
214 MigrationIncomingState *migration_incoming_get_current(void)
215 {
216 assert(current_incoming);
217 return current_incoming;
218 }
219
220 void migration_incoming_state_destroy(void)
221 {
222 struct MigrationIncomingState *mis = migration_incoming_get_current();
223
224 if (mis->to_src_file) {
225 /* Tell source that we are done */
226 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
227 qemu_fclose(mis->to_src_file);
228 mis->to_src_file = NULL;
229 }
230
231 if (mis->from_src_file) {
232 qemu_fclose(mis->from_src_file);
233 mis->from_src_file = NULL;
234 }
235 if (mis->postcopy_remote_fds) {
236 g_array_free(mis->postcopy_remote_fds, TRUE);
237 mis->postcopy_remote_fds = NULL;
238 }
239
240 qemu_event_reset(&mis->main_thread_load_event);
241
242 if (mis->page_requested) {
243 g_tree_destroy(mis->page_requested);
244 mis->page_requested = NULL;
245 }
246
247 if (mis->socket_address_list) {
248 qapi_free_SocketAddressList(mis->socket_address_list);
249 mis->socket_address_list = NULL;
250 }
251 }
252
253 static void migrate_generate_event(int new_state)
254 {
255 if (migrate_use_events()) {
256 qapi_event_send_migration(new_state);
257 }
258 }
259
260 static bool migrate_late_block_activate(void)
261 {
262 MigrationState *s;
263
264 s = migrate_get_current();
265
266 return s->enabled_capabilities[
267 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
268 }
269
270 /*
271 * Called on -incoming with a defer: uri.
272 * The migration can be started later after any parameters have been
273 * changed.
274 */
275 static void deferred_incoming_migration(Error **errp)
276 {
277 if (deferred_incoming) {
278 error_setg(errp, "Incoming migration already deferred");
279 }
280 deferred_incoming = true;
281 }
282
283 /*
284 * Send a message on the return channel back to the source
285 * of the migration.
286 */
287 static int migrate_send_rp_message(MigrationIncomingState *mis,
288 enum mig_rp_message_type message_type,
289 uint16_t len, void *data)
290 {
291 int ret = 0;
292
293 trace_migrate_send_rp_message((int)message_type, len);
294 qemu_mutex_lock(&mis->rp_mutex);
295
296 /*
297 * It's possible that the file handle got lost due to network
298 * failures.
299 */
300 if (!mis->to_src_file) {
301 ret = -EIO;
302 goto error;
303 }
304
305 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
306 qemu_put_be16(mis->to_src_file, len);
307 qemu_put_buffer(mis->to_src_file, data, len);
308 qemu_fflush(mis->to_src_file);
309
310 /* It's possible that qemu file got error during sending */
311 ret = qemu_file_get_error(mis->to_src_file);
312
313 error:
314 qemu_mutex_unlock(&mis->rp_mutex);
315 return ret;
316 }
317
318 /* Request one page from the source VM at the given start address.
319 * rb: the RAMBlock to request the page in
320 * Start: Address offset within the RB
321 * Len: Length in bytes required - must be a multiple of pagesize
322 */
323 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
324 RAMBlock *rb, ram_addr_t start)
325 {
326 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
327 size_t msglen = 12; /* start + len */
328 size_t len = qemu_ram_pagesize(rb);
329 enum mig_rp_message_type msg_type;
330 const char *rbname;
331 int rbname_len;
332
333 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
334 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
335
336 /*
337 * We maintain the last ramblock that we requested for page. Note that we
338 * don't need locking because this function will only be called within the
339 * postcopy ram fault thread.
340 */
341 if (rb != mis->last_rb) {
342 mis->last_rb = rb;
343
344 rbname = qemu_ram_get_idstr(rb);
345 rbname_len = strlen(rbname);
346
347 assert(rbname_len < 256);
348
349 bufc[msglen++] = rbname_len;
350 memcpy(bufc + msglen, rbname, rbname_len);
351 msglen += rbname_len;
352 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
353 } else {
354 msg_type = MIG_RP_MSG_REQ_PAGES;
355 }
356
357 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
358 }
359
360 int migrate_send_rp_req_pages(MigrationIncomingState *mis,
361 RAMBlock *rb, ram_addr_t start, uint64_t haddr)
362 {
363 void *aligned = (void *)(uintptr_t)(haddr & (-qemu_ram_pagesize(rb)));
364 bool received;
365
366 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
367 received = ramblock_recv_bitmap_test_byte_offset(rb, start);
368 if (!received && !g_tree_lookup(mis->page_requested, aligned)) {
369 /*
370 * The page has not been received, and it's not yet in the page
371 * request list. Queue it. Set the value of element to 1, so that
372 * things like g_tree_lookup() will return TRUE (1) when found.
373 */
374 g_tree_insert(mis->page_requested, aligned, (gpointer)1);
375 mis->page_requested_count++;
376 trace_postcopy_page_req_add(aligned, mis->page_requested_count);
377 }
378 }
379
380 /*
381 * If the page is there, skip sending the message. We don't even need the
382 * lock because as long as the page arrived, it'll be there forever.
383 */
384 if (received) {
385 return 0;
386 }
387
388 return migrate_send_rp_message_req_pages(mis, rb, start);
389 }
390
391 static bool migration_colo_enabled;
392 bool migration_incoming_colo_enabled(void)
393 {
394 return migration_colo_enabled;
395 }
396
397 void migration_incoming_disable_colo(void)
398 {
399 ram_block_discard_disable(false);
400 migration_colo_enabled = false;
401 }
402
403 int migration_incoming_enable_colo(void)
404 {
405 if (ram_block_discard_disable(true)) {
406 error_report("COLO: cannot disable RAM discard");
407 return -EBUSY;
408 }
409 migration_colo_enabled = true;
410 return 0;
411 }
412
413 void migrate_add_address(SocketAddress *address)
414 {
415 MigrationIncomingState *mis = migration_incoming_get_current();
416 SocketAddressList *addrs;
417
418 addrs = g_new0(SocketAddressList, 1);
419 addrs->next = mis->socket_address_list;
420 mis->socket_address_list = addrs;
421 addrs->value = QAPI_CLONE(SocketAddress, address);
422 }
423
424 void qemu_start_incoming_migration(const char *uri, Error **errp)
425 {
426 const char *p = NULL;
427
428 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
429 if (!strcmp(uri, "defer")) {
430 deferred_incoming_migration(errp);
431 } else if (strstart(uri, "tcp:", &p) ||
432 strstart(uri, "unix:", NULL) ||
433 strstart(uri, "vsock:", NULL)) {
434 socket_start_incoming_migration(p ? p : uri, errp);
435 #ifdef CONFIG_RDMA
436 } else if (strstart(uri, "rdma:", &p)) {
437 rdma_start_incoming_migration(p, errp);
438 #endif
439 } else if (strstart(uri, "exec:", &p)) {
440 exec_start_incoming_migration(p, errp);
441 } else if (strstart(uri, "fd:", &p)) {
442 fd_start_incoming_migration(p, errp);
443 } else {
444 error_setg(errp, "unknown migration protocol: %s", uri);
445 }
446 }
447
448 static void process_incoming_migration_bh(void *opaque)
449 {
450 Error *local_err = NULL;
451 MigrationIncomingState *mis = opaque;
452
453 /* If capability late_block_activate is set:
454 * Only fire up the block code now if we're going to restart the
455 * VM, else 'cont' will do it.
456 * This causes file locking to happen; so we don't want it to happen
457 * unless we really are starting the VM.
458 */
459 if (!migrate_late_block_activate() ||
460 (autostart && (!global_state_received() ||
461 global_state_get_runstate() == RUN_STATE_RUNNING))) {
462 /* Make sure all file formats flush their mutable metadata.
463 * If we get an error here, just don't restart the VM yet. */
464 bdrv_invalidate_cache_all(&local_err);
465 if (local_err) {
466 error_report_err(local_err);
467 local_err = NULL;
468 autostart = false;
469 }
470 }
471
472 /*
473 * This must happen after all error conditions are dealt with and
474 * we're sure the VM is going to be running on this host.
475 */
476 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
477
478 if (multifd_load_cleanup(&local_err) != 0) {
479 error_report_err(local_err);
480 autostart = false;
481 }
482 /* If global state section was not received or we are in running
483 state, we need to obey autostart. Any other state is set with
484 runstate_set. */
485
486 dirty_bitmap_mig_before_vm_start();
487
488 if (!global_state_received() ||
489 global_state_get_runstate() == RUN_STATE_RUNNING) {
490 if (autostart) {
491 vm_start();
492 } else {
493 runstate_set(RUN_STATE_PAUSED);
494 }
495 } else if (migration_incoming_colo_enabled()) {
496 migration_incoming_disable_colo();
497 vm_start();
498 } else {
499 runstate_set(global_state_get_runstate());
500 }
501 /*
502 * This must happen after any state changes since as soon as an external
503 * observer sees this event they might start to prod at the VM assuming
504 * it's ready to use.
505 */
506 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
507 MIGRATION_STATUS_COMPLETED);
508 qemu_bh_delete(mis->bh);
509 migration_incoming_state_destroy();
510 }
511
512 static void process_incoming_migration_co(void *opaque)
513 {
514 MigrationIncomingState *mis = migration_incoming_get_current();
515 PostcopyState ps;
516 int ret;
517 Error *local_err = NULL;
518
519 assert(mis->from_src_file);
520 mis->migration_incoming_co = qemu_coroutine_self();
521 mis->largest_page_size = qemu_ram_pagesize_largest();
522 postcopy_state_set(POSTCOPY_INCOMING_NONE);
523 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
524 MIGRATION_STATUS_ACTIVE);
525 ret = qemu_loadvm_state(mis->from_src_file);
526
527 ps = postcopy_state_get();
528 trace_process_incoming_migration_co_end(ret, ps);
529 if (ps != POSTCOPY_INCOMING_NONE) {
530 if (ps == POSTCOPY_INCOMING_ADVISE) {
531 /*
532 * Where a migration had postcopy enabled (and thus went to advise)
533 * but managed to complete within the precopy period, we can use
534 * the normal exit.
535 */
536 postcopy_ram_incoming_cleanup(mis);
537 } else if (ret >= 0) {
538 /*
539 * Postcopy was started, cleanup should happen at the end of the
540 * postcopy thread.
541 */
542 trace_process_incoming_migration_co_postcopy_end_main();
543 return;
544 }
545 /* Else if something went wrong then just fall out of the normal exit */
546 }
547
548 /* we get COLO info, and know if we are in COLO mode */
549 if (!ret && migration_incoming_colo_enabled()) {
550 /* Make sure all file formats flush their mutable metadata */
551 bdrv_invalidate_cache_all(&local_err);
552 if (local_err) {
553 error_report_err(local_err);
554 goto fail;
555 }
556
557 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
558 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
559 mis->have_colo_incoming_thread = true;
560 qemu_coroutine_yield();
561
562 /* Wait checkpoint incoming thread exit before free resource */
563 qemu_thread_join(&mis->colo_incoming_thread);
564 /* We hold the global iothread lock, so it is safe here */
565 colo_release_ram_cache();
566 }
567
568 if (ret < 0) {
569 error_report("load of migration failed: %s", strerror(-ret));
570 goto fail;
571 }
572 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
573 qemu_bh_schedule(mis->bh);
574 mis->migration_incoming_co = NULL;
575 return;
576 fail:
577 local_err = NULL;
578 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
579 MIGRATION_STATUS_FAILED);
580 qemu_fclose(mis->from_src_file);
581 if (multifd_load_cleanup(&local_err) != 0) {
582 error_report_err(local_err);
583 }
584 exit(EXIT_FAILURE);
585 }
586
587 /**
588 * @migration_incoming_setup: Setup incoming migration
589 *
590 * Returns 0 for no error or 1 for error
591 *
592 * @f: file for main migration channel
593 * @errp: where to put errors
594 */
595 static int migration_incoming_setup(QEMUFile *f, Error **errp)
596 {
597 MigrationIncomingState *mis = migration_incoming_get_current();
598 Error *local_err = NULL;
599
600 if (multifd_load_setup(&local_err) != 0) {
601 /* We haven't been able to create multifd threads
602 nothing better to do */
603 error_report_err(local_err);
604 exit(EXIT_FAILURE);
605 }
606
607 if (!mis->from_src_file) {
608 mis->from_src_file = f;
609 }
610 qemu_file_set_blocking(f, false);
611 return 0;
612 }
613
614 void migration_incoming_process(void)
615 {
616 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
617 qemu_coroutine_enter(co);
618 }
619
620 /* Returns true if recovered from a paused migration, otherwise false */
621 static bool postcopy_try_recover(QEMUFile *f)
622 {
623 MigrationIncomingState *mis = migration_incoming_get_current();
624
625 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
626 /* Resumed from a paused postcopy migration */
627
628 mis->from_src_file = f;
629 /* Postcopy has standalone thread to do vm load */
630 qemu_file_set_blocking(f, true);
631
632 /* Re-configure the return path */
633 mis->to_src_file = qemu_file_get_return_path(f);
634
635 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
636 MIGRATION_STATUS_POSTCOPY_RECOVER);
637
638 /*
639 * Here, we only wake up the main loading thread (while the
640 * fault thread will still be waiting), so that we can receive
641 * commands from source now, and answer it if needed. The
642 * fault thread will be woken up afterwards until we are sure
643 * that source is ready to reply to page requests.
644 */
645 qemu_sem_post(&mis->postcopy_pause_sem_dst);
646 return true;
647 }
648
649 return false;
650 }
651
652 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
653 {
654 Error *local_err = NULL;
655
656 if (postcopy_try_recover(f)) {
657 return;
658 }
659
660 if (migration_incoming_setup(f, &local_err)) {
661 error_propagate(errp, local_err);
662 return;
663 }
664 migration_incoming_process();
665 }
666
667 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
668 {
669 MigrationIncomingState *mis = migration_incoming_get_current();
670 Error *local_err = NULL;
671 bool start_migration;
672
673 if (!mis->from_src_file) {
674 /* The first connection (multifd may have multiple) */
675 QEMUFile *f = qemu_fopen_channel_input(ioc);
676
677 /* If it's a recovery, we're done */
678 if (postcopy_try_recover(f)) {
679 return;
680 }
681
682 if (migration_incoming_setup(f, &local_err)) {
683 error_propagate(errp, local_err);
684 return;
685 }
686
687 /*
688 * Common migration only needs one channel, so we can start
689 * right now. Multifd needs more than one channel, we wait.
690 */
691 start_migration = !migrate_use_multifd();
692 } else {
693 /* Multiple connections */
694 assert(migrate_use_multifd());
695 start_migration = multifd_recv_new_channel(ioc, &local_err);
696 if (local_err) {
697 error_propagate(errp, local_err);
698 return;
699 }
700 }
701
702 if (start_migration) {
703 migration_incoming_process();
704 }
705 }
706
707 /**
708 * @migration_has_all_channels: We have received all channels that we need
709 *
710 * Returns true when we have got connections to all the channels that
711 * we need for migration.
712 */
713 bool migration_has_all_channels(void)
714 {
715 MigrationIncomingState *mis = migration_incoming_get_current();
716 bool all_channels;
717
718 all_channels = multifd_recv_all_channels_created();
719
720 return all_channels && mis->from_src_file != NULL;
721 }
722
723 /*
724 * Send a 'SHUT' message on the return channel with the given value
725 * to indicate that we've finished with the RP. Non-0 value indicates
726 * error.
727 */
728 void migrate_send_rp_shut(MigrationIncomingState *mis,
729 uint32_t value)
730 {
731 uint32_t buf;
732
733 buf = cpu_to_be32(value);
734 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
735 }
736
737 /*
738 * Send a 'PONG' message on the return channel with the given value
739 * (normally in response to a 'PING')
740 */
741 void migrate_send_rp_pong(MigrationIncomingState *mis,
742 uint32_t value)
743 {
744 uint32_t buf;
745
746 buf = cpu_to_be32(value);
747 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
748 }
749
750 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
751 char *block_name)
752 {
753 char buf[512];
754 int len;
755 int64_t res;
756
757 /*
758 * First, we send the header part. It contains only the len of
759 * idstr, and the idstr itself.
760 */
761 len = strlen(block_name);
762 buf[0] = len;
763 memcpy(buf + 1, block_name, len);
764
765 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
766 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
767 __func__);
768 return;
769 }
770
771 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
772
773 /*
774 * Next, we dump the received bitmap to the stream.
775 *
776 * TODO: currently we are safe since we are the only one that is
777 * using the to_src_file handle (fault thread is still paused),
778 * and it's ok even not taking the mutex. However the best way is
779 * to take the lock before sending the message header, and release
780 * the lock after sending the bitmap.
781 */
782 qemu_mutex_lock(&mis->rp_mutex);
783 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
784 qemu_mutex_unlock(&mis->rp_mutex);
785
786 trace_migrate_send_rp_recv_bitmap(block_name, res);
787 }
788
789 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
790 {
791 uint32_t buf;
792
793 buf = cpu_to_be32(value);
794 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
795 }
796
797 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
798 {
799 MigrationCapabilityStatusList *head = NULL;
800 MigrationCapabilityStatusList *caps;
801 MigrationState *s = migrate_get_current();
802 int i;
803
804 caps = NULL; /* silence compiler warning */
805 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
806 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
807 if (i == MIGRATION_CAPABILITY_BLOCK) {
808 continue;
809 }
810 #endif
811 if (head == NULL) {
812 head = g_malloc0(sizeof(*caps));
813 caps = head;
814 } else {
815 caps->next = g_malloc0(sizeof(*caps));
816 caps = caps->next;
817 }
818 caps->value =
819 g_malloc(sizeof(*caps->value));
820 caps->value->capability = i;
821 caps->value->state = s->enabled_capabilities[i];
822 }
823
824 return head;
825 }
826
827 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
828 {
829 MigrationParameters *params;
830 MigrationState *s = migrate_get_current();
831
832 /* TODO use QAPI_CLONE() instead of duplicating it inline */
833 params = g_malloc0(sizeof(*params));
834 params->has_compress_level = true;
835 params->compress_level = s->parameters.compress_level;
836 params->has_compress_threads = true;
837 params->compress_threads = s->parameters.compress_threads;
838 params->has_compress_wait_thread = true;
839 params->compress_wait_thread = s->parameters.compress_wait_thread;
840 params->has_decompress_threads = true;
841 params->decompress_threads = s->parameters.decompress_threads;
842 params->has_throttle_trigger_threshold = true;
843 params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
844 params->has_cpu_throttle_initial = true;
845 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
846 params->has_cpu_throttle_increment = true;
847 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
848 params->has_cpu_throttle_tailslow = true;
849 params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
850 params->has_tls_creds = true;
851 params->tls_creds = g_strdup(s->parameters.tls_creds);
852 params->has_tls_hostname = true;
853 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
854 params->has_tls_authz = true;
855 params->tls_authz = g_strdup(s->parameters.tls_authz ?
856 s->parameters.tls_authz : "");
857 params->has_max_bandwidth = true;
858 params->max_bandwidth = s->parameters.max_bandwidth;
859 params->has_downtime_limit = true;
860 params->downtime_limit = s->parameters.downtime_limit;
861 params->has_x_checkpoint_delay = true;
862 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
863 params->has_block_incremental = true;
864 params->block_incremental = s->parameters.block_incremental;
865 params->has_multifd_channels = true;
866 params->multifd_channels = s->parameters.multifd_channels;
867 params->has_multifd_compression = true;
868 params->multifd_compression = s->parameters.multifd_compression;
869 params->has_multifd_zlib_level = true;
870 params->multifd_zlib_level = s->parameters.multifd_zlib_level;
871 params->has_multifd_zstd_level = true;
872 params->multifd_zstd_level = s->parameters.multifd_zstd_level;
873 params->has_xbzrle_cache_size = true;
874 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
875 params->has_max_postcopy_bandwidth = true;
876 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
877 params->has_max_cpu_throttle = true;
878 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
879 params->has_announce_initial = true;
880 params->announce_initial = s->parameters.announce_initial;
881 params->has_announce_max = true;
882 params->announce_max = s->parameters.announce_max;
883 params->has_announce_rounds = true;
884 params->announce_rounds = s->parameters.announce_rounds;
885 params->has_announce_step = true;
886 params->announce_step = s->parameters.announce_step;
887
888 if (s->parameters.has_block_bitmap_mapping) {
889 params->has_block_bitmap_mapping = true;
890 params->block_bitmap_mapping =
891 QAPI_CLONE(BitmapMigrationNodeAliasList,
892 s->parameters.block_bitmap_mapping);
893 }
894
895 return params;
896 }
897
898 AnnounceParameters *migrate_announce_params(void)
899 {
900 static AnnounceParameters ap;
901
902 MigrationState *s = migrate_get_current();
903
904 ap.initial = s->parameters.announce_initial;
905 ap.max = s->parameters.announce_max;
906 ap.rounds = s->parameters.announce_rounds;
907 ap.step = s->parameters.announce_step;
908
909 return &ap;
910 }
911
912 /*
913 * Return true if we're already in the middle of a migration
914 * (i.e. any of the active or setup states)
915 */
916 bool migration_is_setup_or_active(int state)
917 {
918 switch (state) {
919 case MIGRATION_STATUS_ACTIVE:
920 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
921 case MIGRATION_STATUS_POSTCOPY_PAUSED:
922 case MIGRATION_STATUS_POSTCOPY_RECOVER:
923 case MIGRATION_STATUS_SETUP:
924 case MIGRATION_STATUS_PRE_SWITCHOVER:
925 case MIGRATION_STATUS_DEVICE:
926 case MIGRATION_STATUS_WAIT_UNPLUG:
927 case MIGRATION_STATUS_COLO:
928 return true;
929
930 default:
931 return false;
932
933 }
934 }
935
936 bool migration_is_running(int state)
937 {
938 switch (state) {
939 case MIGRATION_STATUS_ACTIVE:
940 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
941 case MIGRATION_STATUS_POSTCOPY_PAUSED:
942 case MIGRATION_STATUS_POSTCOPY_RECOVER:
943 case MIGRATION_STATUS_SETUP:
944 case MIGRATION_STATUS_PRE_SWITCHOVER:
945 case MIGRATION_STATUS_DEVICE:
946 case MIGRATION_STATUS_WAIT_UNPLUG:
947 case MIGRATION_STATUS_CANCELLING:
948 return true;
949
950 default:
951 return false;
952
953 }
954 }
955
956 static void populate_time_info(MigrationInfo *info, MigrationState *s)
957 {
958 info->has_status = true;
959 info->has_setup_time = true;
960 info->setup_time = s->setup_time;
961 if (s->state == MIGRATION_STATUS_COMPLETED) {
962 info->has_total_time = true;
963 info->total_time = s->total_time;
964 info->has_downtime = true;
965 info->downtime = s->downtime;
966 } else {
967 info->has_total_time = true;
968 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
969 s->start_time;
970 info->has_expected_downtime = true;
971 info->expected_downtime = s->expected_downtime;
972 }
973 }
974
975 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
976 {
977 info->has_ram = true;
978 info->ram = g_malloc0(sizeof(*info->ram));
979 info->ram->transferred = ram_counters.transferred;
980 info->ram->total = ram_bytes_total();
981 info->ram->duplicate = ram_counters.duplicate;
982 /* legacy value. It is not used anymore */
983 info->ram->skipped = 0;
984 info->ram->normal = ram_counters.normal;
985 info->ram->normal_bytes = ram_counters.normal *
986 qemu_target_page_size();
987 info->ram->mbps = s->mbps;
988 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
989 info->ram->postcopy_requests = ram_counters.postcopy_requests;
990 info->ram->page_size = qemu_target_page_size();
991 info->ram->multifd_bytes = ram_counters.multifd_bytes;
992 info->ram->pages_per_second = s->pages_per_second;
993
994 if (migrate_use_xbzrle()) {
995 info->has_xbzrle_cache = true;
996 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
997 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
998 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
999 info->xbzrle_cache->pages = xbzrle_counters.pages;
1000 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
1001 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
1002 info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
1003 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
1004 }
1005
1006 if (migrate_use_compression()) {
1007 info->has_compression = true;
1008 info->compression = g_malloc0(sizeof(*info->compression));
1009 info->compression->pages = compression_counters.pages;
1010 info->compression->busy = compression_counters.busy;
1011 info->compression->busy_rate = compression_counters.busy_rate;
1012 info->compression->compressed_size =
1013 compression_counters.compressed_size;
1014 info->compression->compression_rate =
1015 compression_counters.compression_rate;
1016 }
1017
1018 if (cpu_throttle_active()) {
1019 info->has_cpu_throttle_percentage = true;
1020 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
1021 }
1022
1023 if (s->state != MIGRATION_STATUS_COMPLETED) {
1024 info->ram->remaining = ram_bytes_remaining();
1025 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
1026 }
1027 }
1028
1029 static void populate_disk_info(MigrationInfo *info)
1030 {
1031 if (blk_mig_active()) {
1032 info->has_disk = true;
1033 info->disk = g_malloc0(sizeof(*info->disk));
1034 info->disk->transferred = blk_mig_bytes_transferred();
1035 info->disk->remaining = blk_mig_bytes_remaining();
1036 info->disk->total = blk_mig_bytes_total();
1037 }
1038 }
1039
1040 static void fill_source_migration_info(MigrationInfo *info)
1041 {
1042 MigrationState *s = migrate_get_current();
1043
1044 switch (s->state) {
1045 case MIGRATION_STATUS_NONE:
1046 /* no migration has happened ever */
1047 /* do not overwrite destination migration status */
1048 return;
1049 case MIGRATION_STATUS_SETUP:
1050 info->has_status = true;
1051 info->has_total_time = false;
1052 break;
1053 case MIGRATION_STATUS_ACTIVE:
1054 case MIGRATION_STATUS_CANCELLING:
1055 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1056 case MIGRATION_STATUS_PRE_SWITCHOVER:
1057 case MIGRATION_STATUS_DEVICE:
1058 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1059 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1060 /* TODO add some postcopy stats */
1061 populate_time_info(info, s);
1062 populate_ram_info(info, s);
1063 populate_disk_info(info);
1064 break;
1065 case MIGRATION_STATUS_COLO:
1066 info->has_status = true;
1067 /* TODO: display COLO specific information (checkpoint info etc.) */
1068 break;
1069 case MIGRATION_STATUS_COMPLETED:
1070 populate_time_info(info, s);
1071 populate_ram_info(info, s);
1072 break;
1073 case MIGRATION_STATUS_FAILED:
1074 info->has_status = true;
1075 if (s->error) {
1076 info->has_error_desc = true;
1077 info->error_desc = g_strdup(error_get_pretty(s->error));
1078 }
1079 break;
1080 case MIGRATION_STATUS_CANCELLED:
1081 info->has_status = true;
1082 break;
1083 case MIGRATION_STATUS_WAIT_UNPLUG:
1084 info->has_status = true;
1085 break;
1086 }
1087 info->status = s->state;
1088 }
1089
1090 /**
1091 * @migration_caps_check - check capability validity
1092 *
1093 * @cap_list: old capability list, array of bool
1094 * @params: new capabilities to be applied soon
1095 * @errp: set *errp if the check failed, with reason
1096 *
1097 * Returns true if check passed, otherwise false.
1098 */
1099 static bool migrate_caps_check(bool *cap_list,
1100 MigrationCapabilityStatusList *params,
1101 Error **errp)
1102 {
1103 MigrationCapabilityStatusList *cap;
1104 bool old_postcopy_cap;
1105 MigrationIncomingState *mis = migration_incoming_get_current();
1106
1107 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1108
1109 for (cap = params; cap; cap = cap->next) {
1110 cap_list[cap->value->capability] = cap->value->state;
1111 }
1112
1113 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1114 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1115 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1116 "block migration");
1117 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1118 return false;
1119 }
1120 #endif
1121
1122 #ifndef CONFIG_REPLICATION
1123 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1124 error_setg(errp, "QEMU compiled without replication module"
1125 " can't enable COLO");
1126 error_append_hint(errp, "Please enable replication before COLO.\n");
1127 return false;
1128 }
1129 #endif
1130
1131 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1132 /* This check is reasonably expensive, so only when it's being
1133 * set the first time, also it's only the destination that needs
1134 * special support.
1135 */
1136 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1137 !postcopy_ram_supported_by_host(mis)) {
1138 /* postcopy_ram_supported_by_host will have emitted a more
1139 * detailed message
1140 */
1141 error_setg(errp, "Postcopy is not supported");
1142 return false;
1143 }
1144
1145 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1146 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1147 return false;
1148 }
1149 }
1150
1151 return true;
1152 }
1153
1154 static void fill_destination_migration_info(MigrationInfo *info)
1155 {
1156 MigrationIncomingState *mis = migration_incoming_get_current();
1157
1158 if (mis->socket_address_list) {
1159 info->has_socket_address = true;
1160 info->socket_address =
1161 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1162 }
1163
1164 switch (mis->state) {
1165 case MIGRATION_STATUS_NONE:
1166 return;
1167 case MIGRATION_STATUS_SETUP:
1168 case MIGRATION_STATUS_CANCELLING:
1169 case MIGRATION_STATUS_CANCELLED:
1170 case MIGRATION_STATUS_ACTIVE:
1171 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1172 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1173 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1174 case MIGRATION_STATUS_FAILED:
1175 case MIGRATION_STATUS_COLO:
1176 info->has_status = true;
1177 break;
1178 case MIGRATION_STATUS_COMPLETED:
1179 info->has_status = true;
1180 fill_destination_postcopy_migration_info(info);
1181 break;
1182 }
1183 info->status = mis->state;
1184 }
1185
1186 MigrationInfo *qmp_query_migrate(Error **errp)
1187 {
1188 MigrationInfo *info = g_malloc0(sizeof(*info));
1189
1190 fill_destination_migration_info(info);
1191 fill_source_migration_info(info);
1192
1193 return info;
1194 }
1195
1196 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1197 Error **errp)
1198 {
1199 MigrationState *s = migrate_get_current();
1200 MigrationCapabilityStatusList *cap;
1201 bool cap_list[MIGRATION_CAPABILITY__MAX];
1202
1203 if (migration_is_running(s->state)) {
1204 error_setg(errp, QERR_MIGRATION_ACTIVE);
1205 return;
1206 }
1207
1208 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1209 if (!migrate_caps_check(cap_list, params, errp)) {
1210 return;
1211 }
1212
1213 for (cap = params; cap; cap = cap->next) {
1214 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1215 }
1216 }
1217
1218 /*
1219 * Check whether the parameters are valid. Error will be put into errp
1220 * (if provided). Return true if valid, otherwise false.
1221 */
1222 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1223 {
1224 if (params->has_compress_level &&
1225 (params->compress_level > 9)) {
1226 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1227 "is invalid, it should be in the range of 0 to 9");
1228 return false;
1229 }
1230
1231 if (params->has_compress_threads && (params->compress_threads < 1)) {
1232 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1233 "compress_threads",
1234 "is invalid, it should be in the range of 1 to 255");
1235 return false;
1236 }
1237
1238 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1239 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1240 "decompress_threads",
1241 "is invalid, it should be in the range of 1 to 255");
1242 return false;
1243 }
1244
1245 if (params->has_throttle_trigger_threshold &&
1246 (params->throttle_trigger_threshold < 1 ||
1247 params->throttle_trigger_threshold > 100)) {
1248 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1249 "throttle_trigger_threshold",
1250 "an integer in the range of 1 to 100");
1251 return false;
1252 }
1253
1254 if (params->has_cpu_throttle_initial &&
1255 (params->cpu_throttle_initial < 1 ||
1256 params->cpu_throttle_initial > 99)) {
1257 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1258 "cpu_throttle_initial",
1259 "an integer in the range of 1 to 99");
1260 return false;
1261 }
1262
1263 if (params->has_cpu_throttle_increment &&
1264 (params->cpu_throttle_increment < 1 ||
1265 params->cpu_throttle_increment > 99)) {
1266 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1267 "cpu_throttle_increment",
1268 "an integer in the range of 1 to 99");
1269 return false;
1270 }
1271
1272 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1273 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1274 "max_bandwidth",
1275 "an integer in the range of 0 to "stringify(SIZE_MAX)
1276 " bytes/second");
1277 return false;
1278 }
1279
1280 if (params->has_downtime_limit &&
1281 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1282 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1283 "downtime_limit",
1284 "an integer in the range of 0 to "
1285 stringify(MAX_MIGRATE_DOWNTIME)" ms");
1286 return false;
1287 }
1288
1289 /* x_checkpoint_delay is now always positive */
1290
1291 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1292 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1293 "multifd_channels",
1294 "is invalid, it should be in the range of 1 to 255");
1295 return false;
1296 }
1297
1298 if (params->has_multifd_zlib_level &&
1299 (params->multifd_zlib_level > 9)) {
1300 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1301 "is invalid, it should be in the range of 0 to 9");
1302 return false;
1303 }
1304
1305 if (params->has_multifd_zstd_level &&
1306 (params->multifd_zstd_level > 20)) {
1307 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1308 "is invalid, it should be in the range of 0 to 20");
1309 return false;
1310 }
1311
1312 if (params->has_xbzrle_cache_size &&
1313 (params->xbzrle_cache_size < qemu_target_page_size() ||
1314 !is_power_of_2(params->xbzrle_cache_size))) {
1315 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1316 "xbzrle_cache_size",
1317 "is invalid, it should be bigger than target page size"
1318 " and a power of 2");
1319 return false;
1320 }
1321
1322 if (params->has_max_cpu_throttle &&
1323 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1324 params->max_cpu_throttle > 99)) {
1325 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1326 "max_cpu_throttle",
1327 "an integer in the range of cpu_throttle_initial to 99");
1328 return false;
1329 }
1330
1331 if (params->has_announce_initial &&
1332 params->announce_initial > 100000) {
1333 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1334 "announce_initial",
1335 "is invalid, it must be less than 100000 ms");
1336 return false;
1337 }
1338 if (params->has_announce_max &&
1339 params->announce_max > 100000) {
1340 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1341 "announce_max",
1342 "is invalid, it must be less than 100000 ms");
1343 return false;
1344 }
1345 if (params->has_announce_rounds &&
1346 params->announce_rounds > 1000) {
1347 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1348 "announce_rounds",
1349 "is invalid, it must be in the range of 0 to 1000");
1350 return false;
1351 }
1352 if (params->has_announce_step &&
1353 (params->announce_step < 1 ||
1354 params->announce_step > 10000)) {
1355 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1356 "announce_step",
1357 "is invalid, it must be in the range of 1 to 10000 ms");
1358 return false;
1359 }
1360
1361 if (params->has_block_bitmap_mapping &&
1362 !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) {
1363 error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: ");
1364 return false;
1365 }
1366
1367 return true;
1368 }
1369
1370 static void migrate_params_test_apply(MigrateSetParameters *params,
1371 MigrationParameters *dest)
1372 {
1373 *dest = migrate_get_current()->parameters;
1374
1375 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1376
1377 if (params->has_compress_level) {
1378 dest->compress_level = params->compress_level;
1379 }
1380
1381 if (params->has_compress_threads) {
1382 dest->compress_threads = params->compress_threads;
1383 }
1384
1385 if (params->has_compress_wait_thread) {
1386 dest->compress_wait_thread = params->compress_wait_thread;
1387 }
1388
1389 if (params->has_decompress_threads) {
1390 dest->decompress_threads = params->decompress_threads;
1391 }
1392
1393 if (params->has_throttle_trigger_threshold) {
1394 dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
1395 }
1396
1397 if (params->has_cpu_throttle_initial) {
1398 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1399 }
1400
1401 if (params->has_cpu_throttle_increment) {
1402 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1403 }
1404
1405 if (params->has_cpu_throttle_tailslow) {
1406 dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1407 }
1408
1409 if (params->has_tls_creds) {
1410 assert(params->tls_creds->type == QTYPE_QSTRING);
1411 dest->tls_creds = params->tls_creds->u.s;
1412 }
1413
1414 if (params->has_tls_hostname) {
1415 assert(params->tls_hostname->type == QTYPE_QSTRING);
1416 dest->tls_hostname = params->tls_hostname->u.s;
1417 }
1418
1419 if (params->has_max_bandwidth) {
1420 dest->max_bandwidth = params->max_bandwidth;
1421 }
1422
1423 if (params->has_downtime_limit) {
1424 dest->downtime_limit = params->downtime_limit;
1425 }
1426
1427 if (params->has_x_checkpoint_delay) {
1428 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1429 }
1430
1431 if (params->has_block_incremental) {
1432 dest->block_incremental = params->block_incremental;
1433 }
1434 if (params->has_multifd_channels) {
1435 dest->multifd_channels = params->multifd_channels;
1436 }
1437 if (params->has_multifd_compression) {
1438 dest->multifd_compression = params->multifd_compression;
1439 }
1440 if (params->has_xbzrle_cache_size) {
1441 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1442 }
1443 if (params->has_max_postcopy_bandwidth) {
1444 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1445 }
1446 if (params->has_max_cpu_throttle) {
1447 dest->max_cpu_throttle = params->max_cpu_throttle;
1448 }
1449 if (params->has_announce_initial) {
1450 dest->announce_initial = params->announce_initial;
1451 }
1452 if (params->has_announce_max) {
1453 dest->announce_max = params->announce_max;
1454 }
1455 if (params->has_announce_rounds) {
1456 dest->announce_rounds = params->announce_rounds;
1457 }
1458 if (params->has_announce_step) {
1459 dest->announce_step = params->announce_step;
1460 }
1461
1462 if (params->has_block_bitmap_mapping) {
1463 dest->has_block_bitmap_mapping = true;
1464 dest->block_bitmap_mapping = params->block_bitmap_mapping;
1465 }
1466 }
1467
1468 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1469 {
1470 MigrationState *s = migrate_get_current();
1471
1472 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1473
1474 if (params->has_compress_level) {
1475 s->parameters.compress_level = params->compress_level;
1476 }
1477
1478 if (params->has_compress_threads) {
1479 s->parameters.compress_threads = params->compress_threads;
1480 }
1481
1482 if (params->has_compress_wait_thread) {
1483 s->parameters.compress_wait_thread = params->compress_wait_thread;
1484 }
1485
1486 if (params->has_decompress_threads) {
1487 s->parameters.decompress_threads = params->decompress_threads;
1488 }
1489
1490 if (params->has_throttle_trigger_threshold) {
1491 s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
1492 }
1493
1494 if (params->has_cpu_throttle_initial) {
1495 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1496 }
1497
1498 if (params->has_cpu_throttle_increment) {
1499 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1500 }
1501
1502 if (params->has_cpu_throttle_tailslow) {
1503 s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1504 }
1505
1506 if (params->has_tls_creds) {
1507 g_free(s->parameters.tls_creds);
1508 assert(params->tls_creds->type == QTYPE_QSTRING);
1509 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1510 }
1511
1512 if (params->has_tls_hostname) {
1513 g_free(s->parameters.tls_hostname);
1514 assert(params->tls_hostname->type == QTYPE_QSTRING);
1515 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1516 }
1517
1518 if (params->has_tls_authz) {
1519 g_free(s->parameters.tls_authz);
1520 assert(params->tls_authz->type == QTYPE_QSTRING);
1521 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1522 }
1523
1524 if (params->has_max_bandwidth) {
1525 s->parameters.max_bandwidth = params->max_bandwidth;
1526 if (s->to_dst_file && !migration_in_postcopy()) {
1527 qemu_file_set_rate_limit(s->to_dst_file,
1528 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1529 }
1530 }
1531
1532 if (params->has_downtime_limit) {
1533 s->parameters.downtime_limit = params->downtime_limit;
1534 }
1535
1536 if (params->has_x_checkpoint_delay) {
1537 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1538 if (migration_in_colo_state()) {
1539 colo_checkpoint_notify(s);
1540 }
1541 }
1542
1543 if (params->has_block_incremental) {
1544 s->parameters.block_incremental = params->block_incremental;
1545 }
1546 if (params->has_multifd_channels) {
1547 s->parameters.multifd_channels = params->multifd_channels;
1548 }
1549 if (params->has_multifd_compression) {
1550 s->parameters.multifd_compression = params->multifd_compression;
1551 }
1552 if (params->has_xbzrle_cache_size) {
1553 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1554 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1555 }
1556 if (params->has_max_postcopy_bandwidth) {
1557 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1558 if (s->to_dst_file && migration_in_postcopy()) {
1559 qemu_file_set_rate_limit(s->to_dst_file,
1560 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1561 }
1562 }
1563 if (params->has_max_cpu_throttle) {
1564 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1565 }
1566 if (params->has_announce_initial) {
1567 s->parameters.announce_initial = params->announce_initial;
1568 }
1569 if (params->has_announce_max) {
1570 s->parameters.announce_max = params->announce_max;
1571 }
1572 if (params->has_announce_rounds) {
1573 s->parameters.announce_rounds = params->announce_rounds;
1574 }
1575 if (params->has_announce_step) {
1576 s->parameters.announce_step = params->announce_step;
1577 }
1578
1579 if (params->has_block_bitmap_mapping) {
1580 qapi_free_BitmapMigrationNodeAliasList(
1581 s->parameters.block_bitmap_mapping);
1582
1583 s->parameters.has_block_bitmap_mapping = true;
1584 s->parameters.block_bitmap_mapping =
1585 QAPI_CLONE(BitmapMigrationNodeAliasList,
1586 params->block_bitmap_mapping);
1587 }
1588 }
1589
1590 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1591 {
1592 MigrationParameters tmp;
1593
1594 /* TODO Rewrite "" to null instead */
1595 if (params->has_tls_creds
1596 && params->tls_creds->type == QTYPE_QNULL) {
1597 qobject_unref(params->tls_creds->u.n);
1598 params->tls_creds->type = QTYPE_QSTRING;
1599 params->tls_creds->u.s = strdup("");
1600 }
1601 /* TODO Rewrite "" to null instead */
1602 if (params->has_tls_hostname
1603 && params->tls_hostname->type == QTYPE_QNULL) {
1604 qobject_unref(params->tls_hostname->u.n);
1605 params->tls_hostname->type = QTYPE_QSTRING;
1606 params->tls_hostname->u.s = strdup("");
1607 }
1608
1609 migrate_params_test_apply(params, &tmp);
1610
1611 if (!migrate_params_check(&tmp, errp)) {
1612 /* Invalid parameter */
1613 return;
1614 }
1615
1616 migrate_params_apply(params, errp);
1617 }
1618
1619
1620 void qmp_migrate_start_postcopy(Error **errp)
1621 {
1622 MigrationState *s = migrate_get_current();
1623
1624 if (!migrate_postcopy()) {
1625 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1626 " the start of migration");
1627 return;
1628 }
1629
1630 if (s->state == MIGRATION_STATUS_NONE) {
1631 error_setg(errp, "Postcopy must be started after migration has been"
1632 " started");
1633 return;
1634 }
1635 /*
1636 * we don't error if migration has finished since that would be racy
1637 * with issuing this command.
1638 */
1639 qatomic_set(&s->start_postcopy, true);
1640 }
1641
1642 /* shared migration helpers */
1643
1644 void migrate_set_state(int *state, int old_state, int new_state)
1645 {
1646 assert(new_state < MIGRATION_STATUS__MAX);
1647 if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
1648 trace_migrate_set_state(MigrationStatus_str(new_state));
1649 migrate_generate_event(new_state);
1650 }
1651 }
1652
1653 static MigrationCapabilityStatusList *migrate_cap_add(
1654 MigrationCapabilityStatusList *list,
1655 MigrationCapability index,
1656 bool state)
1657 {
1658 MigrationCapabilityStatusList *cap;
1659
1660 cap = g_new0(MigrationCapabilityStatusList, 1);
1661 cap->value = g_new0(MigrationCapabilityStatus, 1);
1662 cap->value->capability = index;
1663 cap->value->state = state;
1664 cap->next = list;
1665
1666 return cap;
1667 }
1668
1669 void migrate_set_block_enabled(bool value, Error **errp)
1670 {
1671 MigrationCapabilityStatusList *cap;
1672
1673 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1674 qmp_migrate_set_capabilities(cap, errp);
1675 qapi_free_MigrationCapabilityStatusList(cap);
1676 }
1677
1678 static void migrate_set_block_incremental(MigrationState *s, bool value)
1679 {
1680 s->parameters.block_incremental = value;
1681 }
1682
1683 static void block_cleanup_parameters(MigrationState *s)
1684 {
1685 if (s->must_remove_block_options) {
1686 /* setting to false can never fail */
1687 migrate_set_block_enabled(false, &error_abort);
1688 migrate_set_block_incremental(s, false);
1689 s->must_remove_block_options = false;
1690 }
1691 }
1692
1693 static void migrate_fd_cleanup(MigrationState *s)
1694 {
1695 qemu_bh_delete(s->cleanup_bh);
1696 s->cleanup_bh = NULL;
1697
1698 qemu_savevm_state_cleanup();
1699
1700 if (s->to_dst_file) {
1701 QEMUFile *tmp;
1702
1703 trace_migrate_fd_cleanup();
1704 qemu_mutex_unlock_iothread();
1705 if (s->migration_thread_running) {
1706 qemu_thread_join(&s->thread);
1707 s->migration_thread_running = false;
1708 }
1709 qemu_mutex_lock_iothread();
1710
1711 multifd_save_cleanup();
1712 qemu_mutex_lock(&s->qemu_file_lock);
1713 tmp = s->to_dst_file;
1714 s->to_dst_file = NULL;
1715 qemu_mutex_unlock(&s->qemu_file_lock);
1716 /*
1717 * Close the file handle without the lock to make sure the
1718 * critical section won't block for long.
1719 */
1720 qemu_fclose(tmp);
1721 }
1722
1723 assert(!migration_is_active(s));
1724
1725 if (s->state == MIGRATION_STATUS_CANCELLING) {
1726 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1727 MIGRATION_STATUS_CANCELLED);
1728 }
1729
1730 if (s->error) {
1731 /* It is used on info migrate. We can't free it */
1732 error_report_err(error_copy(s->error));
1733 }
1734 notifier_list_notify(&migration_state_notifiers, s);
1735 block_cleanup_parameters(s);
1736 }
1737
1738 static void migrate_fd_cleanup_schedule(MigrationState *s)
1739 {
1740 /*
1741 * Ref the state for bh, because it may be called when
1742 * there're already no other refs
1743 */
1744 object_ref(OBJECT(s));
1745 qemu_bh_schedule(s->cleanup_bh);
1746 }
1747
1748 static void migrate_fd_cleanup_bh(void *opaque)
1749 {
1750 MigrationState *s = opaque;
1751 migrate_fd_cleanup(s);
1752 object_unref(OBJECT(s));
1753 }
1754
1755 void migrate_set_error(MigrationState *s, const Error *error)
1756 {
1757 QEMU_LOCK_GUARD(&s->error_mutex);
1758 if (!s->error) {
1759 s->error = error_copy(error);
1760 }
1761 }
1762
1763 void migrate_fd_error(MigrationState *s, const Error *error)
1764 {
1765 trace_migrate_fd_error(error_get_pretty(error));
1766 assert(s->to_dst_file == NULL);
1767 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1768 MIGRATION_STATUS_FAILED);
1769 migrate_set_error(s, error);
1770 }
1771
1772 static void migrate_fd_cancel(MigrationState *s)
1773 {
1774 int old_state ;
1775 QEMUFile *f = migrate_get_current()->to_dst_file;
1776 trace_migrate_fd_cancel();
1777
1778 if (s->rp_state.from_dst_file) {
1779 /* shutdown the rp socket, so causing the rp thread to shutdown */
1780 qemu_file_shutdown(s->rp_state.from_dst_file);
1781 }
1782
1783 do {
1784 old_state = s->state;
1785 if (!migration_is_running(old_state)) {
1786 break;
1787 }
1788 /* If the migration is paused, kick it out of the pause */
1789 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1790 qemu_sem_post(&s->pause_sem);
1791 }
1792 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1793 } while (s->state != MIGRATION_STATUS_CANCELLING);
1794
1795 /*
1796 * If we're unlucky the migration code might be stuck somewhere in a
1797 * send/write while the network has failed and is waiting to timeout;
1798 * if we've got shutdown(2) available then we can force it to quit.
1799 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1800 * called in a bh, so there is no race against this cancel.
1801 */
1802 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1803 qemu_file_shutdown(f);
1804 }
1805 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1806 Error *local_err = NULL;
1807
1808 bdrv_invalidate_cache_all(&local_err);
1809 if (local_err) {
1810 error_report_err(local_err);
1811 } else {
1812 s->block_inactive = false;
1813 }
1814 }
1815 }
1816
1817 void add_migration_state_change_notifier(Notifier *notify)
1818 {
1819 notifier_list_add(&migration_state_notifiers, notify);
1820 }
1821
1822 void remove_migration_state_change_notifier(Notifier *notify)
1823 {
1824 notifier_remove(notify);
1825 }
1826
1827 bool migration_in_setup(MigrationState *s)
1828 {
1829 return s->state == MIGRATION_STATUS_SETUP;
1830 }
1831
1832 bool migration_has_finished(MigrationState *s)
1833 {
1834 return s->state == MIGRATION_STATUS_COMPLETED;
1835 }
1836
1837 bool migration_has_failed(MigrationState *s)
1838 {
1839 return (s->state == MIGRATION_STATUS_CANCELLED ||
1840 s->state == MIGRATION_STATUS_FAILED);
1841 }
1842
1843 bool migration_in_postcopy(void)
1844 {
1845 MigrationState *s = migrate_get_current();
1846
1847 switch (s->state) {
1848 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1849 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1850 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1851 return true;
1852 default:
1853 return false;
1854 }
1855 }
1856
1857 bool migration_in_postcopy_after_devices(MigrationState *s)
1858 {
1859 return migration_in_postcopy() && s->postcopy_after_devices;
1860 }
1861
1862 bool migration_in_incoming_postcopy(void)
1863 {
1864 PostcopyState ps = postcopy_state_get();
1865
1866 return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
1867 }
1868
1869 bool migration_is_idle(void)
1870 {
1871 MigrationState *s = current_migration;
1872
1873 if (!s) {
1874 return true;
1875 }
1876
1877 switch (s->state) {
1878 case MIGRATION_STATUS_NONE:
1879 case MIGRATION_STATUS_CANCELLED:
1880 case MIGRATION_STATUS_COMPLETED:
1881 case MIGRATION_STATUS_FAILED:
1882 return true;
1883 case MIGRATION_STATUS_SETUP:
1884 case MIGRATION_STATUS_CANCELLING:
1885 case MIGRATION_STATUS_ACTIVE:
1886 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1887 case MIGRATION_STATUS_COLO:
1888 case MIGRATION_STATUS_PRE_SWITCHOVER:
1889 case MIGRATION_STATUS_DEVICE:
1890 case MIGRATION_STATUS_WAIT_UNPLUG:
1891 return false;
1892 case MIGRATION_STATUS__MAX:
1893 g_assert_not_reached();
1894 }
1895
1896 return false;
1897 }
1898
1899 bool migration_is_active(MigrationState *s)
1900 {
1901 return (s->state == MIGRATION_STATUS_ACTIVE ||
1902 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1903 }
1904
1905 void migrate_init(MigrationState *s)
1906 {
1907 /*
1908 * Reinitialise all migration state, except
1909 * parameters/capabilities that the user set, and
1910 * locks.
1911 */
1912 s->cleanup_bh = 0;
1913 s->to_dst_file = NULL;
1914 s->state = MIGRATION_STATUS_NONE;
1915 s->rp_state.from_dst_file = NULL;
1916 s->rp_state.error = false;
1917 s->mbps = 0.0;
1918 s->pages_per_second = 0.0;
1919 s->downtime = 0;
1920 s->expected_downtime = 0;
1921 s->setup_time = 0;
1922 s->start_postcopy = false;
1923 s->postcopy_after_devices = false;
1924 s->migration_thread_running = false;
1925 error_free(s->error);
1926 s->error = NULL;
1927 s->hostname = NULL;
1928
1929 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1930
1931 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1932 s->total_time = 0;
1933 s->vm_was_running = false;
1934 s->iteration_initial_bytes = 0;
1935 s->threshold_size = 0;
1936 }
1937
1938 static GSList *migration_blockers;
1939
1940 int migrate_add_blocker(Error *reason, Error **errp)
1941 {
1942 if (only_migratable) {
1943 error_propagate_prepend(errp, error_copy(reason),
1944 "disallowing migration blocker "
1945 "(--only-migratable) for: ");
1946 return -EACCES;
1947 }
1948
1949 if (migration_is_idle()) {
1950 migration_blockers = g_slist_prepend(migration_blockers, reason);
1951 return 0;
1952 }
1953
1954 error_propagate_prepend(errp, error_copy(reason),
1955 "disallowing migration blocker "
1956 "(migration in progress) for: ");
1957 return -EBUSY;
1958 }
1959
1960 void migrate_del_blocker(Error *reason)
1961 {
1962 migration_blockers = g_slist_remove(migration_blockers, reason);
1963 }
1964
1965 void qmp_migrate_incoming(const char *uri, Error **errp)
1966 {
1967 Error *local_err = NULL;
1968 static bool once = true;
1969
1970 if (!deferred_incoming) {
1971 error_setg(errp, "For use with '-incoming defer'");
1972 return;
1973 }
1974 if (!once) {
1975 error_setg(errp, "The incoming migration has already been started");
1976 return;
1977 }
1978
1979 qemu_start_incoming_migration(uri, &local_err);
1980
1981 if (local_err) {
1982 error_propagate(errp, local_err);
1983 return;
1984 }
1985
1986 once = false;
1987 }
1988
1989 void qmp_migrate_recover(const char *uri, Error **errp)
1990 {
1991 MigrationIncomingState *mis = migration_incoming_get_current();
1992
1993 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1994 error_setg(errp, "Migrate recover can only be run "
1995 "when postcopy is paused.");
1996 return;
1997 }
1998
1999 if (qatomic_cmpxchg(&mis->postcopy_recover_triggered,
2000 false, true) == true) {
2001 error_setg(errp, "Migrate recovery is triggered already");
2002 return;
2003 }
2004
2005 /*
2006 * Note that this call will never start a real migration; it will
2007 * only re-setup the migration stream and poke existing migration
2008 * to continue using that newly established channel.
2009 */
2010 qemu_start_incoming_migration(uri, errp);
2011 }
2012
2013 void qmp_migrate_pause(Error **errp)
2014 {
2015 MigrationState *ms = migrate_get_current();
2016 MigrationIncomingState *mis = migration_incoming_get_current();
2017 int ret;
2018
2019 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2020 /* Source side, during postcopy */
2021 qemu_mutex_lock(&ms->qemu_file_lock);
2022 ret = qemu_file_shutdown(ms->to_dst_file);
2023 qemu_mutex_unlock(&ms->qemu_file_lock);
2024 if (ret) {
2025 error_setg(errp, "Failed to pause source migration");
2026 }
2027 return;
2028 }
2029
2030 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2031 ret = qemu_file_shutdown(mis->from_src_file);
2032 if (ret) {
2033 error_setg(errp, "Failed to pause destination migration");
2034 }
2035 return;
2036 }
2037
2038 error_setg(errp, "migrate-pause is currently only supported "
2039 "during postcopy-active state");
2040 }
2041
2042 bool migration_is_blocked(Error **errp)
2043 {
2044 if (qemu_savevm_state_blocked(errp)) {
2045 return true;
2046 }
2047
2048 if (migration_blockers) {
2049 error_propagate(errp, error_copy(migration_blockers->data));
2050 return true;
2051 }
2052
2053 return false;
2054 }
2055
2056 /* Returns true if continue to migrate, or false if error detected */
2057 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
2058 bool resume, Error **errp)
2059 {
2060 Error *local_err = NULL;
2061
2062 if (resume) {
2063 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2064 error_setg(errp, "Cannot resume if there is no "
2065 "paused migration");
2066 return false;
2067 }
2068
2069 /*
2070 * Postcopy recovery won't work well with release-ram
2071 * capability since release-ram will drop the page buffer as
2072 * long as the page is put into the send buffer. So if there
2073 * is a network failure happened, any page buffers that have
2074 * not yet reached the destination VM but have already been
2075 * sent from the source VM will be lost forever. Let's refuse
2076 * the client from resuming such a postcopy migration.
2077 * Luckily release-ram was designed to only be used when src
2078 * and destination VMs are on the same host, so it should be
2079 * fine.
2080 */
2081 if (migrate_release_ram()) {
2082 error_setg(errp, "Postcopy recovery cannot work "
2083 "when release-ram capability is set");
2084 return false;
2085 }
2086
2087 /* This is a resume, skip init status */
2088 return true;
2089 }
2090
2091 if (migration_is_running(s->state)) {
2092 error_setg(errp, QERR_MIGRATION_ACTIVE);
2093 return false;
2094 }
2095
2096 if (runstate_check(RUN_STATE_INMIGRATE)) {
2097 error_setg(errp, "Guest is waiting for an incoming migration");
2098 return false;
2099 }
2100
2101 if (migration_is_blocked(errp)) {
2102 return false;
2103 }
2104
2105 if (blk || blk_inc) {
2106 if (migrate_use_block() || migrate_use_block_incremental()) {
2107 error_setg(errp, "Command options are incompatible with "
2108 "current migration capabilities");
2109 return false;
2110 }
2111 migrate_set_block_enabled(true, &local_err);
2112 if (local_err) {
2113 error_propagate(errp, local_err);
2114 return false;
2115 }
2116 s->must_remove_block_options = true;
2117 }
2118
2119 if (blk_inc) {
2120 migrate_set_block_incremental(s, true);
2121 }
2122
2123 migrate_init(s);
2124 /*
2125 * set ram_counters memory to zero for a
2126 * new migration
2127 */
2128 memset(&ram_counters, 0, sizeof(ram_counters));
2129
2130 return true;
2131 }
2132
2133 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2134 bool has_inc, bool inc, bool has_detach, bool detach,
2135 bool has_resume, bool resume, Error **errp)
2136 {
2137 Error *local_err = NULL;
2138 MigrationState *s = migrate_get_current();
2139 const char *p = NULL;
2140
2141 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2142 has_resume && resume, errp)) {
2143 /* Error detected, put into errp */
2144 return;
2145 }
2146
2147 if (strstart(uri, "tcp:", &p) ||
2148 strstart(uri, "unix:", NULL) ||
2149 strstart(uri, "vsock:", NULL)) {
2150 socket_start_outgoing_migration(s, p ? p : uri, &local_err);
2151 #ifdef CONFIG_RDMA
2152 } else if (strstart(uri, "rdma:", &p)) {
2153 rdma_start_outgoing_migration(s, p, &local_err);
2154 #endif
2155 } else if (strstart(uri, "exec:", &p)) {
2156 exec_start_outgoing_migration(s, p, &local_err);
2157 } else if (strstart(uri, "fd:", &p)) {
2158 fd_start_outgoing_migration(s, p, &local_err);
2159 } else {
2160 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2161 "a valid migration protocol");
2162 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2163 MIGRATION_STATUS_FAILED);
2164 block_cleanup_parameters(s);
2165 return;
2166 }
2167
2168 if (local_err) {
2169 migrate_fd_error(s, local_err);
2170 error_propagate(errp, local_err);
2171 return;
2172 }
2173 }
2174
2175 void qmp_migrate_cancel(Error **errp)
2176 {
2177 migrate_fd_cancel(migrate_get_current());
2178 }
2179
2180 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2181 {
2182 MigrationState *s = migrate_get_current();
2183 if (s->state != state) {
2184 error_setg(errp, "Migration not in expected state: %s",
2185 MigrationStatus_str(s->state));
2186 return;
2187 }
2188 qemu_sem_post(&s->pause_sem);
2189 }
2190
2191 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
2192 {
2193 MigrateSetParameters p = {
2194 .has_xbzrle_cache_size = true,
2195 .xbzrle_cache_size = value,
2196 };
2197
2198 qmp_migrate_set_parameters(&p, errp);
2199 }
2200
2201 int64_t qmp_query_migrate_cache_size(Error **errp)
2202 {
2203 return migrate_xbzrle_cache_size();
2204 }
2205
2206 void qmp_migrate_set_speed(int64_t value, Error **errp)
2207 {
2208 MigrateSetParameters p = {
2209 .has_max_bandwidth = true,
2210 .max_bandwidth = value,
2211 };
2212
2213 qmp_migrate_set_parameters(&p, errp);
2214 }
2215
2216 void qmp_migrate_set_downtime(double value, Error **errp)
2217 {
2218 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2219 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
2220 "downtime_limit",
2221 "an integer in the range of 0 to "
2222 stringify(MAX_MIGRATE_DOWNTIME_SECONDS)" seconds");
2223 return;
2224 }
2225
2226 value *= 1000; /* Convert to milliseconds */
2227
2228 MigrateSetParameters p = {
2229 .has_downtime_limit = true,
2230 .downtime_limit = (int64_t)value,
2231 };
2232
2233 qmp_migrate_set_parameters(&p, errp);
2234 }
2235
2236 bool migrate_release_ram(void)
2237 {
2238 MigrationState *s;
2239
2240 s = migrate_get_current();
2241
2242 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2243 }
2244
2245 bool migrate_postcopy_ram(void)
2246 {
2247 MigrationState *s;
2248
2249 s = migrate_get_current();
2250
2251 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2252 }
2253
2254 bool migrate_postcopy(void)
2255 {
2256 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2257 }
2258
2259 bool migrate_auto_converge(void)
2260 {
2261 MigrationState *s;
2262
2263 s = migrate_get_current();
2264
2265 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2266 }
2267
2268 bool migrate_zero_blocks(void)
2269 {
2270 MigrationState *s;
2271
2272 s = migrate_get_current();
2273
2274 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2275 }
2276
2277 bool migrate_postcopy_blocktime(void)
2278 {
2279 MigrationState *s;
2280
2281 s = migrate_get_current();
2282
2283 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2284 }
2285
2286 bool migrate_use_compression(void)
2287 {
2288 MigrationState *s;
2289
2290 s = migrate_get_current();
2291
2292 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2293 }
2294
2295 int migrate_compress_level(void)
2296 {
2297 MigrationState *s;
2298
2299 s = migrate_get_current();
2300
2301 return s->parameters.compress_level;
2302 }
2303
2304 int migrate_compress_threads(void)
2305 {
2306 MigrationState *s;
2307
2308 s = migrate_get_current();
2309
2310 return s->parameters.compress_threads;
2311 }
2312
2313 int migrate_compress_wait_thread(void)
2314 {
2315 MigrationState *s;
2316
2317 s = migrate_get_current();
2318
2319 return s->parameters.compress_wait_thread;
2320 }
2321
2322 int migrate_decompress_threads(void)
2323 {
2324 MigrationState *s;
2325
2326 s = migrate_get_current();
2327
2328 return s->parameters.decompress_threads;
2329 }
2330
2331 bool migrate_dirty_bitmaps(void)
2332 {
2333 MigrationState *s;
2334
2335 s = migrate_get_current();
2336
2337 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2338 }
2339
2340 bool migrate_ignore_shared(void)
2341 {
2342 MigrationState *s;
2343
2344 s = migrate_get_current();
2345
2346 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2347 }
2348
2349 bool migrate_validate_uuid(void)
2350 {
2351 MigrationState *s;
2352
2353 s = migrate_get_current();
2354
2355 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2356 }
2357
2358 bool migrate_use_events(void)
2359 {
2360 MigrationState *s;
2361
2362 s = migrate_get_current();
2363
2364 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2365 }
2366
2367 bool migrate_use_multifd(void)
2368 {
2369 MigrationState *s;
2370
2371 s = migrate_get_current();
2372
2373 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2374 }
2375
2376 bool migrate_pause_before_switchover(void)
2377 {
2378 MigrationState *s;
2379
2380 s = migrate_get_current();
2381
2382 return s->enabled_capabilities[
2383 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2384 }
2385
2386 int migrate_multifd_channels(void)
2387 {
2388 MigrationState *s;
2389
2390 s = migrate_get_current();
2391
2392 return s->parameters.multifd_channels;
2393 }
2394
2395 MultiFDCompression migrate_multifd_compression(void)
2396 {
2397 MigrationState *s;
2398
2399 s = migrate_get_current();
2400
2401 return s->parameters.multifd_compression;
2402 }
2403
2404 int migrate_multifd_zlib_level(void)
2405 {
2406 MigrationState *s;
2407
2408 s = migrate_get_current();
2409
2410 return s->parameters.multifd_zlib_level;
2411 }
2412
2413 int migrate_multifd_zstd_level(void)
2414 {
2415 MigrationState *s;
2416
2417 s = migrate_get_current();
2418
2419 return s->parameters.multifd_zstd_level;
2420 }
2421
2422 int migrate_use_xbzrle(void)
2423 {
2424 MigrationState *s;
2425
2426 s = migrate_get_current();
2427
2428 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2429 }
2430
2431 int64_t migrate_xbzrle_cache_size(void)
2432 {
2433 MigrationState *s;
2434
2435 s = migrate_get_current();
2436
2437 return s->parameters.xbzrle_cache_size;
2438 }
2439
2440 static int64_t migrate_max_postcopy_bandwidth(void)
2441 {
2442 MigrationState *s;
2443
2444 s = migrate_get_current();
2445
2446 return s->parameters.max_postcopy_bandwidth;
2447 }
2448
2449 bool migrate_use_block(void)
2450 {
2451 MigrationState *s;
2452
2453 s = migrate_get_current();
2454
2455 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2456 }
2457
2458 bool migrate_use_return_path(void)
2459 {
2460 MigrationState *s;
2461
2462 s = migrate_get_current();
2463
2464 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2465 }
2466
2467 bool migrate_use_block_incremental(void)
2468 {
2469 MigrationState *s;
2470
2471 s = migrate_get_current();
2472
2473 return s->parameters.block_incremental;
2474 }
2475
2476 /* migration thread support */
2477 /*
2478 * Something bad happened to the RP stream, mark an error
2479 * The caller shall print or trace something to indicate why
2480 */
2481 static void mark_source_rp_bad(MigrationState *s)
2482 {
2483 s->rp_state.error = true;
2484 }
2485
2486 static struct rp_cmd_args {
2487 ssize_t len; /* -1 = variable */
2488 const char *name;
2489 } rp_cmd_args[] = {
2490 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2491 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2492 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2493 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2494 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2495 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2496 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2497 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2498 };
2499
2500 /*
2501 * Process a request for pages received on the return path,
2502 * We're allowed to send more than requested (e.g. to round to our page size)
2503 * and we don't need to send pages that have already been sent.
2504 */
2505 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2506 ram_addr_t start, size_t len)
2507 {
2508 long our_host_ps = qemu_real_host_page_size;
2509
2510 trace_migrate_handle_rp_req_pages(rbname, start, len);
2511
2512 /*
2513 * Since we currently insist on matching page sizes, just sanity check
2514 * we're being asked for whole host pages.
2515 */
2516 if (start & (our_host_ps - 1) ||
2517 (len & (our_host_ps - 1))) {
2518 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2519 " len: %zd", __func__, start, len);
2520 mark_source_rp_bad(ms);
2521 return;
2522 }
2523
2524 if (ram_save_queue_pages(rbname, start, len)) {
2525 mark_source_rp_bad(ms);
2526 }
2527 }
2528
2529 /* Return true to retry, false to quit */
2530 static bool postcopy_pause_return_path_thread(MigrationState *s)
2531 {
2532 trace_postcopy_pause_return_path();
2533
2534 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2535
2536 trace_postcopy_pause_return_path_continued();
2537
2538 return true;
2539 }
2540
2541 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2542 {
2543 RAMBlock *block = qemu_ram_block_by_name(block_name);
2544
2545 if (!block) {
2546 error_report("%s: invalid block name '%s'", __func__, block_name);
2547 return -EINVAL;
2548 }
2549
2550 /* Fetch the received bitmap and refresh the dirty bitmap */
2551 return ram_dirty_bitmap_reload(s, block);
2552 }
2553
2554 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2555 {
2556 trace_source_return_path_thread_resume_ack(value);
2557
2558 if (value != MIGRATION_RESUME_ACK_VALUE) {
2559 error_report("%s: illegal resume_ack value %"PRIu32,
2560 __func__, value);
2561 return -1;
2562 }
2563
2564 /* Now both sides are active. */
2565 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2566 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2567
2568 /* Notify send thread that time to continue send pages */
2569 qemu_sem_post(&s->rp_state.rp_sem);
2570
2571 return 0;
2572 }
2573
2574 /*
2575 * Handles messages sent on the return path towards the source VM
2576 *
2577 */
2578 static void *source_return_path_thread(void *opaque)
2579 {
2580 MigrationState *ms = opaque;
2581 QEMUFile *rp = ms->rp_state.from_dst_file;
2582 uint16_t header_len, header_type;
2583 uint8_t buf[512];
2584 uint32_t tmp32, sibling_error;
2585 ram_addr_t start = 0; /* =0 to silence warning */
2586 size_t len = 0, expected_len;
2587 int res;
2588
2589 trace_source_return_path_thread_entry();
2590 rcu_register_thread();
2591
2592 retry:
2593 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2594 migration_is_setup_or_active(ms->state)) {
2595 trace_source_return_path_thread_loop_top();
2596 header_type = qemu_get_be16(rp);
2597 header_len = qemu_get_be16(rp);
2598
2599 if (qemu_file_get_error(rp)) {
2600 mark_source_rp_bad(ms);
2601 goto out;
2602 }
2603
2604 if (header_type >= MIG_RP_MSG_MAX ||
2605 header_type == MIG_RP_MSG_INVALID) {
2606 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2607 header_type, header_len);
2608 mark_source_rp_bad(ms);
2609 goto out;
2610 }
2611
2612 if ((rp_cmd_args[header_type].len != -1 &&
2613 header_len != rp_cmd_args[header_type].len) ||
2614 header_len > sizeof(buf)) {
2615 error_report("RP: Received '%s' message (0x%04x) with"
2616 "incorrect length %d expecting %zu",
2617 rp_cmd_args[header_type].name, header_type, header_len,
2618 (size_t)rp_cmd_args[header_type].len);
2619 mark_source_rp_bad(ms);
2620 goto out;
2621 }
2622
2623 /* We know we've got a valid header by this point */
2624 res = qemu_get_buffer(rp, buf, header_len);
2625 if (res != header_len) {
2626 error_report("RP: Failed reading data for message 0x%04x"
2627 " read %d expected %d",
2628 header_type, res, header_len);
2629 mark_source_rp_bad(ms);
2630 goto out;
2631 }
2632
2633 /* OK, we have the message and the data */
2634 switch (header_type) {
2635 case MIG_RP_MSG_SHUT:
2636 sibling_error = ldl_be_p(buf);
2637 trace_source_return_path_thread_shut(sibling_error);
2638 if (sibling_error) {
2639 error_report("RP: Sibling indicated error %d", sibling_error);
2640 mark_source_rp_bad(ms);
2641 }
2642 /*
2643 * We'll let the main thread deal with closing the RP
2644 * we could do a shutdown(2) on it, but we're the only user
2645 * anyway, so there's nothing gained.
2646 */
2647 goto out;
2648
2649 case MIG_RP_MSG_PONG:
2650 tmp32 = ldl_be_p(buf);
2651 trace_source_return_path_thread_pong(tmp32);
2652 break;
2653
2654 case MIG_RP_MSG_REQ_PAGES:
2655 start = ldq_be_p(buf);
2656 len = ldl_be_p(buf + 8);
2657 migrate_handle_rp_req_pages(ms, NULL, start, len);
2658 break;
2659
2660 case MIG_RP_MSG_REQ_PAGES_ID:
2661 expected_len = 12 + 1; /* header + termination */
2662
2663 if (header_len >= expected_len) {
2664 start = ldq_be_p(buf);
2665 len = ldl_be_p(buf + 8);
2666 /* Now we expect an idstr */
2667 tmp32 = buf[12]; /* Length of the following idstr */
2668 buf[13 + tmp32] = '\0';
2669 expected_len += tmp32;
2670 }
2671 if (header_len != expected_len) {
2672 error_report("RP: Req_Page_id with length %d expecting %zd",
2673 header_len, expected_len);
2674 mark_source_rp_bad(ms);
2675 goto out;
2676 }
2677 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2678 break;
2679
2680 case MIG_RP_MSG_RECV_BITMAP:
2681 if (header_len < 1) {
2682 error_report("%s: missing block name", __func__);
2683 mark_source_rp_bad(ms);
2684 goto out;
2685 }
2686 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2687 buf[buf[0] + 1] = '\0';
2688 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2689 mark_source_rp_bad(ms);
2690 goto out;
2691 }
2692 break;
2693
2694 case MIG_RP_MSG_RESUME_ACK:
2695 tmp32 = ldl_be_p(buf);
2696 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2697 mark_source_rp_bad(ms);
2698 goto out;
2699 }
2700 break;
2701
2702 default:
2703 break;
2704 }
2705 }
2706
2707 out:
2708 res = qemu_file_get_error(rp);
2709 if (res) {
2710 if (res == -EIO && migration_in_postcopy()) {
2711 /*
2712 * Maybe there is something we can do: it looks like a
2713 * network down issue, and we pause for a recovery.
2714 */
2715 if (postcopy_pause_return_path_thread(ms)) {
2716 /* Reload rp, reset the rest */
2717 if (rp != ms->rp_state.from_dst_file) {
2718 qemu_fclose(rp);
2719 rp = ms->rp_state.from_dst_file;
2720 }
2721 ms->rp_state.error = false;
2722 goto retry;
2723 }
2724 }
2725
2726 trace_source_return_path_thread_bad_end();
2727 mark_source_rp_bad(ms);
2728 }
2729
2730 trace_source_return_path_thread_end();
2731 ms->rp_state.from_dst_file = NULL;
2732 qemu_fclose(rp);
2733 rcu_unregister_thread();
2734 return NULL;
2735 }
2736
2737 static int open_return_path_on_source(MigrationState *ms,
2738 bool create_thread)
2739 {
2740
2741 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2742 if (!ms->rp_state.from_dst_file) {
2743 return -1;
2744 }
2745
2746 trace_open_return_path_on_source();
2747
2748 if (!create_thread) {
2749 /* We're done */
2750 return 0;
2751 }
2752
2753 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2754 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2755
2756 trace_open_return_path_on_source_continue();
2757
2758 return 0;
2759 }
2760
2761 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2762 static int await_return_path_close_on_source(MigrationState *ms)
2763 {
2764 /*
2765 * If this is a normal exit then the destination will send a SHUT and the
2766 * rp_thread will exit, however if there's an error we need to cause
2767 * it to exit.
2768 */
2769 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2770 /*
2771 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2772 * waiting for the destination.
2773 */
2774 qemu_file_shutdown(ms->rp_state.from_dst_file);
2775 mark_source_rp_bad(ms);
2776 }
2777 trace_await_return_path_close_on_source_joining();
2778 qemu_thread_join(&ms->rp_state.rp_thread);
2779 trace_await_return_path_close_on_source_close();
2780 return ms->rp_state.error;
2781 }
2782
2783 /*
2784 * Switch from normal iteration to postcopy
2785 * Returns non-0 on error
2786 */
2787 static int postcopy_start(MigrationState *ms)
2788 {
2789 int ret;
2790 QIOChannelBuffer *bioc;
2791 QEMUFile *fb;
2792 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2793 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2794 bool restart_block = false;
2795 int cur_state = MIGRATION_STATUS_ACTIVE;
2796 if (!migrate_pause_before_switchover()) {
2797 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2798 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2799 }
2800
2801 trace_postcopy_start();
2802 qemu_mutex_lock_iothread();
2803 trace_postcopy_start_set_run();
2804
2805 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2806 global_state_store();
2807 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2808 if (ret < 0) {
2809 goto fail;
2810 }
2811
2812 ret = migration_maybe_pause(ms, &cur_state,
2813 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2814 if (ret < 0) {
2815 goto fail;
2816 }
2817
2818 ret = bdrv_inactivate_all();
2819 if (ret < 0) {
2820 goto fail;
2821 }
2822 restart_block = true;
2823
2824 /*
2825 * Cause any non-postcopiable, but iterative devices to
2826 * send out their final data.
2827 */
2828 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2829
2830 /*
2831 * in Finish migrate and with the io-lock held everything should
2832 * be quiet, but we've potentially still got dirty pages and we
2833 * need to tell the destination to throw any pages it's already received
2834 * that are dirty
2835 */
2836 if (migrate_postcopy_ram()) {
2837 if (ram_postcopy_send_discard_bitmap(ms)) {
2838 error_report("postcopy send discard bitmap failed");
2839 goto fail;
2840 }
2841 }
2842
2843 /*
2844 * send rest of state - note things that are doing postcopy
2845 * will notice we're in POSTCOPY_ACTIVE and not actually
2846 * wrap their state up here
2847 */
2848 /* 0 max-postcopy-bandwidth means unlimited */
2849 if (!bandwidth) {
2850 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2851 } else {
2852 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2853 }
2854 if (migrate_postcopy_ram()) {
2855 /* Ping just for debugging, helps line traces up */
2856 qemu_savevm_send_ping(ms->to_dst_file, 2);
2857 }
2858
2859 /*
2860 * While loading the device state we may trigger page transfer
2861 * requests and the fd must be free to process those, and thus
2862 * the destination must read the whole device state off the fd before
2863 * it starts processing it. Unfortunately the ad-hoc migration format
2864 * doesn't allow the destination to know the size to read without fully
2865 * parsing it through each devices load-state code (especially the open
2866 * coded devices that use get/put).
2867 * So we wrap the device state up in a package with a length at the start;
2868 * to do this we use a qemu_buf to hold the whole of the device state.
2869 */
2870 bioc = qio_channel_buffer_new(4096);
2871 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2872 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2873 object_unref(OBJECT(bioc));
2874
2875 /*
2876 * Make sure the receiver can get incoming pages before we send the rest
2877 * of the state
2878 */
2879 qemu_savevm_send_postcopy_listen(fb);
2880
2881 qemu_savevm_state_complete_precopy(fb, false, false);
2882 if (migrate_postcopy_ram()) {
2883 qemu_savevm_send_ping(fb, 3);
2884 }
2885
2886 qemu_savevm_send_postcopy_run(fb);
2887
2888 /* <><> end of stuff going into the package */
2889
2890 /* Last point of recovery; as soon as we send the package the destination
2891 * can open devices and potentially start running.
2892 * Lets just check again we've not got any errors.
2893 */
2894 ret = qemu_file_get_error(ms->to_dst_file);
2895 if (ret) {
2896 error_report("postcopy_start: Migration stream errored (pre package)");
2897 goto fail_closefb;
2898 }
2899
2900 restart_block = false;
2901
2902 /* Now send that blob */
2903 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2904 goto fail_closefb;
2905 }
2906 qemu_fclose(fb);
2907
2908 /* Send a notify to give a chance for anything that needs to happen
2909 * at the transition to postcopy and after the device state; in particular
2910 * spice needs to trigger a transition now
2911 */
2912 ms->postcopy_after_devices = true;
2913 notifier_list_notify(&migration_state_notifiers, ms);
2914
2915 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2916
2917 qemu_mutex_unlock_iothread();
2918
2919 if (migrate_postcopy_ram()) {
2920 /*
2921 * Although this ping is just for debug, it could potentially be
2922 * used for getting a better measurement of downtime at the source.
2923 */
2924 qemu_savevm_send_ping(ms->to_dst_file, 4);
2925 }
2926
2927 if (migrate_release_ram()) {
2928 ram_postcopy_migrated_memory_release(ms);
2929 }
2930
2931 ret = qemu_file_get_error(ms->to_dst_file);
2932 if (ret) {
2933 error_report("postcopy_start: Migration stream errored");
2934 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2935 MIGRATION_STATUS_FAILED);
2936 }
2937
2938 return ret;
2939
2940 fail_closefb:
2941 qemu_fclose(fb);
2942 fail:
2943 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2944 MIGRATION_STATUS_FAILED);
2945 if (restart_block) {
2946 /* A failure happened early enough that we know the destination hasn't
2947 * accessed block devices, so we're safe to recover.
2948 */
2949 Error *local_err = NULL;
2950
2951 bdrv_invalidate_cache_all(&local_err);
2952 if (local_err) {
2953 error_report_err(local_err);
2954 }
2955 }
2956 qemu_mutex_unlock_iothread();
2957 return -1;
2958 }
2959
2960 /**
2961 * migration_maybe_pause: Pause if required to by
2962 * migrate_pause_before_switchover called with the iothread locked
2963 * Returns: 0 on success
2964 */
2965 static int migration_maybe_pause(MigrationState *s,
2966 int *current_active_state,
2967 int new_state)
2968 {
2969 if (!migrate_pause_before_switchover()) {
2970 return 0;
2971 }
2972
2973 /* Since leaving this state is not atomic with posting the semaphore
2974 * it's possible that someone could have issued multiple migrate_continue
2975 * and the semaphore is incorrectly positive at this point;
2976 * the docs say it's undefined to reinit a semaphore that's already
2977 * init'd, so use timedwait to eat up any existing posts.
2978 */
2979 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2980 /* This block intentionally left blank */
2981 }
2982
2983 /*
2984 * If the migration is cancelled when it is in the completion phase,
2985 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2986 * So we don't need to wait a semaphore, otherwise we would always
2987 * wait for the 'pause_sem' semaphore.
2988 */
2989 if (s->state != MIGRATION_STATUS_CANCELLING) {
2990 qemu_mutex_unlock_iothread();
2991 migrate_set_state(&s->state, *current_active_state,
2992 MIGRATION_STATUS_PRE_SWITCHOVER);
2993 qemu_sem_wait(&s->pause_sem);
2994 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2995 new_state);
2996 *current_active_state = new_state;
2997 qemu_mutex_lock_iothread();
2998 }
2999
3000 return s->state == new_state ? 0 : -EINVAL;
3001 }
3002
3003 /**
3004 * migration_completion: Used by migration_thread when there's not much left.
3005 * The caller 'breaks' the loop when this returns.
3006 *
3007 * @s: Current migration state
3008 */
3009 static void migration_completion(MigrationState *s)
3010 {
3011 int ret;
3012 int current_active_state = s->state;
3013
3014 if (s->state == MIGRATION_STATUS_ACTIVE) {
3015 qemu_mutex_lock_iothread();
3016 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3017 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3018 s->vm_was_running = runstate_is_running();
3019 ret = global_state_store();
3020
3021 if (!ret) {
3022 bool inactivate = !migrate_colo_enabled();
3023 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
3024 if (ret >= 0) {
3025 ret = migration_maybe_pause(s, &current_active_state,
3026 MIGRATION_STATUS_DEVICE);
3027 }
3028 if (ret >= 0) {
3029 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3030 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
3031 inactivate);
3032 }
3033 if (inactivate && ret >= 0) {
3034 s->block_inactive = true;
3035 }
3036 }
3037 qemu_mutex_unlock_iothread();
3038
3039 if (ret < 0) {
3040 goto fail;
3041 }
3042 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3043 trace_migration_completion_postcopy_end();
3044
3045 qemu_savevm_state_complete_postcopy(s->to_dst_file);
3046 trace_migration_completion_postcopy_end_after_complete();
3047 }
3048
3049 /*
3050 * If rp was opened we must clean up the thread before
3051 * cleaning everything else up (since if there are no failures
3052 * it will wait for the destination to send it's status in
3053 * a SHUT command).
3054 */
3055 if (s->rp_state.from_dst_file) {
3056 int rp_error;
3057 trace_migration_return_path_end_before();
3058 rp_error = await_return_path_close_on_source(s);
3059 trace_migration_return_path_end_after(rp_error);
3060 if (rp_error) {
3061 goto fail_invalidate;
3062 }
3063 }
3064
3065 if (qemu_file_get_error(s->to_dst_file)) {
3066 trace_migration_completion_file_err();
3067 goto fail_invalidate;
3068 }
3069
3070 if (!migrate_colo_enabled()) {
3071 migrate_set_state(&s->state, current_active_state,
3072 MIGRATION_STATUS_COMPLETED);
3073 }
3074
3075 return;
3076
3077 fail_invalidate:
3078 /* If not doing postcopy, vm_start() will be called: let's regain
3079 * control on images.
3080 */
3081 if (s->state == MIGRATION_STATUS_ACTIVE ||
3082 s->state == MIGRATION_STATUS_DEVICE) {
3083 Error *local_err = NULL;
3084
3085 qemu_mutex_lock_iothread();
3086 bdrv_invalidate_cache_all(&local_err);
3087 if (local_err) {
3088 error_report_err(local_err);
3089 } else {
3090 s->block_inactive = false;
3091 }
3092 qemu_mutex_unlock_iothread();
3093 }
3094
3095 fail:
3096 migrate_set_state(&s->state, current_active_state,
3097 MIGRATION_STATUS_FAILED);
3098 }
3099
3100 bool migrate_colo_enabled(void)
3101 {
3102 MigrationState *s = migrate_get_current();
3103 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
3104 }
3105
3106 typedef enum MigThrError {
3107 /* No error detected */
3108 MIG_THR_ERR_NONE = 0,
3109 /* Detected error, but resumed successfully */
3110 MIG_THR_ERR_RECOVERED = 1,
3111 /* Detected fatal error, need to exit */
3112 MIG_THR_ERR_FATAL = 2,
3113 } MigThrError;
3114
3115 static int postcopy_resume_handshake(MigrationState *s)
3116 {
3117 qemu_savevm_send_postcopy_resume(s->to_dst_file);
3118
3119 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3120 qemu_sem_wait(&s->rp_state.rp_sem);
3121 }
3122
3123 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3124 return 0;
3125 }
3126
3127 return -1;
3128 }
3129
3130 /* Return zero if success, or <0 for error */
3131 static int postcopy_do_resume(MigrationState *s)
3132 {
3133 int ret;
3134
3135 /*
3136 * Call all the resume_prepare() hooks, so that modules can be
3137 * ready for the migration resume.
3138 */
3139 ret = qemu_savevm_state_resume_prepare(s);
3140 if (ret) {
3141 error_report("%s: resume_prepare() failure detected: %d",
3142 __func__, ret);
3143 return ret;
3144 }
3145
3146 /*
3147 * Last handshake with destination on the resume (destination will
3148 * switch to postcopy-active afterwards)
3149 */
3150 ret = postcopy_resume_handshake(s);
3151 if (ret) {
3152 error_report("%s: handshake failed: %d", __func__, ret);
3153 return ret;
3154 }
3155
3156 return 0;
3157 }
3158
3159 /*
3160 * We don't return until we are in a safe state to continue current
3161 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3162 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3163 */
3164 static MigThrError postcopy_pause(MigrationState *s)
3165 {
3166 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3167
3168 while (true) {
3169 QEMUFile *file;
3170
3171 /* Current channel is possibly broken. Release it. */
3172 assert(s->to_dst_file);
3173 qemu_mutex_lock(&s->qemu_file_lock);
3174 file = s->to_dst_file;
3175 s->to_dst_file = NULL;
3176 qemu_mutex_unlock(&s->qemu_file_lock);
3177
3178 qemu_file_shutdown(file);
3179 qemu_fclose(file);
3180
3181 migrate_set_state(&s->state, s->state,
3182 MIGRATION_STATUS_POSTCOPY_PAUSED);
3183
3184 error_report("Detected IO failure for postcopy. "
3185 "Migration paused.");
3186
3187 /*
3188 * We wait until things fixed up. Then someone will setup the
3189 * status back for us.
3190 */
3191 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3192 qemu_sem_wait(&s->postcopy_pause_sem);
3193 }
3194
3195 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3196 /* Woken up by a recover procedure. Give it a shot */
3197
3198 /*
3199 * Firstly, let's wake up the return path now, with a new
3200 * return path channel.
3201 */
3202 qemu_sem_post(&s->postcopy_pause_rp_sem);
3203
3204 /* Do the resume logic */
3205 if (postcopy_do_resume(s) == 0) {
3206 /* Let's continue! */
3207 trace_postcopy_pause_continued();
3208 return MIG_THR_ERR_RECOVERED;
3209 } else {
3210 /*
3211 * Something wrong happened during the recovery, let's
3212 * pause again. Pause is always better than throwing
3213 * data away.
3214 */
3215 continue;
3216 }
3217 } else {
3218 /* This is not right... Time to quit. */
3219 return MIG_THR_ERR_FATAL;
3220 }
3221 }
3222 }
3223
3224 static MigThrError migration_detect_error(MigrationState *s)
3225 {
3226 int ret;
3227 int state = s->state;
3228 Error *local_error = NULL;
3229
3230 if (state == MIGRATION_STATUS_CANCELLING ||
3231 state == MIGRATION_STATUS_CANCELLED) {
3232 /* End the migration, but don't set the state to failed */
3233 return MIG_THR_ERR_FATAL;
3234 }
3235
3236 /* Try to detect any file errors */
3237 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3238 if (!ret) {
3239 /* Everything is fine */
3240 assert(!local_error);
3241 return MIG_THR_ERR_NONE;
3242 }
3243
3244 if (local_error) {
3245 migrate_set_error(s, local_error);
3246 error_free(local_error);
3247 }
3248
3249 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3250 /*
3251 * For postcopy, we allow the network to be down for a
3252 * while. After that, it can be continued by a
3253 * recovery phase.
3254 */
3255 return postcopy_pause(s);
3256 } else {
3257 /*
3258 * For precopy (or postcopy with error outside IO), we fail
3259 * with no time.
3260 */
3261 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3262 trace_migration_thread_file_err();
3263
3264 /* Time to stop the migration, now. */
3265 return MIG_THR_ERR_FATAL;
3266 }
3267 }
3268
3269 /* How many bytes have we transferred since the beginning of the migration */
3270 static uint64_t migration_total_bytes(MigrationState *s)
3271 {
3272 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3273 }
3274
3275 static void migration_calculate_complete(MigrationState *s)
3276 {
3277 uint64_t bytes = migration_total_bytes(s);
3278 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3279 int64_t transfer_time;
3280
3281 s->total_time = end_time - s->start_time;
3282 if (!s->downtime) {
3283 /*
3284 * It's still not set, so we are precopy migration. For
3285 * postcopy, downtime is calculated during postcopy_start().
3286 */
3287 s->downtime = end_time - s->downtime_start;
3288 }
3289
3290 transfer_time = s->total_time - s->setup_time;
3291 if (transfer_time) {
3292 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3293 }
3294 }
3295
3296 static void update_iteration_initial_status(MigrationState *s)
3297 {
3298 /*
3299 * Update these three fields at the same time to avoid mismatch info lead
3300 * wrong speed calculation.
3301 */
3302 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3303 s->iteration_initial_bytes = migration_total_bytes(s);
3304 s->iteration_initial_pages = ram_get_total_transferred_pages();
3305 }
3306
3307 static void migration_update_counters(MigrationState *s,
3308 int64_t current_time)
3309 {
3310 uint64_t transferred, transferred_pages, time_spent;
3311 uint64_t current_bytes; /* bytes transferred since the beginning */
3312 double bandwidth;
3313
3314 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3315 return;
3316 }
3317
3318 current_bytes = migration_total_bytes(s);
3319 transferred = current_bytes - s->iteration_initial_bytes;
3320 time_spent = current_time - s->iteration_start_time;
3321 bandwidth = (double)transferred / time_spent;
3322 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3323
3324 s->mbps = (((double) transferred * 8.0) /
3325 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3326
3327 transferred_pages = ram_get_total_transferred_pages() -
3328 s->iteration_initial_pages;
3329 s->pages_per_second = (double) transferred_pages /
3330 (((double) time_spent / 1000.0));
3331
3332 /*
3333 * if we haven't sent anything, we don't want to
3334 * recalculate. 10000 is a small enough number for our purposes
3335 */
3336 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3337 s->expected_downtime = ram_counters.remaining / bandwidth;
3338 }
3339
3340 qemu_file_reset_rate_limit(s->to_dst_file);
3341
3342 update_iteration_initial_status(s);
3343
3344 trace_migrate_transferred(transferred, time_spent,
3345 bandwidth, s->threshold_size);
3346 }
3347
3348 /* Migration thread iteration status */
3349 typedef enum {
3350 MIG_ITERATE_RESUME, /* Resume current iteration */
3351 MIG_ITERATE_SKIP, /* Skip current iteration */
3352 MIG_ITERATE_BREAK, /* Break the loop */
3353 } MigIterateState;
3354
3355 /*
3356 * Return true if continue to the next iteration directly, false
3357 * otherwise.
3358 */
3359 static MigIterateState migration_iteration_run(MigrationState *s)
3360 {
3361 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3362 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3363
3364 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3365 &pend_compat, &pend_post);
3366 pending_size = pend_pre + pend_compat + pend_post;
3367
3368 trace_migrate_pending(pending_size, s->threshold_size,
3369 pend_pre, pend_compat, pend_post);
3370
3371 if (pending_size && pending_size >= s->threshold_size) {
3372 /* Still a significant amount to transfer */
3373 if (!in_postcopy && pend_pre <= s->threshold_size &&
3374 qatomic_read(&s->start_postcopy)) {
3375 if (postcopy_start(s)) {
3376 error_report("%s: postcopy failed to start", __func__);
3377 }
3378 return MIG_ITERATE_SKIP;
3379 }
3380 /* Just another iteration step */
3381 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3382 } else {
3383 trace_migration_thread_low_pending(pending_size);
3384 migration_completion(s);
3385 return MIG_ITERATE_BREAK;
3386 }
3387
3388 return MIG_ITERATE_RESUME;
3389 }
3390
3391 static void migration_iteration_finish(MigrationState *s)
3392 {
3393 /* If we enabled cpu throttling for auto-converge, turn it off. */
3394 cpu_throttle_stop();
3395
3396 qemu_mutex_lock_iothread();
3397 switch (s->state) {
3398 case MIGRATION_STATUS_COMPLETED:
3399 migration_calculate_complete(s);
3400 runstate_set(RUN_STATE_POSTMIGRATE);
3401 break;
3402
3403 case MIGRATION_STATUS_ACTIVE:
3404 /*
3405 * We should really assert here, but since it's during
3406 * migration, let's try to reduce the usage of assertions.
3407 */
3408 if (!migrate_colo_enabled()) {
3409 error_report("%s: critical error: calling COLO code without "
3410 "COLO enabled", __func__);
3411 }
3412 migrate_start_colo_process(s);
3413 /*
3414 * Fixme: we will run VM in COLO no matter its old running state.
3415 * After exited COLO, we will keep running.
3416 */
3417 s->vm_was_running = true;
3418 /* Fallthrough */
3419 case MIGRATION_STATUS_FAILED:
3420 case MIGRATION_STATUS_CANCELLED:
3421 case MIGRATION_STATUS_CANCELLING:
3422 if (s->vm_was_running) {
3423 vm_start();
3424 } else {
3425 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3426 runstate_set(RUN_STATE_POSTMIGRATE);
3427 }
3428 }
3429 break;
3430
3431 default:
3432 /* Should not reach here, but if so, forgive the VM. */
3433 error_report("%s: Unknown ending state %d", __func__, s->state);
3434 break;
3435 }
3436 migrate_fd_cleanup_schedule(s);
3437 qemu_mutex_unlock_iothread();
3438 }
3439
3440 void migration_make_urgent_request(void)
3441 {