s390x/pci: use a PCI Function structure
[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 #ifdef CONFIG_VFIO
61 #include "hw/vfio/vfio-common.h"
62 #endif
63
64 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
65
66 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
67 * data. */
68 #define BUFFER_DELAY 100
69 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
70
71 /* Time in milliseconds we are allowed to stop the source,
72 * for sending the last part */
73 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
74
75 /* Maximum migrate downtime set to 2000 seconds */
76 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
77 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
78
79 /* Default compression thread count */
80 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
81 /* Default decompression thread count, usually decompression is at
82 * least 4 times as fast as compression.*/
83 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
84 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
85 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
86 /* Define default autoconverge cpu throttle migration parameters */
87 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
88 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
89 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
90 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
91
92 /* Migration XBZRLE default cache size */
93 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
94
95 /* The delay time (in ms) between two COLO checkpoints */
96 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
97 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
98 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
99 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
100 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
101 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
102 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
103
104 /* Background transfer rate for postcopy, 0 means unlimited, note
105 * that page requests can still exceed this limit.
106 */
107 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
108
109 /*
110 * Parameters for self_announce_delay giving a stream of RARP/ARP
111 * packets after migration.
112 */
113 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
114 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
115 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
116 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
117
118 static NotifierList migration_state_notifiers =
119 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
120
121 static bool deferred_incoming;
122
123 /* Messages sent on the return path from destination to source */
124 enum mig_rp_message_type {
125 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
126 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
127 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
128
129 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
130 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
131 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
132 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
133
134 MIG_RP_MSG_MAX
135 };
136
137 /* When we add fault tolerance, we could have several
138 migrations at once. For now we don't need to add
139 dynamic creation of migration */
140
141 static MigrationState *current_migration;
142 static MigrationIncomingState *current_incoming;
143
144 static bool migration_object_check(MigrationState *ms, Error **errp);
145 static int migration_maybe_pause(MigrationState *s,
146 int *current_active_state,
147 int new_state);
148 static void migrate_fd_cancel(MigrationState *s);
149
150 static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp)
151 {
152 uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp;
153
154 return (a > b) - (a < b);
155 }
156
157 void migration_object_init(void)
158 {
159 Error *err = NULL;
160
161 /* This can only be called once. */
162 assert(!current_migration);
163 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
164
165 /*
166 * Init the migrate incoming object as well no matter whether
167 * we'll use it or not.
168 */
169 assert(!current_incoming);
170 current_incoming = g_new0(MigrationIncomingState, 1);
171 current_incoming->state = MIGRATION_STATUS_NONE;
172 current_incoming->postcopy_remote_fds =
173 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
174 qemu_mutex_init(&current_incoming->rp_mutex);
175 qemu_event_init(&current_incoming->main_thread_load_event, false);
176 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
177 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
178 qemu_mutex_init(&current_incoming->page_request_mutex);
179 current_incoming->page_requested = g_tree_new(page_request_addr_cmp);
180
181 if (!migration_object_check(current_migration, &err)) {
182 error_report_err(err);
183 exit(1);
184 }
185 }
186
187 void migration_shutdown(void)
188 {
189 /*
190 * Cancel the current migration - that will (eventually)
191 * stop the migration using this structure
192 */
193 migrate_fd_cancel(current_migration);
194 object_unref(OBJECT(current_migration));
195
196 /*
197 * Cancel outgoing migration of dirty bitmaps. It should
198 * at least unref used block nodes.
199 */
200 dirty_bitmap_mig_cancel_outgoing();
201
202 /*
203 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
204 * are non-critical data, and their loss never considered as
205 * something serious.
206 */
207 dirty_bitmap_mig_cancel_incoming();
208 }
209
210 /* For outgoing */
211 MigrationState *migrate_get_current(void)
212 {
213 /* This can only be called after the object created. */
214 assert(current_migration);
215 return current_migration;
216 }
217
218 MigrationIncomingState *migration_incoming_get_current(void)
219 {
220 assert(current_incoming);
221 return current_incoming;
222 }
223
224 void migration_incoming_state_destroy(void)
225 {
226 struct MigrationIncomingState *mis = migration_incoming_get_current();
227
228 if (mis->to_src_file) {
229 /* Tell source that we are done */
230 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
231 qemu_fclose(mis->to_src_file);
232 mis->to_src_file = NULL;
233 }
234
235 if (mis->from_src_file) {
236 qemu_fclose(mis->from_src_file);
237 mis->from_src_file = NULL;
238 }
239 if (mis->postcopy_remote_fds) {
240 g_array_free(mis->postcopy_remote_fds, TRUE);
241 mis->postcopy_remote_fds = NULL;
242 }
243
244 qemu_event_reset(&mis->main_thread_load_event);
245
246 if (mis->page_requested) {
247 g_tree_destroy(mis->page_requested);
248 mis->page_requested = NULL;
249 }
250
251 if (mis->socket_address_list) {
252 qapi_free_SocketAddressList(mis->socket_address_list);
253 mis->socket_address_list = NULL;
254 }
255 }
256
257 static void migrate_generate_event(int new_state)
258 {
259 if (migrate_use_events()) {
260 qapi_event_send_migration(new_state);
261 }
262 }
263
264 static bool migrate_late_block_activate(void)
265 {
266 MigrationState *s;
267
268 s = migrate_get_current();
269
270 return s->enabled_capabilities[
271 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
272 }
273
274 /*
275 * Called on -incoming with a defer: uri.
276 * The migration can be started later after any parameters have been
277 * changed.
278 */
279 static void deferred_incoming_migration(Error **errp)
280 {
281 if (deferred_incoming) {
282 error_setg(errp, "Incoming migration already deferred");
283 }
284 deferred_incoming = true;
285 }
286
287 /*
288 * Send a message on the return channel back to the source
289 * of the migration.
290 */
291 static int migrate_send_rp_message(MigrationIncomingState *mis,
292 enum mig_rp_message_type message_type,
293 uint16_t len, void *data)
294 {
295 int ret = 0;
296
297 trace_migrate_send_rp_message((int)message_type, len);
298 qemu_mutex_lock(&mis->rp_mutex);
299
300 /*
301 * It's possible that the file handle got lost due to network
302 * failures.
303 */
304 if (!mis->to_src_file) {
305 ret = -EIO;
306 goto error;
307 }
308
309 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
310 qemu_put_be16(mis->to_src_file, len);
311 qemu_put_buffer(mis->to_src_file, data, len);
312 qemu_fflush(mis->to_src_file);
313
314 /* It's possible that qemu file got error during sending */
315 ret = qemu_file_get_error(mis->to_src_file);
316
317 error:
318 qemu_mutex_unlock(&mis->rp_mutex);
319 return ret;
320 }
321
322 /* Request one page from the source VM at the given start address.
323 * rb: the RAMBlock to request the page in
324 * Start: Address offset within the RB
325 * Len: Length in bytes required - must be a multiple of pagesize
326 */
327 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
328 RAMBlock *rb, ram_addr_t start)
329 {
330 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
331 size_t msglen = 12; /* start + len */
332 size_t len = qemu_ram_pagesize(rb);
333 enum mig_rp_message_type msg_type;
334 const char *rbname;
335 int rbname_len;
336
337 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
338 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
339
340 /*
341 * We maintain the last ramblock that we requested for page. Note that we
342 * don't need locking because this function will only be called within the
343 * postcopy ram fault thread.
344 */
345 if (rb != mis->last_rb) {
346 mis->last_rb = rb;
347
348 rbname = qemu_ram_get_idstr(rb);
349 rbname_len = strlen(rbname);
350
351 assert(rbname_len < 256);
352
353 bufc[msglen++] = rbname_len;
354 memcpy(bufc + msglen, rbname, rbname_len);
355 msglen += rbname_len;
356 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
357 } else {
358 msg_type = MIG_RP_MSG_REQ_PAGES;
359 }
360
361 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
362 }
363
364 int migrate_send_rp_req_pages(MigrationIncomingState *mis,
365 RAMBlock *rb, ram_addr_t start, uint64_t haddr)
366 {
367 void *aligned = (void *)(uintptr_t)(haddr & (-qemu_ram_pagesize(rb)));
368 bool received;
369
370 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
371 received = ramblock_recv_bitmap_test_byte_offset(rb, start);
372 if (!received && !g_tree_lookup(mis->page_requested, aligned)) {
373 /*
374 * The page has not been received, and it's not yet in the page
375 * request list. Queue it. Set the value of element to 1, so that
376 * things like g_tree_lookup() will return TRUE (1) when found.
377 */
378 g_tree_insert(mis->page_requested, aligned, (gpointer)1);
379 mis->page_requested_count++;
380 trace_postcopy_page_req_add(aligned, mis->page_requested_count);
381 }
382 }
383
384 /*
385 * If the page is there, skip sending the message. We don't even need the
386 * lock because as long as the page arrived, it'll be there forever.
387 */
388 if (received) {
389 return 0;
390 }
391
392 return migrate_send_rp_message_req_pages(mis, rb, start);
393 }
394
395 static bool migration_colo_enabled;
396 bool migration_incoming_colo_enabled(void)
397 {
398 return migration_colo_enabled;
399 }
400
401 void migration_incoming_disable_colo(void)
402 {
403 ram_block_discard_disable(false);
404 migration_colo_enabled = false;
405 }
406
407 int migration_incoming_enable_colo(void)
408 {
409 if (ram_block_discard_disable(true)) {
410 error_report("COLO: cannot disable RAM discard");
411 return -EBUSY;
412 }
413 migration_colo_enabled = true;
414 return 0;
415 }
416
417 void migrate_add_address(SocketAddress *address)
418 {
419 MigrationIncomingState *mis = migration_incoming_get_current();
420 SocketAddressList *addrs;
421
422 addrs = g_new0(SocketAddressList, 1);
423 addrs->next = mis->socket_address_list;
424 mis->socket_address_list = addrs;
425 addrs->value = QAPI_CLONE(SocketAddress, address);
426 }
427
428 void qemu_start_incoming_migration(const char *uri, Error **errp)
429 {
430 const char *p = NULL;
431
432 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
433 if (!strcmp(uri, "defer")) {
434 deferred_incoming_migration(errp);
435 } else if (strstart(uri, "tcp:", &p) ||
436 strstart(uri, "unix:", NULL) ||
437 strstart(uri, "vsock:", NULL)) {
438 socket_start_incoming_migration(p ? p : uri, errp);
439 #ifdef CONFIG_RDMA
440 } else if (strstart(uri, "rdma:", &p)) {
441 rdma_start_incoming_migration(p, errp);
442 #endif
443 } else if (strstart(uri, "exec:", &p)) {
444 exec_start_incoming_migration(p, errp);
445 } else if (strstart(uri, "fd:", &p)) {
446 fd_start_incoming_migration(p, errp);
447 } else {
448 error_setg(errp, "unknown migration protocol: %s", uri);
449 }
450 }
451
452 static void process_incoming_migration_bh(void *opaque)
453 {
454 Error *local_err = NULL;
455 MigrationIncomingState *mis = opaque;
456
457 /* If capability late_block_activate is set:
458 * Only fire up the block code now if we're going to restart the
459 * VM, else 'cont' will do it.
460 * This causes file locking to happen; so we don't want it to happen
461 * unless we really are starting the VM.
462 */
463 if (!migrate_late_block_activate() ||
464 (autostart && (!global_state_received() ||
465 global_state_get_runstate() == RUN_STATE_RUNNING))) {
466 /* Make sure all file formats flush their mutable metadata.
467 * If we get an error here, just don't restart the VM yet. */
468 bdrv_invalidate_cache_all(&local_err);
469 if (local_err) {
470 error_report_err(local_err);
471 local_err = NULL;
472 autostart = false;
473 }
474 }
475
476 /*
477 * This must happen after all error conditions are dealt with and
478 * we're sure the VM is going to be running on this host.
479 */
480 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
481
482 if (multifd_load_cleanup(&local_err) != 0) {
483 error_report_err(local_err);
484 autostart = false;
485 }
486 /* If global state section was not received or we are in running
487 state, we need to obey autostart. Any other state is set with
488 runstate_set. */
489
490 dirty_bitmap_mig_before_vm_start();
491
492 if (!global_state_received() ||
493 global_state_get_runstate() == RUN_STATE_RUNNING) {
494 if (autostart) {
495 vm_start();
496 } else {
497 runstate_set(RUN_STATE_PAUSED);
498 }
499 } else if (migration_incoming_colo_enabled()) {
500 migration_incoming_disable_colo();
501 vm_start();
502 } else {
503 runstate_set(global_state_get_runstate());
504 }
505 /*
506 * This must happen after any state changes since as soon as an external
507 * observer sees this event they might start to prod at the VM assuming
508 * it's ready to use.
509 */
510 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
511 MIGRATION_STATUS_COMPLETED);
512 qemu_bh_delete(mis->bh);
513 migration_incoming_state_destroy();
514 }
515
516 static void process_incoming_migration_co(void *opaque)
517 {
518 MigrationIncomingState *mis = migration_incoming_get_current();
519 PostcopyState ps;
520 int ret;
521 Error *local_err = NULL;
522
523 assert(mis->from_src_file);
524 mis->migration_incoming_co = qemu_coroutine_self();
525 mis->largest_page_size = qemu_ram_pagesize_largest();
526 postcopy_state_set(POSTCOPY_INCOMING_NONE);
527 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
528 MIGRATION_STATUS_ACTIVE);
529 ret = qemu_loadvm_state(mis->from_src_file);
530
531 ps = postcopy_state_get();
532 trace_process_incoming_migration_co_end(ret, ps);
533 if (ps != POSTCOPY_INCOMING_NONE) {
534 if (ps == POSTCOPY_INCOMING_ADVISE) {
535 /*
536 * Where a migration had postcopy enabled (and thus went to advise)
537 * but managed to complete within the precopy period, we can use
538 * the normal exit.
539 */
540 postcopy_ram_incoming_cleanup(mis);
541 } else if (ret >= 0) {
542 /*
543 * Postcopy was started, cleanup should happen at the end of the
544 * postcopy thread.
545 */
546 trace_process_incoming_migration_co_postcopy_end_main();
547 return;
548 }
549 /* Else if something went wrong then just fall out of the normal exit */
550 }
551
552 /* we get COLO info, and know if we are in COLO mode */
553 if (!ret && migration_incoming_colo_enabled()) {
554 /* Make sure all file formats flush their mutable metadata */
555 bdrv_invalidate_cache_all(&local_err);
556 if (local_err) {
557 error_report_err(local_err);
558 goto fail;
559 }
560
561 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
562 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
563 mis->have_colo_incoming_thread = true;
564 qemu_coroutine_yield();
565
566 /* Wait checkpoint incoming thread exit before free resource */
567 qemu_thread_join(&mis->colo_incoming_thread);
568 /* We hold the global iothread lock, so it is safe here */
569 colo_release_ram_cache();
570 }
571
572 if (ret < 0) {
573 error_report("load of migration failed: %s", strerror(-ret));
574 goto fail;
575 }
576 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
577 qemu_bh_schedule(mis->bh);
578 mis->migration_incoming_co = NULL;
579 return;
580 fail:
581 local_err = NULL;
582 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
583 MIGRATION_STATUS_FAILED);
584 qemu_fclose(mis->from_src_file);
585 if (multifd_load_cleanup(&local_err) != 0) {
586 error_report_err(local_err);
587 }
588 exit(EXIT_FAILURE);
589 }
590
591 /**
592 * @migration_incoming_setup: Setup incoming migration
593 *
594 * Returns 0 for no error or 1 for error
595 *
596 * @f: file for main migration channel
597 * @errp: where to put errors
598 */
599 static int migration_incoming_setup(QEMUFile *f, Error **errp)
600 {
601 MigrationIncomingState *mis = migration_incoming_get_current();
602 Error *local_err = NULL;
603
604 if (multifd_load_setup(&local_err) != 0) {
605 /* We haven't been able to create multifd threads
606 nothing better to do */
607 error_report_err(local_err);
608 exit(EXIT_FAILURE);
609 }
610
611 if (!mis->from_src_file) {
612 mis->from_src_file = f;
613 }
614 qemu_file_set_blocking(f, false);
615 return 0;
616 }
617
618 void migration_incoming_process(void)
619 {
620 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
621 qemu_coroutine_enter(co);
622 }
623
624 /* Returns true if recovered from a paused migration, otherwise false */
625 static bool postcopy_try_recover(QEMUFile *f)
626 {
627 MigrationIncomingState *mis = migration_incoming_get_current();
628
629 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
630 /* Resumed from a paused postcopy migration */
631
632 mis->from_src_file = f;
633 /* Postcopy has standalone thread to do vm load */
634 qemu_file_set_blocking(f, true);
635
636 /* Re-configure the return path */
637 mis->to_src_file = qemu_file_get_return_path(f);
638
639 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
640 MIGRATION_STATUS_POSTCOPY_RECOVER);
641
642 /*
643 * Here, we only wake up the main loading thread (while the
644 * fault thread will still be waiting), so that we can receive
645 * commands from source now, and answer it if needed. The
646 * fault thread will be woken up afterwards until we are sure
647 * that source is ready to reply to page requests.
648 */
649 qemu_sem_post(&mis->postcopy_pause_sem_dst);
650 return true;
651 }
652
653 return false;
654 }
655
656 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
657 {
658 Error *local_err = NULL;
659
660 if (postcopy_try_recover(f)) {
661 return;
662 }
663
664 if (migration_incoming_setup(f, &local_err)) {
665 error_propagate(errp, local_err);
666 return;
667 }
668 migration_incoming_process();
669 }
670
671 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
672 {
673 MigrationIncomingState *mis = migration_incoming_get_current();
674 Error *local_err = NULL;
675 bool start_migration;
676
677 if (!mis->from_src_file) {
678 /* The first connection (multifd may have multiple) */
679 QEMUFile *f = qemu_fopen_channel_input(ioc);
680
681 /* If it's a recovery, we're done */
682 if (postcopy_try_recover(f)) {
683 return;
684 }
685
686 if (migration_incoming_setup(f, &local_err)) {
687 error_propagate(errp, local_err);
688 return;
689 }
690
691 /*
692 * Common migration only needs one channel, so we can start
693 * right now. Multifd needs more than one channel, we wait.
694 */
695 start_migration = !migrate_use_multifd();
696 } else {
697 /* Multiple connections */
698 assert(migrate_use_multifd());
699 start_migration = multifd_recv_new_channel(ioc, &local_err);
700 if (local_err) {
701 error_propagate(errp, local_err);
702 return;
703 }
704 }
705
706 if (start_migration) {
707 migration_incoming_process();
708 }
709 }
710
711 /**
712 * @migration_has_all_channels: We have received all channels that we need
713 *
714 * Returns true when we have got connections to all the channels that
715 * we need for migration.
716 */
717 bool migration_has_all_channels(void)
718 {
719 MigrationIncomingState *mis = migration_incoming_get_current();
720 bool all_channels;
721
722 all_channels = multifd_recv_all_channels_created();
723
724 return all_channels && mis->from_src_file != NULL;
725 }
726
727 /*
728 * Send a 'SHUT' message on the return channel with the given value
729 * to indicate that we've finished with the RP. Non-0 value indicates
730 * error.
731 */
732 void migrate_send_rp_shut(MigrationIncomingState *mis,
733 uint32_t value)
734 {
735 uint32_t buf;
736
737 buf = cpu_to_be32(value);
738 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
739 }
740
741 /*
742 * Send a 'PONG' message on the return channel with the given value
743 * (normally in response to a 'PING')
744 */
745 void migrate_send_rp_pong(MigrationIncomingState *mis,
746 uint32_t value)
747 {
748 uint32_t buf;
749
750 buf = cpu_to_be32(value);
751 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
752 }
753
754 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
755 char *block_name)
756 {
757 char buf[512];
758 int len;
759 int64_t res;
760
761 /*
762 * First, we send the header part. It contains only the len of
763 * idstr, and the idstr itself.
764 */
765 len = strlen(block_name);
766 buf[0] = len;
767 memcpy(buf + 1, block_name, len);
768
769 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
770 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
771 __func__);
772 return;
773 }
774
775 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
776
777 /*
778 * Next, we dump the received bitmap to the stream.
779 *
780 * TODO: currently we are safe since we are the only one that is
781 * using the to_src_file handle (fault thread is still paused),
782 * and it's ok even not taking the mutex. However the best way is
783 * to take the lock before sending the message header, and release
784 * the lock after sending the bitmap.
785 */
786 qemu_mutex_lock(&mis->rp_mutex);
787 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
788 qemu_mutex_unlock(&mis->rp_mutex);
789
790 trace_migrate_send_rp_recv_bitmap(block_name, res);
791 }
792
793 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
794 {
795 uint32_t buf;
796
797 buf = cpu_to_be32(value);
798 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
799 }
800
801 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
802 {
803 MigrationCapabilityStatusList *head = NULL;
804 MigrationCapabilityStatusList *caps;
805 MigrationState *s = migrate_get_current();
806 int i;
807
808 caps = NULL; /* silence compiler warning */
809 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
810 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
811 if (i == MIGRATION_CAPABILITY_BLOCK) {
812 continue;
813 }
814 #endif
815 if (head == NULL) {
816 head = g_malloc0(sizeof(*caps));
817 caps = head;
818 } else {
819 caps->next = g_malloc0(sizeof(*caps));
820 caps = caps->next;
821 }
822 caps->value =
823 g_malloc(sizeof(*caps->value));
824 caps->value->capability = i;
825 caps->value->state = s->enabled_capabilities[i];
826 }
827
828 return head;
829 }
830
831 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
832 {
833 MigrationParameters *params;
834 MigrationState *s = migrate_get_current();
835
836 /* TODO use QAPI_CLONE() instead of duplicating it inline */
837 params = g_malloc0(sizeof(*params));
838 params->has_compress_level = true;
839 params->compress_level = s->parameters.compress_level;
840 params->has_compress_threads = true;
841 params->compress_threads = s->parameters.compress_threads;
842 params->has_compress_wait_thread = true;
843 params->compress_wait_thread = s->parameters.compress_wait_thread;
844 params->has_decompress_threads = true;
845 params->decompress_threads = s->parameters.decompress_threads;
846 params->has_throttle_trigger_threshold = true;
847 params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
848 params->has_cpu_throttle_initial = true;
849 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
850 params->has_cpu_throttle_increment = true;
851 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
852 params->has_cpu_throttle_tailslow = true;
853 params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
854 params->has_tls_creds = true;
855 params->tls_creds = g_strdup(s->parameters.tls_creds);
856 params->has_tls_hostname = true;
857 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
858 params->has_tls_authz = true;
859 params->tls_authz = g_strdup(s->parameters.tls_authz ?
860 s->parameters.tls_authz : "");
861 params->has_max_bandwidth = true;
862 params->max_bandwidth = s->parameters.max_bandwidth;
863 params->has_downtime_limit = true;
864 params->downtime_limit = s->parameters.downtime_limit;
865 params->has_x_checkpoint_delay = true;
866 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
867 params->has_block_incremental = true;
868 params->block_incremental = s->parameters.block_incremental;
869 params->has_multifd_channels = true;
870 params->multifd_channels = s->parameters.multifd_channels;
871 params->has_multifd_compression = true;
872 params->multifd_compression = s->parameters.multifd_compression;
873 params->has_multifd_zlib_level = true;
874 params->multifd_zlib_level = s->parameters.multifd_zlib_level;
875 params->has_multifd_zstd_level = true;
876 params->multifd_zstd_level = s->parameters.multifd_zstd_level;
877 params->has_xbzrle_cache_size = true;
878 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
879 params->has_max_postcopy_bandwidth = true;
880 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
881 params->has_max_cpu_throttle = true;
882 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
883 params->has_announce_initial = true;
884 params->announce_initial = s->parameters.announce_initial;
885 params->has_announce_max = true;
886 params->announce_max = s->parameters.announce_max;
887 params->has_announce_rounds = true;
888 params->announce_rounds = s->parameters.announce_rounds;
889 params->has_announce_step = true;
890 params->announce_step = s->parameters.announce_step;
891
892 if (s->parameters.has_block_bitmap_mapping) {
893 params->has_block_bitmap_mapping = true;
894 params->block_bitmap_mapping =
895 QAPI_CLONE(BitmapMigrationNodeAliasList,
896 s->parameters.block_bitmap_mapping);
897 }
898
899 return params;
900 }
901
902 AnnounceParameters *migrate_announce_params(void)
903 {
904 static AnnounceParameters ap;
905
906 MigrationState *s = migrate_get_current();
907
908 ap.initial = s->parameters.announce_initial;
909 ap.max = s->parameters.announce_max;
910 ap.rounds = s->parameters.announce_rounds;
911 ap.step = s->parameters.announce_step;
912
913 return &ap;
914 }
915
916 /*
917 * Return true if we're already in the middle of a migration
918 * (i.e. any of the active or setup states)
919 */
920 bool migration_is_setup_or_active(int state)
921 {
922 switch (state) {
923 case MIGRATION_STATUS_ACTIVE:
924 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
925 case MIGRATION_STATUS_POSTCOPY_PAUSED:
926 case MIGRATION_STATUS_POSTCOPY_RECOVER:
927 case MIGRATION_STATUS_SETUP:
928 case MIGRATION_STATUS_PRE_SWITCHOVER:
929 case MIGRATION_STATUS_DEVICE:
930 case MIGRATION_STATUS_WAIT_UNPLUG:
931 case MIGRATION_STATUS_COLO:
932 return true;
933
934 default:
935 return false;
936
937 }
938 }
939
940 bool migration_is_running(int state)
941 {
942 switch (state) {
943 case MIGRATION_STATUS_ACTIVE:
944 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
945 case MIGRATION_STATUS_POSTCOPY_PAUSED:
946 case MIGRATION_STATUS_POSTCOPY_RECOVER:
947 case MIGRATION_STATUS_SETUP:
948 case MIGRATION_STATUS_PRE_SWITCHOVER:
949 case MIGRATION_STATUS_DEVICE:
950 case MIGRATION_STATUS_WAIT_UNPLUG:
951 case MIGRATION_STATUS_CANCELLING:
952 return true;
953
954 default:
955 return false;
956
957 }
958 }
959
960 static void populate_time_info(MigrationInfo *info, MigrationState *s)
961 {
962 info->has_status = true;
963 info->has_setup_time = true;
964 info->setup_time = s->setup_time;
965 if (s->state == MIGRATION_STATUS_COMPLETED) {
966 info->has_total_time = true;
967 info->total_time = s->total_time;
968 info->has_downtime = true;
969 info->downtime = s->downtime;
970 } else {
971 info->has_total_time = true;
972 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
973 s->start_time;
974 info->has_expected_downtime = true;
975 info->expected_downtime = s->expected_downtime;
976 }
977 }
978
979 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
980 {
981 info->has_ram = true;
982 info->ram = g_malloc0(sizeof(*info->ram));
983 info->ram->transferred = ram_counters.transferred;
984 info->ram->total = ram_bytes_total();
985 info->ram->duplicate = ram_counters.duplicate;
986 /* legacy value. It is not used anymore */
987 info->ram->skipped = 0;
988 info->ram->normal = ram_counters.normal;
989 info->ram->normal_bytes = ram_counters.normal *
990 qemu_target_page_size();
991 info->ram->mbps = s->mbps;
992 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
993 info->ram->postcopy_requests = ram_counters.postcopy_requests;
994 info->ram->page_size = qemu_target_page_size();
995 info->ram->multifd_bytes = ram_counters.multifd_bytes;
996 info->ram->pages_per_second = s->pages_per_second;
997
998 if (migrate_use_xbzrle()) {
999 info->has_xbzrle_cache = true;
1000 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
1001 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
1002 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
1003 info->xbzrle_cache->pages = xbzrle_counters.pages;
1004 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
1005 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
1006 info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
1007 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
1008 }
1009
1010 if (migrate_use_compression()) {
1011 info->has_compression = true;
1012 info->compression = g_malloc0(sizeof(*info->compression));
1013 info->compression->pages = compression_counters.pages;
1014 info->compression->busy = compression_counters.busy;
1015 info->compression->busy_rate = compression_counters.busy_rate;
1016 info->compression->compressed_size =
1017 compression_counters.compressed_size;
1018 info->compression->compression_rate =
1019 compression_counters.compression_rate;
1020 }
1021
1022 if (cpu_throttle_active()) {
1023 info->has_cpu_throttle_percentage = true;
1024 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
1025 }
1026
1027 if (s->state != MIGRATION_STATUS_COMPLETED) {
1028 info->ram->remaining = ram_bytes_remaining();
1029 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
1030 }
1031 }
1032
1033 static void populate_disk_info(MigrationInfo *info)
1034 {
1035 if (blk_mig_active()) {
1036 info->has_disk = true;
1037 info->disk = g_malloc0(sizeof(*info->disk));
1038 info->disk->transferred = blk_mig_bytes_transferred();
1039 info->disk->remaining = blk_mig_bytes_remaining();
1040 info->disk->total = blk_mig_bytes_total();
1041 }
1042 }
1043
1044 static void populate_vfio_info(MigrationInfo *info)
1045 {
1046 #ifdef CONFIG_VFIO
1047 if (vfio_mig_active()) {
1048 info->has_vfio = true;
1049 info->vfio = g_malloc0(sizeof(*info->vfio));
1050 info->vfio->transferred = vfio_mig_bytes_transferred();
1051 }
1052 #endif
1053 }
1054
1055 static void fill_source_migration_info(MigrationInfo *info)
1056 {
1057 MigrationState *s = migrate_get_current();
1058
1059 switch (s->state) {
1060 case MIGRATION_STATUS_NONE:
1061 /* no migration has happened ever */
1062 /* do not overwrite destination migration status */
1063 return;
1064 case MIGRATION_STATUS_SETUP:
1065 info->has_status = true;
1066 info->has_total_time = false;
1067 break;
1068 case MIGRATION_STATUS_ACTIVE:
1069 case MIGRATION_STATUS_CANCELLING:
1070 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1071 case MIGRATION_STATUS_PRE_SWITCHOVER:
1072 case MIGRATION_STATUS_DEVICE:
1073 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1074 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1075 /* TODO add some postcopy stats */
1076 populate_time_info(info, s);
1077 populate_ram_info(info, s);
1078 populate_disk_info(info);
1079 populate_vfio_info(info);
1080 break;
1081 case MIGRATION_STATUS_COLO:
1082 info->has_status = true;
1083 /* TODO: display COLO specific information (checkpoint info etc.) */
1084 break;
1085 case MIGRATION_STATUS_COMPLETED:
1086 populate_time_info(info, s);
1087 populate_ram_info(info, s);
1088 populate_vfio_info(info);
1089 break;
1090 case MIGRATION_STATUS_FAILED:
1091 info->has_status = true;
1092 if (s->error) {
1093 info->has_error_desc = true;
1094 info->error_desc = g_strdup(error_get_pretty(s->error));
1095 }
1096 break;
1097 case MIGRATION_STATUS_CANCELLED:
1098 info->has_status = true;
1099 break;
1100 case MIGRATION_STATUS_WAIT_UNPLUG:
1101 info->has_status = true;
1102 break;
1103 }
1104 info->status = s->state;
1105 }
1106
1107 /**
1108 * @migration_caps_check - check capability validity
1109 *
1110 * @cap_list: old capability list, array of bool
1111 * @params: new capabilities to be applied soon
1112 * @errp: set *errp if the check failed, with reason
1113 *
1114 * Returns true if check passed, otherwise false.
1115 */
1116 static bool migrate_caps_check(bool *cap_list,
1117 MigrationCapabilityStatusList *params,
1118 Error **errp)
1119 {
1120 MigrationCapabilityStatusList *cap;
1121 bool old_postcopy_cap;
1122 MigrationIncomingState *mis = migration_incoming_get_current();
1123
1124 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1125
1126 for (cap = params; cap; cap = cap->next) {
1127 cap_list[cap->value->capability] = cap->value->state;
1128 }
1129
1130 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1131 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1132 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1133 "block migration");
1134 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1135 return false;
1136 }
1137 #endif
1138
1139 #ifndef CONFIG_REPLICATION
1140 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1141 error_setg(errp, "QEMU compiled without replication module"
1142 " can't enable COLO");
1143 error_append_hint(errp, "Please enable replication before COLO.\n");
1144 return false;
1145 }
1146 #endif
1147
1148 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1149 /* This check is reasonably expensive, so only when it's being
1150 * set the first time, also it's only the destination that needs
1151 * special support.
1152 */
1153 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1154 !postcopy_ram_supported_by_host(mis)) {
1155 /* postcopy_ram_supported_by_host will have emitted a more
1156 * detailed message
1157 */
1158 error_setg(errp, "Postcopy is not supported");
1159 return false;
1160 }
1161
1162 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1163 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1164 return false;
1165 }
1166 }
1167
1168 return true;
1169 }
1170
1171 static void fill_destination_migration_info(MigrationInfo *info)
1172 {
1173 MigrationIncomingState *mis = migration_incoming_get_current();
1174
1175 if (mis->socket_address_list) {
1176 info->has_socket_address = true;
1177 info->socket_address =
1178 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1179 }
1180
1181 switch (mis->state) {
1182 case MIGRATION_STATUS_NONE:
1183 return;
1184 case MIGRATION_STATUS_SETUP:
1185 case MIGRATION_STATUS_CANCELLING:
1186 case MIGRATION_STATUS_CANCELLED:
1187 case MIGRATION_STATUS_ACTIVE:
1188 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1189 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1190 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1191 case MIGRATION_STATUS_FAILED:
1192 case MIGRATION_STATUS_COLO:
1193 info->has_status = true;
1194 break;
1195 case MIGRATION_STATUS_COMPLETED:
1196 info->has_status = true;
1197 fill_destination_postcopy_migration_info(info);
1198 break;
1199 }
1200 info->status = mis->state;
1201 }
1202
1203 MigrationInfo *qmp_query_migrate(Error **errp)
1204 {
1205 MigrationInfo *info = g_malloc0(sizeof(*info));
1206
1207 fill_destination_migration_info(info);
1208 fill_source_migration_info(info);
1209
1210 return info;
1211 }
1212
1213 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1214 Error **errp)
1215 {
1216 MigrationState *s = migrate_get_current();
1217 MigrationCapabilityStatusList *cap;
1218 bool cap_list[MIGRATION_CAPABILITY__MAX];
1219
1220 if (migration_is_running(s->state)) {
1221 error_setg(errp, QERR_MIGRATION_ACTIVE);
1222 return;
1223 }
1224
1225 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1226 if (!migrate_caps_check(cap_list, params, errp)) {
1227 return;
1228 }
1229
1230 for (cap = params; cap; cap = cap->next) {
1231 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1232 }
1233 }
1234
1235 /*
1236 * Check whether the parameters are valid. Error will be put into errp
1237 * (if provided). Return true if valid, otherwise false.
1238 */
1239 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1240 {
1241 if (params->has_compress_level &&
1242 (params->compress_level > 9)) {
1243 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1244 "is invalid, it should be in the range of 0 to 9");
1245 return false;
1246 }
1247
1248 if (params->has_compress_threads && (params->compress_threads < 1)) {
1249 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1250 "compress_threads",
1251 "is invalid, it should be in the range of 1 to 255");
1252 return false;
1253 }
1254
1255 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1256 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1257 "decompress_threads",
1258 "is invalid, it should be in the range of 1 to 255");
1259 return false;
1260 }
1261
1262 if (params->has_throttle_trigger_threshold &&
1263 (params->throttle_trigger_threshold < 1 ||
1264 params->throttle_trigger_threshold > 100)) {
1265 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1266 "throttle_trigger_threshold",
1267 "an integer in the range of 1 to 100");
1268 return false;
1269 }
1270
1271 if (params->has_cpu_throttle_initial &&
1272 (params->cpu_throttle_initial < 1 ||
1273 params->cpu_throttle_initial > 99)) {
1274 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1275 "cpu_throttle_initial",
1276 "an integer in the range of 1 to 99");
1277 return false;
1278 }
1279
1280 if (params->has_cpu_throttle_increment &&
1281 (params->cpu_throttle_increment < 1 ||
1282 params->cpu_throttle_increment > 99)) {
1283 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1284 "cpu_throttle_increment",
1285 "an integer in the range of 1 to 99");
1286 return false;
1287 }
1288
1289 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1290 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1291 "max_bandwidth",
1292 "an integer in the range of 0 to "stringify(SIZE_MAX)
1293 " bytes/second");
1294 return false;
1295 }
1296
1297 if (params->has_downtime_limit &&
1298 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1299 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1300 "downtime_limit",
1301 "an integer in the range of 0 to "
1302 stringify(MAX_MIGRATE_DOWNTIME)" ms");
1303 return false;
1304 }
1305
1306 /* x_checkpoint_delay is now always positive */
1307
1308 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1309 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1310 "multifd_channels",
1311 "is invalid, it should be in the range of 1 to 255");
1312 return false;
1313 }
1314
1315 if (params->has_multifd_zlib_level &&
1316 (params->multifd_zlib_level > 9)) {
1317 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1318 "is invalid, it should be in the range of 0 to 9");
1319 return false;
1320 }
1321
1322 if (params->has_multifd_zstd_level &&
1323 (params->multifd_zstd_level > 20)) {
1324 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1325 "is invalid, it should be in the range of 0 to 20");
1326 return false;
1327 }
1328
1329 if (params->has_xbzrle_cache_size &&
1330 (params->xbzrle_cache_size < qemu_target_page_size() ||
1331 !is_power_of_2(params->xbzrle_cache_size))) {
1332 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1333 "xbzrle_cache_size",
1334 "is invalid, it should be bigger than target page size"
1335 " and a power of 2");
1336 return false;
1337 }
1338
1339 if (params->has_max_cpu_throttle &&
1340 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1341 params->max_cpu_throttle > 99)) {
1342 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1343 "max_cpu_throttle",
1344 "an integer in the range of cpu_throttle_initial to 99");
1345 return false;
1346 }
1347
1348 if (params->has_announce_initial &&
1349 params->announce_initial > 100000) {
1350 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1351 "announce_initial",
1352 "is invalid, it must be less than 100000 ms");
1353 return false;
1354 }
1355 if (params->has_announce_max &&
1356 params->announce_max > 100000) {
1357 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1358 "announce_max",
1359 "is invalid, it must be less than 100000 ms");
1360 return false;
1361 }
1362 if (params->has_announce_rounds &&
1363 params->announce_rounds > 1000) {
1364 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1365 "announce_rounds",
1366 "is invalid, it must be in the range of 0 to 1000");
1367 return false;
1368 }
1369 if (params->has_announce_step &&
1370 (params->announce_step < 1 ||
1371 params->announce_step > 10000)) {
1372 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1373 "announce_step",
1374 "is invalid, it must be in the range of 1 to 10000 ms");
1375 return false;
1376 }
1377
1378 if (params->has_block_bitmap_mapping &&
1379 !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) {
1380 error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: ");
1381 return false;
1382 }
1383
1384 return true;
1385 }
1386
1387 static void migrate_params_test_apply(MigrateSetParameters *params,
1388 MigrationParameters *dest)
1389 {
1390 *dest = migrate_get_current()->parameters;
1391
1392 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1393
1394 if (params->has_compress_level) {
1395 dest->compress_level = params->compress_level;
1396 }
1397
1398 if (params->has_compress_threads) {
1399 dest->compress_threads = params->compress_threads;
1400 }
1401
1402 if (params->has_compress_wait_thread) {
1403 dest->compress_wait_thread = params->compress_wait_thread;
1404 }
1405
1406 if (params->has_decompress_threads) {
1407 dest->decompress_threads = params->decompress_threads;
1408 }
1409
1410 if (params->has_throttle_trigger_threshold) {
1411 dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
1412 }
1413
1414 if (params->has_cpu_throttle_initial) {
1415 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1416 }
1417
1418 if (params->has_cpu_throttle_increment) {
1419 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1420 }
1421
1422 if (params->has_cpu_throttle_tailslow) {
1423 dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1424 }
1425
1426 if (params->has_tls_creds) {
1427 assert(params->tls_creds->type == QTYPE_QSTRING);
1428 dest->tls_creds = params->tls_creds->u.s;
1429 }
1430
1431 if (params->has_tls_hostname) {
1432 assert(params->tls_hostname->type == QTYPE_QSTRING);
1433 dest->tls_hostname = params->tls_hostname->u.s;
1434 }
1435
1436 if (params->has_max_bandwidth) {
1437 dest->max_bandwidth = params->max_bandwidth;
1438 }
1439
1440 if (params->has_downtime_limit) {
1441 dest->downtime_limit = params->downtime_limit;
1442 }
1443
1444 if (params->has_x_checkpoint_delay) {
1445 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1446 }
1447
1448 if (params->has_block_incremental) {
1449 dest->block_incremental = params->block_incremental;
1450 }
1451 if (params->has_multifd_channels) {
1452 dest->multifd_channels = params->multifd_channels;
1453 }
1454 if (params->has_multifd_compression) {
1455 dest->multifd_compression = params->multifd_compression;
1456 }
1457 if (params->has_xbzrle_cache_size) {
1458 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1459 }
1460 if (params->has_max_postcopy_bandwidth) {
1461 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1462 }
1463 if (params->has_max_cpu_throttle) {
1464 dest->max_cpu_throttle = params->max_cpu_throttle;
1465 }
1466 if (params->has_announce_initial) {
1467 dest->announce_initial = params->announce_initial;
1468 }
1469 if (params->has_announce_max) {
1470 dest->announce_max = params->announce_max;
1471 }
1472 if (params->has_announce_rounds) {
1473 dest->announce_rounds = params->announce_rounds;
1474 }
1475 if (params->has_announce_step) {
1476 dest->announce_step = params->announce_step;
1477 }
1478
1479 if (params->has_block_bitmap_mapping) {
1480 dest->has_block_bitmap_mapping = true;
1481 dest->block_bitmap_mapping = params->block_bitmap_mapping;
1482 }
1483 }
1484
1485 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1486 {
1487 MigrationState *s = migrate_get_current();
1488
1489 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1490
1491 if (params->has_compress_level) {
1492 s->parameters.compress_level = params->compress_level;
1493 }
1494
1495 if (params->has_compress_threads) {
1496 s->parameters.compress_threads = params->compress_threads;
1497 }
1498
1499 if (params->has_compress_wait_thread) {
1500 s->parameters.compress_wait_thread = params->compress_wait_thread;
1501 }
1502
1503 if (params->has_decompress_threads) {
1504 s->parameters.decompress_threads = params->decompress_threads;
1505 }
1506
1507 if (params->has_throttle_trigger_threshold) {
1508 s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
1509 }
1510
1511 if (params->has_cpu_throttle_initial) {
1512 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1513 }
1514
1515 if (params->has_cpu_throttle_increment) {
1516 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1517 }
1518
1519 if (params->has_cpu_throttle_tailslow) {
1520 s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1521 }
1522
1523 if (params->has_tls_creds) {
1524 g_free(s->parameters.tls_creds);
1525 assert(params->tls_creds->type == QTYPE_QSTRING);
1526 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1527 }
1528
1529 if (params->has_tls_hostname) {
1530 g_free(s->parameters.tls_hostname);
1531 assert(params->tls_hostname->type == QTYPE_QSTRING);
1532 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1533 }
1534
1535 if (params->has_tls_authz) {
1536 g_free(s->parameters.tls_authz);
1537 assert(params->tls_authz->type == QTYPE_QSTRING);
1538 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1539 }
1540
1541 if (params->has_max_bandwidth) {
1542 s->parameters.max_bandwidth = params->max_bandwidth;
1543 if (s->to_dst_file && !migration_in_postcopy()) {
1544 qemu_file_set_rate_limit(s->to_dst_file,
1545 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1546 }
1547 }
1548
1549 if (params->has_downtime_limit) {
1550 s->parameters.downtime_limit = params->downtime_limit;
1551 }
1552
1553 if (params->has_x_checkpoint_delay) {
1554 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1555 if (migration_in_colo_state()) {
1556 colo_checkpoint_notify(s);
1557 }
1558 }
1559
1560 if (params->has_block_incremental) {
1561 s->parameters.block_incremental = params->block_incremental;
1562 }
1563 if (params->has_multifd_channels) {
1564 s->parameters.multifd_channels = params->multifd_channels;
1565 }
1566 if (params->has_multifd_compression) {
1567 s->parameters.multifd_compression = params->multifd_compression;
1568 }
1569 if (params->has_xbzrle_cache_size) {
1570 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1571 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1572 }
1573 if (params->has_max_postcopy_bandwidth) {
1574 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1575 if (s->to_dst_file && migration_in_postcopy()) {
1576 qemu_file_set_rate_limit(s->to_dst_file,
1577 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1578 }
1579 }
1580 if (params->has_max_cpu_throttle) {
1581 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1582 }
1583 if (params->has_announce_initial) {
1584 s->parameters.announce_initial = params->announce_initial;
1585 }
1586 if (params->has_announce_max) {
1587 s->parameters.announce_max = params->announce_max;
1588 }
1589 if (params->has_announce_rounds) {
1590 s->parameters.announce_rounds = params->announce_rounds;
1591 }
1592 if (params->has_announce_step) {
1593 s->parameters.announce_step = params->announce_step;
1594 }
1595
1596 if (params->has_block_bitmap_mapping) {
1597 qapi_free_BitmapMigrationNodeAliasList(
1598 s->parameters.block_bitmap_mapping);
1599
1600 s->parameters.has_block_bitmap_mapping = true;
1601 s->parameters.block_bitmap_mapping =
1602 QAPI_CLONE(BitmapMigrationNodeAliasList,
1603 params->block_bitmap_mapping);
1604 }
1605 }
1606
1607 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1608 {
1609 MigrationParameters tmp;
1610
1611 /* TODO Rewrite "" to null instead */
1612 if (params->has_tls_creds
1613 && params->tls_creds->type == QTYPE_QNULL) {
1614 qobject_unref(params->tls_creds->u.n);
1615 params->tls_creds->type = QTYPE_QSTRING;
1616 params->tls_creds->u.s = strdup("");
1617 }
1618 /* TODO Rewrite "" to null instead */
1619 if (params->has_tls_hostname
1620 && params->tls_hostname->type == QTYPE_QNULL) {
1621 qobject_unref(params->tls_hostname->u.n);
1622 params->tls_hostname->type = QTYPE_QSTRING;
1623 params->tls_hostname->u.s = strdup("");
1624 }
1625
1626 migrate_params_test_apply(params, &tmp);
1627
1628 if (!migrate_params_check(&tmp, errp)) {
1629 /* Invalid parameter */
1630 return;
1631 }
1632
1633 migrate_params_apply(params, errp);
1634 }
1635
1636
1637 void qmp_migrate_start_postcopy(Error **errp)
1638 {
1639 MigrationState *s = migrate_get_current();
1640
1641 if (!migrate_postcopy()) {
1642 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1643 " the start of migration");
1644 return;
1645 }
1646
1647 if (s->state == MIGRATION_STATUS_NONE) {
1648 error_setg(errp, "Postcopy must be started after migration has been"
1649 " started");
1650 return;
1651 }
1652 /*
1653 * we don't error if migration has finished since that would be racy
1654 * with issuing this command.
1655 */
1656 qatomic_set(&s->start_postcopy, true);
1657 }
1658
1659 /* shared migration helpers */
1660
1661 void migrate_set_state(int *state, int old_state, int new_state)
1662 {
1663 assert(new_state < MIGRATION_STATUS__MAX);
1664 if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
1665 trace_migrate_set_state(MigrationStatus_str(new_state));
1666 migrate_generate_event(new_state);
1667 }
1668 }
1669
1670 static MigrationCapabilityStatusList *migrate_cap_add(
1671 MigrationCapabilityStatusList *list,
1672 MigrationCapability index,
1673 bool state)
1674 {
1675 MigrationCapabilityStatusList *cap;
1676
1677 cap = g_new0(MigrationCapabilityStatusList, 1);
1678 cap->value = g_new0(MigrationCapabilityStatus, 1);
1679 cap->value->capability = index;
1680 cap->value->state = state;
1681 cap->next = list;
1682
1683 return cap;
1684 }
1685
1686 void migrate_set_block_enabled(bool value, Error **errp)
1687 {
1688 MigrationCapabilityStatusList *cap;
1689
1690 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1691 qmp_migrate_set_capabilities(cap, errp);
1692 qapi_free_MigrationCapabilityStatusList(cap);
1693 }
1694
1695 static void migrate_set_block_incremental(MigrationState *s, bool value)
1696 {
1697 s->parameters.block_incremental = value;
1698 }
1699
1700 static void block_cleanup_parameters(MigrationState *s)
1701 {
1702 if (s->must_remove_block_options) {
1703 /* setting to false can never fail */
1704 migrate_set_block_enabled(false, &error_abort);
1705 migrate_set_block_incremental(s, false);
1706 s->must_remove_block_options = false;
1707 }
1708 }
1709
1710 static void migrate_fd_cleanup(MigrationState *s)
1711 {
1712 qemu_bh_delete(s->cleanup_bh);
1713 s->cleanup_bh = NULL;
1714
1715 qemu_savevm_state_cleanup();
1716
1717 if (s->to_dst_file) {
1718 QEMUFile *tmp;
1719
1720 trace_migrate_fd_cleanup();
1721 qemu_mutex_unlock_iothread();
1722 if (s->migration_thread_running) {
1723 qemu_thread_join(&s->thread);
1724 s->migration_thread_running = false;
1725 }
1726 qemu_mutex_lock_iothread();
1727
1728 multifd_save_cleanup();
1729 qemu_mutex_lock(&s->qemu_file_lock);
1730 tmp = s->to_dst_file;
1731 s->to_dst_file = NULL;
1732 qemu_mutex_unlock(&s->qemu_file_lock);
1733 /*
1734 * Close the file handle without the lock to make sure the
1735 * critical section won't block for long.
1736 */
1737 qemu_fclose(tmp);
1738 }
1739
1740 assert(!migration_is_active(s));
1741
1742 if (s->state == MIGRATION_STATUS_CANCELLING) {
1743 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1744 MIGRATION_STATUS_CANCELLED);
1745 }
1746
1747 if (s->error) {
1748 /* It is used on info migrate. We can't free it */
1749 error_report_err(error_copy(s->error));
1750 }
1751 notifier_list_notify(&migration_state_notifiers, s);
1752 block_cleanup_parameters(s);
1753 }
1754
1755 static void migrate_fd_cleanup_schedule(MigrationState *s)
1756 {
1757 /*
1758 * Ref the state for bh, because it may be called when
1759 * there're already no other refs
1760 */
1761 object_ref(OBJECT(s));
1762 qemu_bh_schedule(s->cleanup_bh);
1763 }
1764
1765 static void migrate_fd_cleanup_bh(void *opaque)
1766 {
1767 MigrationState *s = opaque;
1768 migrate_fd_cleanup(s);
1769 object_unref(OBJECT(s));
1770 }
1771
1772 void migrate_set_error(MigrationState *s, const Error *error)
1773 {
1774 QEMU_LOCK_GUARD(&s->error_mutex);
1775 if (!s->error) {
1776 s->error = error_copy(error);
1777 }
1778 }
1779
1780 void migrate_fd_error(MigrationState *s, const Error *error)
1781 {
1782 trace_migrate_fd_error(error_get_pretty(error));
1783 assert(s->to_dst_file == NULL);
1784 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1785 MIGRATION_STATUS_FAILED);
1786 migrate_set_error(s, error);
1787 }
1788
1789 static void migrate_fd_cancel(MigrationState *s)
1790 {
1791 int old_state ;
1792 QEMUFile *f = migrate_get_current()->to_dst_file;
1793 trace_migrate_fd_cancel();
1794
1795 if (s->rp_state.from_dst_file) {
1796 /* shutdown the rp socket, so causing the rp thread to shutdown */
1797 qemu_file_shutdown(s->rp_state.from_dst_file);
1798 }
1799
1800 do {
1801 old_state = s->state;
1802 if (!migration_is_running(old_state)) {
1803 break;
1804 }
1805 /* If the migration is paused, kick it out of the pause */
1806 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1807 qemu_sem_post(&s->pause_sem);
1808 }
1809 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1810 } while (s->state != MIGRATION_STATUS_CANCELLING);
1811
1812 /*
1813 * If we're unlucky the migration code might be stuck somewhere in a
1814 * send/write while the network has failed and is waiting to timeout;
1815 * if we've got shutdown(2) available then we can force it to quit.
1816 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1817 * called in a bh, so there is no race against this cancel.
1818 */
1819 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1820 qemu_file_shutdown(f);
1821 }
1822 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1823 Error *local_err = NULL;
1824
1825 bdrv_invalidate_cache_all(&local_err);
1826 if (local_err) {
1827 error_report_err(local_err);
1828 } else {
1829 s->block_inactive = false;
1830 }
1831 }
1832 }
1833
1834 void add_migration_state_change_notifier(Notifier *notify)
1835 {
1836 notifier_list_add(&migration_state_notifiers, notify);
1837 }
1838
1839 void remove_migration_state_change_notifier(Notifier *notify)
1840 {
1841 notifier_remove(notify);
1842 }
1843
1844 bool migration_in_setup(MigrationState *s)
1845 {
1846 return s->state == MIGRATION_STATUS_SETUP;
1847 }
1848
1849 bool migration_has_finished(MigrationState *s)
1850 {
1851 return s->state == MIGRATION_STATUS_COMPLETED;
1852 }
1853
1854 bool migration_has_failed(MigrationState *s)
1855 {
1856 return (s->state == MIGRATION_STATUS_CANCELLED ||
1857 s->state == MIGRATION_STATUS_FAILED);
1858 }
1859
1860 bool migration_in_postcopy(void)
1861 {
1862 MigrationState *s = migrate_get_current();
1863
1864 switch (s->state) {
1865 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1866 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1867 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1868 return true;
1869 default:
1870 return false;
1871 }
1872 }
1873
1874 bool migration_in_postcopy_after_devices(MigrationState *s)
1875 {
1876 return migration_in_postcopy() && s->postcopy_after_devices;
1877 }
1878
1879 bool migration_in_incoming_postcopy(void)
1880 {
1881 PostcopyState ps = postcopy_state_get();
1882
1883 return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
1884 }
1885
1886 bool migration_is_idle(void)
1887 {
1888 MigrationState *s = current_migration;
1889
1890 if (!s) {
1891 return true;
1892 }
1893
1894 switch (s->state) {
1895 case MIGRATION_STATUS_NONE:
1896 case MIGRATION_STATUS_CANCELLED:
1897 case MIGRATION_STATUS_COMPLETED:
1898 case MIGRATION_STATUS_FAILED:
1899 return true;
1900 case MIGRATION_STATUS_SETUP:
1901 case MIGRATION_STATUS_CANCELLING:
1902 case MIGRATION_STATUS_ACTIVE:
1903 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1904 case MIGRATION_STATUS_COLO:
1905 case MIGRATION_STATUS_PRE_SWITCHOVER:
1906 case MIGRATION_STATUS_DEVICE:
1907 case MIGRATION_STATUS_WAIT_UNPLUG:
1908 return false;
1909 case MIGRATION_STATUS__MAX:
1910 g_assert_not_reached();
1911 }
1912
1913 return false;
1914 }
1915
1916 bool migration_is_active(MigrationState *s)
1917 {
1918 return (s->state == MIGRATION_STATUS_ACTIVE ||
1919 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1920 }
1921
1922 void migrate_init(MigrationState *s)
1923 {
1924 /*
1925 * Reinitialise all migration state, except
1926 * parameters/capabilities that the user set, and
1927 * locks.
1928 */
1929 s->cleanup_bh = 0;
1930 s->to_dst_file = NULL;
1931 s->state = MIGRATION_STATUS_NONE;
1932 s->rp_state.from_dst_file = NULL;
1933 s->rp_state.error = false;
1934 s->mbps = 0.0;
1935 s->pages_per_second = 0.0;
1936 s->downtime = 0;
1937 s->expected_downtime = 0;
1938 s->setup_time = 0;
1939 s->start_postcopy = false;
1940 s->postcopy_after_devices = false;
1941 s->migration_thread_running = false;
1942 error_free(s->error);
1943 s->error = NULL;
1944 s->hostname = NULL;
1945
1946 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1947
1948 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1949 s->total_time = 0;
1950 s->vm_was_running = false;
1951 s->iteration_initial_bytes = 0;
1952 s->threshold_size = 0;
1953 }
1954
1955 static GSList *migration_blockers;
1956
1957 int migrate_add_blocker(Error *reason, Error **errp)
1958 {
1959 if (only_migratable) {
1960 error_propagate_prepend(errp, error_copy(reason),
1961 "disallowing migration blocker "
1962 "(--only-migratable) for: ");
1963 return -EACCES;
1964 }
1965
1966 if (migration_is_idle()) {
1967 migration_blockers = g_slist_prepend(migration_blockers, reason);
1968 return 0;
1969 }
1970
1971 error_propagate_prepend(errp, error_copy(reason),
1972 "disallowing migration blocker "
1973 "(migration in progress) for: ");
1974 return -EBUSY;
1975 }
1976
1977 void migrate_del_blocker(Error *reason)
1978 {
1979 migration_blockers = g_slist_remove(migration_blockers, reason);
1980 }
1981
1982 void qmp_migrate_incoming(const char *uri, Error **errp)
1983 {
1984 Error *local_err = NULL;
1985 static bool once = true;
1986
1987 if (!deferred_incoming) {
1988 error_setg(errp, "For use with '-incoming defer'");
1989 return;
1990 }
1991 if (!once) {
1992 error_setg(errp, "The incoming migration has already been started");
1993 return;
1994 }
1995
1996 qemu_start_incoming_migration(uri, &local_err);
1997
1998 if (local_err) {
1999 error_propagate(errp, local_err);
2000 return;
2001 }
2002
2003 once = false;
2004 }
2005
2006 void qmp_migrate_recover(const char *uri, Error **errp)
2007 {
2008 MigrationIncomingState *mis = migration_incoming_get_current();
2009
2010 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2011 error_setg(errp, "Migrate recover can only be run "
2012 "when postcopy is paused.");
2013 return;
2014 }
2015
2016 if (qatomic_cmpxchg(&mis->postcopy_recover_triggered,
2017 false, true) == true) {
2018 error_setg(errp, "Migrate recovery is triggered already");
2019 return;
2020 }
2021
2022 /*
2023 * Note that this call will never start a real migration; it will
2024 * only re-setup the migration stream and poke existing migration
2025 * to continue using that newly established channel.
2026 */
2027 qemu_start_incoming_migration(uri, errp);
2028 }
2029
2030 void qmp_migrate_pause(Error **errp)
2031 {
2032 MigrationState *ms = migrate_get_current();
2033 MigrationIncomingState *mis = migration_incoming_get_current();
2034 int ret;
2035
2036 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2037 /* Source side, during postcopy */
2038 qemu_mutex_lock(&ms->qemu_file_lock);
2039 ret = qemu_file_shutdown(ms->to_dst_file);
2040 qemu_mutex_unlock(&ms->qemu_file_lock);
2041 if (ret) {
2042 error_setg(errp, "Failed to pause source migration");
2043 }
2044 return;
2045 }
2046
2047 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2048 ret = qemu_file_shutdown(mis->from_src_file);
2049 if (ret) {
2050 error_setg(errp, "Failed to pause destination migration");
2051 }
2052 return;
2053 }
2054
2055 error_setg(errp, "migrate-pause is currently only supported "
2056 "during postcopy-active state");
2057 }
2058
2059 bool migration_is_blocked(Error **errp)
2060 {
2061 if (qemu_savevm_state_blocked(errp)) {
2062 return true;
2063 }
2064
2065 if (migration_blockers) {
2066 error_propagate(errp, error_copy(migration_blockers->data));
2067 return true;
2068 }
2069
2070 return false;
2071 }
2072
2073 /* Returns true if continue to migrate, or false if error detected */
2074 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
2075 bool resume, Error **errp)
2076 {
2077 Error *local_err = NULL;
2078
2079 if (resume) {
2080 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2081 error_setg(errp, "Cannot resume if there is no "
2082 "paused migration");
2083 return false;
2084 }
2085
2086 /*
2087 * Postcopy recovery won't work well with release-ram
2088 * capability since release-ram will drop the page buffer as
2089 * long as the page is put into the send buffer. So if there
2090 * is a network failure happened, any page buffers that have
2091 * not yet reached the destination VM but have already been
2092 * sent from the source VM will be lost forever. Let's refuse
2093 * the client from resuming such a postcopy migration.
2094 * Luckily release-ram was designed to only be used when src
2095 * and destination VMs are on the same host, so it should be
2096 * fine.
2097 */
2098 if (migrate_release_ram()) {
2099 error_setg(errp, "Postcopy recovery cannot work "
2100 "when release-ram capability is set");
2101 return false;
2102 }
2103
2104 /* This is a resume, skip init status */
2105 return true;
2106 }
2107
2108 if (migration_is_running(s->state)) {
2109 error_setg(errp, QERR_MIGRATION_ACTIVE);
2110 return false;
2111 }
2112
2113 if (runstate_check(RUN_STATE_INMIGRATE)) {
2114 error_setg(errp, "Guest is waiting for an incoming migration");
2115 return false;
2116 }
2117
2118 if (migration_is_blocked(errp)) {
2119 return false;
2120 }
2121
2122 if (blk || blk_inc) {
2123 if (migrate_use_block() || migrate_use_block_incremental()) {
2124 error_setg(errp, "Command options are incompatible with "
2125 "current migration capabilities");
2126 return false;
2127 }
2128 migrate_set_block_enabled(true, &local_err);
2129 if (local_err) {
2130 error_propagate(errp, local_err);
2131 return false;
2132 }
2133 s->must_remove_block_options = true;
2134 }
2135
2136 if (blk_inc) {
2137 migrate_set_block_incremental(s, true);
2138 }
2139
2140 migrate_init(s);
2141 /*
2142 * set ram_counters memory to zero for a
2143 * new migration
2144 */
2145 memset(&ram_counters, 0, sizeof(ram_counters));
2146
2147 return true;
2148 }
2149
2150 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2151 bool has_inc, bool inc, bool has_detach, bool detach,
2152 bool has_resume, bool resume, Error **errp)
2153 {
2154 Error *local_err = NULL;
2155 MigrationState *s = migrate_get_current();
2156 const char *p = NULL;
2157
2158 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2159 has_resume && resume, errp)) {
2160 /* Error detected, put into errp */
2161 return;
2162 }
2163
2164 if (strstart(uri, "tcp:", &p) ||
2165 strstart(uri, "unix:", NULL) ||
2166 strstart(uri, "vsock:", NULL)) {
2167 socket_start_outgoing_migration(s, p ? p : uri, &local_err);
2168 #ifdef CONFIG_RDMA
2169 } else if (strstart(uri, "rdma:", &p)) {
2170 rdma_start_outgoing_migration(s, p, &local_err);
2171 #endif
2172 } else if (strstart(uri, "exec:", &p)) {
2173 exec_start_outgoing_migration(s, p, &local_err);
2174 } else if (strstart(uri, "fd:", &p)) {
2175 fd_start_outgoing_migration(s, p, &local_err);
2176 } else {
2177 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2178 "a valid migration protocol");
2179 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2180 MIGRATION_STATUS_FAILED);
2181 block_cleanup_parameters(s);
2182 return;
2183 }
2184
2185 if (local_err) {
2186 migrate_fd_error(s, local_err);
2187 error_propagate(errp, local_err);
2188 return;
2189 }
2190 }
2191
2192 void qmp_migrate_cancel(Error **errp)
2193 {
2194 migrate_fd_cancel(migrate_get_current());
2195 }
2196
2197 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2198 {
2199 MigrationState *s = migrate_get_current();
2200 if (s->state != state) {
2201 error_setg(errp, "Migration not in expected state: %s",
2202 MigrationStatus_str(s->state));
2203 return;
2204 }
2205 qemu_sem_post(&s->pause_sem);
2206 }
2207
2208 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
2209 {
2210 MigrateSetParameters p = {
2211 .has_xbzrle_cache_size = true,
2212 .xbzrle_cache_size = value,
2213 };
2214
2215 qmp_migrate_set_parameters(&p, errp);
2216 }
2217
2218 int64_t qmp_query_migrate_cache_size(Error **errp)
2219 {
2220 return migrate_xbzrle_cache_size();
2221 }
2222
2223 void qmp_migrate_set_speed(int64_t value, Error **errp)
2224 {
2225 MigrateSetParameters p = {
2226 .has_max_bandwidth = true,
2227 .max_bandwidth = value,
2228 };
2229
2230 qmp_migrate_set_parameters(&p, errp);
2231 }
2232
2233 void qmp_migrate_set_downtime(double value, Error **errp)
2234 {
2235 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2236 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
2237 "downtime_limit",
2238 "an integer in the range of 0 to "
2239 stringify(MAX_MIGRATE_DOWNTIME_SECONDS)" seconds");
2240 return;
2241 }
2242
2243 value *= 1000; /* Convert to milliseconds */
2244
2245 MigrateSetParameters p = {
2246 .has_downtime_limit = true,
2247 .downtime_limit = (int64_t)value,
2248 };
2249
2250 qmp_migrate_set_parameters(&p, errp);
2251 }
2252
2253 bool migrate_release_ram(void)
2254 {
2255 MigrationState *s;
2256
2257 s = migrate_get_current();
2258
2259 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2260 }
2261
2262 bool migrate_postcopy_ram(void)
2263 {
2264 MigrationState *s;
2265
2266 s = migrate_get_current();
2267
2268 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2269 }
2270
2271 bool migrate_postcopy(void)
2272 {
2273 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2274 }
2275
2276 bool migrate_auto_converge(void)
2277 {
2278 MigrationState *s;
2279
2280 s = migrate_get_current();
2281
2282 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2283 }
2284
2285 bool migrate_zero_blocks(void)
2286 {
2287 MigrationState *s;
2288
2289 s = migrate_get_current();
2290
2291 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2292 }
2293
2294 bool migrate_postcopy_blocktime(void)
2295 {
2296 MigrationState *s;
2297
2298 s = migrate_get_current();
2299
2300 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2301 }
2302
2303 bool migrate_use_compression(void)
2304 {
2305 MigrationState *s;
2306
2307 s = migrate_get_current();
2308
2309 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2310 }
2311
2312 int migrate_compress_level(void)
2313 {
2314 MigrationState *s;
2315
2316 s = migrate_get_current();
2317
2318 return s->parameters.compress_level;
2319 }
2320
2321 int migrate_compress_threads(void)
2322 {
2323 MigrationState *s;
2324
2325 s = migrate_get_current();
2326
2327 return s->parameters.compress_threads;
2328 }
2329
2330 int migrate_compress_wait_thread(void)
2331 {
2332 MigrationState *s;
2333
2334 s = migrate_get_current();
2335
2336 return s->parameters.compress_wait_thread;
2337 }
2338
2339 int migrate_decompress_threads(void)
2340 {
2341 MigrationState *s;
2342
2343 s = migrate_get_current();
2344
2345 return s->parameters.decompress_threads;
2346 }
2347
2348 bool migrate_dirty_bitmaps(void)
2349 {
2350 MigrationState *s;
2351
2352 s = migrate_get_current();
2353
2354 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2355 }
2356
2357 bool migrate_ignore_shared(void)
2358 {
2359 MigrationState *s;
2360
2361 s = migrate_get_current();
2362
2363 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2364 }
2365
2366 bool migrate_validate_uuid(void)
2367 {
2368 MigrationState *s;
2369
2370 s = migrate_get_current();
2371
2372 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2373 }
2374
2375 bool migrate_use_events(void)
2376 {
2377 MigrationState *s;
2378
2379 s = migrate_get_current();
2380
2381 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2382 }
2383
2384 bool migrate_use_multifd(void)
2385 {
2386 MigrationState *s;
2387
2388 s = migrate_get_current();
2389
2390 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2391 }
2392
2393 bool migrate_pause_before_switchover(void)
2394 {
2395 MigrationState *s;
2396
2397 s = migrate_get_current();
2398
2399 return s->enabled_capabilities[
2400 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2401 }
2402
2403 int migrate_multifd_channels(void)
2404 {
2405 MigrationState *s;
2406
2407 s = migrate_get_current();
2408
2409 return s->parameters.multifd_channels;
2410 }
2411
2412 MultiFDCompression migrate_multifd_compression(void)
2413 {
2414 MigrationState *s;
2415
2416 s = migrate_get_current();
2417
2418 return s->parameters.multifd_compression;
2419 }
2420
2421 int migrate_multifd_zlib_level(void)
2422 {
2423 MigrationState *s;
2424
2425 s = migrate_get_current();
2426
2427 return s->parameters.multifd_zlib_level;
2428 }
2429
2430 int migrate_multifd_zstd_level(void)
2431 {
2432 MigrationState *s;
2433
2434 s = migrate_get_current();
2435
2436 return s->parameters.multifd_zstd_level;
2437 }
2438
2439 int migrate_use_xbzrle(void)
2440 {
2441 MigrationState *s;
2442
2443 s = migrate_get_current();
2444
2445 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2446 }
2447
2448 int64_t migrate_xbzrle_cache_size(void)
2449 {
2450 MigrationState *s;
2451
2452 s = migrate_get_current();
2453
2454 return s->parameters.xbzrle_cache_size;
2455 }
2456
2457 static int64_t migrate_max_postcopy_bandwidth(void)
2458 {
2459 MigrationState *s;
2460
2461 s = migrate_get_current();
2462
2463 return s->parameters.max_postcopy_bandwidth;
2464 }
2465
2466 bool migrate_use_block(void)
2467 {
2468 MigrationState *s;
2469
2470 s = migrate_get_current();
2471
2472 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2473 }
2474
2475 bool migrate_use_return_path(void)
2476 {
2477 MigrationState *s;
2478
2479 s = migrate_get_current();
2480
2481 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2482 }
2483
2484 bool migrate_use_block_incremental(void)
2485 {
2486 MigrationState *s;
2487
2488 s = migrate_get_current();
2489
2490 return s->parameters.block_incremental;
2491 }
2492
2493 /* migration thread support */
2494 /*
2495 * Something bad happened to the RP stream, mark an error
2496 * The caller shall print or trace something to indicate why
2497 */
2498 static void mark_source_rp_bad(MigrationState *s)
2499 {
2500 s->rp_state.error = true;
2501 }
2502
2503 static struct rp_cmd_args {
2504 ssize_t len; /* -1 = variable */
2505 const char *name;
2506 } rp_cmd_args[] = {
2507 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2508 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2509 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2510 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2511 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2512 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2513 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2514 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2515 };
2516
2517 /*
2518 * Process a request for pages received on the return path,
2519 * We're allowed to send more than requested (e.g. to round to our page size)
2520 * and we don't need to send pages that have already been sent.
2521 */
2522 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2523 ram_addr_t start, size_t len)
2524 {
2525 long our_host_ps = qemu_real_host_page_size;
2526
2527 trace_migrate_handle_rp_req_pages(rbname, start, len);
2528
2529 /*
2530 * Since we currently insist on matching page sizes, just sanity check
2531 * we're being asked for whole host pages.
2532 */
2533 if (start & (our_host_ps - 1) ||
2534 (len & (our_host_ps - 1))) {
2535 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2536 " len: %zd", __func__, start, len);
2537 mark_source_rp_bad(ms);
2538 return;
2539 }
2540
2541 if (ram_save_queue_pages(rbname, start, len)) {
2542 mark_source_rp_bad(ms);
2543 }
2544 }
2545
2546 /* Return true to retry, false to quit */
2547 static bool postcopy_pause_return_path_thread(MigrationState *s)
2548 {
2549 trace_postcopy_pause_return_path();
2550
2551 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2552
2553 trace_postcopy_pause_return_path_continued();
2554
2555 return true;
2556 }
2557
2558 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2559 {
2560 RAMBlock *block = qemu_ram_block_by_name(block_name);
2561
2562 if (!block) {
2563 error_report("%s: invalid block name '%s'", __func__, block_name);
2564 return -EINVAL;
2565 }
2566
2567 /* Fetch the received bitmap and refresh the dirty bitmap */
2568 return ram_dirty_bitmap_reload(s, block);
2569 }
2570
2571 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2572 {
2573 trace_source_return_path_thread_resume_ack(value);
2574
2575 if (value != MIGRATION_RESUME_ACK_VALUE) {
2576 error_report("%s: illegal resume_ack value %"PRIu32,
2577 __func__, value);
2578 return -1;
2579 }
2580
2581 /* Now both sides are active. */
2582 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2583 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2584
2585 /* Notify send thread that time to continue send pages */
2586 qemu_sem_post(&s->rp_state.rp_sem);
2587
2588 return 0;
2589 }
2590
2591 /*
2592 * Handles messages sent on the return path towards the source VM
2593 *
2594 */
2595 static void *source_return_path_thread(void *opaque)
2596 {
2597 MigrationState *ms = opaque;
2598 QEMUFile *rp = ms->rp_state.from_dst_file;
2599 uint16_t header_len, header_type;
2600 uint8_t buf[512];
2601 uint32_t tmp32, sibling_error;
2602 ram_addr_t start = 0; /* =0 to silence warning */
2603 size_t len = 0, expected_len;
2604 int res;
2605
2606 trace_source_return_path_thread_entry();
2607 rcu_register_thread();
2608
2609 retry:
2610 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2611 migration_is_setup_or_active(ms->state)) {
2612 trace_source_return_path_thread_loop_top();
2613 header_type = qemu_get_be16(rp);
2614 header_len = qemu_get_be16(rp);
2615
2616 if (qemu_file_get_error(rp)) {
2617 mark_source_rp_bad(ms);
2618 goto out;
2619 }
2620
2621 if (header_type >= MIG_RP_MSG_MAX ||
2622 header_type == MIG_RP_MSG_INVALID) {
2623 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2624 header_type, header_len);
2625 mark_source_rp_bad(ms);
2626 goto out;
2627 }
2628
2629 if ((rp_cmd_args[header_type].len != -1 &&
2630 header_len != rp_cmd_args[header_type].len) ||
2631 header_len > sizeof(buf)) {
2632 error_report("RP: Received '%s' message (0x%04x) with"
2633 "incorrect length %d expecting %zu",
2634 rp_cmd_args[header_type].name, header_type, header_len,
2635 (size_t)rp_cmd_args[header_type].len);
2636 mark_source_rp_bad(ms);
2637 goto out;
2638 }
2639
2640 /* We know we've got a valid header by this point */
2641 res = qemu_get_buffer(rp, buf, header_len);
2642 if (res != header_len) {
2643 error_report("RP: Failed reading data for message 0x%04x"
2644 " read %d expected %d",
2645 header_type, res, header_len);
2646 mark_source_rp_bad(ms);
2647 goto out;
2648 }
2649
2650 /* OK, we have the message and the data */
2651 switch (header_type) {
2652 case MIG_RP_MSG_SHUT:
2653 sibling_error = ldl_be_p(buf);
2654 trace_source_return_path_thread_shut(sibling_error);
2655 if (sibling_error) {
2656 error_report("RP: Sibling indicated error %d", sibling_error);
2657 mark_source_rp_bad(ms);
2658 }
2659 /*
2660 * We'll let the main thread deal with closing the RP
2661 * we could do a shutdown(2) on it, but we're the only user
2662 * anyway, so there's nothing gained.
2663 */
2664 goto out;
2665
2666 case MIG_RP_MSG_PONG:
2667 tmp32 = ldl_be_p(buf);
2668 trace_source_return_path_thread_pong(tmp32);
2669 break;
2670
2671 case MIG_RP_MSG_REQ_PAGES:
2672 start = ldq_be_p(buf);
2673 len = ldl_be_p(buf + 8);
2674 migrate_handle_rp_req_pages(ms, NULL, start, len);
2675 break;
2676
2677 case MIG_RP_MSG_REQ_PAGES_ID:
2678 expected_len = 12 + 1; /* header + termination */
2679
2680 if (header_len >= expected_len) {
2681 start = ldq_be_p(buf);
2682 len = ldl_be_p(buf + 8);
2683 /* Now we expect an idstr */
2684 tmp32 = buf[12]; /* Length of the following idstr */
2685 buf[13 + tmp32] = '\0';
2686 expected_len += tmp32;
2687 }
2688 if (header_len != expected_len) {
2689 error_report("RP: Req_Page_id with length %d expecting %zd",
2690 header_len, expected_len);
2691 mark_source_rp_bad(ms);
2692 goto out;
2693 }
2694 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2695 break;
2696
2697 case MIG_RP_MSG_RECV_BITMAP:
2698 if (header_len < 1) {
2699 error_report("%s: missing block name", __func__);
2700 mark_source_rp_bad(ms);
2701 goto out;
2702 }
2703 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2704 buf[buf[0] + 1] = '\0';
2705 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2706 mark_source_rp_bad(ms);
2707 goto out;
2708 }
2709 break;
2710
2711 case MIG_RP_MSG_RESUME_ACK:
2712 tmp32 = ldl_be_p(buf);
2713 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2714 mark_source_rp_bad(ms);
2715 goto out;
2716 }
2717 break;
2718
2719 default:
2720 break;
2721 }
2722 }
2723
2724 out:
2725 res = qemu_file_get_error(rp);
2726 if (res) {
2727 if (res == -EIO && migration_in_postcopy()) {
2728 /*
2729 * Maybe there is something we can do: it looks like a
2730 * network down issue, and we pause for a recovery.
2731 */
2732 if (postcopy_pause_return_path_thread(ms)) {
2733 /* Reload rp, reset the rest */
2734 if (rp != ms->rp_state.from_dst_file) {
2735 qemu_fclose(rp);
2736 rp = ms->rp_state.from_dst_file;
2737 }
2738 ms->rp_state.error = false;
2739 goto retry;
2740 }
2741 }
2742
2743 trace_source_return_path_thread_bad_end();
2744 mark_source_rp_bad(ms);
2745 }
2746
2747 trace_source_return_path_thread_end();
2748 ms->rp_state.from_dst_file = NULL;
2749 qemu_fclose(rp);
2750 rcu_unregister_thread();
2751 return NULL;
2752 }
2753
2754 static int open_return_path_on_source(MigrationState *ms,
2755 bool create_thread)
2756 {
2757
2758 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2759 if (!ms->rp_state.from_dst_file) {
2760 return -1;
2761 }
2762
2763 trace_open_return_path_on_source();
2764
2765 if (!create_thread) {
2766 /* We're done */
2767 return 0;
2768 }
2769
2770 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2771 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2772
2773 trace_open_return_path_on_source_continue();
2774
2775 return 0;
2776 }
2777
2778 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2779 static int await_return_path_close_on_source(MigrationState *ms)
2780 {
2781 /*
2782 * If this is a normal exit then the destination will send a SHUT and the
2783 * rp_thread will exit, however if there's an error we need to cause
2784 * it to exit.
2785 */
2786 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2787 /*
2788 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2789 * waiting for the destination.
2790 */
2791 qemu_file_shutdown(ms->rp_state.from_dst_file);
2792 mark_source_rp_bad(ms);
2793 }
2794 trace_await_return_path_close_on_source_joining();
2795 qemu_thread_join(&ms->rp_state.rp_thread);
2796 trace_await_return_path_close_on_source_close();
2797 return ms->rp_state.error;
2798 }
2799
2800 /*
2801 * Switch from normal iteration to postcopy
2802 * Returns non-0 on error
2803 */
2804 static int postcopy_start(MigrationState *ms)
2805 {
2806 int ret;
2807 QIOChannelBuffer *bioc;
2808 QEMUFile *fb;
2809 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2810 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2811 bool restart_block = false;
2812 int cur_state = MIGRATION_STATUS_ACTIVE;
2813 if (!migrate_pause_before_switchover()) {
2814 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2815 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2816 }
2817
2818 trace_postcopy_start();
2819 qemu_mutex_lock_iothread();
2820 trace_postcopy_start_set_run();
2821
2822 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2823 global_state_store();
2824 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2825 if (ret < 0) {
2826 goto fail;
2827 }
2828
2829 ret = migration_maybe_pause(ms, &cur_state,
2830 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2831 if (ret < 0) {
2832 goto fail;
2833 }
2834
2835 ret = bdrv_inactivate_all();
2836 if (ret < 0) {
2837 goto fail;
2838 }
2839 restart_block = true;
2840
2841 /*
2842 * Cause any non-postcopiable, but iterative devices to
2843 * send out their final data.
2844 */
2845 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2846
2847 /*
2848 * in Finish migrate and with the io-lock held everything should
2849 * be quiet, but we've potentially still got dirty pages and we
2850 * need to tell the destination to throw any pages it's already received
2851 * that are dirty
2852 */
2853 if (migrate_postcopy_ram()) {
2854 if (ram_postcopy_send_discard_bitmap(ms)) {
2855 error_report("postcopy send discard bitmap failed");
2856 goto fail;
2857 }
2858 }
2859
2860 /*
2861 * send rest of state - note things that are doing postcopy
2862 * will notice we're in POSTCOPY_ACTIVE and not actually
2863 * wrap their state up here
2864 */
2865 /* 0 max-postcopy-bandwidth means unlimited */
2866 if (!bandwidth) {
2867 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2868 } else {
2869 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2870 }
2871 if (migrate_postcopy_ram()) {
2872 /* Ping just for debugging, helps line traces up */
2873 qemu_savevm_send_ping(ms->to_dst_file, 2);
2874 }
2875
2876 /*
2877 * While loading the device state we may trigger page transfer
2878 * requests and the fd must be free to process those, and thus
2879 * the destination must read the whole device state off the fd before
2880 * it starts processing it. Unfortunately the ad-hoc migration format
2881 * doesn't allow the destination to know the size to read without fully
2882 * parsing it through each devices load-state code (especially the open
2883 * coded devices that use get/put).
2884 * So we wrap the device state up in a package with a length at the start;
2885 * to do this we use a qemu_buf to hold the whole of the device state.
2886 */
2887 bioc = qio_channel_buffer_new(4096);
2888 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2889 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2890 object_unref(OBJECT(bioc));
2891
2892 /*
2893 * Make sure the receiver can get incoming pages before we send the rest
2894 * of the state
2895 */
2896 qemu_savevm_send_postcopy_listen(fb);
2897
2898 qemu_savevm_state_complete_precopy(fb, false, false);
2899 if (migrate_postcopy_ram()) {
2900 qemu_savevm_send_ping(fb, 3);
2901 }
2902
2903 qemu_savevm_send_postcopy_run(fb);
2904
2905 /* <><> end of stuff going into the package */
2906
2907 /* Last point of recovery; as soon as we send the package the destination
2908 * can open devices and potentially start running.
2909 * Lets just check again we've not got any errors.
2910 */
2911 ret = qemu_file_get_error(ms->to_dst_file);
2912 if (ret) {
2913 error_report("postcopy_start: Migration stream errored (pre package)");
2914 goto fail_closefb;
2915 }
2916
2917 restart_block = false;
2918
2919 /* Now send that blob */
2920 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2921 goto fail_closefb;
2922 }
2923 qemu_fclose(fb);
2924
2925 /* Send a notify to give a chance for anything that needs to happen
2926 * at the transition to postcopy and after the device state; in particular
2927 * spice needs to trigger a transition now
2928 */
2929 ms->postcopy_after_devices = true;
2930 notifier_list_notify(&migration_state_notifiers, ms);
2931
2932 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2933
2934 qemu_mutex_unlock_iothread();
2935
2936 if (migrate_postcopy_ram()) {
2937 /*
2938 * Although this ping is just for debug, it could potentially be
2939 * used for getting a better measurement of downtime at the source.
2940 */
2941 qemu_savevm_send_ping(ms->to_dst_file, 4);
2942 }
2943
2944 if (migrate_release_ram()) {
2945 ram_postcopy_migrated_memory_release(ms);
2946 }
2947
2948 ret = qemu_file_get_error(ms->to_dst_file);
2949 if (ret) {
2950 error_report("postcopy_start: Migration stream errored");
2951 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2952 MIGRATION_STATUS_FAILED);
2953 }
2954
2955 return ret;
2956
2957 fail_closefb:
2958 qemu_fclose(fb);
2959 fail:
2960 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2961 MIGRATION_STATUS_FAILED);
2962 if (restart_block) {
2963 /* A failure happened early enough that we know the destination hasn't
2964 * accessed block devices, so we're safe to recover.
2965 */
2966 Error *local_err = NULL;
2967
2968 bdrv_invalidate_cache_all(&local_err);
2969 if (local_err) {
2970 error_report_err(local_err);
2971 }
2972 }
2973 qemu_mutex_unlock_iothread();
2974 return -1;
2975 }
2976
2977 /**
2978 * migration_maybe_pause: Pause if required to by
2979 * migrate_pause_before_switchover called with the iothread locked
2980 * Returns: 0 on success
2981 */
2982 static int migration_maybe_pause(MigrationState *s,
2983 int *current_active_state,
2984 int new_state)
2985 {
2986 if (!migrate_pause_before_switchover()) {
2987 return 0;
2988 }
2989
2990 /* Since leaving this state is not atomic with posting the semaphore
2991 * it's possible that someone could have issued multiple migrate_continue
2992 * and the semaphore is incorrectly positive at this point;
2993 * the docs say it's undefined to reinit a semaphore that's already
2994 * init'd, so use timedwait to eat up any existing posts.
2995 */
2996 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2997 /* This block intentionally left blank */
2998 }
2999
3000 /*
3001 * If the migration is cancelled when it is in the completion phase,
3002 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3003 * So we don't need to wait a semaphore, otherwise we would always
3004 * wait for the 'pause_sem' semaphore.
3005 */
3006 if (s->state != MIGRATION_STATUS_CANCELLING) {
3007 qemu_mutex_unlock_iothread();
3008 migrate_set_state(&s->state, *current_active_state,
3009 MIGRATION_STATUS_PRE_SWITCHOVER);
3010 qemu_sem_wait(&s->pause_sem);
3011 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
3012 new_state);
3013 *current_active_state = new_state;
3014 qemu_mutex_lock_iothread();
3015 }
3016
3017 return s->state == new_state ? 0 : -EINVAL;
3018 }
3019
3020 /**
3021 * migration_completion: Used by migration_thread when there's not much left.
3022 * The caller 'breaks' the loop when this returns.
3023 *
3024 * @s: Current migration state
3025 */
3026 static void migration_completion(MigrationState *s)
3027 {
3028 int ret;
3029 int current_active_state = s->state;
3030
3031 if (s->state == MIGRATION_STATUS_ACTIVE) {
3032 qemu_mutex_lock_iothread();
3033 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3034 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3035 s->vm_was_running = runstate_is_running();
3036 ret = global_state_store();
3037
3038 if (!ret) {
3039 bool inactivate = !migrate_colo_enabled();
3040 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
3041 if (ret >= 0) {
3042 ret = migration_maybe_pause(s, &current_active_state,
3043 MIGRATION_STATUS_DEVICE);
3044 }
3045 if (ret >= 0) {
3046 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3047 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
3048 inactivate);
3049 }
3050 if (inactivate && ret >= 0) {
3051 s->block_inactive = true;
3052 }
3053 }
3054 qemu_mutex_unlock_iothread();
3055
3056 if (ret < 0) {
3057 goto fail;
3058 }
3059 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3060 trace_migration_completion_postcopy_end();
3061
3062 qemu_savevm_state_complete_postcopy(s->to_dst_file);
3063 trace_migration_completion_postcopy_end_after_complete();
3064 }
3065
3066 /*
3067 * If rp was opened we must clean up the thread before
3068 * cleaning everything else up (since if there are no failures
3069 * it will wait for the destination to send it's status in
3070 * a SHUT command).
3071 */
3072 if (s->rp_state.from_dst_file) {
3073 int rp_error;
3074 trace_migration_return_path_end_before();
3075 rp_error = await_return_path_close_on_source(s);
3076 trace_migration_return_path_end_after(rp_error);
3077 if (rp_error) {
3078 goto fail_invalidate;
3079 }
3080 }
3081
3082 if (qemu_file_get_error(s->to_dst_file)) {
3083 trace_migration_completion_file_err();
3084 goto fail_invalidate;
3085 }
3086
3087 if (!migrate_colo_enabled()) {
3088 migrate_set_state(&s->state, current_active_state,
3089 MIGRATION_STATUS_COMPLETED);
3090 }
3091
3092 return;
3093
3094 fail_invalidate:
3095 /* If not doing postcopy, vm_start() will be called: let's regain
3096 * control on images.
3097 */
3098 if (s->state == MIGRATION_STATUS_ACTIVE ||
3099 s->state == MIGRATION_STATUS_DEVICE) {
3100 Error *local_err = NULL;
3101
3102 qemu_mutex_lock_iothread();
3103 bdrv_invalidate_cache_all(&local_err);
3104 if (local_err) {
3105 error_report_err(local_err);
3106 } else {
3107 s->block_inactive = false;
3108 }
3109 qemu_mutex_unlock_iothread();
3110 }
3111
3112 fail:
3113 migrate_set_state(&s->state, current_active_state,
3114 MIGRATION_STATUS_FAILED);
3115 }
3116
3117 bool migrate_colo_enabled(void)
3118 {
3119 MigrationState *s = migrate_get_current();
3120 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
3121 }
3122
3123 typedef enum MigThrError {
3124 /* No error detected */
3125 MIG_THR_ERR_NONE = 0,
3126 /* Detected error, but resumed successfully */
3127 MIG_THR_ERR_RECOVERED = 1,
3128 /* Detected fatal error, need to exit */
3129 MIG_THR_ERR_FATAL = 2,
3130 } MigThrError;
3131
3132 static int postcopy_resume_handshake(MigrationState *s)
3133 {
3134 qemu_savevm_send_postcopy_resume(s->to_dst_file);
3135
3136 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3137 qemu_sem_wait(&s->rp_state.rp_sem);
3138 }
3139
3140 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3141 return 0;
3142 }
3143
3144 return -1;
3145 }
3146
3147 /* Return zero if success, or <0 for error */
3148 static int postcopy_do_resume(MigrationState *s)
3149 {
3150 int ret;
3151
3152 /*
3153 * Call all the resume_prepare() hooks, so that modules can be
3154 * ready for the migration resume.
3155 */
3156 ret = qemu_savevm_state_resume_prepare(s);
3157 if (ret) {
3158 error_report("%s: resume_prepare() failure detected: %d",
3159 __func__, ret);
3160 return ret;
3161 }
3162
3163 /*
3164 * Last handshake with destination on the resume (destination will
3165 * switch to postcopy-active afterwards)
3166 */
3167 ret = postcopy_resume_handshake(s);
3168 if (ret) {
3169 error_report("%s: handshake failed: %d", __func__, ret);
3170 return ret;
3171 }
3172
3173 return 0;
3174 }
3175
3176 /*
3177 * We don't return until we are in a safe state to continue current
3178 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3179 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3180 */
3181 static MigThrError postcopy_pause(MigrationState *s)
3182 {
3183 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3184
3185 while (true) {
3186 QEMUFile *file;
3187
3188 /* Current channel is possibly broken. Release it. */
3189 assert(s->to_dst_file);
3190 qemu_mutex_lock(&s->qemu_file_lock);
3191 file = s->to_dst_file;
3192 s->to_dst_file = NULL;
3193 qemu_mutex_unlock(&s->qemu_file_lock);
3194
3195 qemu_file_shutdown(file);
3196 qemu_fclose(file);
3197
3198 migrate_set_state(&s->state, s->state,
3199 MIGRATION_STATUS_POSTCOPY_PAUSED);
3200
3201 error_report("Detected IO failure for postcopy. "
3202 "Migration paused.");
3203
3204 /*
3205 * We wait until things fixed up. Then someone will setup the
3206 * status back for us.
3207 */
3208 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3209 qemu_sem_wait(&s->postcopy_pause_sem);
3210 }
3211
3212 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3213 /* Woken up by a recover procedure. Give it a shot */
3214
3215 /*
3216 * Firstly, let's wake up the return path now, with a new
3217 * return path channel.
3218 */
3219 qemu_sem_post(&s->postcopy_pause_rp_sem);
3220
3221 /* Do the resume logic */
3222 if (postcopy_do_resume(s) == 0) {
3223 /* Let's continue! */
3224 trace_postcopy_pause_continued();
3225 return MIG_THR_ERR_RECOVERED;
3226 } else {
3227 /*
3228 * Something wrong happened during the recovery, let's
3229 * pause again. Pause is always better than throwing
3230 * data away.
3231 */
3232 continue;
3233 }
3234 } else {
3235 /* This is not right... Time to quit. */
3236 return MIG_THR_ERR_FATAL;
3237 }
3238 }
3239 }
3240
3241 static MigThrError migration_detect_error(MigrationState *s)
3242 {
3243 int ret;
3244 int state = s->state;
3245 Error *local_error = NULL;
3246
3247 if (state == MIGRATION_STATUS_CANCELLING ||
3248 state == MIGRATION_STATUS_CANCELLED) {
3249 /* End the migration, but don't set the state to failed */
3250 return MIG_THR_ERR_FATAL;
3251 }
3252
3253 /* Try to detect any file errors */
3254 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3255 if (!ret) {
3256 /* Everything is fine */
3257 assert(!local_error);
3258 return MIG_THR_ERR_NONE;
3259 }
3260
3261 if (local_error) {
3262 migrate_set_error(s, local_error);
3263 error_free(local_error);
3264 }
3265
3266 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3267 /*
3268 * For postcopy, we allow the network to be down for a
3269 * while. After that, it can be continued by a
3270 * recovery phase.
3271 */
3272 return postcopy_pause(s);
3273 } else {
3274 /*
3275 * For precopy (or postcopy with error outside IO), we fail
3276 * with no time.
3277 */
3278 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3279 trace_migration_thread_file_err();
3280
3281 /* Time to stop the migration, now. */
3282 return MIG_THR_ERR_FATAL;
3283 }
3284 }
3285
3286 /* How many bytes have we transferred since the beginning of the migration */
3287 static uint64_t migration_total_bytes(MigrationState *s)
3288 {
3289 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3290 }
3291
3292 static void migration_calculate_complete(MigrationState *s)
3293 {
3294 uint64_t bytes = migration_total_bytes(s);
3295 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3296 int64_t transfer_time;
3297
3298 s->total_time = end_time - s->start_time;
3299 if (!s->downtime) {
3300 /*
3301 * It's still not set, so we are precopy migration. For
3302 * postcopy, downtime is calculated during postcopy_start().
3303 */
3304 s->downtime = end_time - s->downtime_start;
3305 }
3306
3307 transfer_time = s->total_time - s->setup_time;
3308 if (transfer_time) {
3309 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3310 }
3311 }
3312
3313 static void update_iteration_initial_status(MigrationState *s)
3314 {
3315 /*
3316 * Update these three fields at the same time to avoid mismatch info lead
3317 * wrong speed calculation.
3318 */
3319 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3320 s->iteration_initial_bytes = migration_total_bytes(s);
3321 s->iteration_initial_pages = ram_get_total_transferred_pages();
3322 }
3323
3324 static void migration_update_counters(MigrationState *s,
3325 int64_t current_time)
3326 {
3327 uint64_t transferred, transferred_pages, time_spent;
3328 uint64_t current_bytes; /* bytes transferred since the beginning */
3329 double bandwidth;
3330
3331 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3332 return;
3333 }
3334
3335 current_bytes = migration_total_bytes(s);
3336 transferred = current_bytes - s->iteration_initial_bytes;
3337 time_spent = current_time - s->iteration_start_time;
3338 bandwidth = (double)transferred / time_spent;
3339 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3340
3341 s->mbps = (((double) transferred * 8.0) /
3342 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3343
3344 transferred_pages = ram_get_total_transferred_pages() -
3345 s->iteration_initial_pages;
3346 s->pages_per_second = (double) transferred_pages /
3347 (((double) time_spent / 1000.0));
3348
3349 /*
3350 * if we haven't sent anything, we don't want to
3351 * recalculate. 10000 is a small enough number for our purposes
3352 */
3353 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3354 s->expected_downtime = ram_counters.remaining / bandwidth;
3355 }
3356
3357 qemu_file_reset_rate_limit(s->to_dst_file);
3358
3359 update_iteration_initial_status(s);
3360
3361 trace_migrate_transferred(transferred, time_spent,
3362 bandwidth, s->threshold_size);
3363 }
3364
3365 /* Migration thread iteration status */
3366 typedef enum {
3367 MIG_ITERATE_RESUME, /* Resume current iteration */
3368 MIG_ITERATE_SKIP, /* Skip current iteration */
3369 MIG_ITERATE_BREAK, /* Break the loop */
3370 } MigIterateState;
3371
3372 /*
3373 * Return true if continue to the next iteration directly, false
3374 * otherwise.
3375 */
3376 static MigIterateState migration_iteration_run(MigrationState *s)
3377 {
3378 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3379 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3380
3381 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3382 &pend_compat, &pend_post);
3383 pending_size = pend_pre + pend_compat + pend_post;
3384
3385 trace_migrate_pending(pending_size, s->threshold_size,
3386 pend_pre, pend_compat, pend_post);
3387
3388 if (pending_size && pending_size >= s->threshold_size) {
3389 /* Still a significant amount to transfer */
3390 if (!in_postcopy && pend_pre <= s->threshold_size &&
3391 qatomic_read(&s->start_postcopy)) {
3392 if (postcopy_start(s)) {
3393 error_report("%s: postcopy failed to start", __func__);
3394 }
3395 return MIG_ITERATE_SKIP;
3396 }
3397 /* Just another iteration step */
3398 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3399 } else {
3400 trace_migration_thread_low_pending(pending_size);
3401 migration_completion(s);
3402 return MIG_ITERATE_BREAK;
3403 }
3404
3405 return MIG_ITERATE_RESUME;
3406 }
3407
3408 static void migration_iteration_finish(MigrationState *s)
3409 {
3410 /* If we enabled cpu throttling for auto-converge, turn it off. */
3411 cpu_throttle_stop();
3412
3413 qemu_mutex_lock_iothread();
3414 switch (s->state) {
3415 case MIGRATION_STATUS_COMPLETED:
3416 migration_calculate_complete(s);
3417 runstate_set(RUN_STATE_POSTMIGRATE);
3418 break;
3419
3420 case MIGRATION_STATUS_ACTIVE:
3421 /*
3422 * We should really assert here, but since it's during
3423 * migration, let's try to reduce the usage of assertions.
3424 */
3425 if (!migrate_colo_enabled()) {
3426 error_report("%s: critical error: calling COLO code without "
3427 "COLO enabled", __func__);
3428 }
3429 migrate_start_colo_process(s);
3430 /*
3431 * Fixme: we will run VM in COLO no matter its old running state.
3432 * After exited COLO, we will keep running.
3433 */
3434 s->vm_was_running = true;
3435 /* Fallthrough */
3436 case MIGRATION_STATUS_FAILED:
3437 case MIGRATION_STATUS_CANCELLED:
3438 case MIGRATION_STATUS_CANCELLING:
3439 if (s->vm_was_running) {
3440 vm_start();
3441 } else {
3442 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3443 runstate_set(RUN_STATE_POSTMIGRATE);
3444 }
3445 }
3446 break;
3447
3448 default:
3449 /* Should not reach here, but if so, forgive the VM. */
3450 error_report("%s: Unknown ending state %d", __func__, s->state);
3451 break;
3452 }
3453 migrate_fd_cleanup_schedule(s);
3454 qemu_mutex_unlock_iothread();
3455 }
3456
3457 void migration_make_urgent_request(void)
3458 {
3459 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3460 }
3461
3462 void migration_consume_urgent_request(void)
3463 {
3464 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3465 }
3466
3467 /* Returns true if the rate limiting was broken by an urgent request */
3468 bool migration_rate_limit(void)
3469 {
3470 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3471 MigrationState *s = migrate_get_current();
3472
3473 bool urgent = false;
3474 migration_update_counters(s, now);
3475 if (qemu_file_rate_limit(s->to_dst_file)) {
3476
3477 if (qemu_file_get_error(s->to_dst_file)) {
3478 return false;
3479 }
3480 /*
3481 * Wait for a delay to do rate limiting OR
3482 * something urgent to post the semaphore.
3483 */
3484 int ms = s->iteration_start_time + BUFFER_DELAY - now;
3485 trace_migration_rate_limit_pre(ms);
3486 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3487 /*
3488 * We were woken by one or more urgent things but
3489 * the timedwait will have consumed one of them.
3490 * The service routine for the urgent wake will dec
3491 * the semaphore itself for each item it consumes,
3492 * so add this one we just eat back.
3493 */
3494 qemu_sem_post(&s->rate_limit_sem);
3495 urgent = true;
3496 }
3497 trace_migration_rate_limit_post(urgent);
3498 }
3499 return urgent;
3500 }
3501
3502 /*
3503 * Master migration thread on the source VM.
3504 * It drives the migration and pumps the data down the outgoing channel.
3505 */
3506 static void *migration_thread(void *opaque)
3507 {
3508 MigrationState *s = opaque;
3509 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3510 MigThrError thr_error;
3511 bool urgent = false;
3512
3513 rcu_register_thread();
3514
3515 object_ref(OBJECT(s));
3516 update_iteration_initial_status(s);
3517
3518 qemu_savevm_state_header(s->to_dst_file);
3519
3520 /*
3521 * If we opened the return path, we need to make sure dst has it
3522 * opened as well.
3523 */
3524 if (s->rp_state.from_dst_file) {
3525 /* Now tell the dest that it should open its end so it can reply */
3526 qemu_savevm_send_open_return_path(s->to_dst_file);
3527
3528 /* And do a ping that will make stuff easier to debug */
3529 qemu_savevm_send_ping(s->to_dst_file, 1);
3530 }
3531
3532 if (migrate_postcopy()) {
3533 /*
3534 * Tell the destination that we *might* want to do postcopy later;
3535 * if the other end can't do postcopy it should fail now, nice and
3536 * early.
3537 */
3538 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3539 }
3540
3541 if (migrate_colo_enabled()) {
3542 /* Notify migration destination that we enable COLO */
3543 qemu_savevm_send_colo_enable(s->to_dst_file);
3544 }
3545
3546 qemu_savevm_state_setup(s->to_dst_file);
3547
3548 if (qemu_savevm_state_guest_unplug_pending()) {
3549 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3550 MIGRATION_STATUS_WAIT_UNPLUG);
3551
3552 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3553 qemu_savevm_state_guest_unplug_pending()) {
3554 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3555 }
3556
3557 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3558 MIGRATION_STATUS_ACTIVE);
3559 }
3560
3561 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3562 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3563 MIGRATION_STATUS_ACTIVE);
3564
3565 trace_migration_thread_setup_complete();
3566
3567 while (migration_is_active(s)) {
3568 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3569 MigIterateState iter_state = migration_iteration_run(s);
3570 if (iter_state == MIG_ITERATE_SKIP) {
3571 continue;
3572 } else if (iter_state == MIG_ITERATE_BREAK) {
3573 break;
3574 }
3575 }
3576
3577 /*
3578 * Try to detect any kind of failures, and see whether we
3579 * should stop the migration now.
3580 */
3581 thr_error = migration_detect_error(s);
3582 if (thr_error == MIG_THR_ERR_FATAL) {
3583 /* Stop migration */
3584 break;
3585 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3586 /*
3587 * Just recovered from a e.g. network failure, reset all
3588 * the local variables. This is important to avoid
3589 * breaking transferred_bytes and bandwidth calculation
3590 */
3591 update_iteration_initial_status(s);
3592 }
3593
3594 urgent = migration_rate_limit();
3595 }
3596
3597 trace_migration_thread_after_loop();
3598 migration_iteration_finish(s);
3599 object_unref(OBJECT(s));
3600 rcu_unregister_thread();
3601 return NULL;
3602 }
3603
3604 void migrate_fd_connect(MigrationState *s, Error *error_in)
3605 {
3606 Error *local_err = NULL;
3607 int64_t rate_limit;
3608 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3609
3610 s->expected_downtime = s->parameters.downtime_limit;
3611 if (resume) {
3612 assert(s->cleanup_bh);
3613 } else {
3614 assert(!s->cleanup_bh);
3615 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3616 }
3617 if (error_in) {
3618 migrate_fd_error(s, error_in);
3619 migrate_fd_cleanup(s);
3620 return;
3621 }
3622
3623 if (resume) {
3624 /* This is a resumed migration */
3625 rate_limit = s->parameters.max_postcopy_bandwidth /
3626 XFER_LIMIT_RATIO;
3627 } else {
3628 /* This is a fresh new migration */
3629 rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3630
3631 /* Notify before starting migration thread */
3632 notifier_list_notify(&migration_state_notifiers, s);
3633 }
3634
3635 qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3636 qemu_file_set_blocking(s->to_dst_file, true);
3637
3638 /*
3639 * Open the return path. For postcopy, it is used exclusively. For
3640 * precopy, only if user specified "return-path" capability would
3641 * QEMU uses the return path.
3642 */
3643 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3644 if (open_return_path_on_source(s, !resume)) {
3645 error_report("Unable to open return-path for postcopy");
3646 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3647 migrate_fd_cleanup(s);
3648 return;
3649 }
3650 }
3651
3652 if (resume) {
3653 /* Wakeup the main migration thread to do the recovery */
3654 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3655 MIGRATION_STATUS_POSTCOPY_RECOVER);
3656 qemu_sem_post(&s->postcopy_pause_sem);
3657 return;
3658 }
3659
3660 if (multifd_save_setup(&local_err) != 0) {
3661 error_report_err(local_err);
3662 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3663 MIGRATION_STATUS_FAILED);
3664 migrate_fd_cleanup(s);
3665 return;
3666 }
3667 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3668 QEMU_THREAD_JOINABLE);
3669 s->migration_thread_running = true;
3670 }
3671
3672 void migration_global_dump(Monitor *mon)
3673 {
3674 MigrationState *ms = migrate_get_current();
3675
3676 monitor_printf(mon, "globals:\n");
3677 monitor_printf(mon, "store-global-state: %s\n",
3678 ms->store_global_state ? "on" : "off");
3679 monitor_printf(mon, "only-migratable: %s\n",
3680 only_migratable ? "on" : "off");
3681 monitor_printf(mon, "send-configuration: %s\n",
3682 ms->send_configuration ? "on" : "off");
3683 monitor_printf(mon, "send-section-footer: %s\n",
3684 ms->send_section_footer ? "on" : "off");
3685 monitor_printf(mon, "decompress-error-check: %s\n",
3686 ms->decompress_error_check ? "on" : "off");
3687 monitor_printf(mon, "clear-bitmap-shift: %u\n",
3688 ms->clear_bitmap_shift);
3689 }
3690
3691 #define DEFINE_PROP_MIG_CAP(name, x) \
3692 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3693
3694 static Property migration_properties[] = {
3695 DEFINE_PROP_BOOL("store-global-state", MigrationState,
3696 store_global_state, true),
3697 DEFINE_PROP_BOOL("send-configuration", MigrationState,
3698 send_configuration, true),
3699 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3700 send_section_footer, true),
3701 DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3702 decompress_error_check, true),
3703 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
3704 clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
3705
3706 /* Migration parameters */
3707 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3708 parameters.compress_level,
3709 DEFAULT_MIGRATE_COMPRESS_LEVEL),
3710 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3711 parameters.compress_threads,
3712 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3713 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3714 parameters.compress_wait_thread, true),
3715 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3716 parameters.decompress_threads,
3717 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3718 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState,
3719 parameters.throttle_trigger_threshold,
3720 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD),
3721 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3722 parameters.cpu_throttle_initial,
3723 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3724 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3725 parameters.cpu_throttle_increment,
3726 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3727 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState,
3728 parameters.cpu_throttle_tailslow, false),
3729 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3730 parameters.max_bandwidth, MAX_THROTTLE),
3731 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3732 parameters.downtime_limit,
3733 DEFAULT_MIGRATE_SET_DOWNTIME),
3734 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3735 parameters.x_checkpoint_delay,
3736 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3737 DEFINE_PROP_UINT8("multifd-channels", MigrationState,
3738 parameters.multifd_channels,
3739 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3740 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState,
3741 parameters.multifd_compression,
3742 DEFAULT_MIGRATE_MULTIFD_COMPRESSION),
3743 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState,
3744 parameters.multifd_zlib_level,
3745 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL),
3746 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState,
3747 parameters.multifd_zstd_level,
3748 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL),
3749 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3750 parameters.xbzrle_cache_size,
3751 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3752 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3753 parameters.max_postcopy_bandwidth,
3754 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3755 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3756 parameters.max_cpu_throttle,
3757 DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3758 DEFINE_PROP_SIZE("announce-initial", MigrationState,
3759 parameters.announce_initial,
3760 DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
3761 DEFINE_PROP_SIZE("announce-max", MigrationState,
3762 parameters.announce_max,
3763 DEFAULT_MIGRATE_ANNOUNCE_MAX),
3764 DEFINE_PROP_SIZE("announce-rounds", MigrationState,
3765 parameters.announce_rounds,
3766 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
3767 DEFINE_PROP_SIZE("announce-step", MigrationState,
3768 parameters.announce_step,
3769 DEFAULT_MIGRATE_ANNOUNCE_STEP),
3770
3771 /* Migration capabilities */
3772 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3773 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3774 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3775 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),