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