]> git.proxmox.com Git - mirror_qemu.git/blame - migration/migration.c
projects / mirror_qemu.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
migration: Synchronize multifd threads with main thread
[mirror_qemu.git] / migration / migration.c
CommitLineData
5bb7910a
AL		 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 *
6b620ca3
PB		 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.
5bb7910a
AL		 14 */
15
1393a485 16#include "qemu/osdep.h"
f348b6d1 17#include "qemu/cutils.h"
d49b6836 18#include "qemu/error-report.h"
795c40b8 19#include "migration/blocker.h"
f4dbe1bf 20#include "exec.h"
7fcac4a2 21#include "fd.h"
61e8b148 22#include "socket.h"
e1a3ecee 23#include "rdma.h"
7b1e1a22 24#include "ram.h"
84a899de 25#include "migration/global_state.h"
c4b63b7c 26#include "migration/misc.h"
6666c96a 27#include "migration.h"
20a519a0 28#include "savevm.h"
40014d81 29#include "qemu-file-channel.h"
08a0aee1 30#include "qemu-file.h"
987772d9 31#include "migration/vmstate.h"
737e150e 32#include "block/block.h"
e688df6b 33#include "qapi/error.h"
9af23989
MA		 34#include "qapi/qapi-commands-migration.h"
35#include "qapi/qapi-events-migration.h"
cc7a8ea7 36#include "qapi/qmp/qerror.h"
15280c36 37#include "qapi/qmp/qnull.h"
ab28bd23 38#include "qemu/rcu.h"
2c9e6fec 39#include "block.h"
be07b0ac 40#include "postcopy-ram.h"
766bd176 41#include "qemu/thread.h"
c09e5bb1 42#include "trace.h"
51180423 43#include "exec/target_page.h"
61b67d47 44#include "io/channel-buffer.h"
35a6ed4f 45#include "migration/colo.h"
4ffdb337 46#include "hw/boards.h"
9d18af93 47#include "monitor/monitor.h"
065e2813 48
dc325627 49#define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
5bb7910a 50
5b4e1eb7
JQ		 51/* Amount of time to allocate to each "chunk" of bandwidth-throttled
52 * data. */
53#define BUFFER_DELAY 100
54#define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
55
2ff30257
AA		 56/* Time in milliseconds we are allowed to stop the source,
57 * for sending the last part */
58#define DEFAULT_MIGRATE_SET_DOWNTIME 300
59
87c9cc1c
DHB		 60/* Maximum migrate downtime set to 2000 seconds */
61#define MAX_MIGRATE_DOWNTIME_SECONDS 2000
62#define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
63
8706d2d5
LL		 64/* Default compression thread count */
65#define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
3fcb38c2
LL		 66/* Default decompression thread count, usually decompression is at
67 * least 4 times as fast as compression.*/
68#define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
8706d2d5
LL		 69/*0: means nocompress, 1: best speed, ... 9: best compress ratio */
70#define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
1626fee3 71/* Define default autoconverge cpu throttle migration parameters */
d85a31d1
JH		 72#define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
73#define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
8706d2d5 74
17ad9b35 75/* Migration XBZRLE default cache size */
73af8dd8 76#define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
17ad9b35 77
68b53591
HZ		 78/* The delay time (in ms) between two COLO checkpoints
79 * Note: Please change this default value to 10000 when we support hybrid mode.
80 */
81#define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
4075fb1c 82#define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
0fb86605 83#define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
68b53591 84
7e555c6c
DDAG		 85/* Background transfer rate for postcopy, 0 means unlimited, note
86 * that page requests can still exceed this limit.
87 */
88#define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
89
99a0db9b
GH		 90static NotifierList migration_state_notifiers =
91 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
92
adde220a
DDAG		 93static bool deferred_incoming;
94
da6f1790
JQ		 95/* Messages sent on the return path from destination to source */
96enum mig_rp_message_type {
97 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
98 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
99 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
100
101 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
102 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
a335debb 103 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
13955b89 104 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
da6f1790
JQ		 105
106 MIG_RP_MSG_MAX
107};
108
17549e84
JQ		 109/* When we add fault tolerance, we could have several
110 migrations at once. For now we don't need to add
111 dynamic creation of migration */
112
e5cb7e76 113static MigrationState *current_migration;
e1b1b1bc 114static MigrationIncomingState *current_incoming;
e5cb7e76 115
8b0b29dc 116static bool migration_object_check(MigrationState *ms, Error **errp);
0331c8ca
DDAG		 117static int migration_maybe_pause(MigrationState *s,
118 int *current_active_state,
119 int new_state);
8b0b29dc 120
e5cb7e76
PX		 121void migration_object_init(void)
122{
4ffdb337 123 MachineState *ms = MACHINE(qdev_get_machine());
8b0b29dc 124 Error *err = NULL;
4ffdb337 125
e5cb7e76
PX		 126 /* This can only be called once. */
127 assert(!current_migration);
128 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
4ffdb337 129
e1b1b1bc
PX		 130 /*
131 * Init the migrate incoming object as well no matter whether
132 * we'll use it or not.
133 */
134 assert(!current_incoming);
135 current_incoming = g_new0(MigrationIncomingState, 1);
136 current_incoming->state = MIGRATION_STATUS_NONE;
137 current_incoming->postcopy_remote_fds =
138 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
139 qemu_mutex_init(&current_incoming->rp_mutex);
140 qemu_event_init(&current_incoming->main_thread_load_event, false);
141 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
142 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
143
144 init_dirty_bitmap_incoming_migration();
145
8b0b29dc
PX		 146 if (!migration_object_check(current_migration, &err)) {
147 error_report_err(err);
148 exit(1);
149 }
150
4ffdb337
PX		 151 /*
152 * We cannot really do this in migration_instance_init() since at
153 * that time global properties are not yet applied, then this
154 * value will be definitely replaced by something else.
155 */
156 if (ms->enforce_config_section) {
157 current_migration->send_configuration = true;
158 }
e5cb7e76
PX		 159}
160
1f895604
VSO		 161void migration_object_finalize(void)
162{
163 object_unref(OBJECT(current_migration));
164}
165
bca7856a 166/* For outgoing */
859bc756 167MigrationState *migrate_get_current(void)
17549e84 168{
e5cb7e76
PX		 169 /* This can only be called after the object created. */
170 assert(current_migration);
171 return current_migration;
17549e84
JQ		 172}
173
bca7856a
DDAG		 174MigrationIncomingState *migration_incoming_get_current(void)
175{
e1b1b1bc
PX		 176 assert(current_incoming);
177 return current_incoming;
bca7856a
DDAG		 178}
179
180void migration_incoming_state_destroy(void)
181{
b4b076da
JQ		 182 struct MigrationIncomingState *mis = migration_incoming_get_current();
183
3482655b 184 if (mis->to_src_file) {
660819b1
PX		 185 /* Tell source that we are done */
186 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
3482655b
PX		 187 qemu_fclose(mis->to_src_file);
188 mis->to_src_file = NULL;
189 }
190
660819b1
PX		 191 if (mis->from_src_file) {
192 qemu_fclose(mis->from_src_file);
193 mis->from_src_file = NULL;
194 }
00fa4fc8
DDAG		 195 if (mis->postcopy_remote_fds) {
196 g_array_free(mis->postcopy_remote_fds, TRUE);
197 mis->postcopy_remote_fds = NULL;
198 }
660819b1 199
5089e186 200 qemu_event_reset(&mis->main_thread_load_event);
bca7856a
DDAG		 201}
202
b05dc723
JQ		 203static void migrate_generate_event(int new_state)
204{
205 if (migrate_use_events()) {
206 qapi_event_send_migration(new_state, &error_abort);
b05dc723
JQ		 207 }
208}
209
0f073f44
DDAG		 210static bool migrate_late_block_activate(void)
211{
212 MigrationState *s;
213
214 s = migrate_get_current();
215
216 return s->enabled_capabilities[
217 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
218}
219
adde220a
DDAG		 220/*
221 * Called on -incoming with a defer: uri.
222 * The migration can be started later after any parameters have been
223 * changed.
224 */
225static void deferred_incoming_migration(Error **errp)
226{
227 if (deferred_incoming) {
228 error_setg(errp, "Incoming migration already deferred");
229 }
230 deferred_incoming = true;
231}
232
da6f1790
JQ		 233/*
234 * Send a message on the return channel back to the source
235 * of the migration.
236 */
d6208e35
PX		 237static int migrate_send_rp_message(MigrationIncomingState *mis,
238 enum mig_rp_message_type message_type,
239 uint16_t len, void *data)
da6f1790 240{
d6208e35
PX		 241 int ret = 0;
242
da6f1790
JQ		 243 trace_migrate_send_rp_message((int)message_type, len);
244 qemu_mutex_lock(&mis->rp_mutex);
d6208e35
PX		 245
246 /*
247 * It's possible that the file handle got lost due to network
248 * failures.
249 */
250 if (!mis->to_src_file) {
251 ret = -EIO;
252 goto error;
253 }
254
da6f1790
JQ		 255 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
256 qemu_put_be16(mis->to_src_file, len);
257 qemu_put_buffer(mis->to_src_file, data, len);
258 qemu_fflush(mis->to_src_file);
d6208e35
PX		 259
260 /* It's possible that qemu file got error during sending */
261 ret = qemu_file_get_error(mis->to_src_file);
262
263error:
da6f1790 264 qemu_mutex_unlock(&mis->rp_mutex);
d6208e35 265 return ret;
da6f1790
JQ		 266}
267
1e2d90eb
DDAG		 268/* Request a range of pages from the source VM at the given
269 * start address.
270 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
271 * as the last request (a name must have been given previously)
272 * Start: Address offset within the RB
273 * Len: Length in bytes required - must be a multiple of pagesize
274 */
d6208e35
PX		 275int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
276 ram_addr_t start, size_t len)
1e2d90eb 277{
cb8d4c8f 278 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
1e2d90eb 279 size_t msglen = 12; /* start + len */
d6208e35 280 enum mig_rp_message_type msg_type;
1e2d90eb
DDAG		 281
282 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
283 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
284
285 if (rbname) {
286 int rbname_len = strlen(rbname);
287 assert(rbname_len < 256);
288
289 bufc[msglen++] = rbname_len;
290 memcpy(bufc + msglen, rbname, rbname_len);
291 msglen += rbname_len;
d6208e35 292 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
1e2d90eb 293 } else {
d6208e35 294 msg_type = MIG_RP_MSG_REQ_PAGES;
1e2d90eb 295 }
d6208e35
PX		 296
297 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
1e2d90eb
DDAG		 298}
299
43eaae28 300void qemu_start_incoming_migration(const char *uri, Error **errp)
5bb7910a 301{
34c9dd8e
AL		 302 const char *p;
303
7cf1fe6d 304 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
adde220a
DDAG		 305 if (!strcmp(uri, "defer")) {
306 deferred_incoming_migration(errp);
307 } else if (strstart(uri, "tcp:", &p)) {
43eaae28 308 tcp_start_incoming_migration(p, errp);
2da776db 309#ifdef CONFIG_RDMA
adde220a 310 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 311 rdma_start_incoming_migration(p, errp);
312#endif
adde220a 313 } else if (strstart(uri, "exec:", &p)) {
43eaae28 314 exec_start_incoming_migration(p, errp);
adde220a 315 } else if (strstart(uri, "unix:", &p)) {
43eaae28 316 unix_start_incoming_migration(p, errp);
adde220a 317 } else if (strstart(uri, "fd:", &p)) {
43eaae28 318 fd_start_incoming_migration(p, errp);
adde220a 319 } else {
312fd5f2 320 error_setg(errp, "unknown migration protocol: %s", uri);
8ca5e801 321 }
5bb7910a
AL		 322}
323
0aa6aefc
DL		 324static void process_incoming_migration_bh(void *opaque)
325{
326 Error *local_err = NULL;
327 MigrationIncomingState *mis = opaque;
328
0f073f44
DDAG		 329 /* If capability late_block_activate is set:
330 * Only fire up the block code now if we're going to restart the
331 * VM, else 'cont' will do it.
332 * This causes file locking to happen; so we don't want it to happen
333 * unless we really are starting the VM.
334 */
335 if (!migrate_late_block_activate() ||
336 (autostart && (!global_state_received() ||
337 global_state_get_runstate() == RUN_STATE_RUNNING))) {
338 /* Make sure all file formats flush their mutable metadata.
339 * If we get an error here, just don't restart the VM yet. */
340 bdrv_invalidate_cache_all(&local_err);
341 if (local_err) {
342 error_report_err(local_err);
343 local_err = NULL;
344 autostart = false;
345 }
d35ff5e6
KW		 346 }
347
0aa6aefc
DL		 348 /*
349 * This must happen after all error conditions are dealt with and
350 * we're sure the VM is going to be running on this host.
351 */
352 qemu_announce_self();
353
f986c3d2
JQ		 354 if (multifd_load_cleanup(&local_err) != 0) {
355 error_report_err(local_err);
356 autostart = false;
357 }
0aa6aefc
DL		 358 /* If global state section was not received or we are in running
359 state, we need to obey autostart. Any other state is set with
360 runstate_set. */
361
b35ebdf0
VSO		 362 dirty_bitmap_mig_before_vm_start();
363
0aa6aefc
DL		 364 if (!global_state_received() ||
365 global_state_get_runstate() == RUN_STATE_RUNNING) {
366 if (autostart) {
367 vm_start();
368 } else {
369 runstate_set(RUN_STATE_PAUSED);
370 }
371 } else {
372 runstate_set(global_state_get_runstate());
373 }
0aa6aefc
DL		 374 /*
375 * This must happen after any state changes since as soon as an external
376 * observer sees this event they might start to prod at the VM assuming
377 * it's ready to use.
378 */
379 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
380 MIGRATION_STATUS_COMPLETED);
381 qemu_bh_delete(mis->bh);
382 migration_incoming_state_destroy();
383}
384
82a4da79 385static void process_incoming_migration_co(void *opaque)
511c0231 386{
b4b076da 387 MigrationIncomingState *mis = migration_incoming_get_current();
e9bef235 388 PostcopyState ps;
1c12e1f5
PB		 389 int ret;
390
4f0fae7f 391 assert(mis->from_src_file);
67f11b5c 392 mis->largest_page_size = qemu_ram_pagesize_largest();
093e3c42 393 postcopy_state_set(POSTCOPY_INCOMING_NONE);
93d7af6f
HZ		 394 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
395 MIGRATION_STATUS_ACTIVE);
4f0fae7f 396 ret = qemu_loadvm_state(mis->from_src_file);
bca7856a 397
e9bef235
DDAG		 398 ps = postcopy_state_get();
399 trace_process_incoming_migration_co_end(ret, ps);
400 if (ps != POSTCOPY_INCOMING_NONE) {
401 if (ps == POSTCOPY_INCOMING_ADVISE) {
402 /*
403 * Where a migration had postcopy enabled (and thus went to advise)
404 * but managed to complete within the precopy period, we can use
405 * the normal exit.
406 */
407 postcopy_ram_incoming_cleanup(mis);
408 } else if (ret >= 0) {
409 /*
410 * Postcopy was started, cleanup should happen at the end of the
411 * postcopy thread.
412 */
413 trace_process_incoming_migration_co_postcopy_end_main();
414 return;
415 }
416 /* Else if something went wrong then just fall out of the normal exit */
417 }
418
25d0c16f
HZ		 419 /* we get COLO info, and know if we are in COLO mode */
420 if (!ret && migration_incoming_enable_colo()) {
421 mis->migration_incoming_co = qemu_coroutine_self();
422 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
423 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
424 mis->have_colo_incoming_thread = true;
425 qemu_coroutine_yield();
426
427 /* Wait checkpoint incoming thread exit before free resource */
428 qemu_thread_join(&mis->colo_incoming_thread);
429 }
430
1c12e1f5 431 if (ret < 0) {
f986c3d2
JQ		 432 Error *local_err = NULL;
433
93d7af6f
HZ		 434 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
435 MIGRATION_STATUS_FAILED);
db80face 436 error_report("load of migration failed: %s", strerror(-ret));
3a0f2cea 437 qemu_fclose(mis->from_src_file);
f986c3d2
JQ		 438 if (multifd_load_cleanup(&local_err) != 0) {
439 error_report_err(local_err);
440 }
4aead692 441 exit(EXIT_FAILURE);
511c0231 442 }
0aa6aefc
DL		 443 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
444 qemu_bh_schedule(mis->bh);
511c0231
JQ		 445}
446
e595a01a 447static void migration_incoming_setup(QEMUFile *f)
82a4da79 448{
4f0fae7f 449 MigrationIncomingState *mis = migration_incoming_get_current();
82a4da79 450
f986c3d2
JQ		 451 if (multifd_load_setup() != 0) {
452 /* We haven't been able to create multifd threads
453 nothing better to do */
454 exit(EXIT_FAILURE);
455 }
456
4f0fae7f
JQ		 457 if (!mis->from_src_file) {
458 mis->from_src_file = f;
459 }
06ad5135 460 qemu_file_set_blocking(f, false);
e595a01a
JQ		 461}
462
36c2f8be 463void migration_incoming_process(void)
e595a01a
JQ		 464{
465 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
0b8b8753 466 qemu_coroutine_enter(co);
82a4da79
PB		 467}
468
e595a01a
JQ		 469void migration_fd_process_incoming(QEMUFile *f)
470{
d96c9e8d
PX		 471 MigrationIncomingState *mis = migration_incoming_get_current();
472
473 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
474 /* Resumed from a paused postcopy migration */
475
476 mis->from_src_file = f;
477 /* Postcopy has standalone thread to do vm load */
478 qemu_file_set_blocking(f, true);
479
480 /* Re-configure the return path */
481 mis->to_src_file = qemu_file_get_return_path(f);
482
483 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
484 MIGRATION_STATUS_POSTCOPY_RECOVER);
485
486 /*
487 * Here, we only wake up the main loading thread (while the
488 * fault thread will still be waiting), so that we can receive
489 * commands from source now, and answer it if needed. The
490 * fault thread will be woken up afterwards until we are sure
491 * that source is ready to reply to page requests.
492 */
493 qemu_sem_post(&mis->postcopy_pause_sem_dst);
494 } else {
495 /* New incoming migration */
496 migration_incoming_setup(f);
497 migration_incoming_process();
498 }
e595a01a
JQ		 499}
500
4f0fae7f
JQ		 501void migration_ioc_process_incoming(QIOChannel *ioc)
502{
503 MigrationIncomingState *mis = migration_incoming_get_current();
504
505 if (!mis->from_src_file) {
506 QEMUFile *f = qemu_fopen_channel_input(ioc);
36c2f8be 507 migration_incoming_setup(f);
71bb07db 508 return;
4f0fae7f 509 }
71bb07db 510 multifd_recv_new_channel(ioc);
4f0fae7f
JQ		 511}
512
428d8908
JQ		 513/**
514 * @migration_has_all_channels: We have received all channels that we need
515 *
516 * Returns true when we have got connections to all the channels that
517 * we need for migration.
518 */
519bool migration_has_all_channels(void)
520{
62c1e0ca
JQ		 521 bool all_channels;
522
523 all_channels = multifd_recv_all_channels_created();
524
525 return all_channels;
428d8908
JQ		 526}
527
6decec93
DDAG		 528/*
529 * Send a 'SHUT' message on the return channel with the given value
530 * to indicate that we've finished with the RP. Non-0 value indicates
531 * error.
532 */
533void migrate_send_rp_shut(MigrationIncomingState *mis,
534 uint32_t value)
535{
536 uint32_t buf;
537
538 buf = cpu_to_be32(value);
539 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
540}
541
542/*
543 * Send a 'PONG' message on the return channel with the given value
544 * (normally in response to a 'PING')
545 */
546void migrate_send_rp_pong(MigrationIncomingState *mis,
547 uint32_t value)
548{
549 uint32_t buf;
550
551 buf = cpu_to_be32(value);
552 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
553}
554
a335debb
PX		 555void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
556 char *block_name)
557{
558 char buf[512];
559 int len;
560 int64_t res;
561
562 /*
563 * First, we send the header part. It contains only the len of
564 * idstr, and the idstr itself.
565 */
566 len = strlen(block_name);
567 buf[0] = len;
568 memcpy(buf + 1, block_name, len);
569
570 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
571 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
572 __func__);
573 return;
574 }
575
576 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
577
578 /*
579 * Next, we dump the received bitmap to the stream.
580 *
581 * TODO: currently we are safe since we are the only one that is
582 * using the to_src_file handle (fault thread is still paused),
583 * and it's ok even not taking the mutex. However the best way is
584 * to take the lock before sending the message header, and release
585 * the lock after sending the bitmap.
586 */
587 qemu_mutex_lock(&mis->rp_mutex);
588 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
589 qemu_mutex_unlock(&mis->rp_mutex);
590
591 trace_migrate_send_rp_recv_bitmap(block_name, res);
592}
593
13955b89
PX		 594void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
595{
596 uint32_t buf;
597
598 buf = cpu_to_be32(value);
599 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
600}
601
bbf6da32
OW		 602MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
603{
604 MigrationCapabilityStatusList *head = NULL;
605 MigrationCapabilityStatusList *caps;
606 MigrationState *s = migrate_get_current();
607 int i;
608
387eedeb 609 caps = NULL; /* silence compiler warning */
7fb1cf16 610 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
ed1701c6
DDAG		 611#ifndef CONFIG_LIVE_BLOCK_MIGRATION
612 if (i == MIGRATION_CAPABILITY_BLOCK) {
613 continue;
614 }
615#endif
bbf6da32
OW		 616 if (head == NULL) {
617 head = g_malloc0(sizeof(*caps));
618 caps = head;
619 } else {
620 caps->next = g_malloc0(sizeof(*caps));
621 caps = caps->next;
622 }
623 caps->value =
624 g_malloc(sizeof(*caps->value));
625 caps->value->capability = i;
626 caps->value->state = s->enabled_capabilities[i];
627 }
628
629 return head;
630}
631
85de8323
LL		 632MigrationParameters *qmp_query_migrate_parameters(Error **errp)
633{
634 MigrationParameters *params;
635 MigrationState *s = migrate_get_current();
636
e87fae4c 637 /* TODO use QAPI_CLONE() instead of duplicating it inline */
85de8323 638 params = g_malloc0(sizeof(*params));
de63ab61 639 params->has_compress_level = true;
2594f56d 640 params->compress_level = s->parameters.compress_level;
de63ab61 641 params->has_compress_threads = true;
2594f56d 642 params->compress_threads = s->parameters.compress_threads;
de63ab61 643 params->has_decompress_threads = true;
2594f56d 644 params->decompress_threads = s->parameters.decompress_threads;
de63ab61 645 params->has_cpu_throttle_initial = true;
2594f56d 646 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
de63ab61 647 params->has_cpu_throttle_increment = true;
2594f56d 648 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
8cc99dcd 649 params->has_tls_creds = true;
69ef1f36 650 params->tls_creds = g_strdup(s->parameters.tls_creds);
8cc99dcd 651 params->has_tls_hostname = true;
69ef1f36 652 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
2ff30257
AA		 653 params->has_max_bandwidth = true;
654 params->max_bandwidth = s->parameters.max_bandwidth;
655 params->has_downtime_limit = true;
656 params->downtime_limit = s->parameters.downtime_limit;
fe39a4d4 657 params->has_x_checkpoint_delay = true;
68b53591 658 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
2833c59b
JQ		 659 params->has_block_incremental = true;
660 params->block_incremental = s->parameters.block_incremental;
4075fb1c
JQ		 661 params->has_x_multifd_channels = true;
662 params->x_multifd_channels = s->parameters.x_multifd_channels;
0fb86605
JQ		 663 params->has_x_multifd_page_count = true;
664 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
73af8dd8
JQ		 665 params->has_xbzrle_cache_size = true;
666 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
7e555c6c
DDAG		 667 params->has_max_postcopy_bandwidth = true;
668 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
85de8323
LL		 669
670 return params;
671}
672
f6844b99
DDAG		 673/*
674 * Return true if we're already in the middle of a migration
675 * (i.e. any of the active or setup states)
676 */
677static bool migration_is_setup_or_active(int state)
678{
679 switch (state) {
680 case MIGRATION_STATUS_ACTIVE:
9ec055ae 681 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
a688d2c1 682 case MIGRATION_STATUS_POSTCOPY_PAUSED:
135b87b4 683 case MIGRATION_STATUS_POSTCOPY_RECOVER:
f6844b99 684 case MIGRATION_STATUS_SETUP:
31e06077
DDAG		 685 case MIGRATION_STATUS_PRE_SWITCHOVER:
686 case MIGRATION_STATUS_DEVICE:
f6844b99
DDAG		 687 return true;
688
689 default:
690 return false;
691
692 }
693}
694
a22463a5
DDAG		 695static void populate_ram_info(MigrationInfo *info, MigrationState *s)
696{
697 info->has_ram = true;
698 info->ram = g_malloc0(sizeof(*info->ram));
9360447d 699 info->ram->transferred = ram_counters.transferred;
a22463a5 700 info->ram->total = ram_bytes_total();
9360447d 701 info->ram->duplicate = ram_counters.duplicate;
bedf53c1
JQ		 702 /* legacy value. It is not used anymore */
703 info->ram->skipped = 0;
9360447d
JQ		 704 info->ram->normal = ram_counters.normal;
705 info->ram->normal_bytes = ram_counters.normal *
20afaed9 706 qemu_target_page_size();
a22463a5 707 info->ram->mbps = s->mbps;
9360447d
JQ		 708 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
709 info->ram->postcopy_requests = ram_counters.postcopy_requests;
030ce1f8 710 info->ram->page_size = qemu_target_page_size();
a22463a5 711
114f5aee
JQ		 712 if (migrate_use_xbzrle()) {
713 info->has_xbzrle_cache = true;
714 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
715 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
9360447d
JQ		 716 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
717 info->xbzrle_cache->pages = xbzrle_counters.pages;
718 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
719 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
720 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
114f5aee
JQ		 721 }
722
338182c8
JQ		 723 if (cpu_throttle_active()) {
724 info->has_cpu_throttle_percentage = true;
725 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
726 }
727
a22463a5
DDAG		 728 if (s->state != MIGRATION_STATUS_COMPLETED) {
729 info->ram->remaining = ram_bytes_remaining();
9360447d 730 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
a22463a5
DDAG		 731 }
732}
733
930ac04c
JQ		 734static void populate_disk_info(MigrationInfo *info)
735{
736 if (blk_mig_active()) {
737 info->has_disk = true;
738 info->disk = g_malloc0(sizeof(*info->disk));
739 info->disk->transferred = blk_mig_bytes_transferred();
740 info->disk->remaining = blk_mig_bytes_remaining();
741 info->disk->total = blk_mig_bytes_total();
742 }
743}
744
65ace060 745static void fill_source_migration_info(MigrationInfo *info)
5bb7910a 746{
17549e84
JQ		 747 MigrationState *s = migrate_get_current();
748
749 switch (s->state) {
31194731 750 case MIGRATION_STATUS_NONE:
17549e84 751 /* no migration has happened ever */
65ace060
AP		 752 /* do not overwrite destination migration status */
753 return;
17549e84 754 break;
31194731 755 case MIGRATION_STATUS_SETUP:
29ae8a41 756 info->has_status = true;
ed4fbd10 757 info->has_total_time = false;
29ae8a41 758 break;
31194731
HZ		 759 case MIGRATION_STATUS_ACTIVE:
760 case MIGRATION_STATUS_CANCELLING:
9ec055ae 761 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
31e06077
DDAG		 762 case MIGRATION_STATUS_PRE_SWITCHOVER:
763 case MIGRATION_STATUS_DEVICE:
a688d2c1 764 case MIGRATION_STATUS_POSTCOPY_PAUSED:
135b87b4 765 case MIGRATION_STATUS_POSTCOPY_RECOVER:
c8f9f4f4 766 /* TODO add some postcopy stats */
9ec055ae
DDAG		 767 info->has_status = true;
768 info->has_total_time = true;
769 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
4af246a3 770 - s->start_time;
9ec055ae
DDAG		 771 info->has_expected_downtime = true;
772 info->expected_downtime = s->expected_downtime;
773 info->has_setup_time = true;
774 info->setup_time = s->setup_time;
775
a22463a5 776 populate_ram_info(info, s);
930ac04c 777 populate_disk_info(info);
17549e84 778 break;
0b827d5e
HZ		 779 case MIGRATION_STATUS_COLO:
780 info->has_status = true;
781 /* TODO: display COLO specific information (checkpoint info etc.) */
782 break;
31194731 783 case MIGRATION_STATUS_COMPLETED:
791e7c82 784 info->has_status = true;
00c14997 785 info->has_total_time = true;
7aa939af 786 info->total_time = s->total_time;
9c5a9fcf
JQ		 787 info->has_downtime = true;
788 info->downtime = s->downtime;
ed4fbd10
MH		 789 info->has_setup_time = true;
790 info->setup_time = s->setup_time;
d5f8a570 791
a22463a5 792 populate_ram_info(info, s);
17549e84 793 break;
31194731 794 case MIGRATION_STATUS_FAILED:
791e7c82 795 info->has_status = true;
d59ce6f3
DB		 796 if (s->error) {
797 info->has_error_desc = true;
798 info->error_desc = g_strdup(error_get_pretty(s->error));
799 }
17549e84 800 break;
31194731 801 case MIGRATION_STATUS_CANCELLED:
791e7c82 802 info->has_status = true;
17549e84 803 break;
5bb7910a 804 }
cde63fbe 805 info->status = s->state;
5bb7910a
AL		 806}
807
4a84214e
PX		 808/**
809 * @migration_caps_check - check capability validity
810 *
811 * @cap_list: old capability list, array of bool
812 * @params: new capabilities to be applied soon
813 * @errp: set *errp if the check failed, with reason
814 *
815 * Returns true if check passed, otherwise false.
816 */
817static bool migrate_caps_check(bool *cap_list,
818 MigrationCapabilityStatusList *params,
819 Error **errp)
00458433 820{
00458433 821 MigrationCapabilityStatusList *cap;
4a84214e 822 bool old_postcopy_cap;
d7651f15 823 MigrationIncomingState *mis = migration_incoming_get_current();
00458433 824
4a84214e 825 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
00458433
OW		 826
827 for (cap = params; cap; cap = cap->next) {
4a84214e
PX		 828 cap_list[cap->value->capability] = cap->value->state;
829 }
830
ed1701c6 831#ifndef CONFIG_LIVE_BLOCK_MIGRATION
4a84214e
PX		 832 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
833 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
834 "block migration");
835 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
836 return false;
00458433 837 }
4a84214e 838#endif
53dd370c 839
4a84214e
PX		 840 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
841 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
53dd370c
DDAG		 842 /* The decompression threads asynchronously write into RAM
843 * rather than use the atomic copies needed to avoid
844 * userfaulting. It should be possible to fix the decompression
845 * threads for compatibility in future.
846 */
4a84214e
PX		 847 error_setg(errp, "Postcopy is not currently compatible "
848 "with compression");
849 return false;
53dd370c 850 }
4a84214e 851
096631bd
DDAG		 852 /* This check is reasonably expensive, so only when it's being
853 * set the first time, also it's only the destination that needs
854 * special support.
855 */
856 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
d7651f15 857 !postcopy_ram_supported_by_host(mis)) {
096631bd
DDAG		 858 /* postcopy_ram_supported_by_host will have emitted a more
859 * detailed message
860 */
4a84214e
PX		 861 error_setg(errp, "Postcopy is not supported");
862 return false;
096631bd 863 }
53dd370c 864 }
4a84214e
PX		 865
866 return true;
867}
868
65ace060
AP		 869static void fill_destination_migration_info(MigrationInfo *info)
870{
871 MigrationIncomingState *mis = migration_incoming_get_current();
872
873 switch (mis->state) {
874 case MIGRATION_STATUS_NONE:
875 return;
876 break;
877 case MIGRATION_STATUS_SETUP:
878 case MIGRATION_STATUS_CANCELLING:
879 case MIGRATION_STATUS_CANCELLED:
880 case MIGRATION_STATUS_ACTIVE:
881 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
882 case MIGRATION_STATUS_FAILED:
883 case MIGRATION_STATUS_COLO:
884 info->has_status = true;
885 break;
886 case MIGRATION_STATUS_COMPLETED:
887 info->has_status = true;
888 fill_destination_postcopy_migration_info(info);
889 break;
890 }
891 info->status = mis->state;
892}
893
894MigrationInfo *qmp_query_migrate(Error **errp)
895{
896 MigrationInfo *info = g_malloc0(sizeof(*info));
897
898 fill_destination_migration_info(info);
899 fill_source_migration_info(info);
900
901 return info;
902}
903
4a84214e
PX		 904void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
905 Error **errp)
906{
907 MigrationState *s = migrate_get_current();
908 MigrationCapabilityStatusList *cap;
dd0ee30c 909 bool cap_list[MIGRATION_CAPABILITY__MAX];
4a84214e
PX		 910
911 if (migration_is_setup_or_active(s->state)) {
912 error_setg(errp, QERR_MIGRATION_ACTIVE);
913 return;
914 }
915
dd0ee30c
PX		 916 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
917 if (!migrate_caps_check(cap_list, params, errp)) {
4a84214e
PX		 918 return;
919 }
920
921 for (cap = params; cap; cap = cap->next) {
922 s->enabled_capabilities[cap->value->capability] = cap->value->state;
923 }
00458433
OW		 924}
925
16d063bc
PX		 926/*
927 * Check whether the parameters are valid. Error will be put into errp
928 * (if provided). Return true if valid, otherwise false.
929 */
930static bool migrate_params_check(MigrationParameters *params, Error **errp)
85de8323 931{
7f375e04 932 if (params->has_compress_level &&
741d4086 933 (params->compress_level > 9)) {
c6bd8c70
MA		 934 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
935 "is invalid, it should be in the range of 0 to 9");
16d063bc 936 return false;
85de8323 937 }
16d063bc 938
741d4086 939 if (params->has_compress_threads && (params->compress_threads < 1)) {
c6bd8c70
MA		 940 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
941 "compress_threads",
942 "is invalid, it should be in the range of 1 to 255");
16d063bc 943 return false;
85de8323 944 }
16d063bc 945
741d4086 946 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
c6bd8c70
MA		 947 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
948 "decompress_threads",
949 "is invalid, it should be in the range of 1 to 255");
16d063bc 950 return false;
85de8323 951 }
16d063bc 952
7f375e04
EB		 953 if (params->has_cpu_throttle_initial &&
954 (params->cpu_throttle_initial < 1 ||
955 params->cpu_throttle_initial > 99)) {
1626fee3 956 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 957 "cpu_throttle_initial",
1626fee3 958 "an integer in the range of 1 to 99");
16d063bc 959 return false;
1626fee3 960 }
16d063bc 961
7f375e04
EB		 962 if (params->has_cpu_throttle_increment &&
963 (params->cpu_throttle_increment < 1 ||
964 params->cpu_throttle_increment > 99)) {
1626fee3 965 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 966 "cpu_throttle_increment",
1626fee3 967 "an integer in the range of 1 to 99");
16d063bc 968 return false;
1626fee3 969 }
16d063bc 970
741d4086 971 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
2ff30257
AA		 972 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
973 " range of 0 to %zu bytes/second", SIZE_MAX);
16d063bc 974 return false;
2ff30257 975 }
16d063bc 976
2ff30257 977 if (params->has_downtime_limit &&
741d4086 978 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
87c9cc1c
DHB		 979 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
980 "the range of 0 to %d milliseconds",
981 MAX_MIGRATE_DOWNTIME);
16d063bc 982 return false;
2ff30257 983 }
16d063bc 984
741d4086
JQ		 985 /* x_checkpoint_delay is now always positive */
986
987 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
4075fb1c
JQ		 988 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
989 "multifd_channels",
990 "is invalid, it should be in the range of 1 to 255");
991 return false;
992 }
0fb86605 993 if (params->has_x_multifd_page_count &&
741d4086
JQ		 994 (params->x_multifd_page_count < 1 ||
995 params->x_multifd_page_count > 10000)) {
0fb86605
JQ		 996 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
997 "multifd_page_count",
998 "is invalid, it should be in the range of 1 to 10000");
999 return false;
1000 }
16d063bc 1001
73af8dd8
JQ		 1002 if (params->has_xbzrle_cache_size &&
1003 (params->xbzrle_cache_size < qemu_target_page_size() ||
1004 !is_power_of_2(params->xbzrle_cache_size))) {
1005 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1006 "xbzrle_cache_size",
1007 "is invalid, it should be bigger than target page size"
1008 " and a power of two");
1009 return false;
1010 }
1011
16d063bc
PX		 1012 return true;
1013}
1014
1bda8b3c
MA		 1015static void migrate_params_test_apply(MigrateSetParameters *params,
1016 MigrationParameters *dest)
1017{
1018 *dest = migrate_get_current()->parameters;
1019
1020 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1021
1022 if (params->has_compress_level) {
1023 dest->compress_level = params->compress_level;
1024 }
1025
1026 if (params->has_compress_threads) {
1027 dest->compress_threads = params->compress_threads;
1028 }
1029
1030 if (params->has_decompress_threads) {
1031 dest->decompress_threads = params->decompress_threads;
1032 }
1033
1034 if (params->has_cpu_throttle_initial) {
1035 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1036 }
1037
1038 if (params->has_cpu_throttle_increment) {
1039 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1040 }
1041
1042 if (params->has_tls_creds) {
01fa5598
MA		 1043 assert(params->tls_creds->type == QTYPE_QSTRING);
1044 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1bda8b3c
MA		 1045 }
1046
1047 if (params->has_tls_hostname) {
01fa5598
MA		 1048 assert(params->tls_hostname->type == QTYPE_QSTRING);
1049 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1bda8b3c
MA		 1050 }
1051
1052 if (params->has_max_bandwidth) {
1053 dest->max_bandwidth = params->max_bandwidth;
1054 }
1055
1056 if (params->has_downtime_limit) {
1057 dest->downtime_limit = params->downtime_limit;
1058 }
1059
1060 if (params->has_x_checkpoint_delay) {
1061 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1062 }
1063
1064 if (params->has_block_incremental) {
1065 dest->block_incremental = params->block_incremental;
1066 }
5e7577a1
JQ		 1067 if (params->has_x_multifd_channels) {
1068 dest->x_multifd_channels = params->x_multifd_channels;
1069 }
1070 if (params->has_x_multifd_page_count) {
1071 dest->x_multifd_page_count = params->x_multifd_page_count;
1072 }
73af8dd8
JQ		 1073 if (params->has_xbzrle_cache_size) {
1074 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1075 }
7e555c6c
DDAG		 1076 if (params->has_max_postcopy_bandwidth) {
1077 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1078 }
1bda8b3c
MA		 1079}
1080
73af8dd8 1081static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
16d063bc
PX		 1082{
1083 MigrationState *s = migrate_get_current();
1084
e87fae4c
MA		 1085 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1086
7f375e04
EB		 1087 if (params->has_compress_level) {
1088 s->parameters.compress_level = params->compress_level;
85de8323 1089 }
476c72aa 1090
7f375e04
EB		 1091 if (params->has_compress_threads) {
1092 s->parameters.compress_threads = params->compress_threads;
85de8323 1093 }
476c72aa 1094
7f375e04
EB		 1095 if (params->has_decompress_threads) {
1096 s->parameters.decompress_threads = params->decompress_threads;
85de8323 1097 }
476c72aa 1098
7f375e04
EB		 1099 if (params->has_cpu_throttle_initial) {
1100 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1626fee3 1101 }
476c72aa 1102
7f375e04
EB		 1103 if (params->has_cpu_throttle_increment) {
1104 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1626fee3 1105 }
476c72aa 1106
7f375e04 1107 if (params->has_tls_creds) {
69ef1f36 1108 g_free(s->parameters.tls_creds);
01fa5598
MA		 1109 assert(params->tls_creds->type == QTYPE_QSTRING);
1110 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
69ef1f36 1111 }
476c72aa 1112
7f375e04 1113 if (params->has_tls_hostname) {
69ef1f36 1114 g_free(s->parameters.tls_hostname);
01fa5598
MA		 1115 assert(params->tls_hostname->type == QTYPE_QSTRING);
1116 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
69ef1f36 1117 }
476c72aa 1118
2ff30257
AA		 1119 if (params->has_max_bandwidth) {
1120 s->parameters.max_bandwidth = params->max_bandwidth;
1121 if (s->to_dst_file) {
1122 qemu_file_set_rate_limit(s->to_dst_file,
1123 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1124 }
1125 }
476c72aa 1126
2ff30257
AA		 1127 if (params->has_downtime_limit) {
1128 s->parameters.downtime_limit = params->downtime_limit;
1129 }
68b53591
HZ		 1130
1131 if (params->has_x_checkpoint_delay) {
1132 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
479125d5
HZ		 1133 if (migration_in_colo_state()) {
1134 colo_checkpoint_notify(s);
1135 }
68b53591 1136 }
476c72aa 1137
2833c59b
JQ		 1138 if (params->has_block_incremental) {
1139 s->parameters.block_incremental = params->block_incremental;
1140 }
4075fb1c
JQ		 1141 if (params->has_x_multifd_channels) {
1142 s->parameters.x_multifd_channels = params->x_multifd_channels;
1143 }
0fb86605
JQ		 1144 if (params->has_x_multifd_page_count) {
1145 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
1146 }
73af8dd8
JQ		 1147 if (params->has_xbzrle_cache_size) {
1148 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1149 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1150 }
7e555c6c
DDAG		 1151 if (params->has_max_postcopy_bandwidth) {
1152 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1153 }
85de8323
LL		 1154}
1155
1bda8b3c 1156void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
476c72aa 1157{
1bda8b3c
MA		 1158 MigrationParameters tmp;
1159
01fa5598
MA		 1160 /* TODO Rewrite "" to null instead */
1161 if (params->has_tls_creds
1162 && params->tls_creds->type == QTYPE_QNULL) {
cb3e7f08 1163 qobject_unref(params->tls_creds->u.n);
01fa5598
MA		 1164 params->tls_creds->type = QTYPE_QSTRING;
1165 params->tls_creds->u.s = strdup("");
1166 }
1167 /* TODO Rewrite "" to null instead */
1168 if (params->has_tls_hostname
1169 && params->tls_hostname->type == QTYPE_QNULL) {
cb3e7f08 1170 qobject_unref(params->tls_hostname->u.n);
01fa5598
MA		 1171 params->tls_hostname->type = QTYPE_QSTRING;
1172 params->tls_hostname->u.s = strdup("");
1173 }
1174
1bda8b3c
MA		 1175 migrate_params_test_apply(params, &tmp);
1176
1177 if (!migrate_params_check(&tmp, errp)) {
476c72aa
PX		 1178 /* Invalid parameter */
1179 return;
1180 }
1181
73af8dd8 1182 migrate_params_apply(params, errp);
476c72aa
PX		 1183}
1184
2594f56d 1185
4886a1bc
DDAG		 1186void qmp_migrate_start_postcopy(Error **errp)
1187{
1188 MigrationState *s = migrate_get_current();
1189
16b0fd32 1190 if (!migrate_postcopy()) {
a54d340b 1191 error_setg(errp, "Enable postcopy with migrate_set_capability before"
4886a1bc
DDAG		 1192 " the start of migration");
1193 return;
1194 }
1195
1196 if (s->state == MIGRATION_STATUS_NONE) {
1197 error_setg(errp, "Postcopy must be started after migration has been"
1198 " started");
1199 return;
1200 }
1201 /*
1202 * we don't error if migration has finished since that would be racy
1203 * with issuing this command.
1204 */
1205 atomic_set(&s->start_postcopy, true);
1206}
1207
065e2813
AL		 1208/* shared migration helpers */
1209
48781e5b 1210void migrate_set_state(int *state, int old_state, int new_state)
51cf4c1a 1211{
a31fedee 1212 assert(new_state < MIGRATION_STATUS__MAX);
48781e5b 1213 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
a31fedee 1214 trace_migrate_set_state(MigrationStatus_str(new_state));
b05dc723 1215 migrate_generate_event(new_state);
51cf4c1a
Z		 1216 }
1217}
1218
4e4a3d3a
PX		 1219static MigrationCapabilityStatusList *migrate_cap_add(
1220 MigrationCapabilityStatusList *list,
1221 MigrationCapability index,
1222 bool state)
2833c59b
JQ		 1223{
1224 MigrationCapabilityStatusList *cap;
1225
1226 cap = g_new0(MigrationCapabilityStatusList, 1);
1227 cap->value = g_new0(MigrationCapabilityStatus, 1);
4e4a3d3a
PX		 1228 cap->value->capability = index;
1229 cap->value->state = state;
1230 cap->next = list;
1231
1232 return cap;
1233}
1234
1235void migrate_set_block_enabled(bool value, Error **errp)
1236{
1237 MigrationCapabilityStatusList *cap;
1238
1239 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
2833c59b
JQ		 1240 qmp_migrate_set_capabilities(cap, errp);
1241 qapi_free_MigrationCapabilityStatusList(cap);
1242}
1243
1244static void migrate_set_block_incremental(MigrationState *s, bool value)
1245{
1246 s->parameters.block_incremental = value;
1247}
1248
1249static void block_cleanup_parameters(MigrationState *s)
1250{
1251 if (s->must_remove_block_options) {
1252 /* setting to false can never fail */
1253 migrate_set_block_enabled(false, &error_abort);
1254 migrate_set_block_incremental(s, false);
1255 s->must_remove_block_options = false;
1256 }
1257}
1258
bb1fadc4 1259static void migrate_fd_cleanup(void *opaque)
065e2813 1260{
bb1fadc4
PB		 1261 MigrationState *s = opaque;
1262
1263 qemu_bh_delete(s->cleanup_bh);
1264 s->cleanup_bh = NULL;
1265
0ceccd85
PX		 1266 qemu_savevm_state_cleanup();
1267
89a02a9f 1268 if (s->to_dst_file) {
f986c3d2 1269 Error *local_err = NULL;
62df066f 1270 QEMUFile *tmp;
f986c3d2 1271
9013dca5 1272 trace_migrate_fd_cleanup();
404a7c05 1273 qemu_mutex_unlock_iothread();
1d34e4bf
DDAG		 1274 if (s->migration_thread_running) {
1275 qemu_thread_join(&s->thread);
1276 s->migration_thread_running = false;
1277 }
404a7c05
PB		 1278 qemu_mutex_lock_iothread();
1279
f986c3d2
JQ		 1280 if (multifd_save_cleanup(&local_err) != 0) {
1281 error_report_err(local_err);
1282 }
62df066f
PX		 1283 qemu_mutex_lock(&s->qemu_file_lock);
1284 tmp = s->to_dst_file;
89a02a9f 1285 s->to_dst_file = NULL;
62df066f
PX		 1286 qemu_mutex_unlock(&s->qemu_file_lock);
1287 /*
1288 * Close the file handle without the lock to make sure the
1289 * critical section won't block for long.
1290 */
1291 qemu_fclose(tmp);
065e2813
AL		 1292 }
1293
9ec055ae
DDAG		 1294 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1295 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
7a2c1721 1296
94f5a437 1297 if (s->state == MIGRATION_STATUS_CANCELLING) {
48781e5b 1298 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
94f5a437 1299 MIGRATION_STATUS_CANCELLED);
7a2c1721 1300 }
a3fa1d78 1301
87db1a7d
JQ		 1302 if (s->error) {
1303 /* It is used on info migrate. We can't free it */
1304 error_report_err(error_copy(s->error));
1305 }
a3fa1d78 1306 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1307 block_cleanup_parameters(s);
065e2813
AL		 1308}
1309
87db1a7d
JQ		 1310void migrate_set_error(MigrationState *s, const Error *error)
1311{
1312 qemu_mutex_lock(&s->error_mutex);
1313 if (!s->error) {
1314 s->error = error_copy(error);
1315 }
1316 qemu_mutex_unlock(&s->error_mutex);
1317}
1318
d59ce6f3 1319void migrate_fd_error(MigrationState *s, const Error *error)
065e2813 1320{
25174055 1321 trace_migrate_fd_error(error_get_pretty(error));
89a02a9f 1322 assert(s->to_dst_file == NULL);
48781e5b
HZ		 1323 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1324 MIGRATION_STATUS_FAILED);
87db1a7d 1325 migrate_set_error(s, error);
458cf28e
JQ		 1326}
1327
0edda1c4 1328static void migrate_fd_cancel(MigrationState *s)
065e2813 1329{
6f2b811a 1330 int old_state ;
89a02a9f 1331 QEMUFile *f = migrate_get_current()->to_dst_file;
9013dca5 1332 trace_migrate_fd_cancel();
065e2813 1333
70b20477
DDAG		 1334 if (s->rp_state.from_dst_file) {
1335 /* shutdown the rp socket, so causing the rp thread to shutdown */
1336 qemu_file_shutdown(s->rp_state.from_dst_file);
1337 }
1338
6f2b811a
Z		 1339 do {
1340 old_state = s->state;
f6844b99 1341 if (!migration_is_setup_or_active(old_state)) {
6f2b811a
Z		 1342 break;
1343 }
a7b36b48
DDAG		 1344 /* If the migration is paused, kick it out of the pause */
1345 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1346 qemu_sem_post(&s->pause_sem);
1347 }
48781e5b 1348 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
31194731 1349 } while (s->state != MIGRATION_STATUS_CANCELLING);
a26ba26e
DDAG		 1350
1351 /*
1352 * If we're unlucky the migration code might be stuck somewhere in a
1353 * send/write while the network has failed and is waiting to timeout;
1354 * if we've got shutdown(2) available then we can force it to quit.
1355 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1356 * called in a bh, so there is no race against this cancel.
1357 */
31194731 1358 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
a26ba26e
DDAG		 1359 qemu_file_shutdown(f);
1360 }
1d2acc31
HZ		 1361 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1362 Error *local_err = NULL;
1363
1364 bdrv_invalidate_cache_all(&local_err);
1365 if (local_err) {
1366 error_report_err(local_err);
1367 } else {
1368 s->block_inactive = false;
1369 }
1370 }
065e2813
AL		 1371}
1372
99a0db9b
GH		 1373void add_migration_state_change_notifier(Notifier *notify)
1374{
1375 notifier_list_add(&migration_state_notifiers, notify);
1376}
1377
1378void remove_migration_state_change_notifier(Notifier *notify)
1379{
31552529 1380 notifier_remove(notify);
99a0db9b
GH		 1381}
1382
02edd2e7 1383bool migration_in_setup(MigrationState *s)
afe2df69 1384{
31194731 1385 return s->state == MIGRATION_STATUS_SETUP;
afe2df69
GH		 1386}
1387
7073693b 1388bool migration_has_finished(MigrationState *s)
99a0db9b 1389{
31194731 1390 return s->state == MIGRATION_STATUS_COMPLETED;
99a0db9b 1391}
0edda1c4 1392
afe2df69
GH		 1393bool migration_has_failed(MigrationState *s)
1394{
31194731
HZ		 1395 return (s->state == MIGRATION_STATUS_CANCELLED ||
1396 s->state == MIGRATION_STATUS_FAILED);
afe2df69
GH		 1397}
1398
5727309d 1399bool migration_in_postcopy(void)
9ec055ae 1400{
5727309d
JQ		 1401 MigrationState *s = migrate_get_current();
1402
9ec055ae
DDAG		 1403 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1404}
1405
b82fc321
DDAG		 1406bool migration_in_postcopy_after_devices(MigrationState *s)
1407{
5727309d 1408 return migration_in_postcopy() && s->postcopy_after_devices;
b82fc321
DDAG		 1409}
1410
fab35005 1411bool migration_is_idle(void)
fe44dc91 1412{
fab35005 1413 MigrationState *s = migrate_get_current();
fe44dc91
AA		 1414
1415 switch (s->state) {
1416 case MIGRATION_STATUS_NONE:
1417 case MIGRATION_STATUS_CANCELLED:
1418 case MIGRATION_STATUS_COMPLETED:
1419 case MIGRATION_STATUS_FAILED:
1420 return true;
1421 case MIGRATION_STATUS_SETUP:
1422 case MIGRATION_STATUS_CANCELLING:
1423 case MIGRATION_STATUS_ACTIVE:
1424 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1425 case MIGRATION_STATUS_COLO:
31e06077
DDAG		 1426 case MIGRATION_STATUS_PRE_SWITCHOVER:
1427 case MIGRATION_STATUS_DEVICE:
fe44dc91
AA		 1428 return false;
1429 case MIGRATION_STATUS__MAX:
1430 g_assert_not_reached();
1431 }
1432
1433 return false;
1434}
1435
3e0c8050 1436void migrate_init(MigrationState *s)
0edda1c4 1437{
389775d1
DDAG		 1438 /*
1439 * Reinitialise all migration state, except
1440 * parameters/capabilities that the user set, and
1441 * locks.
1442 */
1443 s->bytes_xfer = 0;
1444 s->xfer_limit = 0;
1445 s->cleanup_bh = 0;
89a02a9f 1446 s->to_dst_file = NULL;
389775d1 1447 s->state = MIGRATION_STATUS_NONE;
389775d1
DDAG		 1448 s->rp_state.from_dst_file = NULL;
1449 s->rp_state.error = false;
1450 s->mbps = 0.0;
1451 s->downtime = 0;
1452 s->expected_downtime = 0;
389775d1 1453 s->setup_time = 0;
389775d1 1454 s->start_postcopy = false;
b82fc321 1455 s->postcopy_after_devices = false;
389775d1 1456 s->migration_thread_running = false;
d59ce6f3
DB		 1457 error_free(s->error);
1458 s->error = NULL;
389775d1 1459
48781e5b 1460 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
0edda1c4 1461
4af246a3
PX		 1462 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1463 s->total_time = 0;
7287cbd4 1464 s->vm_was_running = false;
b15df1ae
PX		 1465 s->iteration_initial_bytes = 0;
1466 s->threshold_size = 0;
0edda1c4 1467}
cab30143 1468
fa2756b7
AL		 1469static GSList *migration_blockers;
1470
fe44dc91 1471int migrate_add_blocker(Error *reason, Error **errp)
fa2756b7 1472{
3df663e5 1473 if (migrate_get_current()->only_migratable) {
b67b8c3a
AA		 1474 error_propagate(errp, error_copy(reason));
1475 error_prepend(errp, "disallowing migration blocker "
1476 "(--only_migratable) for: ");
1477 return -EACCES;
1478 }
1479
fab35005 1480 if (migration_is_idle()) {
fe44dc91
AA		 1481 migration_blockers = g_slist_prepend(migration_blockers, reason);
1482 return 0;
1483 }
1484
1485 error_propagate(errp, error_copy(reason));
1486 error_prepend(errp, "disallowing migration blocker (migration in "
1487 "progress) for: ");
1488 return -EBUSY;
fa2756b7
AL		 1489}
1490
1491void migrate_del_blocker(Error *reason)
1492{
1493 migration_blockers = g_slist_remove(migration_blockers, reason);
1494}
1495
bf1ae1f4
DDAG		 1496void qmp_migrate_incoming(const char *uri, Error **errp)
1497{
1498 Error *local_err = NULL;
4debb5f5 1499 static bool once = true;
bf1ae1f4
DDAG		 1500
1501 if (!deferred_incoming) {
4debb5f5 1502 error_setg(errp, "For use with '-incoming defer'");
bf1ae1f4
DDAG		 1503 return;
1504 }
4debb5f5
DDAG		 1505 if (!once) {
1506 error_setg(errp, "The incoming migration has already been started");
1507 }
bf1ae1f4
DDAG		 1508
1509 qemu_start_incoming_migration(uri, &local_err);
1510
1511 if (local_err) {
1512 error_propagate(errp, local_err);
1513 return;
1514 }
1515
4debb5f5 1516 once = false;
bf1ae1f4
DDAG		 1517}
1518
02affd41
PX		 1519void qmp_migrate_recover(const char *uri, Error **errp)
1520{
1521 MigrationIncomingState *mis = migration_incoming_get_current();
1522
1523 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1524 error_setg(errp, "Migrate recover can only be run "
1525 "when postcopy is paused.");
1526 return;
1527 }
1528
1529 if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1530 false, true) == true) {
1531 error_setg(errp, "Migrate recovery is triggered already");
1532 return;
1533 }
1534
1535 /*
1536 * Note that this call will never start a real migration; it will
1537 * only re-setup the migration stream and poke existing migration
1538 * to continue using that newly established channel.
1539 */
1540 qemu_start_incoming_migration(uri, errp);
1541}
1542
bfbf89c2
PX		 1543void qmp_migrate_pause(Error **errp)
1544{
1545 MigrationState *ms = migrate_get_current();
1546 MigrationIncomingState *mis = migration_incoming_get_current();
1547 int ret;
1548
1549 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1550 /* Source side, during postcopy */
1551 qemu_mutex_lock(&ms->qemu_file_lock);
1552 ret = qemu_file_shutdown(ms->to_dst_file);
1553 qemu_mutex_unlock(&ms->qemu_file_lock);
1554 if (ret) {
1555 error_setg(errp, "Failed to pause source migration");
1556 }
1557 return;
1558 }
1559
1560 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1561 ret = qemu_file_shutdown(mis->from_src_file);
1562 if (ret) {
1563 error_setg(errp, "Failed to pause destination migration");
1564 }
1565 return;
1566 }
1567
1568 error_setg(errp, "migrate-pause is currently only supported "
1569 "during postcopy-active state");
1570}
1571
24f3902b
GK		 1572bool migration_is_blocked(Error **errp)
1573{
1574 if (qemu_savevm_state_blocked(errp)) {
1575 return true;
1576 }
1577
1578 if (migration_blockers) {
250561e1 1579 error_propagate(errp, error_copy(migration_blockers->data));
24f3902b
GK		 1580 return true;
1581 }
1582
1583 return false;
1584}
1585
d3e35b8f
PX		 1586/* Returns true if continue to migrate, or false if error detected */
1587static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1588 bool resume, Error **errp)
cab30143 1589{
be7059cd 1590 Error *local_err = NULL;
d3e35b8f
PX		 1591
1592 if (resume) {
1593 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1594 error_setg(errp, "Cannot resume if there is no "
1595 "paused migration");
1596 return false;
1597 }
1598 /* This is a resume, skip init status */
1599 return true;
1600 }
cab30143 1601
f6844b99 1602 if (migration_is_setup_or_active(s->state) ||
0b827d5e
HZ		 1603 s->state == MIGRATION_STATUS_CANCELLING ||
1604 s->state == MIGRATION_STATUS_COLO) {
c6bd8c70 1605 error_setg(errp, QERR_MIGRATION_ACTIVE);
d3e35b8f 1606 return false;
cab30143 1607 }
d3e35b8f 1608
ca99993a
DDAG		 1609 if (runstate_check(RUN_STATE_INMIGRATE)) {
1610 error_setg(errp, "Guest is waiting for an incoming migration");
d3e35b8f 1611 return false;
ca99993a
DDAG		 1612 }
1613
24f3902b 1614 if (migration_is_blocked(errp)) {
d3e35b8f 1615 return false;
fa2756b7
AL		 1616 }
1617
d3e35b8f 1618 if (blk || blk_inc) {
2833c59b
JQ		 1619 if (migrate_use_block() || migrate_use_block_incremental()) {
1620 error_setg(errp, "Command options are incompatible with "
1621 "current migration capabilities");
d3e35b8f 1622 return false;
2833c59b
JQ		 1623 }
1624 migrate_set_block_enabled(true, &local_err);
1625 if (local_err) {
1626 error_propagate(errp, local_err);
d3e35b8f 1627 return false;
2833c59b
JQ		 1628 }
1629 s->must_remove_block_options = true;
1630 }
1631
d3e35b8f 1632 if (blk_inc) {
2833c59b
JQ		 1633 migrate_set_block_incremental(s, true);
1634 }
1635
3e0c8050 1636 migrate_init(s);
cab30143 1637
d3e35b8f
PX		 1638 return true;
1639}
1640
1641void qmp_migrate(const char *uri, bool has_blk, bool blk,
1642 bool has_inc, bool inc, bool has_detach, bool detach,
1643 bool has_resume, bool resume, Error **errp)
1644{
1645 Error *local_err = NULL;
1646 MigrationState *s = migrate_get_current();
1647 const char *p;
1648
1649 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1650 has_resume && resume, errp)) {
1651 /* Error detected, put into errp */
1652 return;
1653 }
1654
cab30143 1655 if (strstart(uri, "tcp:", &p)) {
f37afb5a 1656 tcp_start_outgoing_migration(s, p, &local_err);
2da776db 1657#ifdef CONFIG_RDMA
41310c68 1658 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 1659 rdma_start_outgoing_migration(s, p, &local_err);
1660#endif
cab30143 1661 } else if (strstart(uri, "exec:", &p)) {
f37afb5a 1662 exec_start_outgoing_migration(s, p, &local_err);
cab30143 1663 } else if (strstart(uri, "unix:", &p)) {
f37afb5a 1664 unix_start_outgoing_migration(s, p, &local_err);
cab30143 1665 } else if (strstart(uri, "fd:", &p)) {
f37afb5a 1666 fd_start_outgoing_migration(s, p, &local_err);
99a0db9b 1667 } else {
c6bd8c70
MA		 1668 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1669 "a valid migration protocol");
48781e5b
HZ		 1670 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1671 MIGRATION_STATUS_FAILED);
09576e74 1672 block_cleanup_parameters(s);
e1c37d0e 1673 return;
cab30143
JQ		 1674 }
1675
f37afb5a 1676 if (local_err) {
d59ce6f3 1677 migrate_fd_error(s, local_err);
f37afb5a 1678 error_propagate(errp, local_err);
e1c37d0e 1679 return;
1299c631 1680 }
cab30143
JQ		 1681}
1682
6cdedb07 1683void qmp_migrate_cancel(Error **errp)
cab30143 1684{
17549e84 1685 migrate_fd_cancel(migrate_get_current());
cab30143
JQ		 1686}
1687
89cfc02c
DDAG		 1688void qmp_migrate_continue(MigrationStatus state, Error **errp)
1689{
1690 MigrationState *s = migrate_get_current();
1691 if (s->state != state) {
1692 error_setg(errp, "Migration not in expected state: %s",
1693 MigrationStatus_str(s->state));
1694 return;
1695 }
1696 qemu_sem_post(&s->pause_sem);
1697}
1698
9e1ba4cc
OW		 1699void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1700{
73af8dd8
JQ		 1701 MigrateSetParameters p = {
1702 .has_xbzrle_cache_size = true,
1703 .xbzrle_cache_size = value,
1704 };
c91e681a 1705
73af8dd8 1706 qmp_migrate_set_parameters(&p, errp);
9e1ba4cc
OW		 1707}
1708
1709int64_t qmp_query_migrate_cache_size(Error **errp)
1710{
1711 return migrate_xbzrle_cache_size();
1712}
1713
3dc85383 1714void qmp_migrate_set_speed(int64_t value, Error **errp)
cab30143 1715{
1bda8b3c 1716 MigrateSetParameters p = {
2ff30257
AA		 1717 .has_max_bandwidth = true,
1718 .max_bandwidth = value,
1719 };
cab30143 1720
2ff30257 1721 qmp_migrate_set_parameters(&p, errp);
cab30143
JQ		 1722}
1723
4f0a993b 1724void qmp_migrate_set_downtime(double value, Error **errp)
cab30143 1725{
87c9cc1c
DHB		 1726 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1727 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1728 "the range of 0 to %d seconds",
1729 MAX_MIGRATE_DOWNTIME_SECONDS);
1730 return;
1731 }
1732
2ff30257
AA		 1733 value *= 1000; /* Convert to milliseconds */
1734 value = MAX(0, MIN(INT64_MAX, value));
1735
1bda8b3c 1736 MigrateSetParameters p = {
2ff30257
AA		 1737 .has_downtime_limit = true,
1738 .downtime_limit = value,
1739 };
1740
1741 qmp_migrate_set_parameters(&p, errp);
99a0db9b 1742}
17ad9b35 1743
53f09a10
PB		 1744bool migrate_release_ram(void)
1745{
1746 MigrationState *s;
1747
1748 s = migrate_get_current();
1749
1750 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1751}
1752
53dd370c
DDAG		 1753bool migrate_postcopy_ram(void)
1754{
1755 MigrationState *s;
1756
1757 s = migrate_get_current();
1758
32c3db5b 1759 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
53dd370c
DDAG		 1760}
1761
58110f0a
VSO		 1762bool migrate_postcopy(void)
1763{
dd6bb914 1764 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
58110f0a
VSO		 1765}
1766
bde1e2ec
CV		 1767bool migrate_auto_converge(void)
1768{
1769 MigrationState *s;
1770
1771 s = migrate_get_current();
1772
1773 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1774}
1775
323004a3
PL		 1776bool migrate_zero_blocks(void)
1777{
1778 MigrationState *s;
1779
1780 s = migrate_get_current();
1781
1782 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1783}
1784
f22f928e
AP		 1785bool migrate_postcopy_blocktime(void)
1786{
1787 MigrationState *s;
1788
1789 s = migrate_get_current();
1790
1791 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
1792}
1793
8706d2d5
LL		 1794bool migrate_use_compression(void)
1795{
dde4e694
LL		 1796 MigrationState *s;
1797
1798 s = migrate_get_current();
1799
1800 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
8706d2d5
LL		 1801}
1802
1803int migrate_compress_level(void)
1804{
1805 MigrationState *s;
1806
1807 s = migrate_get_current();
1808
2594f56d 1809 return s->parameters.compress_level;
8706d2d5
LL		 1810}
1811
1812int migrate_compress_threads(void)
1813{
1814 MigrationState *s;
1815
1816 s = migrate_get_current();
1817
2594f56d 1818 return s->parameters.compress_threads;
8706d2d5
LL		 1819}
1820
3fcb38c2
LL		 1821int migrate_decompress_threads(void)
1822{
1823 MigrationState *s;
1824
1825 s = migrate_get_current();
1826
2594f56d 1827 return s->parameters.decompress_threads;
3fcb38c2
LL		 1828}
1829
55efc8c2
VSO		 1830bool migrate_dirty_bitmaps(void)
1831{
1832 MigrationState *s;
1833
1834 s = migrate_get_current();
1835
1836 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
1837}
1838
b05dc723
JQ		 1839bool migrate_use_events(void)
1840{
1841 MigrationState *s;
1842
1843 s = migrate_get_current();
1844
1845 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1846}
1847
30126bbf
JQ		 1848bool migrate_use_multifd(void)
1849{
1850 MigrationState *s;
1851
1852 s = migrate_get_current();
1853
1854 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1855}
1856
93fbd031
DDAG		 1857bool migrate_pause_before_switchover(void)
1858{
1859 MigrationState *s;
1860
1861 s = migrate_get_current();
1862
1863 return s->enabled_capabilities[
1864 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1865}
1866
4075fb1c
JQ		 1867int migrate_multifd_channels(void)
1868{
1869 MigrationState *s;
1870
1871 s = migrate_get_current();
1872
1873 return s->parameters.x_multifd_channels;
1874}
1875
0fb86605
JQ		 1876int migrate_multifd_page_count(void)
1877{
1878 MigrationState *s;
1879
1880 s = migrate_get_current();
1881
1882 return s->parameters.x_multifd_page_count;
1883}
1884
17ad9b35
OW		 1885int migrate_use_xbzrle(void)
1886{
1887 MigrationState *s;
1888
1889 s = migrate_get_current();
1890
1891 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1892}
1893
1894int64_t migrate_xbzrle_cache_size(void)
1895{
1896 MigrationState *s;
1897
1898 s = migrate_get_current();
1899
73af8dd8 1900 return s->parameters.xbzrle_cache_size;
17ad9b35 1901}
0d82d0e8 1902
7e555c6c
DDAG		 1903static int64_t migrate_max_postcopy_bandwidth(void)
1904{
1905 MigrationState *s;
1906
1907 s = migrate_get_current();
1908
1909 return s->parameters.max_postcopy_bandwidth;
1910}
1911
1912
2833c59b
JQ		 1913bool migrate_use_block(void)
1914{
1915 MigrationState *s;
1916
1917 s = migrate_get_current();
1918
1919 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1920}
1921
c788ada8
PX		 1922bool migrate_use_return_path(void)
1923{
1924 MigrationState *s;
1925
1926 s = migrate_get_current();
1927
1928 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1929}
1930
2833c59b
JQ		 1931bool migrate_use_block_incremental(void)
1932{
1933 MigrationState *s;
1934
1935 s = migrate_get_current();
1936
1937 return s->parameters.block_incremental;
1938}
1939
70b20477
DDAG		 1940/* migration thread support */
1941/*
1942 * Something bad happened to the RP stream, mark an error
1943 * The caller shall print or trace something to indicate why
1944 */
1945static void mark_source_rp_bad(MigrationState *s)
1946{
1947 s->rp_state.error = true;
1948}
1949
1950static struct rp_cmd_args {
1951 ssize_t len; /* -1 = variable */
1952 const char *name;
1953} rp_cmd_args[] = {
1954 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1955 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1956 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1e2d90eb
DDAG		 1957 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1958 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
a335debb 1959 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
13955b89 1960 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
70b20477
DDAG		 1961 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1962};
1963
1e2d90eb
DDAG		 1964/*
1965 * Process a request for pages received on the return path,
1966 * We're allowed to send more than requested (e.g. to round to our page size)
1967 * and we don't need to send pages that have already been sent.
1968 */
1969static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1970 ram_addr_t start, size_t len)
1971{
6c595cde
DDAG		 1972 long our_host_ps = getpagesize();
1973
1e2d90eb 1974 trace_migrate_handle_rp_req_pages(rbname, start, len);
6c595cde
DDAG		 1975
1976 /*
1977 * Since we currently insist on matching page sizes, just sanity check
1978 * we're being asked for whole host pages.
1979 */
1980 if (start & (our_host_ps-1) ||
1981 (len & (our_host_ps-1))) {
1982 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1983 " len: %zd", __func__, start, len);
1984 mark_source_rp_bad(ms);
1985 return;
1986 }
1987
96506894 1988 if (ram_save_queue_pages(rbname, start, len)) {
6c595cde
DDAG		 1989 mark_source_rp_bad(ms);
1990 }
1e2d90eb
DDAG		 1991}
1992
14b1742e
PX		 1993/* Return true to retry, false to quit */
1994static bool postcopy_pause_return_path_thread(MigrationState *s)
1995{
1996 trace_postcopy_pause_return_path();
1997
1998 qemu_sem_wait(&s->postcopy_pause_rp_sem);
1999
2000 trace_postcopy_pause_return_path_continued();
2001
2002 return true;
2003}
2004
a335debb
PX		 2005static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2006{
2007 RAMBlock *block = qemu_ram_block_by_name(block_name);
2008
2009 if (!block) {
2010 error_report("%s: invalid block name '%s'", __func__, block_name);
2011 return -EINVAL;
2012 }
2013
2014 /* Fetch the received bitmap and refresh the dirty bitmap */
2015 return ram_dirty_bitmap_reload(s, block);
2016}
2017
13955b89
PX		 2018static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2019{
2020 trace_source_return_path_thread_resume_ack(value);
2021
2022 if (value != MIGRATION_RESUME_ACK_VALUE) {
2023 error_report("%s: illegal resume_ack value %"PRIu32,
2024 __func__, value);
2025 return -1;
2026 }
2027
2028 /* Now both sides are active. */
2029 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2030 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2031
94190696
PX		 2032 /* Notify send thread that time to continue send pages */
2033 qemu_sem_post(&s->rp_state.rp_sem);
13955b89
PX		 2034
2035 return 0;
2036}
2037
70b20477
DDAG		 2038/*
2039 * Handles messages sent on the return path towards the source VM
2040 *
2041 */
2042static void *source_return_path_thread(void *opaque)
2043{
2044 MigrationState *ms = opaque;
2045 QEMUFile *rp = ms->rp_state.from_dst_file;
2046 uint16_t header_len, header_type;
568b01ca 2047 uint8_t buf[512];
70b20477 2048 uint32_t tmp32, sibling_error;
1e2d90eb
DDAG		 2049 ram_addr_t start = 0; /* =0 to silence warning */
2050 size_t len = 0, expected_len;
70b20477
DDAG		 2051 int res;
2052
2053 trace_source_return_path_thread_entry();
14b1742e
PX		 2054
2055retry:
70b20477
DDAG		 2056 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2057 migration_is_setup_or_active(ms->state)) {
2058 trace_source_return_path_thread_loop_top();
2059 header_type = qemu_get_be16(rp);
2060 header_len = qemu_get_be16(rp);
2061
7a9ddfbf
PX		 2062 if (qemu_file_get_error(rp)) {
2063 mark_source_rp_bad(ms);
2064 goto out;
2065 }
2066
70b20477
DDAG		 2067 if (header_type >= MIG_RP_MSG_MAX ||
2068 header_type == MIG_RP_MSG_INVALID) {
2069 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2070 header_type, header_len);
2071 mark_source_rp_bad(ms);
2072 goto out;
2073 }
2074
2075 if ((rp_cmd_args[header_type].len != -1 &&
2076 header_len != rp_cmd_args[header_type].len) ||
568b01ca 2077 header_len > sizeof(buf)) {
70b20477
DDAG		 2078 error_report("RP: Received '%s' message (0x%04x) with"
2079 "incorrect length %d expecting %zu",
2080 rp_cmd_args[header_type].name, header_type, header_len,
2081 (size_t)rp_cmd_args[header_type].len);
2082 mark_source_rp_bad(ms);
2083 goto out;
2084 }
2085
2086 /* We know we've got a valid header by this point */
2087 res = qemu_get_buffer(rp, buf, header_len);
2088 if (res != header_len) {
2089 error_report("RP: Failed reading data for message 0x%04x"
2090 " read %d expected %d",
2091 header_type, res, header_len);
2092 mark_source_rp_bad(ms);
2093 goto out;
2094 }
2095
2096 /* OK, we have the message and the data */
2097 switch (header_type) {
2098 case MIG_RP_MSG_SHUT:
4d885131 2099 sibling_error = ldl_be_p(buf);
70b20477
DDAG		 2100 trace_source_return_path_thread_shut(sibling_error);
2101 if (sibling_error) {
2102 error_report("RP: Sibling indicated error %d", sibling_error);
2103 mark_source_rp_bad(ms);
2104 }
2105 /*
2106 * We'll let the main thread deal with closing the RP
2107 * we could do a shutdown(2) on it, but we're the only user
2108 * anyway, so there's nothing gained.
2109 */
2110 goto out;
2111
2112 case MIG_RP_MSG_PONG:
4d885131 2113 tmp32 = ldl_be_p(buf);
70b20477
DDAG		 2114 trace_source_return_path_thread_pong(tmp32);
2115 break;
2116
1e2d90eb 2117 case MIG_RP_MSG_REQ_PAGES:
4d885131
PM		 2118 start = ldq_be_p(buf);
2119 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 2120 migrate_handle_rp_req_pages(ms, NULL, start, len);
2121 break;
2122
2123 case MIG_RP_MSG_REQ_PAGES_ID:
2124 expected_len = 12 + 1; /* header + termination */
2125
2126 if (header_len >= expected_len) {
4d885131
PM		 2127 start = ldq_be_p(buf);
2128 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 2129 /* Now we expect an idstr */
2130 tmp32 = buf[12]; /* Length of the following idstr */
2131 buf[13 + tmp32] = '\0';
2132 expected_len += tmp32;
2133 }
2134 if (header_len != expected_len) {
2135 error_report("RP: Req_Page_id with length %d expecting %zd",
2136 header_len, expected_len);
2137 mark_source_rp_bad(ms);
2138 goto out;
2139 }
2140 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2141 break;
2142
a335debb
PX		 2143 case MIG_RP_MSG_RECV_BITMAP:
2144 if (header_len < 1) {
2145 error_report("%s: missing block name", __func__);
2146 mark_source_rp_bad(ms);
2147 goto out;
2148 }
2149 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2150 buf[buf[0] + 1] = '\0';
2151 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2152 mark_source_rp_bad(ms);
2153 goto out;
2154 }
2155 break;
2156
13955b89
PX		 2157 case MIG_RP_MSG_RESUME_ACK:
2158 tmp32 = ldl_be_p(buf);
2159 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2160 mark_source_rp_bad(ms);
2161 goto out;
2162 }
2163 break;
2164
70b20477
DDAG		 2165 default:
2166 break;
2167 }
2168 }
14b1742e
PX		 2169
2170out:
2171 res = qemu_file_get_error(rp);
2172 if (res) {
2173 if (res == -EIO) {
2174 /*
2175 * Maybe there is something we can do: it looks like a
2176 * network down issue, and we pause for a recovery.
2177 */
2178 if (postcopy_pause_return_path_thread(ms)) {
2179 /* Reload rp, reset the rest */
2180 rp = ms->rp_state.from_dst_file;
2181 ms->rp_state.error = false;
2182 goto retry;
2183 }
2184 }
2185
70b20477
DDAG		 2186 trace_source_return_path_thread_bad_end();
2187 mark_source_rp_bad(ms);
2188 }
2189
2190 trace_source_return_path_thread_end();
70b20477
DDAG		 2191 ms->rp_state.from_dst_file = NULL;
2192 qemu_fclose(rp);
2193 return NULL;
2194}
2195
d3e35b8f
PX		 2196static int open_return_path_on_source(MigrationState *ms,
2197 bool create_thread)
70b20477
DDAG		 2198{
2199
89a02a9f 2200 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
70b20477
DDAG		 2201 if (!ms->rp_state.from_dst_file) {
2202 return -1;
2203 }
2204
2205 trace_open_return_path_on_source();
d3e35b8f
PX		 2206
2207 if (!create_thread) {
2208 /* We're done */
2209 return 0;
2210 }
2211
70b20477
DDAG		 2212 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2213 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2214
2215 trace_open_return_path_on_source_continue();
2216
2217 return 0;
2218}
2219
70b20477
DDAG		 2220/* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2221static int await_return_path_close_on_source(MigrationState *ms)
2222{
2223 /*
2224 * If this is a normal exit then the destination will send a SHUT and the
2225 * rp_thread will exit, however if there's an error we need to cause
2226 * it to exit.
2227 */
89a02a9f 2228 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
70b20477
DDAG		 2229 /*
2230 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2231 * waiting for the destination.
2232 */
2233 qemu_file_shutdown(ms->rp_state.from_dst_file);
2234 mark_source_rp_bad(ms);
2235 }
2236 trace_await_return_path_close_on_source_joining();
2237 qemu_thread_join(&ms->rp_state.rp_thread);
2238 trace_await_return_path_close_on_source_close();
2239 return ms->rp_state.error;
2240}
2241
1d34e4bf
DDAG		 2242/*
2243 * Switch from normal iteration to postcopy
2244 * Returns non-0 on error
2245 */
7287cbd4 2246static int postcopy_start(MigrationState *ms)
1d34e4bf
DDAG		 2247{
2248 int ret;
61b67d47
DB		 2249 QIOChannelBuffer *bioc;
2250 QEMUFile *fb;
1d34e4bf 2251 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
7e555c6c 2252 int64_t bandwidth = migrate_max_postcopy_bandwidth();
ef8d6488 2253 bool restart_block = false;
0331c8ca
DDAG		 2254 int cur_state = MIGRATION_STATUS_ACTIVE;
2255 if (!migrate_pause_before_switchover()) {
2256 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2257 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2258 }
1d34e4bf
DDAG		 2259
2260 trace_postcopy_start();
2261 qemu_mutex_lock_iothread();
2262 trace_postcopy_start_set_run();
2263
2264 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1d34e4bf
DDAG		 2265 global_state_store();
2266 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
76b1c7fe
KW		 2267 if (ret < 0) {
2268 goto fail;
2269 }
1d34e4bf 2270
0331c8ca
DDAG		 2271 ret = migration_maybe_pause(ms, &cur_state,
2272 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2273 if (ret < 0) {
2274 goto fail;
2275 }
2276
76b1c7fe 2277 ret = bdrv_inactivate_all();
1d34e4bf
DDAG		 2278 if (ret < 0) {
2279 goto fail;
2280 }
ef8d6488 2281 restart_block = true;
1d34e4bf 2282
1c0d249d
DDAG		 2283 /*
2284 * Cause any non-postcopiable, but iterative devices to
2285 * send out their final data.
2286 */
a1fbe750 2287 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1c0d249d 2288
1d34e4bf
DDAG		 2289 /*
2290 * in Finish migrate and with the io-lock held everything should
2291 * be quiet, but we've potentially still got dirty pages and we
2292 * need to tell the destination to throw any pages it's already received
2293 * that are dirty
2294 */
58110f0a
VSO		 2295 if (migrate_postcopy_ram()) {
2296 if (ram_postcopy_send_discard_bitmap(ms)) {
2297 error_report("postcopy send discard bitmap failed");
2298 goto fail;
2299 }
1d34e4bf
DDAG		 2300 }
2301
2302 /*
2303 * send rest of state - note things that are doing postcopy
2304 * will notice we're in POSTCOPY_ACTIVE and not actually
2305 * wrap their state up here
2306 */
7e555c6c
DDAG		 2307 /* 0 max-postcopy-bandwidth means unlimited */
2308 if (!bandwidth) {
2309 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2310 } else {
2311 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2312 }
58110f0a
VSO		 2313 if (migrate_postcopy_ram()) {
2314 /* Ping just for debugging, helps line traces up */
2315 qemu_savevm_send_ping(ms->to_dst_file, 2);
2316 }
1d34e4bf
DDAG		 2317
2318 /*
2319 * While loading the device state we may trigger page transfer
2320 * requests and the fd must be free to process those, and thus
2321 * the destination must read the whole device state off the fd before
2322 * it starts processing it. Unfortunately the ad-hoc migration format
2323 * doesn't allow the destination to know the size to read without fully
2324 * parsing it through each devices load-state code (especially the open
2325 * coded devices that use get/put).
2326 * So we wrap the device state up in a package with a length at the start;
2327 * to do this we use a qemu_buf to hold the whole of the device state.
2328 */
61b67d47 2329 bioc = qio_channel_buffer_new(4096);
6f01f136 2330 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
61b67d47
DB		 2331 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2332 object_unref(OBJECT(bioc));
1d34e4bf 2333
c76201ab
DDAG		 2334 /*
2335 * Make sure the receiver can get incoming pages before we send the rest
2336 * of the state
2337 */
2338 qemu_savevm_send_postcopy_listen(fb);
2339
a1fbe750 2340 qemu_savevm_state_complete_precopy(fb, false, false);
58110f0a
VSO		 2341 if (migrate_postcopy_ram()) {
2342 qemu_savevm_send_ping(fb, 3);
2343 }
1d34e4bf
DDAG		 2344
2345 qemu_savevm_send_postcopy_run(fb);
2346
2347 /* <><> end of stuff going into the package */
1d34e4bf 2348
ef8d6488
DDAG		 2349 /* Last point of recovery; as soon as we send the package the destination
2350 * can open devices and potentially start running.
2351 * Lets just check again we've not got any errors.
2352 */
2353 ret = qemu_file_get_error(ms->to_dst_file);
2354 if (ret) {
2355 error_report("postcopy_start: Migration stream errored (pre package)");
2356 goto fail_closefb;
2357 }
2358
2359 restart_block = false;
2360
1d34e4bf 2361 /* Now send that blob */
61b67d47 2362 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1d34e4bf
DDAG		 2363 goto fail_closefb;
2364 }
2365 qemu_fclose(fb);
b82fc321
DDAG		 2366
2367 /* Send a notify to give a chance for anything that needs to happen
2368 * at the transition to postcopy and after the device state; in particular
2369 * spice needs to trigger a transition now
2370 */
2371 ms->postcopy_after_devices = true;
2372 notifier_list_notify(&migration_state_notifiers, ms);
2373
1d34e4bf
DDAG		 2374 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2375
2376 qemu_mutex_unlock_iothread();
2377
58110f0a
VSO		 2378 if (migrate_postcopy_ram()) {
2379 /*
2380 * Although this ping is just for debug, it could potentially be
2381 * used for getting a better measurement of downtime at the source.
2382 */
2383 qemu_savevm_send_ping(ms->to_dst_file, 4);
2384 }
1d34e4bf 2385
ced1c616
PB		 2386 if (migrate_release_ram()) {
2387 ram_postcopy_migrated_memory_release(ms);
2388 }
2389
89a02a9f 2390 ret = qemu_file_get_error(ms->to_dst_file);
1d34e4bf
DDAG		 2391 if (ret) {
2392 error_report("postcopy_start: Migration stream errored");
48781e5b 2393 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf
DDAG		 2394 MIGRATION_STATUS_FAILED);
2395 }
2396
2397 return ret;
2398
2399fail_closefb:
2400 qemu_fclose(fb);
2401fail:
48781e5b 2402 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf 2403 MIGRATION_STATUS_FAILED);
ef8d6488
DDAG		 2404 if (restart_block) {
2405 /* A failure happened early enough that we know the destination hasn't
2406 * accessed block devices, so we're safe to recover.
2407 */
2408 Error *local_err = NULL;
2409
2410 bdrv_invalidate_cache_all(&local_err);
2411 if (local_err) {
2412 error_report_err(local_err);
2413 }
2414 }
1d34e4bf
DDAG		 2415 qemu_mutex_unlock_iothread();
2416 return -1;
2417}
2418
e91d8951
DDAG		 2419/**
2420 * migration_maybe_pause: Pause if required to by
2421 * migrate_pause_before_switchover called with the iothread locked
2422 * Returns: 0 on success
2423 */
0331c8ca
DDAG		 2424static int migration_maybe_pause(MigrationState *s,
2425 int *current_active_state,
2426 int new_state)
e91d8951
DDAG		 2427{
2428 if (!migrate_pause_before_switchover()) {
2429 return 0;
2430 }
2431
2432 /* Since leaving this state is not atomic with posting the semaphore
2433 * it's possible that someone could have issued multiple migrate_continue
2434 * and the semaphore is incorrectly positive at this point;
2435 * the docs say it's undefined to reinit a semaphore that's already
2436 * init'd, so use timedwait to eat up any existing posts.
2437 */
2438 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2439 /* This block intentionally left blank */
2440 }
2441
2442 qemu_mutex_unlock_iothread();
2443 migrate_set_state(&s->state, *current_active_state,
2444 MIGRATION_STATUS_PRE_SWITCHOVER);
2445 qemu_sem_wait(&s->pause_sem);
2446 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
0331c8ca
DDAG		 2447 new_state);
2448 *current_active_state = new_state;
e91d8951
DDAG		 2449 qemu_mutex_lock_iothread();
2450
0331c8ca 2451 return s->state == new_state ? 0 : -EINVAL;
e91d8951
DDAG		 2452}
2453
09f6c85e
DDAG		 2454/**
2455 * migration_completion: Used by migration_thread when there's not much left.
2456 * The caller 'breaks' the loop when this returns.
2457 *
2458 * @s: Current migration state
09f6c85e 2459 */
2ad87305 2460static void migration_completion(MigrationState *s)
09f6c85e
DDAG		 2461{
2462 int ret;
2ad87305 2463 int current_active_state = s->state;
09f6c85e 2464
b10ac0c4
DDAG		 2465 if (s->state == MIGRATION_STATUS_ACTIVE) {
2466 qemu_mutex_lock_iothread();
64909f97 2467 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
b10ac0c4 2468 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
7287cbd4 2469 s->vm_was_running = runstate_is_running();
b10ac0c4
DDAG		 2470 ret = global_state_store();
2471
2472 if (!ret) {
a1fbe750 2473 bool inactivate = !migrate_colo_enabled();
b10ac0c4 2474 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
e91d8951 2475 if (ret >= 0) {
0331c8ca
DDAG		 2476 ret = migration_maybe_pause(s, &current_active_state,
2477 MIGRATION_STATUS_DEVICE);
e91d8951 2478 }
f07fa4cb
KW		 2479 if (ret >= 0) {
2480 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
a1fbe750
FZ		 2481 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2482 inactivate);
f07fa4cb 2483 }
a1fbe750
FZ		 2484 if (inactivate && ret >= 0) {
2485 s->block_inactive = true;
b10ac0c4
DDAG		 2486 }
2487 }
2488 qemu_mutex_unlock_iothread();
09f6c85e 2489
b10ac0c4
DDAG		 2490 if (ret < 0) {
2491 goto fail;
09f6c85e 2492 }
b10ac0c4
DDAG		 2493 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2494 trace_migration_completion_postcopy_end();
2495
89a02a9f 2496 qemu_savevm_state_complete_postcopy(s->to_dst_file);
b10ac0c4 2497 trace_migration_completion_postcopy_end_after_complete();
09f6c85e 2498 }
09f6c85e 2499
b10ac0c4
DDAG		 2500 /*
2501 * If rp was opened we must clean up the thread before
2502 * cleaning everything else up (since if there are no failures
2503 * it will wait for the destination to send it's status in
2504 * a SHUT command).
b10ac0c4 2505 */
0425dc97 2506 if (s->rp_state.from_dst_file) {
b10ac0c4 2507 int rp_error;
0425dc97 2508 trace_migration_return_path_end_before();
b10ac0c4 2509 rp_error = await_return_path_close_on_source(s);
0425dc97 2510 trace_migration_return_path_end_after(rp_error);
b10ac0c4 2511 if (rp_error) {
fe904ea8 2512 goto fail_invalidate;
b10ac0c4 2513 }
09f6c85e
DDAG		 2514 }
2515
89a02a9f 2516 if (qemu_file_get_error(s->to_dst_file)) {
09f6c85e 2517 trace_migration_completion_file_err();
fe904ea8 2518 goto fail_invalidate;
09f6c85e
DDAG		 2519 }
2520
0b827d5e
HZ		 2521 if (!migrate_colo_enabled()) {
2522 migrate_set_state(&s->state, current_active_state,
2523 MIGRATION_STATUS_COMPLETED);
2524 }
2525
09f6c85e
DDAG		 2526 return;
2527
fe904ea8
GK		 2528fail_invalidate:
2529 /* If not doing postcopy, vm_start() will be called: let's regain
2530 * control on images.
2531 */
6039dd5b
DDAG		 2532 if (s->state == MIGRATION_STATUS_ACTIVE ||
2533 s->state == MIGRATION_STATUS_DEVICE) {
fe904ea8
GK		 2534 Error *local_err = NULL;
2535
1d2acc31 2536 qemu_mutex_lock_iothread();
fe904ea8
GK		 2537 bdrv_invalidate_cache_all(&local_err);
2538 if (local_err) {
2539 error_report_err(local_err);
1d2acc31
HZ		 2540 } else {
2541 s->block_inactive = false;
fe904ea8 2542 }
1d2acc31 2543 qemu_mutex_unlock_iothread();
fe904ea8
GK		 2544 }
2545
09f6c85e 2546fail:
48781e5b
HZ		 2547 migrate_set_state(&s->state, current_active_state,
2548 MIGRATION_STATUS_FAILED);
09f6c85e
DDAG		 2549}
2550
35a6ed4f
HZ		 2551bool migrate_colo_enabled(void)
2552{
2553 MigrationState *s = migrate_get_current();
2554 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2555}
2556
b23c2ade
PX		 2557typedef enum MigThrError {
2558 /* No error detected */
2559 MIG_THR_ERR_NONE = 0,
2560 /* Detected error, but resumed successfully */
2561 MIG_THR_ERR_RECOVERED = 1,
2562 /* Detected fatal error, need to exit */
2563 MIG_THR_ERR_FATAL = 2,
2564} MigThrError;
2565
94190696
PX		 2566static int postcopy_resume_handshake(MigrationState *s)
2567{
2568 qemu_savevm_send_postcopy_resume(s->to_dst_file);
2569
2570 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2571 qemu_sem_wait(&s->rp_state.rp_sem);
2572 }
2573
2574 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2575 return 0;
2576 }
2577
2578 return -1;
2579}
2580
135b87b4
PX		 2581/* Return zero if success, or <0 for error */
2582static int postcopy_do_resume(MigrationState *s)
2583{
d1b8eadb
PX		 2584 int ret;
2585
2586 /*
2587 * Call all the resume_prepare() hooks, so that modules can be
2588 * ready for the migration resume.
2589 */
2590 ret = qemu_savevm_state_resume_prepare(s);
2591 if (ret) {
2592 error_report("%s: resume_prepare() failure detected: %d",
2593 __func__, ret);
2594 return ret;
2595 }
2596
2597 /*
94190696
PX		 2598 * Last handshake with destination on the resume (destination will
2599 * switch to postcopy-active afterwards)
d1b8eadb 2600 */
94190696
PX		 2601 ret = postcopy_resume_handshake(s);
2602 if (ret) {
2603 error_report("%s: handshake failed: %d", __func__, ret);
2604 return ret;
2605 }
d1b8eadb 2606
135b87b4
PX		 2607 return 0;
2608}
2609
b23c2ade
PX		 2610/*
2611 * We don't return until we are in a safe state to continue current
2612 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2613 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2614 */
2615static MigThrError postcopy_pause(MigrationState *s)
2616{
2617 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
b23c2ade 2618
135b87b4 2619 while (true) {
62df066f
PX		 2620 QEMUFile *file;
2621
135b87b4
PX		 2622 migrate_set_state(&s->state, s->state,
2623 MIGRATION_STATUS_POSTCOPY_PAUSED);
b23c2ade 2624
135b87b4
PX		 2625 /* Current channel is possibly broken. Release it. */
2626 assert(s->to_dst_file);
62df066f
PX		 2627 qemu_mutex_lock(&s->qemu_file_lock);
2628 file = s->to_dst_file;
135b87b4 2629 s->to_dst_file = NULL;
62df066f
PX		 2630 qemu_mutex_unlock(&s->qemu_file_lock);
2631
2632 qemu_file_shutdown(file);
2633 qemu_fclose(file);
b23c2ade 2634
135b87b4
PX		 2635 error_report("Detected IO failure for postcopy. "
2636 "Migration paused.");
b23c2ade 2637
135b87b4
PX		 2638 /*
2639 * We wait until things fixed up. Then someone will setup the
2640 * status back for us.
2641 */
2642 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2643 qemu_sem_wait(&s->postcopy_pause_sem);
2644 }
2645
2646 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2647 /* Woken up by a recover procedure. Give it a shot */
b23c2ade 2648
135b87b4
PX		 2649 /*
2650 * Firstly, let's wake up the return path now, with a new
2651 * return path channel.
2652 */
2653 qemu_sem_post(&s->postcopy_pause_rp_sem);
2654
2655 /* Do the resume logic */
2656 if (postcopy_do_resume(s) == 0) {
2657 /* Let's continue! */
2658 trace_postcopy_pause_continued();
2659 return MIG_THR_ERR_RECOVERED;
2660 } else {
2661 /*
2662 * Something wrong happened during the recovery, let's
2663 * pause again. Pause is always better than throwing
2664 * data away.
2665 */
2666 continue;
2667 }
2668 } else {
2669 /* This is not right... Time to quit. */
2670 return MIG_THR_ERR_FATAL;
2671 }
2672 }
b23c2ade
PX		 2673}
2674
2675static MigThrError migration_detect_error(MigrationState *s)
2676{
2677 int ret;
2678
2679 /* Try to detect any file errors */
2680 ret = qemu_file_get_error(s->to_dst_file);
2681
2682 if (!ret) {
2683 /* Everything is fine */
2684 return MIG_THR_ERR_NONE;
2685 }
2686
2687 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
2688 /*
2689 * For postcopy, we allow the network to be down for a
2690 * while. After that, it can be continued by a
2691 * recovery phase.
2692 */
2693 return postcopy_pause(s);
2694 } else {
2695 /*
2696 * For precopy (or postcopy with error outside IO), we fail
2697 * with no time.
2698 */
2699 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
2700 trace_migration_thread_file_err();
2701
2702 /* Time to stop the migration, now. */
2703 return MIG_THR_ERR_FATAL;
2704 }
2705}
2706
0c8f0efd
JQ		 2707/* How many bytes have we transferred since the beggining of the migration */
2708static uint64_t migration_total_bytes(MigrationState *s)
2709{
2710 return qemu_ftell(s->to_dst_file);
2711}
2712
cf011f08
PX		 2713static void migration_calculate_complete(MigrationState *s)
2714{
0c8f0efd 2715 uint64_t bytes = migration_total_bytes(s);
cf011f08 2716 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
6cde6fbe 2717 int64_t transfer_time;
cf011f08
PX		 2718
2719 s->total_time = end_time - s->start_time;
2720 if (!s->downtime) {
2721 /*
2722 * It's still not set, so we are precopy migration. For
2723 * postcopy, downtime is calculated during postcopy_start().
2724 */
2725 s->downtime = end_time - s->downtime_start;
2726 }
2727
6cde6fbe
JQ		 2728 transfer_time = s->total_time - s->setup_time;
2729 if (transfer_time) {
2730 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
cf011f08
PX		 2731 }
2732}
2733
b15df1ae
PX		 2734static void migration_update_counters(MigrationState *s,
2735 int64_t current_time)
2736{
2737 uint64_t transferred, time_spent;
0c8f0efd 2738 uint64_t current_bytes; /* bytes transferred since the beginning */
b15df1ae
PX		 2739 double bandwidth;
2740
2741 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2742 return;
2743 }
2744
0c8f0efd
JQ		 2745 current_bytes = migration_total_bytes(s);
2746 transferred = current_bytes - s->iteration_initial_bytes;
b15df1ae
PX		 2747 time_spent = current_time - s->iteration_start_time;
2748 bandwidth = (double)transferred / time_spent;
0781c1ed 2749 s->threshold_size = bandwidth * s->parameters.downtime_limit;
b15df1ae
PX		 2750
2751 s->mbps = (((double) transferred * 8.0) /
2752 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2753
2754 /*
2755 * if we haven't sent anything, we don't want to
2756 * recalculate. 10000 is a small enough number for our purposes
2757 */
2758 if (ram_counters.dirty_pages_rate && transferred > 10000) {
650af890 2759 s->expected_downtime = ram_counters.remaining / bandwidth;
b15df1ae
PX		 2760 }
2761
2762 qemu_file_reset_rate_limit(s->to_dst_file);
2763
2764 s->iteration_start_time = current_time;
0c8f0efd 2765 s->iteration_initial_bytes = current_bytes;
b15df1ae
PX		 2766
2767 trace_migrate_transferred(transferred, time_spent,
0781c1ed 2768 bandwidth, s->threshold_size);
b15df1ae
PX		 2769}
2770
2ad87305
PX		 2771/* Migration thread iteration status */
2772typedef enum {
2773 MIG_ITERATE_RESUME, /* Resume current iteration */
2774 MIG_ITERATE_SKIP, /* Skip current iteration */
2775 MIG_ITERATE_BREAK, /* Break the loop */
2776} MigIterateState;
2777
2778/*
2779 * Return true if continue to the next iteration directly, false
2780 * otherwise.
2781 */
2782static MigIterateState migration_iteration_run(MigrationState *s)
2783{
47995026 2784 uint64_t pending_size, pend_pre, pend_compat, pend_post;
2ad87305
PX		 2785 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2786
47995026
VSO		 2787 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
2788 &pend_compat, &pend_post);
2789 pending_size = pend_pre + pend_compat + pend_post;
2ad87305
PX		 2790
2791 trace_migrate_pending(pending_size, s->threshold_size,
47995026 2792 pend_pre, pend_compat, pend_post);
2ad87305
PX		 2793
2794 if (pending_size && pending_size >= s->threshold_size) {
2795 /* Still a significant amount to transfer */
2796 if (migrate_postcopy() && !in_postcopy &&
47995026 2797 pend_pre <= s->threshold_size &&
2ad87305
PX		 2798 atomic_read(&s->start_postcopy)) {
2799 if (postcopy_start(s)) {
2800 error_report("%s: postcopy failed to start", __func__);
2801 }
2802 return MIG_ITERATE_SKIP;
2803 }
2804 /* Just another iteration step */
2805 qemu_savevm_state_iterate(s->to_dst_file,
2806 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2807 } else {
2808 trace_migration_thread_low_pending(pending_size);
2809 migration_completion(s);
2810 return MIG_ITERATE_BREAK;
2811 }
2812
2813 return MIG_ITERATE_RESUME;
2814}
2815
199aa6d4
PX		 2816static void migration_iteration_finish(MigrationState *s)
2817{
2818 /* If we enabled cpu throttling for auto-converge, turn it off. */
2819 cpu_throttle_stop();
2820
2821 qemu_mutex_lock_iothread();
2822 switch (s->state) {
2823 case MIGRATION_STATUS_COMPLETED:
2824 migration_calculate_complete(s);
2825 runstate_set(RUN_STATE_POSTMIGRATE);
2826 break;
2827
2828 case MIGRATION_STATUS_ACTIVE:
2829 /*
2830 * We should really assert here, but since it's during
2831 * migration, let's try to reduce the usage of assertions.
2832 */
2833 if (!migrate_colo_enabled()) {
2834 error_report("%s: critical error: calling COLO code without "
2835 "COLO enabled", __func__);
2836 }
2837 migrate_start_colo_process(s);
2838 /*
2839 * Fixme: we will run VM in COLO no matter its old running state.
2840 * After exited COLO, we will keep running.
2841 */
2842 s->vm_was_running = true;
2843 /* Fallthrough */
2844 case MIGRATION_STATUS_FAILED:
2845 case MIGRATION_STATUS_CANCELLED:
2846 if (s->vm_was_running) {
2847 vm_start();
2848 } else {
2849 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2850 runstate_set(RUN_STATE_POSTMIGRATE);
2851 }
2852 }
2853 break;
2854
2855 default:
2856 /* Should not reach here, but if so, forgive the VM. */
2857 error_report("%s: Unknown ending state %d", __func__, s->state);
2858 break;
2859 }
2860 qemu_bh_schedule(s->cleanup_bh);
2861 qemu_mutex_unlock_iothread();
2862}
2863
ad767bed
DDAG		 2864void migration_make_urgent_request(void)
2865{
2866 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
2867}
2868
2869void migration_consume_urgent_request(void)
2870{
2871 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
2872}
2873
70b20477
DDAG		 2874/*
2875 * Master migration thread on the source VM.
2876 * It drives the migration and pumps the data down the outgoing channel.
2877 */
5f496a1b 2878static void *migration_thread(void *opaque)
0d82d0e8 2879{
9848a404 2880 MigrationState *s = opaque;
bc72ad67 2881 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
b23c2ade 2882 MigThrError thr_error;
ad767bed 2883 bool urgent = false;
76f5933a 2884
ab28bd23
PB		 2885 rcu_register_thread();
2886
b15df1ae
PX		 2887 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2888
89a02a9f 2889 qemu_savevm_state_header(s->to_dst_file);
1d34e4bf 2890
62a02658
PX		 2891 /*
2892 * If we opened the return path, we need to make sure dst has it
2893 * opened as well.
2894 */
2895 if (s->rp_state.from_dst_file) {
1d34e4bf 2896 /* Now tell the dest that it should open its end so it can reply */
89a02a9f 2897 qemu_savevm_send_open_return_path(s->to_dst_file);
1d34e4bf
DDAG		 2898
2899 /* And do a ping that will make stuff easier to debug */
89a02a9f 2900 qemu_savevm_send_ping(s->to_dst_file, 1);
0425dc97 2901 }
1d34e4bf 2902
58110f0a 2903 if (migrate_postcopy()) {
1d34e4bf
DDAG		 2904 /*
2905 * Tell the destination that we *might* want to do postcopy later;
2906 * if the other end can't do postcopy it should fail now, nice and
2907 * early.
2908 */
89a02a9f 2909 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1d34e4bf
DDAG		 2910 }
2911
9907e842 2912 qemu_savevm_state_setup(s->to_dst_file);
0d82d0e8 2913
bc72ad67 2914 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
48781e5b
HZ		 2915 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2916 MIGRATION_STATUS_ACTIVE);
29ae8a41 2917
9ec055ae
DDAG		 2918 trace_migration_thread_setup_complete();
2919
2920 while (s->state == MIGRATION_STATUS_ACTIVE ||
2921 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
a3e879cd 2922 int64_t current_time;
0d82d0e8 2923
ad767bed 2924 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
2ad87305
PX		 2925 MigIterateState iter_state = migration_iteration_run(s);
2926 if (iter_state == MIG_ITERATE_SKIP) {
2927 continue;
2928 } else if (iter_state == MIG_ITERATE_BREAK) {
09f6c85e 2929 break;
c369f40d
JQ		 2930 }
2931 }
f4410a5d 2932
b23c2ade
PX		 2933 /*
2934 * Try to detect any kind of failures, and see whether we
2935 * should stop the migration now.
2936 */
2937 thr_error = migration_detect_error(s);
2938 if (thr_error == MIG_THR_ERR_FATAL) {
2939 /* Stop migration */
fd45ee2c 2940 break;
b23c2ade
PX		 2941 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
2942 /*
2943 * Just recovered from a e.g. network failure, reset all
2944 * the local variables. This is important to avoid
2945 * breaking transferred_bytes and bandwidth calculation
2946 */
2947 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2948 s->iteration_initial_bytes = 0;
fd45ee2c 2949 }
b15df1ae 2950
bc72ad67 2951 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0d82d0e8 2952
b15df1ae
PX		 2953 migration_update_counters(s, current_time);
2954
ad767bed 2955 urgent = false;
89a02a9f 2956 if (qemu_file_rate_limit(s->to_dst_file)) {
ad767bed
DDAG		 2957 /* Wait for a delay to do rate limiting OR
2958 * something urgent to post the semaphore.
2959 */
2960 int ms = s->iteration_start_time + BUFFER_DELAY - current_time;
2961 trace_migration_thread_ratelimit_pre(ms);
2962 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
2963 /* We were worken by one or more urgent things but
2964 * the timedwait will have consumed one of them.
2965 * The service routine for the urgent wake will dec
2966 * the semaphore itself for each item it consumes,
2967 * so add this one we just eat back.
2968 */
2969 qemu_sem_post(&s->rate_limit_sem);
2970 urgent = true;
2971 }
2972 trace_migration_thread_ratelimit_post(urgent);
0d82d0e8 2973 }
a3fa1d78
PB		 2974 }
2975
1d34e4bf 2976 trace_migration_thread_after_loop();
199aa6d4 2977 migration_iteration_finish(s);
ab28bd23 2978 rcu_unregister_thread();
0d82d0e8
JQ		 2979 return NULL;
2980}
2981
cce8040b 2982void migrate_fd_connect(MigrationState *s, Error *error_in)
0d82d0e8 2983{
d3e35b8f
PX		 2984 int64_t rate_limit;
2985 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
2986
2ff30257 2987 s->expected_downtime = s->parameters.downtime_limit;
bb1fadc4 2988 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
cce8040b
DDAG		 2989 if (error_in) {
2990 migrate_fd_error(s, error_in);
2991 migrate_fd_cleanup(s);
2992 return;
2993 }
0d82d0e8 2994
d3e35b8f
PX		 2995 if (resume) {
2996 /* This is a resumed migration */
2997 rate_limit = INT64_MAX;
2998 } else {
2999 /* This is a fresh new migration */
3000 rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3001 s->expected_downtime = s->parameters.downtime_limit;
3002 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
442773ce 3003
d3e35b8f
PX		 3004 /* Notify before starting migration thread */
3005 notifier_list_notify(&migration_state_notifiers, s);
3006 }
3007
3008 qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3009 qemu_file_set_blocking(s->to_dst_file, true);
9287ac27 3010
1d34e4bf 3011 /*
c788ada8
PX		 3012 * Open the return path. For postcopy, it is used exclusively. For
3013 * precopy, only if user specified "return-path" capability would
3014 * QEMU uses the return path.
1d34e4bf 3015 */
c788ada8 3016 if (migrate_postcopy_ram() || migrate_use_return_path()) {
d3e35b8f 3017 if (open_return_path_on_source(s, !resume)) {
1d34e4bf 3018 error_report("Unable to open return-path for postcopy");
d3e35b8f 3019 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
1d34e4bf
DDAG		 3020 migrate_fd_cleanup(s);
3021 return;
3022 }
3023 }
3024
d3e35b8f 3025 if (resume) {
135b87b4
PX		 3026 /* Wakeup the main migration thread to do the recovery */
3027 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3028 MIGRATION_STATUS_POSTCOPY_RECOVER);
3029 qemu_sem_post(&s->postcopy_pause_sem);
d3e35b8f
PX		 3030 return;
3031 }
3032
f986c3d2
JQ		 3033 if (multifd_save_setup() != 0) {
3034 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3035 MIGRATION_STATUS_FAILED);
3036 migrate_fd_cleanup(s);
3037 return;
3038 }
009fad7f 3039 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
bb1fadc4 3040 QEMU_THREAD_JOINABLE);
1d34e4bf 3041 s->migration_thread_running = true;
0d82d0e8 3042}
093e3c42 3043
9d18af93
PX		 3044void migration_global_dump(Monitor *mon)
3045{
3046 MigrationState *ms = migrate_get_current();
3047
6f0f6428
JQ		 3048 monitor_printf(mon, "globals:\n");
3049 monitor_printf(mon, "store-global-state: %s\n",
3050 ms->store_global_state ? "on" : "off");
3051 monitor_printf(mon, "only-migratable: %s\n",
3052 ms->only_migratable ? "on" : "off");
3053 monitor_printf(mon, "send-configuration: %s\n",
3054 ms->send_configuration ? "on" : "off");
3055 monitor_printf(mon, "send-section-footer: %s\n",
3056 ms->send_section_footer ? "on" : "off");
f548222c
XG		 3057 monitor_printf(mon, "decompress-error-check: %s\n",
3058 ms->decompress_error_check ? "on" : "off");
9d18af93
PX		 3059}
3060
20814758
PX		 3061#define DEFINE_PROP_MIG_CAP(name, x) \
3062 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3063
5272298c
PX		 3064static Property migration_properties[] = {
3065 DEFINE_PROP_BOOL("store-global-state", MigrationState,
3066 store_global_state, true),
3df663e5 3067 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
71dd4c1a
PX		 3068 DEFINE_PROP_BOOL("send-configuration", MigrationState,
3069 send_configuration, true),
15c38503
PX		 3070 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3071 send_section_footer, true),
f548222c
XG		 3072 DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3073 decompress_error_check, true),
89632faf
PX		 3074
3075 /* Migration parameters */
741d4086 3076 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
89632faf
PX		 3077 parameters.compress_level,
3078 DEFAULT_MIGRATE_COMPRESS_LEVEL),
741d4086 3079 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
89632faf
PX		 3080 parameters.compress_threads,
3081 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
741d4086 3082 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
89632faf
PX		 3083 parameters.decompress_threads,
3084 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
741d4086 3085 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
89632faf
PX		 3086 parameters.cpu_throttle_initial,
3087 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
741d4086 3088 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
89632faf
PX		 3089 parameters.cpu_throttle_increment,
3090 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
741d4086 3091 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
89632faf 3092 parameters.max_bandwidth, MAX_THROTTLE),
741d4086 3093 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
89632faf
PX		 3094 parameters.downtime_limit,
3095 DEFAULT_MIGRATE_SET_DOWNTIME),
741d4086 3096 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
89632faf
PX		 3097 parameters.x_checkpoint_delay,
3098 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
741d4086 3099 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
4075fb1c
JQ		 3100 parameters.x_multifd_channels,
3101 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
741d4086 3102 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
0fb86605
JQ		 3103 parameters.x_multifd_page_count,
3104 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
73af8dd8
JQ		 3105 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3106 parameters.xbzrle_cache_size,
3107 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
7e555c6c
DDAG		 3108 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3109 parameters.max_postcopy_bandwidth,
3110 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
20814758
PX		 3111
3112 /* Migration capabilities */
3113 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3114 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3115 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3116 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3117 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3118 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3119 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3120 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3121 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3122 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3123 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
30126bbf 3124 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
20814758 3125
5272298c
PX		 3126 DEFINE_PROP_END_OF_LIST(),
3127};
3128
e5cb7e76
PX		 3129static void migration_class_init(ObjectClass *klass, void *data)
3130{
3131 DeviceClass *dc = DEVICE_CLASS(klass);
3132
3133 dc->user_creatable = false;
5272298c 3134 dc->props = migration_properties;
e5cb7e76
PX		 3135}
3136
b91bf5e4
MAL		 3137static void migration_instance_finalize(Object *obj)
3138{
3139 MigrationState *ms = MIGRATION_OBJ(obj);
3140 MigrationParameters *params = &ms->parameters;
3141
87db1a7d 3142 qemu_mutex_destroy(&ms->error_mutex);
62df066f 3143 qemu_mutex_destroy(&ms->qemu_file_lock);
b91bf5e4
MAL		 3144 g_free(params->tls_hostname);
3145 g_free(params->tls_creds);
ad767bed 3146 qemu_sem_destroy(&ms->rate_limit_sem);
e91d8951 3147 qemu_sem_destroy(&ms->pause_sem);
b23c2ade 3148 qemu_sem_destroy(&ms->postcopy_pause_sem);
14b1742e 3149 qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
edd090c7 3150 qemu_sem_destroy(&ms->rp_state.rp_sem);
ab105cc1 3151 error_free(ms->error);
b91bf5e4
MAL		 3152}
3153
e5cb7e76
PX		 3154static void migration_instance_init(Object *obj)
3155{
3156 MigrationState *ms = MIGRATION_OBJ(obj);
8b0b29dc 3157 MigrationParameters *params = &ms->parameters;
e5cb7e76
PX		 3158
3159 ms->state = MIGRATION_STATUS_NONE;
e5cb7e76 3160 ms->mbps = -1;
e91d8951 3161 qemu_sem_init(&ms->pause_sem, 0);
87db1a7d 3162 qemu_mutex_init(&ms->error_mutex);
8b0b29dc
PX		 3163
3164 params->tls_hostname = g_strdup("");
3165 params->tls_creds = g_strdup("");
3166
3167 /* Set has_* up only for parameter checks */
3168 params->has_compress_level = true;
3169 params->has_compress_threads = true;
3170 params->has_decompress_threads = true;
3171 params->has_cpu_throttle_initial = true;
3172 params->has_cpu_throttle_increment = true;
3173 params->has_max_bandwidth = true;
3174 params->has_downtime_limit = true;
3175 params->has_x_checkpoint_delay = true;
3176 params->has_block_incremental = true;
4075fb1c 3177 params->has_x_multifd_channels = true;
0fb86605 3178 params->has_x_multifd_page_count = true;
73af8dd8 3179 params->has_xbzrle_cache_size = true;
7e555c6c 3180 params->has_max_postcopy_bandwidth = true;
b23c2ade
PX		 3181
3182 qemu_sem_init(&ms->postcopy_pause_sem, 0);
14b1742e 3183 qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
edd090c7 3184 qemu_sem_init(&ms->rp_state.rp_sem, 0);
ad767bed 3185 qemu_sem_init(&ms->rate_limit_sem, 0);
62df066f 3186 qemu_mutex_init(&ms->qemu_file_lock);
8b0b29dc
PX		 3187}
3188
3189/*
3190 * Return true if check pass, false otherwise. Error will be put
3191 * inside errp if provided.
3192 */
3193static bool migration_object_check(MigrationState *ms, Error **errp)
3194{
6b19a7d9
PX		 3195 MigrationCapabilityStatusList *head = NULL;
3196 /* Assuming all off */
3197 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3198 int i;
3199
8b0b29dc
PX		 3200 if (!migrate_params_check(&ms->parameters, errp)) {
3201 return false;
3202 }
3203
6b19a7d9
PX		 3204 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3205 if (ms->enabled_capabilities[i]) {
3206 head = migrate_cap_add(head, i, true);
3207 }
3208 }
3209
3210 ret = migrate_caps_check(cap_list, head, errp);
3211
3212 /* It works with head == NULL */
3213 qapi_free_MigrationCapabilityStatusList(head);
3214
3215 return ret;
e5cb7e76
PX		 3216}
3217
3218static const TypeInfo migration_type = {
3219 .name = TYPE_MIGRATION,
01f6e14c 3220 /*
c8d3ff38
PX		 3221 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3222 * not created using qdev_create(), it is not attached to the qdev
3223 * device tree, and it is never realized.
3224 *
3225 * TODO: Make this TYPE_OBJECT once QOM provides something like
3226 * TYPE_DEVICE's "-global" properties.
01f6e14c 3227 */
e5cb7e76
PX		 3228 .parent = TYPE_DEVICE,
3229 .class_init = migration_class_init,
3230 .class_size = sizeof(MigrationClass),
3231 .instance_size = sizeof(MigrationState),
3232 .instance_init = migration_instance_init,
b91bf5e4 3233 .instance_finalize = migration_instance_finalize,
e5cb7e76
PX		 3234};
3235
3236static void register_migration_types(void)
3237{
3238 type_register_static(&migration_type);
3239}
3240
3241type_init(register_migration_types);
QEMU mirror