migration/colo.c

   1 /*
   2  * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO)
   3  * (a.k.a. Fault Tolerance or Continuous Replication)
   4  *
   5  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
   6  * Copyright (c) 2016 FUJITSU LIMITED
   7  * Copyright (c) 2016 Intel Corporation
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or
  10  * later.  See the COPYING file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "qemu/timer.h"
  15 #include "sysemu/sysemu.h"
  16 #include "qemu-file-channel.h"
  17 #include "migration/migration.h"
  18 #include "migration/qemu-file.h"
  19 #include "migration/colo.h"
  20 #include "migration/block.h"
  21 #include "io/channel-buffer.h"
  22 #include "trace.h"
  23 #include "qemu/error-report.h"
  24 #include "qapi/error.h"
  25 #include "migration/failover.h"
  26 #include "replication.h"
  27 #include "qmp-commands.h"
  28
  29 static bool vmstate_loading;
  30
  31 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
  32
  33 bool colo_supported(void)
  34 {
  35     return true;
  36 }
  37
  38 bool migration_in_colo_state(void)
  39 {
  40     MigrationState *s = migrate_get_current();
  41
  42     return (s->state == MIGRATION_STATUS_COLO);
  43 }
  44
  45 bool migration_incoming_in_colo_state(void)
  46 {
  47     MigrationIncomingState *mis = migration_incoming_get_current();
  48
  49     return mis && (mis->state == MIGRATION_STATUS_COLO);
  50 }
  51
  52 static bool colo_runstate_is_stopped(void)
  53 {
  54     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
  55 }
  56
  57 static void secondary_vm_do_failover(void)
  58 {
  59     int old_state;
  60     MigrationIncomingState *mis = migration_incoming_get_current();
  61
  62     /* Can not do failover during the process of VM's loading VMstate, Or
  63      * it will break the secondary VM.
  64      */
  65     if (vmstate_loading) {
  66         old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
  67                         FAILOVER_STATUS_RELAUNCH);
  68         if (old_state != FAILOVER_STATUS_ACTIVE) {
  69             error_report("Unknown error while do failover for secondary VM,"
  70                          "old_state: %s", FailoverStatus_lookup[old_state]);
  71         }
  72         return;
  73     }
  74
  75     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
  76                       MIGRATION_STATUS_COMPLETED);
  77
  78     if (!autostart) {
  79         error_report("\"-S\" qemu option will be ignored in secondary side");
  80         /* recover runstate to normal migration finish state */
  81         autostart = true;
  82     }
  83     /*
  84      * Make sure COLO incoming thread not block in recv or send,
  85      * If mis->from_src_file and mis->to_src_file use the same fd,
  86      * The second shutdown() will return -1, we ignore this value,
  87      * It is harmless.
  88      */
  89     if (mis->from_src_file) {
  90         qemu_file_shutdown(mis->from_src_file);
  91     }
  92     if (mis->to_src_file) {
  93         qemu_file_shutdown(mis->to_src_file);
  94     }
  95
  96     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
  97                                    FAILOVER_STATUS_COMPLETED);
  98     if (old_state != FAILOVER_STATUS_ACTIVE) {
  99         error_report("Incorrect state (%s) while doing failover for "
 100                      "secondary VM", FailoverStatus_lookup[old_state]);
 101         return;
 102     }
 103     /* Notify COLO incoming thread that failover work is finished */
 104     qemu_sem_post(&mis->colo_incoming_sem);
 105     /* For Secondary VM, jump to incoming co */
 106     if (mis->migration_incoming_co) {
 107         qemu_coroutine_enter(mis->migration_incoming_co);
 108     }
 109 }
 110
 111 static void primary_vm_do_failover(void)
 112 {
 113     MigrationState *s = migrate_get_current();
 114     int old_state;
 115
 116     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
 117                       MIGRATION_STATUS_COMPLETED);
 118
 119     /*
 120      * Wake up COLO thread which may blocked in recv() or send(),
 121      * The s->rp_state.from_dst_file and s->to_dst_file may use the
 122      * same fd, but we still shutdown the fd for twice, it is harmless.
 123      */
 124     if (s->to_dst_file) {
 125         qemu_file_shutdown(s->to_dst_file);
 126     }
 127     if (s->rp_state.from_dst_file) {
 128         qemu_file_shutdown(s->rp_state.from_dst_file);
 129     }
 130
 131     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 132                                    FAILOVER_STATUS_COMPLETED);
 133     if (old_state != FAILOVER_STATUS_ACTIVE) {
 134         error_report("Incorrect state (%s) while doing failover for Primary VM",
 135                      FailoverStatus_lookup[old_state]);
 136         return;
 137     }
 138     /* Notify COLO thread that failover work is finished */
 139     qemu_sem_post(&s->colo_exit_sem);
 140 }
 141
 142 void colo_do_failover(MigrationState *s)
 143 {
 144     /* Make sure VM stopped while failover happened. */
 145     if (!colo_runstate_is_stopped()) {
 146         vm_stop_force_state(RUN_STATE_COLO);
 147     }
 148
 149     if (get_colo_mode() == COLO_MODE_PRIMARY) {
 150         primary_vm_do_failover();
 151     } else {
 152         secondary_vm_do_failover();
 153     }
 154 }
 155
 156 void qmp_xen_set_replication(bool enable, bool primary,
 157                              bool has_failover, bool failover,
 158                              Error **errp)
 159 {
 160 #ifdef CONFIG_REPLICATION
 161     ReplicationMode mode = primary ?
 162                            REPLICATION_MODE_PRIMARY :
 163                            REPLICATION_MODE_SECONDARY;
 164
 165     if (has_failover && enable) {
 166         error_setg(errp, "Parameter 'failover' is only for"
 167                    " stopping replication");
 168         return;
 169     }
 170
 171     if (enable) {
 172         replication_start_all(mode, errp);
 173     } else {
 174         if (!has_failover) {
 175             failover = NULL;
 176         }
 177         replication_stop_all(failover, failover ? NULL : errp);
 178     }
 179 #else
 180     abort();
 181 #endif
 182 }
 183
 184 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
 185 {
 186 #ifdef CONFIG_REPLICATION
 187     Error *err = NULL;
 188     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
 189
 190     replication_get_error_all(&err);
 191     if (err) {
 192         s->error = true;
 193         s->has_desc = true;
 194         s->desc = g_strdup(error_get_pretty(err));
 195     } else {
 196         s->error = false;
 197     }
 198
 199     error_free(err);
 200     return s;
 201 #else
 202     abort();
 203 #endif
 204 }
 205
 206 void qmp_xen_colo_do_checkpoint(Error **errp)
 207 {
 208 #ifdef CONFIG_REPLICATION
 209     replication_do_checkpoint_all(errp);
 210 #else
 211     abort();
 212 #endif
 213 }
 214
 215 static void colo_send_message(QEMUFile *f, COLOMessage msg,
 216                               Error **errp)
 217 {
 218     int ret;
 219
 220     if (msg >= COLO_MESSAGE__MAX) {
 221         error_setg(errp, "%s: Invalid message", __func__);
 222         return;
 223     }
 224     qemu_put_be32(f, msg);
 225     qemu_fflush(f);
 226
 227     ret = qemu_file_get_error(f);
 228     if (ret < 0) {
 229         error_setg_errno(errp, -ret, "Can't send COLO message");
 230     }
 231     trace_colo_send_message(COLOMessage_lookup[msg]);
 232 }
 233
 234 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
 235                                     uint64_t value, Error **errp)
 236 {
 237     Error *local_err = NULL;
 238     int ret;
 239
 240     colo_send_message(f, msg, &local_err);
 241     if (local_err) {
 242         error_propagate(errp, local_err);
 243         return;
 244     }
 245     qemu_put_be64(f, value);
 246     qemu_fflush(f);
 247
 248     ret = qemu_file_get_error(f);
 249     if (ret < 0) {
 250         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
 251                          COLOMessage_lookup[msg]);
 252     }
 253 }
 254
 255 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
 256 {
 257     COLOMessage msg;
 258     int ret;
 259
 260     msg = qemu_get_be32(f);
 261     ret = qemu_file_get_error(f);
 262     if (ret < 0) {
 263         error_setg_errno(errp, -ret, "Can't receive COLO message");
 264         return msg;
 265     }
 266     if (msg >= COLO_MESSAGE__MAX) {
 267         error_setg(errp, "%s: Invalid message", __func__);
 268         return msg;
 269     }
 270     trace_colo_receive_message(COLOMessage_lookup[msg]);
 271     return msg;
 272 }
 273
 274 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
 275                                        Error **errp)
 276 {
 277     COLOMessage msg;
 278     Error *local_err = NULL;
 279
 280     msg = colo_receive_message(f, &local_err);
 281     if (local_err) {
 282         error_propagate(errp, local_err);
 283         return;
 284     }
 285     if (msg != expect_msg) {
 286         error_setg(errp, "Unexpected COLO message %d, expected %d",
 287                           msg, expect_msg);
 288     }
 289 }
 290
 291 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
 292                                            Error **errp)
 293 {
 294     Error *local_err = NULL;
 295     uint64_t value;
 296     int ret;
 297
 298     colo_receive_check_message(f, expect_msg, &local_err);
 299     if (local_err) {
 300         error_propagate(errp, local_err);
 301         return 0;
 302     }
 303
 304     value = qemu_get_be64(f);
 305     ret = qemu_file_get_error(f);
 306     if (ret < 0) {
 307         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
 308                          COLOMessage_lookup[expect_msg]);
 309     }
 310     return value;
 311 }
 312
 313 static int colo_do_checkpoint_transaction(MigrationState *s,
 314                                           QIOChannelBuffer *bioc,
 315                                           QEMUFile *fb)
 316 {
 317     Error *local_err = NULL;
 318     int ret = -1;
 319
 320     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
 321                       &local_err);
 322     if (local_err) {
 323         goto out;
 324     }
 325
 326     colo_receive_check_message(s->rp_state.from_dst_file,
 327                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
 328     if (local_err) {
 329         goto out;
 330     }
 331     /* Reset channel-buffer directly */
 332     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 333     bioc->usage = 0;
 334
 335     qemu_mutex_lock_iothread();
 336     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 337         qemu_mutex_unlock_iothread();
 338         goto out;
 339     }
 340     vm_stop_force_state(RUN_STATE_COLO);
 341     qemu_mutex_unlock_iothread();
 342     trace_colo_vm_state_change("run", "stop");
 343     /*
 344      * Failover request bh could be called after vm_stop_force_state(),
 345      * So we need check failover_request_is_active() again.
 346      */
 347     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 348         goto out;
 349     }
 350
 351     /* Disable block migration */
 352     migrate_set_block_enabled(false, &local_err);
 353     qemu_savevm_state_header(fb);
 354     qemu_savevm_state_begin(fb);
 355     qemu_mutex_lock_iothread();
 356     qemu_savevm_state_complete_precopy(fb, false);
 357     qemu_mutex_unlock_iothread();
 358
 359     qemu_fflush(fb);
 360
 361     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
 362     if (local_err) {
 363         goto out;
 364     }
 365     /*
 366      * We need the size of the VMstate data in Secondary side,
 367      * With which we can decide how much data should be read.
 368      */
 369     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
 370                             bioc->usage, &local_err);
 371     if (local_err) {
 372         goto out;
 373     }
 374
 375     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
 376     qemu_fflush(s->to_dst_file);
 377     ret = qemu_file_get_error(s->to_dst_file);
 378     if (ret < 0) {
 379         goto out;
 380     }
 381
 382     colo_receive_check_message(s->rp_state.from_dst_file,
 383                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
 384     if (local_err) {
 385         goto out;
 386     }
 387
 388     colo_receive_check_message(s->rp_state.from_dst_file,
 389                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
 390     if (local_err) {
 391         goto out;
 392     }
 393
 394     ret = 0;
 395
 396     qemu_mutex_lock_iothread();
 397     vm_start();
 398     qemu_mutex_unlock_iothread();
 399     trace_colo_vm_state_change("stop", "run");
 400
 401 out:
 402     if (local_err) {
 403         error_report_err(local_err);
 404     }
 405     return ret;
 406 }
 407
 408 static void colo_process_checkpoint(MigrationState *s)
 409 {
 410     QIOChannelBuffer *bioc;
 411     QEMUFile *fb = NULL;
 412     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 413     Error *local_err = NULL;
 414     int ret;
 415
 416     failover_init_state();
 417
 418     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
 419     if (!s->rp_state.from_dst_file) {
 420         error_report("Open QEMUFile from_dst_file failed");
 421         goto out;
 422     }
 423
 424     /*
 425      * Wait for Secondary finish loading VM states and enter COLO
 426      * restore.
 427      */
 428     colo_receive_check_message(s->rp_state.from_dst_file,
 429                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
 430     if (local_err) {
 431         goto out;
 432     }
 433     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 434     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
 435     object_unref(OBJECT(bioc));
 436
 437     qemu_mutex_lock_iothread();
 438     vm_start();
 439     qemu_mutex_unlock_iothread();
 440     trace_colo_vm_state_change("stop", "run");
 441
 442     timer_mod(s->colo_delay_timer,
 443             current_time + s->parameters.x_checkpoint_delay);
 444
 445     while (s->state == MIGRATION_STATUS_COLO) {
 446         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 447             error_report("failover request");
 448             goto out;
 449         }
 450
 451         qemu_sem_wait(&s->colo_checkpoint_sem);
 452
 453         ret = colo_do_checkpoint_transaction(s, bioc, fb);
 454         if (ret < 0) {
 455             goto out;
 456         }
 457     }
 458
 459 out:
 460     /* Throw the unreported error message after exited from loop */
 461     if (local_err) {
 462         error_report_err(local_err);
 463     }
 464
 465     if (fb) {
 466         qemu_fclose(fb);
 467     }
 468
 469     timer_del(s->colo_delay_timer);
 470
 471     /* Hope this not to be too long to wait here */
 472     qemu_sem_wait(&s->colo_exit_sem);
 473     qemu_sem_destroy(&s->colo_exit_sem);
 474     /*
 475      * Must be called after failover BH is completed,
 476      * Or the failover BH may shutdown the wrong fd that
 477      * re-used by other threads after we release here.
 478      */
 479     if (s->rp_state.from_dst_file) {
 480         qemu_fclose(s->rp_state.from_dst_file);
 481     }
 482 }
 483
 484 void colo_checkpoint_notify(void *opaque)
 485 {
 486     MigrationState *s = opaque;
 487     int64_t next_notify_time;
 488
 489     qemu_sem_post(&s->colo_checkpoint_sem);
 490     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 491     next_notify_time = s->colo_checkpoint_time +
 492                     s->parameters.x_checkpoint_delay;
 493     timer_mod(s->colo_delay_timer, next_notify_time);
 494 }
 495
 496 void migrate_start_colo_process(MigrationState *s)
 497 {
 498     qemu_mutex_unlock_iothread();
 499     qemu_sem_init(&s->colo_checkpoint_sem, 0);
 500     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
 501                                 colo_checkpoint_notify, s);
 502
 503     qemu_sem_init(&s->colo_exit_sem, 0);
 504     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
 505                       MIGRATION_STATUS_COLO);
 506     colo_process_checkpoint(s);
 507     qemu_mutex_lock_iothread();
 508 }
 509
 510 static void colo_wait_handle_message(QEMUFile *f, int *checkpoint_request,
 511                                      Error **errp)
 512 {
 513     COLOMessage msg;
 514     Error *local_err = NULL;
 515
 516     msg = colo_receive_message(f, &local_err);
 517     if (local_err) {
 518         error_propagate(errp, local_err);
 519         return;
 520     }
 521
 522     switch (msg) {
 523     case COLO_MESSAGE_CHECKPOINT_REQUEST:
 524         *checkpoint_request = 1;
 525         break;
 526     default:
 527         *checkpoint_request = 0;
 528         error_setg(errp, "Got unknown COLO message: %d", msg);
 529         break;
 530     }
 531 }
 532
 533 void *colo_process_incoming_thread(void *opaque)
 534 {
 535     MigrationIncomingState *mis = opaque;
 536     QEMUFile *fb = NULL;
 537     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
 538     uint64_t total_size;
 539     uint64_t value;
 540     Error *local_err = NULL;
 541
 542     qemu_sem_init(&mis->colo_incoming_sem, 0);
 543
 544     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 545                       MIGRATION_STATUS_COLO);
 546
 547     failover_init_state();
 548
 549     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 550     if (!mis->to_src_file) {
 551         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
 552         goto out;
 553     }
 554     /*
 555      * Note: the communication between Primary side and Secondary side
 556      * should be sequential, we set the fd to unblocked in migration incoming
 557      * coroutine, and here we are in the COLO incoming thread, so it is ok to
 558      * set the fd back to blocked.
 559      */
 560     qemu_file_set_blocking(mis->from_src_file, true);
 561
 562     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 563     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
 564     object_unref(OBJECT(bioc));
 565
 566     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
 567                       &local_err);
 568     if (local_err) {
 569         goto out;
 570     }
 571
 572     while (mis->state == MIGRATION_STATUS_COLO) {
 573         int request = 0;
 574
 575         colo_wait_handle_message(mis->from_src_file, &request, &local_err);
 576         if (local_err) {
 577             goto out;
 578         }
 579         assert(request);
 580         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 581             error_report("failover request");
 582             goto out;
 583         }
 584
 585         /* FIXME: This is unnecessary for periodic checkpoint mode */
 586         colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
 587                      &local_err);
 588         if (local_err) {
 589             goto out;
 590         }
 591
 592         colo_receive_check_message(mis->from_src_file,
 593                            COLO_MESSAGE_VMSTATE_SEND, &local_err);
 594         if (local_err) {
 595             goto out;
 596         }
 597
 598         value = colo_receive_message_value(mis->from_src_file,
 599                                  COLO_MESSAGE_VMSTATE_SIZE, &local_err);
 600         if (local_err) {
 601             goto out;
 602         }
 603
 604         /*
 605          * Read VM device state data into channel buffer,
 606          * It's better to re-use the memory allocated.
 607          * Here we need to handle the channel buffer directly.
 608          */
 609         if (value > bioc->capacity) {
 610             bioc->capacity = value;
 611             bioc->data = g_realloc(bioc->data, bioc->capacity);
 612         }
 613         total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
 614         if (total_size != value) {
 615             error_report("Got %" PRIu64 " VMState data, less than expected"
 616                         " %" PRIu64, total_size, value);
 617             goto out;
 618         }
 619         bioc->usage = total_size;
 620         qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 621
 622         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
 623                      &local_err);
 624         if (local_err) {
 625             goto out;
 626         }
 627
 628         qemu_mutex_lock_iothread();
 629         qemu_system_reset(SHUTDOWN_CAUSE_NONE);
 630         vmstate_loading = true;
 631         if (qemu_loadvm_state(fb) < 0) {
 632             error_report("COLO: loadvm failed");
 633             qemu_mutex_unlock_iothread();
 634             goto out;
 635         }
 636
 637         vmstate_loading = false;
 638         qemu_mutex_unlock_iothread();
 639
 640         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 641             failover_set_state(FAILOVER_STATUS_RELAUNCH,
 642                             FAILOVER_STATUS_NONE);
 643             failover_request_active(NULL);
 644             goto out;
 645         }
 646
 647         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
 648                      &local_err);
 649         if (local_err) {
 650             goto out;
 651         }
 652     }
 653
 654 out:
 655     vmstate_loading = false;
 656     /* Throw the unreported error message after exited from loop */
 657     if (local_err) {
 658         error_report_err(local_err);
 659     }
 660
 661     if (fb) {
 662         qemu_fclose(fb);
 663     }
 664
 665     /* Hope this not to be too long to loop here */
 666     qemu_sem_wait(&mis->colo_incoming_sem);
 667     qemu_sem_destroy(&mis->colo_incoming_sem);
 668     /* Must be called after failover BH is completed */
 669     if (mis->to_src_file) {
 670         qemu_fclose(mis->to_src_file);
 671     }
 672     migration_incoming_exit_colo();
 673
 674     return NULL;
 675 }