migration/migration.h

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #ifndef QEMU_MIGRATION_H
  15 #define QEMU_MIGRATION_H
  16
  17 #include "exec/cpu-common.h"
  18 #include "hw/qdev-core.h"
  19 #include "qapi/qapi-types-migration.h"
  20 #include "qapi/qmp/json-writer.h"
  21 #include "qemu/thread.h"
  22 #include "qemu/coroutine_int.h"
  23 #include "io/channel.h"
  24 #include "io/channel-buffer.h"
  25 #include "net/announce.h"
  26 #include "qom/object.h"
  27 #include "postcopy-ram.h"
  28 #include "sysemu/runstate.h"
  29 #include "migration/misc.h"
  30
  31 struct PostcopyBlocktimeContext;
  32
  33 #define  MIGRATION_RESUME_ACK_VALUE  (1)
  34
  35 /*
  36  * 1<<6=64 pages -> 256K chunk when page size is 4K.  This gives us
  37  * the benefit that all the chunks are 64 pages aligned then the
  38  * bitmaps are always aligned to LONG.
  39  */
  40 #define CLEAR_BITMAP_SHIFT_MIN             6
  41 /*
  42  * 1<<18=256K pages -> 1G chunk when page size is 4K.  This is the
  43  * default value to use if no one specified.
  44  */
  45 #define CLEAR_BITMAP_SHIFT_DEFAULT        18
  46 /*
  47  * 1<<31=2G pages -> 8T chunk when page size is 4K.  This should be
  48  * big enough and make sure we won't overflow easily.
  49  */
  50 #define CLEAR_BITMAP_SHIFT_MAX            31
  51
  52 /* This is an abstraction of a "temp huge page" for postcopy's purpose */
  53 typedef struct {
  54     /*
  55      * This points to a temporary huge page as a buffer for UFFDIO_COPY.  It's
  56      * mmap()ed and needs to be freed when cleanup.
  57      */
  58     void *tmp_huge_page;
  59     /*
  60      * This points to the host page we're going to install for this temp page.
  61      * It tells us after we've received the whole page, where we should put it.
  62      */
  63     void *host_addr;
  64     /* Number of small pages copied (in size of TARGET_PAGE_SIZE) */
  65     unsigned int target_pages;
  66     /* Whether this page contains all zeros */
  67     bool all_zero;
  68 } PostcopyTmpPage;
  69
  70 typedef enum {
  71     PREEMPT_THREAD_NONE = 0,
  72     PREEMPT_THREAD_CREATED,
  73     PREEMPT_THREAD_QUIT,
  74 } PreemptThreadStatus;
  75
  76 /* State for the incoming migration */
  77 struct MigrationIncomingState {
  78     QEMUFile *from_src_file;
  79     /* Previously received RAM's RAMBlock pointer */
  80     RAMBlock *last_recv_block[RAM_CHANNEL_MAX];
  81     /* A hook to allow cleanup at the end of incoming migration */
  82     void *transport_data;
  83     void (*transport_cleanup)(void *data);
  84     /*
  85      * Used to sync thread creations.  Note that we can't create threads in
  86      * parallel with this sem.
  87      */
  88     QemuSemaphore  thread_sync_sem;
  89     /*
  90      * Free at the start of the main state load, set as the main thread finishes
  91      * loading state.
  92      */
  93     QemuEvent main_thread_load_event;
  94
  95     /* For network announces */
  96     AnnounceTimer  announce_timer;
  97
  98     size_t         largest_page_size;
  99     bool           have_fault_thread;
 100     QemuThread     fault_thread;
 101     /* Set this when we want the fault thread to quit */
 102     bool           fault_thread_quit;
 103
 104     bool           have_listen_thread;
 105     QemuThread     listen_thread;
 106
 107     /* For the kernel to send us notifications */
 108     int       userfault_fd;
 109     /* To notify the fault_thread to wake, e.g., when need to quit */
 110     int       userfault_event_fd;
 111     QEMUFile *to_src_file;
 112     QemuMutex rp_mutex;    /* We send replies from multiple threads */
 113     /* RAMBlock of last request sent to source */
 114     RAMBlock *last_rb;
 115     /*
 116      * Number of postcopy channels including the default precopy channel, so
 117      * vanilla postcopy will only contain one channel which contain both
 118      * precopy and postcopy streams.
 119      *
 120      * This is calculated when the src requests to enable postcopy but before
 121      * it starts.  Its value can depend on e.g. whether postcopy preemption is
 122      * enabled.
 123      */
 124     unsigned int postcopy_channels;
 125     /* QEMUFile for postcopy only; it'll be handled by a separate thread */
 126     QEMUFile *postcopy_qemufile_dst;
 127     /*
 128      * When postcopy_qemufile_dst is properly setup, this sem is posted.
 129      * One can wait on this semaphore to wait until the preempt channel is
 130      * properly setup.
 131      */
 132     QemuSemaphore postcopy_qemufile_dst_done;
 133     /* Postcopy priority thread is used to receive postcopy requested pages */
 134     QemuThread postcopy_prio_thread;
 135     /*
 136      * Always set by the main vm load thread only, but can be read by the
 137      * postcopy preempt thread.  "volatile" makes sure all reads will be
 138      * up-to-date across cores.
 139      */
 140     volatile PreemptThreadStatus preempt_thread_status;
 141     /*
 142      * Used to sync between the ram load main thread and the fast ram load
 143      * thread.  It protects postcopy_qemufile_dst, which is the postcopy
 144      * fast channel.
 145      *
 146      * The ram fast load thread will take it mostly for the whole lifecycle
 147      * because it needs to continuously read data from the channel, and
 148      * it'll only release this mutex if postcopy is interrupted, so that
 149      * the ram load main thread will take this mutex over and properly
 150      * release the broken channel.
 151      */
 152     QemuMutex postcopy_prio_thread_mutex;
 153     /*
 154      * An array of temp host huge pages to be used, one for each postcopy
 155      * channel.
 156      */
 157     PostcopyTmpPage *postcopy_tmp_pages;
 158     /* This is shared for all postcopy channels */
 159     void     *postcopy_tmp_zero_page;
 160     /* PostCopyFD's for external userfaultfds & handlers of shared memory */
 161     GArray   *postcopy_remote_fds;
 162
 163     int state;
 164
 165     /*
 166      * The incoming migration coroutine, non-NULL during qemu_loadvm_state().
 167      * Used to wake the migration incoming coroutine from rdma code. How much is
 168      * it safe - it's a question.
 169      */
 170     Coroutine *loadvm_co;
 171
 172     /* The coroutine we should enter (back) after failover */
 173     Coroutine *colo_incoming_co;
 174     QemuSemaphore colo_incoming_sem;
 175
 176     /*
 177      * PostcopyBlocktimeContext to keep information for postcopy
 178      * live migration, to calculate vCPU block time
 179      * */
 180     struct PostcopyBlocktimeContext *blocktime_ctx;
 181
 182     /* notify PAUSED postcopy incoming migrations to try to continue */
 183     QemuSemaphore postcopy_pause_sem_dst;
 184     QemuSemaphore postcopy_pause_sem_fault;
 185     /*
 186      * This semaphore is used to allow the ram fast load thread (only when
 187      * postcopy preempt is enabled) fall into sleep when there's network
 188      * interruption detected.  When the recovery is done, the main load
 189      * thread will kick the fast ram load thread using this semaphore.
 190      */
 191     QemuSemaphore postcopy_pause_sem_fast_load;
 192
 193     /* List of listening socket addresses  */
 194     SocketAddressList *socket_address_list;
 195
 196     /* A tree of pages that we requested to the source VM */
 197     GTree *page_requested;
 198     /*
 199      * For postcopy only, count the number of requested page faults that
 200      * still haven't been resolved.
 201      */
 202     int page_requested_count;
 203     /*
 204      * The mutex helps to maintain the requested pages that we sent to the
 205      * source, IOW, to guarantee coherent between the page_requests tree and
 206      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
 207      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
 208      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
 209      * for that page already.  This is intended so that the mutex won't
 210      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
 211      * this should be enough to make sure the page_requested tree always
 212      * contains valid information.
 213      */
 214     QemuMutex page_request_mutex;
 215     /*
 216      * If postcopy preempt is enabled, there is a chance that the main
 217      * thread finished loading its data before the preempt channel has
 218      * finished loading the urgent pages.  If that happens, the two threads
 219      * will use this condvar to synchronize, so the main thread will always
 220      * wait until all pages received.
 221      */
 222     QemuCond page_request_cond;
 223
 224     /*
 225      * Number of devices that have yet to approve switchover. When this reaches
 226      * zero an ACK that it's OK to do switchover is sent to the source. No lock
 227      * is needed as this field is updated serially.
 228      */
 229     unsigned int switchover_ack_pending_num;
 230 };
 231
 232 MigrationIncomingState *migration_incoming_get_current(void);
 233 void migration_incoming_state_destroy(void);
 234 void migration_incoming_transport_cleanup(MigrationIncomingState *mis);
 235 /*
 236  * Functions to work with blocktime context
 237  */
 238 void fill_destination_postcopy_migration_info(MigrationInfo *info);
 239
 240 #define TYPE_MIGRATION "migration"
 241
 242 typedef struct MigrationClass MigrationClass;
 243 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
 244                      MIGRATION_OBJ, TYPE_MIGRATION)
 245
 246 struct MigrationClass {
 247     /*< private >*/
 248     DeviceClass parent_class;
 249 };
 250
 251 struct MigrationState {
 252     /*< private >*/
 253     DeviceState parent_obj;
 254
 255     /*< public >*/
 256     QemuThread thread;
 257     /* Protected by qemu_file_lock */
 258     QEMUFile *to_dst_file;
 259     /* Postcopy specific transfer channel */
 260     QEMUFile *postcopy_qemufile_src;
 261     /*
 262      * It is posted when the preempt channel is established.  Note: this is
 263      * used for both the start or recover of a postcopy migration.  We'll
 264      * post to this sem every time a new preempt channel is created in the
 265      * main thread, and we keep post() and wait() in pair.
 266      */
 267     QemuSemaphore postcopy_qemufile_src_sem;
 268     QIOChannelBuffer *bioc;
 269     /*
 270      * Protects to_dst_file/from_dst_file pointers.  We need to make sure we
 271      * won't yield or hang during the critical section, since this lock will be
 272      * used in OOB command handler.
 273      */
 274     QemuMutex qemu_file_lock;
 275
 276     /*
 277      * Used to allow urgent requests to override rate limiting.
 278      */
 279     QemuSemaphore rate_limit_sem;
 280
 281     /* pages already send at the beginning of current iteration */
 282     uint64_t iteration_initial_pages;
 283
 284     /* pages transferred per second */
 285     double pages_per_second;
 286
 287     /* bytes already send at the beginning of current iteration */
 288     uint64_t iteration_initial_bytes;
 289     /* time at the start of current iteration */
 290     int64_t iteration_start_time;
 291     /*
 292      * The final stage happens when the remaining data is smaller than
 293      * this threshold; it's calculated from the requested downtime and
 294      * measured bandwidth, or avail-switchover-bandwidth if specified.
 295      */
 296     uint64_t threshold_size;
 297
 298     /* params from 'migrate-set-parameters' */
 299     MigrationParameters parameters;
 300
 301     int state;
 302
 303     /* State related to return path */
 304     struct {
 305         /* Protected by qemu_file_lock */
 306         QEMUFile     *from_dst_file;
 307         QemuThread    rp_thread;
 308         /*
 309          * We can also check non-zero of rp_thread, but there's no "official"
 310          * way to do this, so this bool makes it slightly more elegant.
 311          * Checking from_dst_file for this is racy because from_dst_file will
 312          * be cleared in the rp_thread!
 313          */
 314         bool          rp_thread_created;
 315         /*
 316          * Used to synchronize between migration main thread and return
 317          * path thread.  The migration thread can wait() on this sem, while
 318          * other threads (e.g., return path thread) can kick it using a
 319          * post().
 320          */
 321         QemuSemaphore rp_sem;
 322         /*
 323          * We post to this when we got one PONG from dest. So far it's an
 324          * easy way to know the main channel has successfully established
 325          * on dest QEMU.
 326          */
 327         QemuSemaphore rp_pong_acks;
 328     } rp_state;
 329
 330     double mbps;
 331     /* Timestamp when recent migration starts (ms) */
 332     int64_t start_time;
 333     /* Total time used by latest migration (ms) */
 334     int64_t total_time;
 335     /* Timestamp when VM is down (ms) to migrate the last stuff */
 336     int64_t downtime_start;
 337     int64_t downtime;
 338     int64_t expected_downtime;
 339     bool capabilities[MIGRATION_CAPABILITY__MAX];
 340     int64_t setup_time;
 341
 342     /*
 343      * State before stopping the vm by vm_stop_force_state().
 344      * If migration is interrupted by any reason, we need to continue
 345      * running the guest on source if it was running or restore its stopped
 346      * state.
 347      */
 348     RunState vm_old_state;
 349
 350     /* Flag set once the migration has been asked to enter postcopy */
 351     bool start_postcopy;
 352
 353     /* Flag set once the migration thread is running (and needs joining) */
 354     bool migration_thread_running;
 355
 356     /* Flag set once the migration thread called bdrv_inactivate_all */
 357     bool block_inactive;
 358
 359     /* Migration is waiting for guest to unplug device */
 360     QemuSemaphore wait_unplug_sem;
 361
 362     /* Migration is paused due to pause-before-switchover */
 363     QemuSemaphore pause_sem;
 364
 365     /* The semaphore is used to notify COLO thread that failover is finished */
 366     QemuSemaphore colo_exit_sem;
 367
 368     /* The event is used to notify COLO thread to do checkpoint */
 369     QemuEvent colo_checkpoint_event;
 370     int64_t colo_checkpoint_time;
 371     QEMUTimer *colo_delay_timer;
 372
 373     /* The first error that has occurred.
 374        We used the mutex to be able to return the 1st error message */
 375     Error *error;
 376     /* mutex to protect errp */
 377     QemuMutex error_mutex;
 378
 379     /* Do we have to clean up -b/-i from old migrate parameters */
 380     /* This feature is deprecated and will be removed */
 381     bool must_remove_block_options;
 382
 383     /*
 384      * Global switch on whether we need to store the global state
 385      * during migration.
 386      */
 387     bool store_global_state;
 388
 389     /* Whether we send QEMU_VM_CONFIGURATION during migration */
 390     bool send_configuration;
 391     /* Whether we send section footer during migration */
 392     bool send_section_footer;
 393
 394     /* Needed by postcopy-pause state */
 395     QemuSemaphore postcopy_pause_sem;
 396     /*
 397      * Whether we abort the migration if decompression errors are
 398      * detected at the destination. It is left at false for qemu
 399      * older than 3.0, since only newer qemu sends streams that
 400      * do not trigger spurious decompression errors.
 401      */
 402     bool decompress_error_check;
 403     /*
 404      * This variable only affects behavior when postcopy preempt mode is
 405      * enabled.
 406      *
 407      * When set:
 408      *
 409      * - postcopy preempt src QEMU instance will generate an EOS message at
 410      *   the end of migration to shut the preempt channel on dest side.
 411      *
 412      * - postcopy preempt channel will be created at the setup phase on src
 413          QEMU.
 414      *
 415      * When clear:
 416      *
 417      * - postcopy preempt src QEMU instance will _not_ generate an EOS
 418      *   message at the end of migration, the dest qemu will shutdown the
 419      *   channel itself.
 420      *
 421      * - postcopy preempt channel will be created at the switching phase
 422      *   from precopy -> postcopy (to avoid race condition of misordered
 423      *   creation of channels).
 424      *
 425      * NOTE: See message-id <ZBoShWArKDPpX/D7@work-vm> on qemu-devel
 426      * mailing list for more information on the possible race.  Everyone
 427      * should probably just keep this value untouched after set by the
 428      * machine type (or the default).
 429      */
 430     bool preempt_pre_7_2;
 431
 432     /*
 433      * flush every channel after each section sent.
 434      *
 435      * This assures that we can't mix pages from one iteration through
 436      * ram pages with pages for the following iteration.  We really
 437      * only need to do this flush after we have go through all the
 438      * dirty pages.  For historical reasons, we do that after each
 439      * section.  This is suboptimal (we flush too many times).
 440      * Default value is false. (since 8.1)
 441      */
 442     bool multifd_flush_after_each_section;
 443     /*
 444      * This decides the size of guest memory chunk that will be used
 445      * to track dirty bitmap clearing.  The size of memory chunk will
 446      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
 447      * bitmap for each page to send (1<<0=1); N=10 means we will clear
 448      * dirty bitmap only once for 1<<10=1K continuous guest pages
 449      * (which is in 4M chunk).
 450      */
 451     uint8_t clear_bitmap_shift;
 452
 453     /*
 454      * This save hostname when out-going migration starts
 455      */
 456     char *hostname;
 457
 458     /* QEMU_VM_VMDESCRIPTION content filled for all non-iterable devices. */
 459     JSONWriter *vmdesc;
 460
 461     /*
 462      * Indicates whether an ACK from the destination that it's OK to do
 463      * switchover has been received.
 464      */
 465     bool switchover_acked;
 466     /* Is this a rdma migration */
 467     bool rdma_migration;
 468 };
 469
 470 void migrate_set_state(int *state, int old_state, int new_state);
 471
 472 void migration_fd_process_incoming(QEMUFile *f);
 473 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
 474 void migration_incoming_process(void);
 475
 476 bool  migration_has_all_channels(void);
 477
 478 void migrate_set_error(MigrationState *s, const Error *error);
 479 bool migrate_has_error(MigrationState *s);
 480
 481 void migrate_fd_connect(MigrationState *s, Error *error_in);
 482
 483 int migration_call_notifiers(MigrationState *s, MigrationEventType type,
 484                              Error **errp);
 485
 486 int migrate_init(MigrationState *s, Error **errp);
 487 bool migration_is_blocked(Error **errp);
 488 /* True if outgoing migration has entered postcopy phase */
 489 bool migration_in_postcopy(void);
 490 bool migration_postcopy_is_alive(int state);
 491 MigrationState *migrate_get_current(void);
 492 bool migration_has_failed(MigrationState *);
 493 bool migrate_mode_is_cpr(MigrationState *);
 494
 495 uint64_t ram_get_total_transferred_pages(void);
 496
 497 /* Sending on the return path - generic and then for each message type */
 498 void migrate_send_rp_shut(MigrationIncomingState *mis,
 499                           uint32_t value);
 500 void migrate_send_rp_pong(MigrationIncomingState *mis,
 501                           uint32_t value);
 502 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
 503                               ram_addr_t start, uint64_t haddr);
 504 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
 505                                       RAMBlock *rb, ram_addr_t start);
 506 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
 507                                  char *block_name);
 508 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
 509 int migrate_send_rp_switchover_ack(MigrationIncomingState *mis);
 510
 511 void dirty_bitmap_mig_before_vm_start(void);
 512 void dirty_bitmap_mig_cancel_outgoing(void);
 513 void dirty_bitmap_mig_cancel_incoming(void);
 514 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
 515                                       Error **errp);
 516
 517 void migrate_add_address(SocketAddress *address);
 518 bool migrate_uri_parse(const char *uri, MigrationChannel **channel,
 519                        Error **errp);
 520 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
 521
 522 #define qemu_ram_foreach_block \
 523   #warning "Use foreach_not_ignored_block in migration code"
 524
 525 void migration_make_urgent_request(void);
 526 void migration_consume_urgent_request(void);
 527 bool migration_rate_limit(void);
 528 void migration_bh_schedule(QEMUBHFunc *cb, void *opaque);
 529 void migration_cancel(const Error *error);
 530
 531 void migration_populate_vfio_info(MigrationInfo *info);
 532 void migration_reset_vfio_bytes_transferred(void);
 533 void postcopy_temp_page_reset(PostcopyTmpPage *tmp_page);
 534
 535 /*
 536  * Migration thread waiting for return path thread.  Return non-zero if an
 537  * error is detected.
 538  */
 539 int migration_rp_wait(MigrationState *s);
 540 /*
 541  * Kick the migration thread waiting for return path messages.  NOTE: the
 542  * name can be slightly confusing (when read as "kick the rp thread"), just
 543  * to remember the target is always the migration thread.
 544  */
 545 void migration_rp_kick(MigrationState *s);
 546
 547 #endif