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 "qemu/thread.h"
  21 #include "qemu/coroutine_int.h"
  22 #include "io/channel.h"
  23 #include "io/channel-buffer.h"
  24 #include "net/announce.h"
  25 #include "qom/object.h"
  26 #include "postcopy-ram.h"
  27
  28 struct PostcopyBlocktimeContext;
  29
  30 #define  MIGRATION_RESUME_ACK_VALUE  (1)
  31
  32 /*
  33  * 1<<6=64 pages -> 256K chunk when page size is 4K.  This gives us
  34  * the benefit that all the chunks are 64 pages aligned then the
  35  * bitmaps are always aligned to LONG.
  36  */
  37 #define CLEAR_BITMAP_SHIFT_MIN             6
  38 /*
  39  * 1<<18=256K pages -> 1G chunk when page size is 4K.  This is the
  40  * default value to use if no one specified.
  41  */
  42 #define CLEAR_BITMAP_SHIFT_DEFAULT        18
  43 /*
  44  * 1<<31=2G pages -> 8T chunk when page size is 4K.  This should be
  45  * big enough and make sure we won't overflow easily.
  46  */
  47 #define CLEAR_BITMAP_SHIFT_MAX            31
  48
  49 /* This is an abstraction of a "temp huge page" for postcopy's purpose */
  50 typedef struct {
  51     /*
  52      * This points to a temporary huge page as a buffer for UFFDIO_COPY.  It's
  53      * mmap()ed and needs to be freed when cleanup.
  54      */
  55     void *tmp_huge_page;
  56     /*
  57      * This points to the host page we're going to install for this temp page.
  58      * It tells us after we've received the whole page, where we should put it.
  59      */
  60     void *host_addr;
  61     /* Number of small pages copied (in size of TARGET_PAGE_SIZE) */
  62     unsigned int target_pages;
  63     /* Whether this page contains all zeros */
  64     bool all_zero;
  65 } PostcopyTmpPage;
  66
  67 /* State for the incoming migration */
  68 struct MigrationIncomingState {
  69     QEMUFile *from_src_file;
  70     /* Previously received RAM's RAMBlock pointer */
  71     RAMBlock *last_recv_block[RAM_CHANNEL_MAX];
  72     /* A hook to allow cleanup at the end of incoming migration */
  73     void *transport_data;
  74     void (*transport_cleanup)(void *data);
  75     /*
  76      * Used to sync thread creations.  Note that we can't create threads in
  77      * parallel with this sem.
  78      */
  79     QemuSemaphore  thread_sync_sem;
  80     /*
  81      * Free at the start of the main state load, set as the main thread finishes
  82      * loading state.
  83      */
  84     QemuEvent main_thread_load_event;
  85
  86     /* For network announces */
  87     AnnounceTimer  announce_timer;
  88
  89     size_t         largest_page_size;
  90     bool           have_fault_thread;
  91     QemuThread     fault_thread;
  92     /* Set this when we want the fault thread to quit */
  93     bool           fault_thread_quit;
  94
  95     bool           have_listen_thread;
  96     QemuThread     listen_thread;
  97
  98     /* For the kernel to send us notifications */
  99     int       userfault_fd;
 100     /* To notify the fault_thread to wake, e.g., when need to quit */
 101     int       userfault_event_fd;
 102     QEMUFile *to_src_file;
 103     QemuMutex rp_mutex;    /* We send replies from multiple threads */
 104     /* RAMBlock of last request sent to source */
 105     RAMBlock *last_rb;
 106     /*
 107      * Number of postcopy channels including the default precopy channel, so
 108      * vanilla postcopy will only contain one channel which contain both
 109      * precopy and postcopy streams.
 110      *
 111      * This is calculated when the src requests to enable postcopy but before
 112      * it starts.  Its value can depend on e.g. whether postcopy preemption is
 113      * enabled.
 114      */
 115     unsigned int postcopy_channels;
 116     /* QEMUFile for postcopy only; it'll be handled by a separate thread */
 117     QEMUFile *postcopy_qemufile_dst;
 118     /* Postcopy priority thread is used to receive postcopy requested pages */
 119     QemuThread postcopy_prio_thread;
 120     bool postcopy_prio_thread_created;
 121     /*
 122      * Used to sync between the ram load main thread and the fast ram load
 123      * thread.  It protects postcopy_qemufile_dst, which is the postcopy
 124      * fast channel.
 125      *
 126      * The ram fast load thread will take it mostly for the whole lifecycle
 127      * because it needs to continuously read data from the channel, and
 128      * it'll only release this mutex if postcopy is interrupted, so that
 129      * the ram load main thread will take this mutex over and properly
 130      * release the broken channel.
 131      */
 132     QemuMutex postcopy_prio_thread_mutex;
 133     /*
 134      * An array of temp host huge pages to be used, one for each postcopy
 135      * channel.
 136      */
 137     PostcopyTmpPage *postcopy_tmp_pages;
 138     /* This is shared for all postcopy channels */
 139     void     *postcopy_tmp_zero_page;
 140     /* PostCopyFD's for external userfaultfds & handlers of shared memory */
 141     GArray   *postcopy_remote_fds;
 142
 143     QEMUBH *bh;
 144
 145     int state;
 146
 147     bool have_colo_incoming_thread;
 148     QemuThread colo_incoming_thread;
 149     /* The coroutine we should enter (back) after failover */
 150     Coroutine *migration_incoming_co;
 151     QemuSemaphore colo_incoming_sem;
 152
 153     /*
 154      * PostcopyBlocktimeContext to keep information for postcopy
 155      * live migration, to calculate vCPU block time
 156      * */
 157     struct PostcopyBlocktimeContext *blocktime_ctx;
 158
 159     /* notify PAUSED postcopy incoming migrations to try to continue */
 160     QemuSemaphore postcopy_pause_sem_dst;
 161     QemuSemaphore postcopy_pause_sem_fault;
 162     /*
 163      * This semaphore is used to allow the ram fast load thread (only when
 164      * postcopy preempt is enabled) fall into sleep when there's network
 165      * interruption detected.  When the recovery is done, the main load
 166      * thread will kick the fast ram load thread using this semaphore.
 167      */
 168     QemuSemaphore postcopy_pause_sem_fast_load;
 169
 170     /* List of listening socket addresses  */
 171     SocketAddressList *socket_address_list;
 172
 173     /* A tree of pages that we requested to the source VM */
 174     GTree *page_requested;
 175     /* For debugging purpose only, but would be nice to keep */
 176     int page_requested_count;
 177     /*
 178      * The mutex helps to maintain the requested pages that we sent to the
 179      * source, IOW, to guarantee coherent between the page_requests tree and
 180      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
 181      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
 182      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
 183      * for that page already.  This is intended so that the mutex won't
 184      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
 185      * this should be enough to make sure the page_requested tree always
 186      * contains valid information.
 187      */
 188     QemuMutex page_request_mutex;
 189 };
 190
 191 MigrationIncomingState *migration_incoming_get_current(void);
 192 void migration_incoming_state_destroy(void);
 193 void migration_incoming_transport_cleanup(MigrationIncomingState *mis);
 194 /*
 195  * Functions to work with blocktime context
 196  */
 197 void fill_destination_postcopy_migration_info(MigrationInfo *info);
 198
 199 #define TYPE_MIGRATION "migration"
 200
 201 typedef struct MigrationClass MigrationClass;
 202 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
 203                      MIGRATION_OBJ, TYPE_MIGRATION)
 204
 205 struct MigrationClass {
 206     /*< private >*/
 207     DeviceClass parent_class;
 208 };
 209
 210 struct MigrationState {
 211     /*< private >*/
 212     DeviceState parent_obj;
 213
 214     /*< public >*/
 215     QemuThread thread;
 216     QEMUBH *vm_start_bh;
 217     QEMUBH *cleanup_bh;
 218     /* Protected by qemu_file_lock */
 219     QEMUFile *to_dst_file;
 220     /* Postcopy specific transfer channel */
 221     QEMUFile *postcopy_qemufile_src;
 222     /*
 223      * It is posted when the preempt channel is established.  Note: this is
 224      * used for both the start or recover of a postcopy migration.  We'll
 225      * post to this sem every time a new preempt channel is created in the
 226      * main thread, and we keep post() and wait() in pair.
 227      */
 228     QemuSemaphore postcopy_qemufile_src_sem;
 229     QIOChannelBuffer *bioc;
 230     /*
 231      * Protects to_dst_file/from_dst_file pointers.  We need to make sure we
 232      * won't yield or hang during the critical section, since this lock will be
 233      * used in OOB command handler.
 234      */
 235     QemuMutex qemu_file_lock;
 236
 237     /*
 238      * Used to allow urgent requests to override rate limiting.
 239      */
 240     QemuSemaphore rate_limit_sem;
 241
 242     /* pages already send at the beginning of current iteration */
 243     uint64_t iteration_initial_pages;
 244
 245     /* pages transferred per second */
 246     double pages_per_second;
 247
 248     /* bytes already send at the beginning of current iteration */
 249     uint64_t iteration_initial_bytes;
 250     /* time at the start of current iteration */
 251     int64_t iteration_start_time;
 252     /*
 253      * The final stage happens when the remaining data is smaller than
 254      * this threshold; it's calculated from the requested downtime and
 255      * measured bandwidth
 256      */
 257     int64_t threshold_size;
 258
 259     /* params from 'migrate-set-parameters' */
 260     MigrationParameters parameters;
 261
 262     int state;
 263
 264     /* State related to return path */
 265     struct {
 266         /* Protected by qemu_file_lock */
 267         QEMUFile     *from_dst_file;
 268         QemuThread    rp_thread;
 269         bool          error;
 270         /*
 271          * We can also check non-zero of rp_thread, but there's no "official"
 272          * way to do this, so this bool makes it slightly more elegant.
 273          * Checking from_dst_file for this is racy because from_dst_file will
 274          * be cleared in the rp_thread!
 275          */
 276         bool          rp_thread_created;
 277         QemuSemaphore rp_sem;
 278     } rp_state;
 279
 280     double mbps;
 281     /* Timestamp when recent migration starts (ms) */
 282     int64_t start_time;
 283     /* Total time used by latest migration (ms) */
 284     int64_t total_time;
 285     /* Timestamp when VM is down (ms) to migrate the last stuff */
 286     int64_t downtime_start;
 287     int64_t downtime;
 288     int64_t expected_downtime;
 289     bool enabled_capabilities[MIGRATION_CAPABILITY__MAX];
 290     int64_t setup_time;
 291     /*
 292      * Whether guest was running when we enter the completion stage.
 293      * If migration is interrupted by any reason, we need to continue
 294      * running the guest on source.
 295      */
 296     bool vm_was_running;
 297
 298     /* Flag set once the migration has been asked to enter postcopy */
 299     bool start_postcopy;
 300     /* Flag set after postcopy has sent the device state */
 301     bool postcopy_after_devices;
 302
 303     /* Flag set once the migration thread is running (and needs joining) */
 304     bool migration_thread_running;
 305
 306     /* Flag set once the migration thread called bdrv_inactivate_all */
 307     bool block_inactive;
 308
 309     /* Migration is waiting for guest to unplug device */
 310     QemuSemaphore wait_unplug_sem;
 311
 312     /* Migration is paused due to pause-before-switchover */
 313     QemuSemaphore pause_sem;
 314
 315     /* The semaphore is used to notify COLO thread that failover is finished */
 316     QemuSemaphore colo_exit_sem;
 317
 318     /* The event is used to notify COLO thread to do checkpoint */
 319     QemuEvent colo_checkpoint_event;
 320     int64_t colo_checkpoint_time;
 321     QEMUTimer *colo_delay_timer;
 322
 323     /* The first error that has occurred.
 324        We used the mutex to be able to return the 1st error message */
 325     Error *error;
 326     /* mutex to protect errp */
 327     QemuMutex error_mutex;
 328
 329     /* Do we have to clean up -b/-i from old migrate parameters */
 330     /* This feature is deprecated and will be removed */
 331     bool must_remove_block_options;
 332
 333     /*
 334      * Global switch on whether we need to store the global state
 335      * during migration.
 336      */
 337     bool store_global_state;
 338
 339     /* Whether we send QEMU_VM_CONFIGURATION during migration */
 340     bool send_configuration;
 341     /* Whether we send section footer during migration */
 342     bool send_section_footer;
 343     /*
 344      * Whether we allow break sending huge pages when postcopy preempt is
 345      * enabled.  When disabled, we won't interrupt precopy within sending a
 346      * host huge page, which is the old behavior of vanilla postcopy.
 347      * NOTE: this parameter is ignored if postcopy preempt is not enabled.
 348      */
 349     bool postcopy_preempt_break_huge;
 350
 351     /* Needed by postcopy-pause state */
 352     QemuSemaphore postcopy_pause_sem;
 353     QemuSemaphore postcopy_pause_rp_sem;
 354     /*
 355      * Whether we abort the migration if decompression errors are
 356      * detected at the destination. It is left at false for qemu
 357      * older than 3.0, since only newer qemu sends streams that
 358      * do not trigger spurious decompression errors.
 359      */
 360     bool decompress_error_check;
 361
 362     /*
 363      * This decides the size of guest memory chunk that will be used
 364      * to track dirty bitmap clearing.  The size of memory chunk will
 365      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
 366      * bitmap for each page to send (1<<0=1); N=10 means we will clear
 367      * dirty bitmap only once for 1<<10=1K continuous guest pages
 368      * (which is in 4M chunk).
 369      */
 370     uint8_t clear_bitmap_shift;
 371
 372     /*
 373      * This save hostname when out-going migration starts
 374      */
 375     char *hostname;
 376 };
 377
 378 void migrate_set_state(int *state, int old_state, int new_state);
 379
 380 void migration_fd_process_incoming(QEMUFile *f, Error **errp);
 381 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
 382 void migration_incoming_process(void);
 383
 384 bool  migration_has_all_channels(void);
 385
 386 uint64_t migrate_max_downtime(void);
 387
 388 void migrate_set_error(MigrationState *s, const Error *error);
 389 void migrate_fd_error(MigrationState *s, const Error *error);
 390
 391 void migrate_fd_connect(MigrationState *s, Error *error_in);
 392
 393 bool migration_is_setup_or_active(int state);
 394 bool migration_is_running(int state);
 395
 396 void migrate_init(MigrationState *s);
 397 bool migration_is_blocked(Error **errp);
 398 /* True if outgoing migration has entered postcopy phase */
 399 bool migration_in_postcopy(void);
 400 MigrationState *migrate_get_current(void);
 401
 402 bool migrate_postcopy(void);
 403
 404 bool migrate_release_ram(void);
 405 bool migrate_postcopy_ram(void);
 406 bool migrate_zero_blocks(void);
 407 bool migrate_dirty_bitmaps(void);
 408 bool migrate_ignore_shared(void);
 409 bool migrate_validate_uuid(void);
 410
 411 bool migrate_auto_converge(void);
 412 bool migrate_use_multifd(void);
 413 bool migrate_pause_before_switchover(void);
 414 int migrate_multifd_channels(void);
 415 MultiFDCompression migrate_multifd_compression(void);
 416 int migrate_multifd_zlib_level(void);
 417 int migrate_multifd_zstd_level(void);
 418
 419 #ifdef CONFIG_LINUX
 420 bool migrate_use_zero_copy_send(void);
 421 #else
 422 #define migrate_use_zero_copy_send() (false)
 423 #endif
 424 int migrate_use_tls(void);
 425 int migrate_use_xbzrle(void);
 426 uint64_t migrate_xbzrle_cache_size(void);
 427 bool migrate_colo_enabled(void);
 428
 429 bool migrate_use_block(void);
 430 bool migrate_use_block_incremental(void);
 431 int migrate_max_cpu_throttle(void);
 432 bool migrate_use_return_path(void);
 433
 434 uint64_t ram_get_total_transferred_pages(void);
 435
 436 bool migrate_use_compression(void);
 437 int migrate_compress_level(void);
 438 int migrate_compress_threads(void);
 439 int migrate_compress_wait_thread(void);
 440 int migrate_decompress_threads(void);
 441 bool migrate_use_events(void);
 442 bool migrate_postcopy_blocktime(void);
 443 bool migrate_background_snapshot(void);
 444 bool migrate_postcopy_preempt(void);
 445
 446 /* Sending on the return path - generic and then for each message type */
 447 void migrate_send_rp_shut(MigrationIncomingState *mis,
 448                           uint32_t value);
 449 void migrate_send_rp_pong(MigrationIncomingState *mis,
 450                           uint32_t value);
 451 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
 452                               ram_addr_t start, uint64_t haddr);
 453 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
 454                                       RAMBlock *rb, ram_addr_t start);
 455 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
 456                                  char *block_name);
 457 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
 458
 459 void dirty_bitmap_mig_before_vm_start(void);
 460 void dirty_bitmap_mig_cancel_outgoing(void);
 461 void dirty_bitmap_mig_cancel_incoming(void);
 462 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
 463                                       Error **errp);
 464
 465 void migrate_add_address(SocketAddress *address);
 466
 467 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
 468
 469 #define qemu_ram_foreach_block \
 470   #warning "Use foreach_not_ignored_block in migration code"
 471
 472 void migration_make_urgent_request(void);
 473 void migration_consume_urgent_request(void);
 474 bool migration_rate_limit(void);
 475 void migration_cancel(const Error *error);
 476
 477 void populate_vfio_info(MigrationInfo *info);
 478 void postcopy_temp_page_reset(PostcopyTmpPage *tmp_page);
 479
 480 bool migrate_multi_channels_is_allowed(void);
 481 void migrate_protocol_allow_multi_channels(bool allow);
 482
 483 #endif