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