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