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 "net/announce.h" 24 #include "qom/object.h" 25 26 struct PostcopyBlocktimeContext; 27 28 #define MIGRATION_RESUME_ACK_VALUE (1) 29 30 /* 31 * 1<<6=64 pages -> 256K chunk when page size is 4K. This gives us 32 * the benefit that all the chunks are 64 pages aligned then the 33 * bitmaps are always aligned to LONG. 34 */ 35 #define CLEAR_BITMAP_SHIFT_MIN 6 36 /* 37 * 1<<18=256K pages -> 1G chunk when page size is 4K. This is the 38 * default value to use if no one specified. 39 */ 40 #define CLEAR_BITMAP_SHIFT_DEFAULT 18 41 /* 42 * 1<<31=2G pages -> 8T chunk when page size is 4K. This should be 43 * big enough and make sure we won't overflow easily. 44 */ 45 #define CLEAR_BITMAP_SHIFT_MAX 31 46 47 /* State for the incoming migration */ 48 struct MigrationIncomingState { 49 QEMUFile *from_src_file; 50 51 /* 52 * Free at the start of the main state load, set as the main thread finishes 53 * loading state. 54 */ 55 QemuEvent main_thread_load_event; 56 57 /* For network announces */ 58 AnnounceTimer announce_timer; 59 60 size_t largest_page_size; 61 bool have_fault_thread; 62 QemuThread fault_thread; 63 QemuSemaphore fault_thread_sem; 64 /* Set this when we want the fault thread to quit */ 65 bool fault_thread_quit; 66 67 bool have_listen_thread; 68 QemuThread listen_thread; 69 QemuSemaphore listen_thread_sem; 70 71 /* For the kernel to send us notifications */ 72 int userfault_fd; 73 /* To notify the fault_thread to wake, e.g., when need to quit */ 74 int userfault_event_fd; 75 QEMUFile *to_src_file; 76 QemuMutex rp_mutex; /* We send replies from multiple threads */ 77 /* RAMBlock of last request sent to source */ 78 RAMBlock *last_rb; 79 void *postcopy_tmp_page; 80 void *postcopy_tmp_zero_page; 81 /* PostCopyFD's for external userfaultfds & handlers of shared memory */ 82 GArray *postcopy_remote_fds; 83 84 QEMUBH *bh; 85 86 int state; 87 88 bool have_colo_incoming_thread; 89 QemuThread colo_incoming_thread; 90 /* The coroutine we should enter (back) after failover */ 91 Coroutine *migration_incoming_co; 92 QemuSemaphore colo_incoming_sem; 93 94 /* 95 * PostcopyBlocktimeContext to keep information for postcopy 96 * live migration, to calculate vCPU block time 97 * */ 98 struct PostcopyBlocktimeContext *blocktime_ctx; 99 100 /* notify PAUSED postcopy incoming migrations to try to continue */ 101 bool postcopy_recover_triggered; 102 QemuSemaphore postcopy_pause_sem_dst; 103 QemuSemaphore postcopy_pause_sem_fault; 104 105 /* List of listening socket addresses */ 106 SocketAddressList *socket_address_list; 107 }; 108 109 MigrationIncomingState *migration_incoming_get_current(void); 110 void migration_incoming_state_destroy(void); 111 /* 112 * Functions to work with blocktime context 113 */ 114 void fill_destination_postcopy_migration_info(MigrationInfo *info); 115 116 #define TYPE_MIGRATION "migration" 117 118 typedef struct MigrationClass MigrationClass; 119 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass, 120 MIGRATION_OBJ, TYPE_MIGRATION) 121 122 struct MigrationClass { 123 /*< private >*/ 124 DeviceClass parent_class; 125 }; 126 127 struct MigrationState 128 { 129 /*< private >*/ 130 DeviceState parent_obj; 131 132 /*< public >*/ 133 QemuThread thread; 134 QEMUBH *cleanup_bh; 135 QEMUFile *to_dst_file; 136 /* 137 * Protects to_dst_file pointer. We need to make sure we won't 138 * yield or hang during the critical section, since this lock will 139 * be used in OOB command handler. 140 */ 141 QemuMutex qemu_file_lock; 142 143 /* 144 * Used to allow urgent requests to override rate limiting. 145 */ 146 QemuSemaphore rate_limit_sem; 147 148 /* pages already send at the beginning of current iteration */ 149 uint64_t iteration_initial_pages; 150 151 /* pages transferred per second */ 152 double pages_per_second; 153 154 /* bytes already send at the beginning of current iteration */ 155 uint64_t iteration_initial_bytes; 156 /* time at the start of current iteration */ 157 int64_t iteration_start_time; 158 /* 159 * The final stage happens when the remaining data is smaller than 160 * this threshold; it's calculated from the requested downtime and 161 * measured bandwidth 162 */ 163 int64_t threshold_size; 164 165 /* params from 'migrate-set-parameters' */ 166 MigrationParameters parameters; 167 168 int state; 169 170 /* State related to return path */ 171 struct { 172 QEMUFile *from_dst_file; 173 QemuThread rp_thread; 174 bool error; 175 QemuSemaphore rp_sem; 176 } rp_state; 177 178 double mbps; 179 /* Timestamp when recent migration starts (ms) */ 180 int64_t start_time; 181 /* Total time used by latest migration (ms) */ 182 int64_t total_time; 183 /* Timestamp when VM is down (ms) to migrate the last stuff */ 184 int64_t downtime_start; 185 int64_t downtime; 186 int64_t expected_downtime; 187 bool enabled_capabilities[MIGRATION_CAPABILITY__MAX]; 188 int64_t setup_time; 189 /* 190 * Whether guest was running when we enter the completion stage. 191 * If migration is interrupted by any reason, we need to continue 192 * running the guest on source. 193 */ 194 bool vm_was_running; 195 196 /* Flag set once the migration has been asked to enter postcopy */ 197 bool start_postcopy; 198 /* Flag set after postcopy has sent the device state */ 199 bool postcopy_after_devices; 200 201 /* Flag set once the migration thread is running (and needs joining) */ 202 bool migration_thread_running; 203 204 /* Flag set once the migration thread called bdrv_inactivate_all */ 205 bool block_inactive; 206 207 /* Migration is waiting for guest to unplug device */ 208 QemuSemaphore wait_unplug_sem; 209 210 /* Migration is paused due to pause-before-switchover */ 211 QemuSemaphore pause_sem; 212 213 /* The semaphore is used to notify COLO thread that failover is finished */ 214 QemuSemaphore colo_exit_sem; 215 216 /* The event is used to notify COLO thread to do checkpoint */ 217 QemuEvent colo_checkpoint_event; 218 int64_t colo_checkpoint_time; 219 QEMUTimer *colo_delay_timer; 220 221 /* The first error that has occurred. 222 We used the mutex to be able to return the 1st error message */ 223 Error *error; 224 /* mutex to protect errp */ 225 QemuMutex error_mutex; 226 227 /* Do we have to clean up -b/-i from old migrate parameters */ 228 /* This feature is deprecated and will be removed */ 229 bool must_remove_block_options; 230 231 /* 232 * Global switch on whether we need to store the global state 233 * during migration. 234 */ 235 bool store_global_state; 236 237 /* Whether we send QEMU_VM_CONFIGURATION during migration */ 238 bool send_configuration; 239 /* Whether we send section footer during migration */ 240 bool send_section_footer; 241 242 /* Needed by postcopy-pause state */ 243 QemuSemaphore postcopy_pause_sem; 244 QemuSemaphore postcopy_pause_rp_sem; 245 /* 246 * Whether we abort the migration if decompression errors are 247 * detected at the destination. It is left at false for qemu 248 * older than 3.0, since only newer qemu sends streams that 249 * do not trigger spurious decompression errors. 250 */ 251 bool decompress_error_check; 252 253 /* 254 * This decides the size of guest memory chunk that will be used 255 * to track dirty bitmap clearing. The size of memory chunk will 256 * be GUEST_PAGE_SIZE << N. Say, N=0 means we will clear dirty 257 * bitmap for each page to send (1<<0=1); N=10 means we will clear 258 * dirty bitmap only once for 1<<10=1K continuous guest pages 259 * (which is in 4M chunk). 260 */ 261 uint8_t clear_bitmap_shift; 262 }; 263 264 void migrate_set_state(int *state, int old_state, int new_state); 265 266 void migration_fd_process_incoming(QEMUFile *f, Error **errp); 267 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp); 268 void migration_incoming_process(void); 269 270 bool migration_has_all_channels(void); 271 272 uint64_t migrate_max_downtime(void); 273 274 void migrate_set_error(MigrationState *s, const Error *error); 275 void migrate_fd_error(MigrationState *s, const Error *error); 276 277 void migrate_fd_connect(MigrationState *s, Error *error_in); 278 279 bool migration_is_setup_or_active(int state); 280 bool migration_is_running(int state); 281 282 void migrate_init(MigrationState *s); 283 bool migration_is_blocked(Error **errp); 284 /* True if outgoing migration has entered postcopy phase */ 285 bool migration_in_postcopy(void); 286 MigrationState *migrate_get_current(void); 287 288 bool migrate_postcopy(void); 289 290 bool migrate_release_ram(void); 291 bool migrate_postcopy_ram(void); 292 bool migrate_zero_blocks(void); 293 bool migrate_dirty_bitmaps(void); 294 bool migrate_ignore_shared(void); 295 bool migrate_validate_uuid(void); 296 297 bool migrate_auto_converge(void); 298 bool migrate_use_multifd(void); 299 bool migrate_pause_before_switchover(void); 300 int migrate_multifd_channels(void); 301 MultiFDCompression migrate_multifd_compression(void); 302 int migrate_multifd_zlib_level(void); 303 int migrate_multifd_zstd_level(void); 304 305 int migrate_use_xbzrle(void); 306 int64_t migrate_xbzrle_cache_size(void); 307 bool migrate_colo_enabled(void); 308 309 bool migrate_use_block(void); 310 bool migrate_use_block_incremental(void); 311 int migrate_max_cpu_throttle(void); 312 bool migrate_use_return_path(void); 313 314 uint64_t ram_get_total_transferred_pages(void); 315 316 bool migrate_use_compression(void); 317 int migrate_compress_level(void); 318 int migrate_compress_threads(void); 319 int migrate_compress_wait_thread(void); 320 int migrate_decompress_threads(void); 321 bool migrate_use_events(void); 322 bool migrate_postcopy_blocktime(void); 323 324 /* Sending on the return path - generic and then for each message type */ 325 void migrate_send_rp_shut(MigrationIncomingState *mis, 326 uint32_t value); 327 void migrate_send_rp_pong(MigrationIncomingState *mis, 328 uint32_t value); 329 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char* rbname, 330 ram_addr_t start, size_t len); 331 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis, 332 char *block_name); 333 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value); 334 335 void dirty_bitmap_mig_before_vm_start(void); 336 void dirty_bitmap_mig_cancel_outgoing(void); 337 void dirty_bitmap_mig_cancel_incoming(void); 338 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm, 339 Error **errp); 340 341 void migrate_add_address(SocketAddress *address); 342 343 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque); 344 345 #define qemu_ram_foreach_block \ 346 #warning "Use foreach_not_ignored_block in migration code" 347 348 void migration_make_urgent_request(void); 349 void migration_consume_urgent_request(void); 350 bool migration_rate_limit(void); 351 352 #endif 353