David 'Digit' Turner | 5d8f37a | 2009-09-14 14:32:27 -0700 | [diff] [blame] | 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 | #include "qemu-common.h" |
| 15 | #include "migration.h" |
| 16 | #include "monitor.h" |
| 17 | #include "buffered_file.h" |
| 18 | #include "sysemu.h" |
| 19 | #include "block.h" |
| 20 | #include "qemu_socket.h" |
| 21 | |
| 22 | //#define DEBUG_MIGRATION |
| 23 | |
| 24 | #ifdef DEBUG_MIGRATION |
| 25 | #define dprintf(fmt, ...) \ |
| 26 | do { printf("migration: " fmt, ## __VA_ARGS__); } while (0) |
| 27 | #else |
| 28 | #define dprintf(fmt, ...) \ |
| 29 | do { } while (0) |
| 30 | #endif |
| 31 | |
| 32 | /* Migration speed throttling */ |
| 33 | static uint32_t max_throttle = (32 << 20); |
| 34 | |
| 35 | static MigrationState *current_migration; |
| 36 | |
| 37 | void qemu_start_incoming_migration(const char *uri) |
| 38 | { |
| 39 | const char *p; |
| 40 | |
| 41 | if (strstart(uri, "tcp:", &p)) |
| 42 | tcp_start_incoming_migration(p); |
| 43 | #if !defined(WIN32) |
| 44 | else if (strstart(uri, "exec:", &p)) |
| 45 | exec_start_incoming_migration(p); |
| 46 | #endif |
| 47 | else |
| 48 | fprintf(stderr, "unknown migration protocol: %s\n", uri); |
| 49 | } |
| 50 | |
| 51 | void do_migrate(Monitor *mon, int detach, const char *uri) |
| 52 | { |
| 53 | MigrationState *s = NULL; |
| 54 | const char *p; |
| 55 | |
| 56 | if (strstart(uri, "tcp:", &p)) |
| 57 | s = tcp_start_outgoing_migration(p, max_throttle, detach); |
| 58 | #if !defined(WIN32) |
| 59 | else if (strstart(uri, "exec:", &p)) |
| 60 | s = exec_start_outgoing_migration(p, max_throttle, detach); |
| 61 | #endif |
| 62 | else |
| 63 | monitor_printf(mon, "unknown migration protocol: %s\n", uri); |
| 64 | |
| 65 | if (s == NULL) |
| 66 | monitor_printf(mon, "migration failed\n"); |
| 67 | else { |
| 68 | if (current_migration) |
| 69 | current_migration->release(current_migration); |
| 70 | |
| 71 | current_migration = s; |
| 72 | } |
| 73 | } |
| 74 | |
| 75 | void do_migrate_cancel(Monitor *mon) |
| 76 | { |
| 77 | MigrationState *s = current_migration; |
| 78 | |
| 79 | if (s) |
| 80 | s->cancel(s); |
| 81 | } |
| 82 | |
| 83 | void do_migrate_set_speed(Monitor *mon, const char *value) |
| 84 | { |
| 85 | double d; |
| 86 | char *ptr; |
| 87 | FdMigrationState *s; |
| 88 | |
| 89 | d = strtod(value, &ptr); |
| 90 | switch (*ptr) { |
| 91 | case 'G': case 'g': |
| 92 | d *= 1024; |
| 93 | case 'M': case 'm': |
| 94 | d *= 1024; |
| 95 | case 'K': case 'k': |
| 96 | d *= 1024; |
| 97 | default: |
| 98 | break; |
| 99 | } |
| 100 | |
| 101 | max_throttle = (uint32_t)d; |
| 102 | s = migrate_to_fms(current_migration); |
| 103 | |
| 104 | if (s) { |
| 105 | qemu_file_set_rate_limit(s->file, max_throttle); |
| 106 | } |
| 107 | |
| 108 | } |
| 109 | |
| 110 | /* amount of nanoseconds we are willing to wait for migration to be down. |
| 111 | * the choice of nanoseconds is because it is the maximum resolution that |
| 112 | * get_clock() can achieve. It is an internal measure. All user-visible |
| 113 | * units must be in seconds */ |
| 114 | static uint64_t max_downtime = 30000000; |
| 115 | |
| 116 | uint64_t migrate_max_downtime(void) |
| 117 | { |
| 118 | return max_downtime; |
| 119 | } |
| 120 | |
| 121 | void do_migrate_set_downtime(Monitor *mon, const char *value) |
| 122 | { |
| 123 | char *ptr; |
| 124 | double d; |
| 125 | |
| 126 | d = strtod(value, &ptr); |
| 127 | if (!strcmp(ptr,"ms")) { |
| 128 | d *= 1000000; |
| 129 | } else if (!strcmp(ptr,"us")) { |
| 130 | d *= 1000; |
| 131 | } else if (!strcmp(ptr,"ns")) { |
| 132 | } else { |
| 133 | /* all else considered to be seconds */ |
| 134 | d *= 1000000000; |
| 135 | } |
| 136 | |
| 137 | max_downtime = (uint64_t)d; |
| 138 | } |
| 139 | |
| 140 | void do_info_migrate(Monitor *mon) |
| 141 | { |
| 142 | MigrationState *s = current_migration; |
| 143 | |
| 144 | if (s) { |
| 145 | monitor_printf(mon, "Migration status: "); |
| 146 | switch (s->get_status(s)) { |
| 147 | case MIG_STATE_ACTIVE: |
| 148 | monitor_printf(mon, "active\n"); |
| 149 | monitor_printf(mon, "transferred ram: %" PRIu64 " kbytes\n", ram_bytes_transferred() >> 10); |
| 150 | monitor_printf(mon, "remaining ram: %" PRIu64 " kbytes\n", ram_bytes_remaining() >> 10); |
| 151 | monitor_printf(mon, "total ram: %" PRIu64 " kbytes\n", ram_bytes_total() >> 10); |
| 152 | break; |
| 153 | case MIG_STATE_COMPLETED: |
| 154 | monitor_printf(mon, "completed\n"); |
| 155 | break; |
| 156 | case MIG_STATE_ERROR: |
| 157 | monitor_printf(mon, "failed\n"); |
| 158 | break; |
| 159 | case MIG_STATE_CANCELLED: |
| 160 | monitor_printf(mon, "cancelled\n"); |
| 161 | break; |
| 162 | } |
| 163 | } |
| 164 | } |
| 165 | |
| 166 | /* shared migration helpers */ |
| 167 | |
| 168 | void migrate_fd_monitor_suspend(FdMigrationState *s) |
| 169 | { |
| 170 | s->mon_resume = cur_mon; |
| 171 | if (monitor_suspend(cur_mon) == 0) |
| 172 | dprintf("suspending monitor\n"); |
| 173 | else |
| 174 | monitor_printf(cur_mon, "terminal does not allow synchronous " |
| 175 | "migration, continuing detached\n"); |
| 176 | } |
| 177 | |
| 178 | void migrate_fd_error(FdMigrationState *s) |
| 179 | { |
| 180 | dprintf("setting error state\n"); |
| 181 | s->state = MIG_STATE_ERROR; |
| 182 | migrate_fd_cleanup(s); |
| 183 | } |
| 184 | |
| 185 | void migrate_fd_cleanup(FdMigrationState *s) |
| 186 | { |
| 187 | qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL); |
| 188 | |
| 189 | if (s->file) { |
| 190 | dprintf("closing file\n"); |
| 191 | qemu_fclose(s->file); |
| 192 | } |
| 193 | |
| 194 | if (s->fd != -1) |
| 195 | close(s->fd); |
| 196 | |
| 197 | /* Don't resume monitor until we've flushed all of the buffers */ |
| 198 | if (s->mon_resume) |
| 199 | monitor_resume(s->mon_resume); |
| 200 | |
| 201 | s->fd = -1; |
| 202 | } |
| 203 | |
| 204 | void migrate_fd_put_notify(void *opaque) |
| 205 | { |
| 206 | FdMigrationState *s = opaque; |
| 207 | |
| 208 | qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL); |
| 209 | qemu_file_put_notify(s->file); |
| 210 | } |
| 211 | |
| 212 | ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size) |
| 213 | { |
| 214 | FdMigrationState *s = opaque; |
| 215 | ssize_t ret; |
| 216 | |
| 217 | do { |
| 218 | ret = s->write(s, data, size); |
| 219 | } while (ret == -1 && ((s->get_error(s)) == EINTR)); |
| 220 | |
| 221 | if (ret == -1) |
| 222 | ret = -(s->get_error(s)); |
| 223 | |
| 224 | if (ret == -EAGAIN) |
| 225 | qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s); |
| 226 | |
| 227 | return ret; |
| 228 | } |
| 229 | |
| 230 | void migrate_fd_connect(FdMigrationState *s) |
| 231 | { |
| 232 | int ret; |
| 233 | |
| 234 | s->file = qemu_fopen_ops_buffered(s, |
| 235 | s->bandwidth_limit, |
| 236 | migrate_fd_put_buffer, |
| 237 | migrate_fd_put_ready, |
| 238 | migrate_fd_wait_for_unfreeze, |
| 239 | migrate_fd_close); |
| 240 | |
| 241 | dprintf("beginning savevm\n"); |
| 242 | ret = qemu_savevm_state_begin(s->file); |
| 243 | if (ret < 0) { |
| 244 | dprintf("failed, %d\n", ret); |
| 245 | migrate_fd_error(s); |
| 246 | return; |
| 247 | } |
| 248 | |
| 249 | migrate_fd_put_ready(s); |
| 250 | } |
| 251 | |
| 252 | void migrate_fd_put_ready(void *opaque) |
| 253 | { |
| 254 | FdMigrationState *s = opaque; |
| 255 | |
| 256 | if (s->state != MIG_STATE_ACTIVE) { |
| 257 | dprintf("put_ready returning because of non-active state\n"); |
| 258 | return; |
| 259 | } |
| 260 | |
| 261 | dprintf("iterate\n"); |
| 262 | if (qemu_savevm_state_iterate(s->file) == 1) { |
| 263 | int state; |
| 264 | dprintf("done iterating\n"); |
| 265 | vm_stop(0); |
| 266 | |
| 267 | bdrv_flush_all(); |
| 268 | if ((qemu_savevm_state_complete(s->file)) < 0) { |
| 269 | vm_start(); |
| 270 | state = MIG_STATE_ERROR; |
| 271 | } else { |
| 272 | state = MIG_STATE_COMPLETED; |
| 273 | } |
| 274 | migrate_fd_cleanup(s); |
| 275 | s->state = state; |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | int migrate_fd_get_status(MigrationState *mig_state) |
| 280 | { |
| 281 | FdMigrationState *s = migrate_to_fms(mig_state); |
| 282 | return s->state; |
| 283 | } |
| 284 | |
| 285 | void migrate_fd_cancel(MigrationState *mig_state) |
| 286 | { |
| 287 | FdMigrationState *s = migrate_to_fms(mig_state); |
| 288 | |
| 289 | if (s->state != MIG_STATE_ACTIVE) |
| 290 | return; |
| 291 | |
| 292 | dprintf("cancelling migration\n"); |
| 293 | |
| 294 | s->state = MIG_STATE_CANCELLED; |
| 295 | |
| 296 | migrate_fd_cleanup(s); |
| 297 | } |
| 298 | |
| 299 | void migrate_fd_release(MigrationState *mig_state) |
| 300 | { |
| 301 | FdMigrationState *s = migrate_to_fms(mig_state); |
| 302 | |
| 303 | dprintf("releasing state\n"); |
| 304 | |
| 305 | if (s->state == MIG_STATE_ACTIVE) { |
| 306 | s->state = MIG_STATE_CANCELLED; |
| 307 | migrate_fd_cleanup(s); |
| 308 | } |
| 309 | free(s); |
| 310 | } |
| 311 | |
| 312 | void migrate_fd_wait_for_unfreeze(void *opaque) |
| 313 | { |
| 314 | FdMigrationState *s = opaque; |
| 315 | int ret; |
| 316 | |
| 317 | dprintf("wait for unfreeze\n"); |
| 318 | if (s->state != MIG_STATE_ACTIVE) |
| 319 | return; |
| 320 | |
| 321 | do { |
| 322 | fd_set wfds; |
| 323 | |
| 324 | FD_ZERO(&wfds); |
| 325 | FD_SET(s->fd, &wfds); |
| 326 | |
| 327 | ret = select(s->fd + 1, NULL, &wfds, NULL, NULL); |
| 328 | } while (ret == -1 && (s->get_error(s)) == EINTR); |
| 329 | } |
| 330 | |
| 331 | int migrate_fd_close(void *opaque) |
| 332 | { |
| 333 | FdMigrationState *s = opaque; |
| 334 | |
| 335 | qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL); |
| 336 | return s->close(s); |
| 337 | } |