migration.c

   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 #include "block-migration.h"
  22
  23 //#define DEBUG_MIGRATION
  24
  25 #ifdef DEBUG_MIGRATION
  26 #define dprintf(fmt, ...) \
  27     do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
  28 #else
  29 #define dprintf(fmt, ...) \
  30     do { } while (0)
  31 #endif
  32
  33 /* Migration speed throttling */
  34 static uint32_t max_throttle = (32 << 20);
  35
  36 static MigrationState *current_migration;
  37
  38 void qemu_start_incoming_migration(const char *uri)
  39 {
  40     const char *p;
  41
  42     if (strstart(uri, "tcp:", &p))
  43         tcp_start_incoming_migration(p);
  44 #if !defined(WIN32)
  45     else if (strstart(uri, "exec:", &p))
  46         exec_start_incoming_migration(p);
  47     else if (strstart(uri, "unix:", &p))
  48         unix_start_incoming_migration(p);
  49     else if (strstart(uri, "fd:", &p))
  50         fd_start_incoming_migration(p);
  51 #endif
  52     else
  53         fprintf(stderr, "unknown migration protocol: %s\n", uri);
  54 }
  55
  56 void do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
  57 {
  58     MigrationState *s = NULL;
  59     const char *p;
  60     int detach = qdict_get_int(qdict, "detach");
  61     const char *uri = qdict_get_str(qdict, "uri");
  62
  63     if (current_migration &&
  64         current_migration->get_status(current_migration) == MIG_STATE_ACTIVE) {
  65         monitor_printf(mon, "migration already in progress\n");
  66         return;
  67     }
  68
  69     if (strstart(uri, "tcp:", &p))
  70         s = tcp_start_outgoing_migration(mon, p, max_throttle, detach,
  71                                          (int)qdict_get_int(qdict, "blk"),
  72                                          (int)qdict_get_int(qdict, "inc"));
  73 #if !defined(WIN32)
  74     else if (strstart(uri, "exec:", &p))
  75         s = exec_start_outgoing_migration(mon, p, max_throttle, detach,
  76                                           (int)qdict_get_int(qdict, "blk"),
  77                                           (int)qdict_get_int(qdict, "inc"));
  78     else if (strstart(uri, "unix:", &p))
  79         s = unix_start_outgoing_migration(mon, p, max_throttle, detach,
  80                                           (int)qdict_get_int(qdict, "blk"),
  81                                           (int)qdict_get_int(qdict, "inc"));
  82     else if (strstart(uri, "fd:", &p))
  83         s = fd_start_outgoing_migration(mon, p, max_throttle, detach,
  84                                         (int)qdict_get_int(qdict, "blk"),
  85                                         (int)qdict_get_int(qdict, "inc"));
  86 #endif
  87     else
  88         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
  89
  90     if (s == NULL)
  91         monitor_printf(mon, "migration failed\n");
  92     else {
  93         if (current_migration)
  94             current_migration->release(current_migration);
  95
  96         current_migration = s;
  97     }
  98 }
  99
 100 void do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 101 {
 102     MigrationState *s = current_migration;
 103
 104     if (s)
 105         s->cancel(s);
 106 }
 107
 108 void do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 109 {
 110     double d;
 111     char *ptr;
 112     FdMigrationState *s;
 113     const char *value = qdict_get_str(qdict, "value");
 114
 115     d = strtod(value, &ptr);
 116     switch (*ptr) {
 117     case 'G': case 'g':
 118         d *= 1024;
 119     case 'M': case 'm':
 120         d *= 1024;
 121     case 'K': case 'k':
 122         d *= 1024;
 123     default:
 124         break;
 125     }
 126
 127     max_throttle = (uint32_t)d;
 128
 129     s = migrate_to_fms(current_migration);
 130     if (s && s->file) {
 131         qemu_file_set_rate_limit(s->file, max_throttle);
 132     }
 133 }
 134
 135 /* amount of nanoseconds we are willing to wait for migration to be down.
 136  * the choice of nanoseconds is because it is the maximum resolution that
 137  * get_clock() can achieve. It is an internal measure. All user-visible
 138  * units must be in seconds */
 139 static uint64_t max_downtime = 30000000;
 140
 141 uint64_t migrate_max_downtime(void)
 142 {
 143     return max_downtime;
 144 }
 145
 146 void do_migrate_set_downtime(Monitor *mon, const QDict *qdict)
 147 {
 148     char *ptr;
 149     double d;
 150     const char *value = qdict_get_str(qdict, "value");
 151
 152     d = strtod(value, &ptr);
 153     if (!strcmp(ptr,"ms")) {
 154         d *= 1000000;
 155     } else if (!strcmp(ptr,"us")) {
 156         d *= 1000;
 157     } else if (!strcmp(ptr,"ns")) {
 158     } else {
 159         /* all else considered to be seconds */
 160         d *= 1000000000;
 161     }
 162
 163     max_downtime = (uint64_t)d;
 164 }
 165
 166 void do_info_migrate(Monitor *mon)
 167 {
 168     MigrationState *s = current_migration;
 169
 170     if (s) {
 171         monitor_printf(mon, "Migration status: ");
 172         switch (s->get_status(s)) {
 173         case MIG_STATE_ACTIVE:
 174             monitor_printf(mon, "active\n");
 175             monitor_printf(mon, "transferred ram: %" PRIu64 " kbytes\n", ram_bytes_transferred() >> 10);
 176             monitor_printf(mon, "remaining ram: %" PRIu64 " kbytes\n", ram_bytes_remaining() >> 10);
 177             monitor_printf(mon, "total ram: %" PRIu64 " kbytes\n", ram_bytes_total() >> 10);
 178             if (blk_mig_active()) {
 179                 monitor_printf(mon, "transferred disk: %" PRIu64 " kbytes\n",
 180                                blk_mig_bytes_transferred() >> 10);
 181                 monitor_printf(mon, "remaining disk: %" PRIu64 " kbytes\n",
 182                                blk_mig_bytes_remaining() >> 10);
 183                 monitor_printf(mon, "total disk: %" PRIu64 " kbytes\n",
 184                                blk_mig_bytes_total() >> 10);
 185             }
 186             break;
 187         case MIG_STATE_COMPLETED:
 188             monitor_printf(mon, "completed\n");
 189             break;
 190         case MIG_STATE_ERROR:
 191             monitor_printf(mon, "failed\n");
 192             break;
 193         case MIG_STATE_CANCELLED:
 194             monitor_printf(mon, "cancelled\n");
 195             break;
 196         }
 197     }
 198 }
 199
 200 /* shared migration helpers */
 201
 202 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
 203 {
 204     s->mon = mon;
 205     if (monitor_suspend(mon) == 0) {
 206         dprintf("suspending monitor\n");
 207     } else {
 208         monitor_printf(mon, "terminal does not allow synchronous "
 209                        "migration, continuing detached\n");
 210     }
 211 }
 212
 213 void migrate_fd_error(FdMigrationState *s)
 214 {
 215     dprintf("setting error state\n");
 216     s->state = MIG_STATE_ERROR;
 217     migrate_fd_cleanup(s);
 218 }
 219
 220 void migrate_fd_cleanup(FdMigrationState *s)
 221 {
 222     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 223
 224     if (s->file) {
 225         dprintf("closing file\n");
 226         qemu_fclose(s->file);
 227         s->file = NULL;
 228     }
 229
 230     if (s->fd != -1)
 231         close(s->fd);
 232
 233     /* Don't resume monitor until we've flushed all of the buffers */
 234     if (s->mon) {
 235         monitor_resume(s->mon);
 236     }
 237
 238     s->fd = -1;
 239 }
 240
 241 void migrate_fd_put_notify(void *opaque)
 242 {
 243     FdMigrationState *s = opaque;
 244
 245     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 246     qemu_file_put_notify(s->file);
 247 }
 248
 249 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 250 {
 251     FdMigrationState *s = opaque;
 252     ssize_t ret;
 253
 254     do {
 255         ret = s->write(s, data, size);
 256     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 257
 258     if (ret == -1)
 259         ret = -(s->get_error(s));
 260
 261     if (ret == -EAGAIN)
 262         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 263
 264     return ret;
 265 }
 266
 267 void migrate_fd_connect(FdMigrationState *s)
 268 {
 269     int ret;
 270
 271     s->file = qemu_fopen_ops_buffered(s,
 272                                       s->bandwidth_limit,
 273                                       migrate_fd_put_buffer,
 274                                       migrate_fd_put_ready,
 275                                       migrate_fd_wait_for_unfreeze,
 276                                       migrate_fd_close);
 277
 278     dprintf("beginning savevm\n");
 279     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
 280                                   s->mig_state.shared);
 281     if (ret < 0) {
 282         dprintf("failed, %d\n", ret);
 283         migrate_fd_error(s);
 284         return;
 285     }
 286
 287     migrate_fd_put_ready(s);
 288 }
 289
 290 void migrate_fd_put_ready(void *opaque)
 291 {
 292     FdMigrationState *s = opaque;
 293
 294     if (s->state != MIG_STATE_ACTIVE) {
 295         dprintf("put_ready returning because of non-active state\n");
 296         return;
 297     }
 298
 299     dprintf("iterate\n");
 300     if (qemu_savevm_state_iterate(s->mon, s->file) == 1) {
 301         int state;
 302         int old_vm_running = vm_running;
 303
 304         dprintf("done iterating\n");
 305         vm_stop(0);
 306
 307         qemu_aio_flush();
 308         bdrv_flush_all();
 309         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 310             if (old_vm_running) {
 311                 vm_start();
 312             }
 313             state = MIG_STATE_ERROR;
 314         } else {
 315             state = MIG_STATE_COMPLETED;
 316         }
 317         migrate_fd_cleanup(s);
 318         s->state = state;
 319     }
 320 }
 321
 322 int migrate_fd_get_status(MigrationState *mig_state)
 323 {
 324     FdMigrationState *s = migrate_to_fms(mig_state);
 325     return s->state;
 326 }
 327
 328 void migrate_fd_cancel(MigrationState *mig_state)
 329 {
 330     FdMigrationState *s = migrate_to_fms(mig_state);
 331
 332     if (s->state != MIG_STATE_ACTIVE)
 333         return;
 334
 335     dprintf("cancelling migration\n");
 336
 337     s->state = MIG_STATE_CANCELLED;
 338     qemu_savevm_state_cancel(s->mon, s->file);
 339
 340     migrate_fd_cleanup(s);
 341 }
 342
 343 void migrate_fd_release(MigrationState *mig_state)
 344 {
 345     FdMigrationState *s = migrate_to_fms(mig_state);
 346
 347     dprintf("releasing state\n");
 348
 349     if (s->state == MIG_STATE_ACTIVE) {
 350         s->state = MIG_STATE_CANCELLED;
 351         migrate_fd_cleanup(s);
 352     }
 353     free(s);
 354 }
 355
 356 void migrate_fd_wait_for_unfreeze(void *opaque)
 357 {
 358     FdMigrationState *s = opaque;
 359     int ret;
 360
 361     dprintf("wait for unfreeze\n");
 362     if (s->state != MIG_STATE_ACTIVE)
 363         return;
 364
 365     do {
 366         fd_set wfds;
 367
 368         FD_ZERO(&wfds);
 369         FD_SET(s->fd, &wfds);
 370
 371         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 372     } while (ret == -1 && (s->get_error(s)) == EINTR);
 373 }
 374
 375 int migrate_fd_close(void *opaque)
 376 {
 377     FdMigrationState *s = opaque;
 378
 379     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 380     return s->close(s);
 381 }