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