debian/patches/pve/0022-internal-snapshot-async.patch

   1 From ab6904d8e0f3a976ade19e8b5f99270738ed6518 Mon Sep 17 00:00:00 2001
   2 From: Wolfgang Bumiller <w.bumiller@proxmox.com>
   3 Date: Wed, 9 Dec 2015 16:04:32 +0100
   4 Subject: [PATCH 22/48] internal snapshot async
   5
   6 ---
   7  Makefile.objs           |   1 +
   8  block.c                 |   2 +-
   9  hmp-commands-info.hx    |  13 ++
  10  hmp-commands.hx         |  32 +++
  11  hmp.c                   |  57 ++++++
  12  hmp.h                   |   5 +
  13  include/block/block.h   |   1 +
  14  include/sysemu/sysemu.h |   5 +-
  15  migration/savevm.c      |  12 +-
  16  qapi-schema.json        |  68 +++++++
  17  qemu-options.hx         |  13 ++
  18  savevm-async.c          | 525 ++++++++++++++++++++++++++++++++++++++++++++++++
  19  vl.c                    |   8 +
  20  13 files changed, 734 insertions(+), 8 deletions(-)
  21  create mode 100644 savevm-async.c
  22
  23 diff --git a/Makefile.objs b/Makefile.objs
  24 index 9b12ee6..f5f8dba 100644
  25 --- a/Makefile.objs
  26 +++ b/Makefile.objs
  27 @@ -51,6 +51,7 @@ common-obj-$(CONFIG_LINUX) += fsdev/
  28
  29  common-obj-y += migration/
  30  common-obj-y += page_cache.o #aio.o
  31 +common-obj-y += savevm-async.o
  32
  33  common-obj-$(CONFIG_SPICE) += spice-qemu-char.o
  34
  35 diff --git a/block.c b/block.c
  36 index 6e906ec..5563a4f 100644
  37 --- a/block.c
  38 +++ b/block.c
  39 @@ -3045,7 +3045,7 @@ out:
  40      bdrv_unref(bs_new);
  41  }
  42
  43 -static void bdrv_delete(BlockDriverState *bs)
  44 +void bdrv_delete(BlockDriverState *bs)
  45  {
  46      assert(!bs->job);
  47      assert(bdrv_op_blocker_is_empty(bs));
  48 diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
  49 index 1a18380..3b5a0f9 100644
  50 --- a/hmp-commands-info.hx
  51 +++ b/hmp-commands-info.hx
  52 @@ -573,6 +573,19 @@ Show current migration xbzrle cache size.
  53  ETEXI
  54
  55      {
  56 +        .name       = "savevm",
  57 +        .args_type  = "",
  58 +        .params     = "",
  59 +        .help       = "show savevm status",
  60 +        .cmd = hmp_info_savevm,
  61 +    },
  62 +
  63 +STEXI
  64 +@item info savevm
  65 +show savevm status
  66 +ETEXI
  67 +
  68 +    {
  69          .name       = "balloon",
  70          .args_type  = "",
  71          .params     = "",
  72 diff --git a/hmp-commands.hx b/hmp-commands.hx
  73 index 7288203..a2867b5 100644
  74 --- a/hmp-commands.hx
  75 +++ b/hmp-commands.hx
  76 @@ -1808,3 +1808,35 @@ ETEXI
  77  STEXI
  78  @end table
  79  ETEXI
  80 +
  81 +    {
  82 +        .name       = "savevm-start",
  83 +        .args_type  = "statefile:s?",
  84 +        .params     = "[statefile]",
  85 +        .help       = "Prepare for snapshot and halt VM. Save VM state to statefile.",
  86 +        .cmd = hmp_savevm_start,
  87 +    },
  88 +
  89 +    {
  90 +        .name       = "snapshot-drive",
  91 +        .args_type  = "device:s,name:s",
  92 +        .params     = "device name",
  93 +        .help       = "Create internal snapshot.",
  94 +        .cmd = hmp_snapshot_drive,
  95 +    },
  96 +
  97 +    {
  98 +        .name       = "delete-drive-snapshot",
  99 +        .args_type  = "device:s,name:s",
 100 +        .params     = "device name",
 101 +        .help       = "Delete internal snapshot.",
 102 +        .cmd = hmp_delete_drive_snapshot,
 103 +    },
 104 +
 105 +    {
 106 +        .name       = "savevm-end",
 107 +        .args_type  = "",
 108 +        .params     = "",
 109 +        .help       = "Resume VM after snaphot.",
 110 +        .cmd = hmp_savevm_end,
 111 +    },
 112 diff --git a/hmp.c b/hmp.c
 113 index 465d7fa..aaf0de1 100644
 114 --- a/hmp.c
 115 +++ b/hmp.c
 116 @@ -2270,6 +2270,63 @@ void hmp_info_memory_devices(Monitor *mon, const QDict *qdict)
 117      qapi_free_MemoryDeviceInfoList(info_list);
 118  }
 119
 120 +void hmp_savevm_start(Monitor *mon, const QDict *qdict)
 121 +{
 122 +    Error *errp = NULL;
 123 +    const char *statefile = qdict_get_try_str(qdict, "statefile");
 124 +
 125 +    qmp_savevm_start(statefile != NULL, statefile, &errp);
 126 +    hmp_handle_error(mon, &errp);
 127 +}
 128 +
 129 +void hmp_snapshot_drive(Monitor *mon, const QDict *qdict)
 130 +{
 131 +    Error *errp = NULL;
 132 +    const char *name = qdict_get_str(qdict, "name");
 133 +    const char *device = qdict_get_str(qdict, "device");
 134 +
 135 +    qmp_snapshot_drive(device, name, &errp);
 136 +    hmp_handle_error(mon, &errp);
 137 +}
 138 +
 139 +void hmp_delete_drive_snapshot(Monitor *mon, const QDict *qdict)
 140 +{
 141 +    Error *errp = NULL;
 142 +    const char *name = qdict_get_str(qdict, "name");
 143 +    const char *device = qdict_get_str(qdict, "device");
 144 +
 145 +    qmp_delete_drive_snapshot(device, name, &errp);
 146 +    hmp_handle_error(mon, &errp);
 147 +}
 148 +
 149 +void hmp_savevm_end(Monitor *mon, const QDict *qdict)
 150 +{
 151 +    Error *errp = NULL;
 152 +
 153 +    qmp_savevm_end(&errp);
 154 +    hmp_handle_error(mon, &errp);
 155 +}
 156 +
 157 +void hmp_info_savevm(Monitor *mon, const QDict *qdict)
 158 +{
 159 +    SaveVMInfo *info;
 160 +    info = qmp_query_savevm(NULL);
 161 +
 162 +    if (info->has_status) {
 163 +        monitor_printf(mon, "savevm status: %s\n", info->status);
 164 +        monitor_printf(mon, "total time: %" PRIu64 " milliseconds\n",
 165 +                       info->total_time);
 166 +    } else {
 167 +        monitor_printf(mon, "savevm status: not running\n");
 168 +    }
 169 +    if (info->has_bytes) {
 170 +        monitor_printf(mon, "Bytes saved: %"PRIu64"\n", info->bytes);
 171 +    }
 172 +    if (info->has_error) {
 173 +        monitor_printf(mon, "Error: %s\n", info->error);
 174 +    }
 175 +}
 176 +
 177  void hmp_info_iothreads(Monitor *mon, const QDict *qdict)
 178  {
 179      IOThreadInfoList *info_list = qmp_query_iothreads(NULL);
 180 diff --git a/hmp.h b/hmp.h
 181 index 17a65b2..8c1b484 100644
 182 --- a/hmp.h
 183 +++ b/hmp.h
 184 @@ -26,6 +26,7 @@ void hmp_info_status(Monitor *mon, const QDict *qdict);
 185  void hmp_info_uuid(Monitor *mon, const QDict *qdict);
 186  void hmp_info_chardev(Monitor *mon, const QDict *qdict);
 187  void hmp_info_mice(Monitor *mon, const QDict *qdict);
 188 +void hmp_info_savevm(Monitor *mon, const QDict *qdict);
 189  void hmp_info_migrate(Monitor *mon, const QDict *qdict);
 190  void hmp_info_migrate_capabilities(Monitor *mon, const QDict *qdict);
 191  void hmp_info_migrate_parameters(Monitor *mon, const QDict *qdict);
 192 @@ -95,6 +96,10 @@ void hmp_netdev_add(Monitor *mon, const QDict *qdict);
 193  void hmp_netdev_del(Monitor *mon, const QDict *qdict);
 194  void hmp_getfd(Monitor *mon, const QDict *qdict);
 195  void hmp_closefd(Monitor *mon, const QDict *qdict);
 196 +void hmp_savevm_start(Monitor *mon, const QDict *qdict);
 197 +void hmp_snapshot_drive(Monitor *mon, const QDict *qdict);
 198 +void hmp_delete_drive_snapshot(Monitor *mon, const QDict *qdict);
 199 +void hmp_savevm_end(Monitor *mon, const QDict *qdict);
 200  void hmp_sendkey(Monitor *mon, const QDict *qdict);
 201  void hmp_screendump(Monitor *mon, const QDict *qdict);
 202  void hmp_nbd_server_start(Monitor *mon, const QDict *qdict);
 203 diff --git a/include/block/block.h b/include/block/block.h
 204 index 5149260..b29c69d 100644
 205 --- a/include/block/block.h
 206 +++ b/include/block/block.h
 207 @@ -295,6 +295,7 @@ BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
 208  int bdrv_get_backing_file_depth(BlockDriverState *bs);
 209  void bdrv_refresh_filename(BlockDriverState *bs);
 210  int bdrv_truncate(BdrvChild *child, int64_t offset);
 211 +void bdrv_delete(BlockDriverState *bs);
 212  int64_t bdrv_nb_sectors(BlockDriverState *bs);
 213  int64_t bdrv_getlength(BlockDriverState *bs);
 214  int64_t bdrv_get_allocated_file_size(BlockDriverState *bs);
 215 diff --git a/include/sysemu/sysemu.h b/include/sysemu/sysemu.h
 216 index 576c7ce..74623de 100644
 217 --- a/include/sysemu/sysemu.h
 218 +++ b/include/sysemu/sysemu.h
 219 @@ -78,6 +78,7 @@ void qemu_remove_machine_init_done_notifier(Notifier *notify);
 220  void hmp_savevm(Monitor *mon, const QDict *qdict);
 221  int save_vmstate(Monitor *mon, const char *name);
 222  int load_vmstate(const char *name);
 223 +int load_state_from_blockdev(const char *filename);
 224  void hmp_delvm(Monitor *mon, const QDict *qdict);
 225  void hmp_info_snapshots(Monitor *mon, const QDict *qdict);
 226
 227 @@ -105,13 +106,13 @@ enum qemu_vm_cmd {
 228  #define MAX_VM_CMD_PACKAGED_SIZE (1ul << 24)
 229
 230  bool qemu_savevm_state_blocked(Error **errp);
 231 -void qemu_savevm_state_begin(QEMUFile *f,
 232 +int qemu_savevm_state_begin(QEMUFile *f,
 233                               const MigrationParams *params);
 234  void qemu_savevm_state_header(QEMUFile *f);
 235  int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy);
 236  void qemu_savevm_state_cleanup(void);
 237  void qemu_savevm_state_complete_postcopy(QEMUFile *f);
 238 -void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only);
 239 +int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only);
 240  void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size,
 241                                 uint64_t *res_non_postcopiable,
 242                                 uint64_t *res_postcopiable);
 243 diff --git a/migration/savevm.c b/migration/savevm.c
 244 index 3b19a4a..feb0dc6 100644
 245 --- a/migration/savevm.c
 246 +++ b/migration/savevm.c
 247 @@ -970,11 +970,11 @@ void qemu_savevm_state_header(QEMUFile *f)
 248
 249  }
 250
 251 -void qemu_savevm_state_begin(QEMUFile *f,
 252 +int qemu_savevm_state_begin(QEMUFile *f,
 253                               const MigrationParams *params)
 254  {
 255      SaveStateEntry *se;
 256 -    int ret;
 257 +    int ret = 0;
 258
 259      trace_savevm_state_begin();
 260      QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 261 @@ -1002,6 +1002,7 @@ void qemu_savevm_state_begin(QEMUFile *f,
 262              break;
 263          }
 264      }
 265 +    return ret;
 266  }
 267
 268  /*
 269 @@ -1105,7 +1106,7 @@ void qemu_savevm_state_complete_postcopy(QEMUFile *f)
 270      qemu_fflush(f);
 271  }
 272
 273 -void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
 274 +int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
 275  {
 276      QJSON *vmdesc;
 277      int vmdesc_len;
 278 @@ -1139,12 +1140,12 @@ void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
 279          save_section_footer(f, se);
 280          if (ret < 0) {
 281              qemu_file_set_error(f, ret);
 282 -            return;
 283 +            return ret;
 284          }
 285      }
 286
 287      if (iterable_only) {
 288 -        return;
 289 +        return ret;
 290      }
 291
 292      vmdesc = qjson_new();
 293 @@ -1191,6 +1192,7 @@ void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
 294      qjson_destroy(vmdesc);
 295
 296      qemu_fflush(f);
 297 +    return qemu_file_get_error(f);
 298  }
 299
 300  /* Give an estimate of the amount left to be transferred,
 301 diff --git a/qapi-schema.json b/qapi-schema.json
 302 index 1127f2c..c33ebb3 100644
 303 --- a/qapi-schema.json
 304 +++ b/qapi-schema.json
 305 @@ -813,6 +813,40 @@
 306             '*error-desc': 'str'} }
 307
 308  ##
 309 +# @SaveVMInfo:
 310 +#
 311 +# Information about current migration process.
 312 +#
 313 +# @status: string describing the current savevm status.
 314 +#          This can be 'active', 'completed', 'failed'.
 315 +#          If this field is not returned, no savevm process
 316 +#          has been initiated
 317 +#
 318 +# @error: string containing error message is status is failed.
 319 +#
 320 +# @total-time: total amount of milliseconds since savevm started.
 321 +#        If savevm has ended, it returns the total save time
 322 +#
 323 +# @bytes: total amount of data transfered
 324 +#
 325 +# Since: 1.3
 326 +##
 327 +{ 'struct': 'SaveVMInfo',
 328 +  'data': {'*status': 'str', '*error': 'str',
 329 +           '*total-time': 'int', '*bytes': 'int'} }
 330 +
 331 +##
 332 +# @query-savevm:
 333 +#
 334 +# Returns information about current savevm process.
 335 +#
 336 +# Returns: @SaveVMInfo
 337 +#
 338 +# Since: 1.3
 339 +##
 340 +{ 'command': 'query-savevm', 'returns': 'SaveVMInfo' }
 341 +
 342 +##
 343  # @query-migrate:
 344  #
 345  # Returns information about current migration process. If migration
 346 @@ -4828,9 +4862,43 @@
 347  #
 348  # Since: 1.2.0
 349  ##
 350 +
 351  { 'command': 'query-target', 'returns': 'TargetInfo' }
 352
 353  ##
 354 +# @savevm-start:
 355 +#
 356 +# Prepare for snapshot and halt VM. Save VM state to statefile.
 357 +#
 358 +##
 359 +{ 'command': 'savevm-start', 'data': { '*statefile': 'str' } }
 360 +
 361 +##
 362 +# @snapshot-drive:
 363 +#
 364 +# Create an internal drive snapshot.
 365 +#
 366 +##
 367 +{ 'command': 'snapshot-drive', 'data': { 'device': 'str', 'name': 'str' } }
 368 +
 369 +##
 370 +# @delete-drive-snapshot:
 371 +#
 372 +# Delete a drive snapshot.
 373 +#
 374 +##
 375 +{ 'command': 'delete-drive-snapshot', 'data': { 'device': 'str', 'name': 'str' } }
 376 +
 377 +##
 378 +# @savevm-end:
 379 +#
 380 +# Resume VM after a snapshot.
 381 +#
 382 +##
 383 +{ 'command': 'savevm-end' }
 384 +
 385 +
 386 +##
 387  # @QKeyCode:
 388  #
 389  # An enumeration of key name.
 390 diff --git a/qemu-options.hx b/qemu-options.hx
 391 index 99af8ed..10f0e81 100644
 392 --- a/qemu-options.hx
 393 +++ b/qemu-options.hx
 394 @@ -3396,6 +3396,19 @@ STEXI
 395  Start right away with a saved state (@code{loadvm} in monitor)
 396  ETEXI
 397
 398 +DEF("loadstate", HAS_ARG, QEMU_OPTION_loadstate, \
 399 +    "-loadstate file\n" \
 400 +    "                start right away with a saved state\n",
 401 +    QEMU_ARCH_ALL)
 402 +STEXI
 403 +@item -loadstate @var{file}
 404 +@findex -loadstate
 405 +Start right away with a saved state. This option does not rollback
 406 +disk state like @code{loadvm}, so user must make sure that disk
 407 +have correct state. @var{file} can be any valid device URL. See the section
 408 +for "Device URL Syntax" for more information.
 409 +ETEXI
 410 +
 411  #ifndef _WIN32
 412  DEF("daemonize", 0, QEMU_OPTION_daemonize, \
 413      "-daemonize      daemonize QEMU after initializing\n", QEMU_ARCH_ALL)
 414 diff --git a/savevm-async.c b/savevm-async.c
 415 new file mode 100644
 416 index 0000000..9704a41
 417 --- /dev/null
 418 +++ b/savevm-async.c
 419 @@ -0,0 +1,525 @@
 420 +#include "qemu/osdep.h"
 421 +#include "qemu-common.h"
 422 +#include "qapi/qmp/qerror.h"
 423 +#include "qemu/error-report.h"
 424 +#include "sysemu/sysemu.h"
 425 +#include "qmp-commands.h"
 426 +#include "qemu-options.h"
 427 +#include "migration/qemu-file.h"
 428 +#include "qom/qom-qobject.h"
 429 +#include "migration/migration.h"
 430 +#include "block/snapshot.h"
 431 +#include "block/qapi.h"
 432 +#include "block/block.h"
 433 +#include "qemu/timer.h"
 434 +#include "sysemu/block-backend.h"
 435 +#include "qapi/qmp/qstring.h"
 436 +#include "qemu/rcu.h"
 437 +#include "qemu/thread.h"
 438 +#include "qemu/cutils.h"
 439 +
 440 +/* #define DEBUG_SAVEVM_STATE */
 441 +
 442 +#ifdef DEBUG_SAVEVM_STATE
 443 +#define DPRINTF(fmt, ...) \
 444 +    do { printf("savevm-async: " fmt, ## __VA_ARGS__); } while (0)
 445 +#else
 446 +#define DPRINTF(fmt, ...) \
 447 +    do { } while (0)
 448 +#endif
 449 +
 450 +enum {
 451 +    SAVE_STATE_DONE,
 452 +    SAVE_STATE_ERROR,
 453 +    SAVE_STATE_ACTIVE,
 454 +    SAVE_STATE_COMPLETED,
 455 +    SAVE_STATE_CANCELLED
 456 +};
 457 +
 458 +
 459 +static struct SnapshotState {
 460 +    BlockDriverState *bs;
 461 +    size_t bs_pos;
 462 +    int state;
 463 +    Error *error;
 464 +    Error *blocker;
 465 +    int saved_vm_running;
 466 +    QEMUFile *file;
 467 +    int64_t total_time;
 468 +} snap_state;
 469 +
 470 +SaveVMInfo *qmp_query_savevm(Error **errp)
 471 +{
 472 +    SaveVMInfo *info = g_malloc0(sizeof(*info));
 473 +    struct SnapshotState *s = &snap_state;
 474 +
 475 +    if (s->state != SAVE_STATE_DONE) {
 476 +        info->has_bytes = true;
 477 +        info->bytes = s->bs_pos;
 478 +        switch (s->state) {
 479 +        case SAVE_STATE_ERROR:
 480 +            info->has_status = true;
 481 +            info->status = g_strdup("failed");
 482 +            info->has_total_time = true;
 483 +            info->total_time = s->total_time;
 484 +            if (s->error) {
 485 +                info->has_error = true;
 486 +                info->error = g_strdup(error_get_pretty(s->error));
 487 +            }
 488 +            break;
 489 +        case SAVE_STATE_ACTIVE:
 490 +            info->has_status = true;
 491 +            info->status = g_strdup("active");
 492 +            info->has_total_time = true;
 493 +            info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 494 +                - s->total_time;
 495 +            break;
 496 +        case SAVE_STATE_COMPLETED:
 497 +            info->has_status = true;
 498 +            info->status = g_strdup("completed");
 499 +            info->has_total_time = true;
 500 +            info->total_time = s->total_time;
 501 +            break;
 502 +        }
 503 +    }
 504 +
 505 +    return info;
 506 +}
 507 +
 508 +static int save_snapshot_cleanup(void)
 509 +{
 510 +    int ret = 0;
 511 +
 512 +    DPRINTF("save_snapshot_cleanup\n");
 513 +
 514 +    snap_state.total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
 515 +        snap_state.total_time;
 516 +
 517 +    if (snap_state.file) {
 518 +        ret = qemu_fclose(snap_state.file);
 519 +    }
 520 +
 521 +    if (snap_state.bs) {
 522 +        /* try to truncate, but ignore errors (will fail on block devices).
 523 +         * note: bdrv_read() need whole blocks, so we round up
 524 +         */
 525 +        size_t size = (snap_state.bs_pos + BDRV_SECTOR_SIZE) & BDRV_SECTOR_MASK;
 526 +        bdrv_truncate(snap_state.bs, size);
 527 +        bdrv_op_unblock_all(snap_state.bs, snap_state.blocker);
 528 +        error_free(snap_state.blocker);
 529 +        snap_state.blocker = NULL;
 530 +        bdrv_unref(snap_state.bs);
 531 +        snap_state.bs = NULL;
 532 +    }
 533 +
 534 +    return ret;
 535 +}
 536 +
 537 +static void save_snapshot_error(const char *fmt, ...)
 538 +{
 539 +    va_list ap;
 540 +    char *msg;
 541 +
 542 +    va_start(ap, fmt);
 543 +    msg = g_strdup_vprintf(fmt, ap);
 544 +    va_end(ap);
 545 +
 546 +    DPRINTF("save_snapshot_error: %s\n", msg);
 547 +
 548 +    if (!snap_state.error) {
 549 +        error_set(&snap_state.error, ERROR_CLASS_GENERIC_ERROR, "%s", msg);
 550 +    }
 551 +
 552 +    g_free (msg);
 553 +
 554 +    snap_state.state = SAVE_STATE_ERROR;
 555 +
 556 +    save_snapshot_cleanup();
 557 +}
 558 +
 559 +static void save_snapshot_completed(void)
 560 +{
 561 +    DPRINTF("save_snapshot_completed\n");
 562 +
 563 +    if (save_snapshot_cleanup() < 0) {
 564 +        snap_state.state = SAVE_STATE_ERROR;
 565 +    } else {
 566 +        snap_state.state = SAVE_STATE_COMPLETED;
 567 +    }
 568 +}
 569 +
 570 +static int block_state_close(void *opaque)
 571 +{
 572 +    snap_state.file = NULL;
 573 +    return bdrv_flush(snap_state.bs);
 574 +}
 575 +
 576 +static int block_state_put_buffer(void *opaque, const uint8_t *buf,
 577 +                                  int64_t pos, int size)
 578 +{
 579 +    int ret;
 580 +
 581 +    assert(pos == snap_state.bs_pos);
 582 +
 583 +    if ((ret = bdrv_pwrite(snap_state.bs, snap_state.bs_pos, buf, size)) > 0) {
 584 +        snap_state.bs_pos += ret;
 585 +    }
 586 +
 587 +    return ret;
 588 +}
 589 +
 590 +static int store_and_stop(void) {
 591 +    if (global_state_store()) {
 592 +        save_snapshot_error("Error saving global state");
 593 +        return 1;
 594 +    }
 595 +    if (runstate_is_running()) {
 596 +        vm_stop(RUN_STATE_SAVE_VM);
 597 +    }
 598 +    return 0;
 599 +}
 600 +
 601 +static void process_savevm_co(void *opaque)
 602 +{
 603 +    int ret;
 604 +    int64_t maxlen;
 605 +    MigrationParams params = {
 606 +        .blk = 0,
 607 +        .shared = 0
 608 +    };
 609 +
 610 +    snap_state.state = SAVE_STATE_ACTIVE;
 611 +
 612 +    qemu_mutex_unlock_iothread();
 613 +    qemu_savevm_state_header(snap_state.file);
 614 +    ret = qemu_savevm_state_begin(snap_state.file, &params);
 615 +    qemu_mutex_lock_iothread();
 616 +
 617 +    if (ret < 0) {
 618 +        save_snapshot_error("qemu_savevm_state_begin failed");
 619 +        return;
 620 +    }
 621 +
 622 +    while (snap_state.state == SAVE_STATE_ACTIVE) {
 623 +        uint64_t pending_size;
 624 +
 625 +        pending_size = qemu_savevm_state_pending(snap_state.file, 0);
 626 +
 627 +        if (pending_size) {
 628 +                ret = qemu_savevm_state_iterate(snap_state.file);
 629 +                if (ret < 0) {
 630 +                    save_snapshot_error("qemu_savevm_state_iterate error %d", ret);
 631 +                    break;
 632 +                }
 633 +                DPRINTF("savevm inerate pending size %lu ret %d\n", pending_size, ret);
 634 +        } else {
 635 +            DPRINTF("done iterating\n");
 636 +            if (store_and_stop())
 637 +                break;
 638 +            DPRINTF("savevm inerate finished\n");
 639 +            qemu_savevm_state_complete_precopy(snap_state.file);
 640 +            DPRINTF("save complete\n");
 641 +            save_snapshot_completed();
 642 +            break;
 643 +        }
 644 +
 645 +        /* stop the VM if we get to the end of available space,
 646 +         * or if pending_size is just a few MB
 647 +         */
 648 +        maxlen = bdrv_getlength(snap_state.bs) - 30*1024*1024;
 649 +        if ((pending_size < 100000) ||
 650 +            ((snap_state.bs_pos + pending_size) >= maxlen)) {
 651 +            if (store_and_stop())
 652 +                break;
 653 +        }
 654 +    }
 655 +
 656 +    if(snap_state.state == SAVE_STATE_CANCELLED) {
 657 +        save_snapshot_completed();
 658 +        Error *errp = NULL;
 659 +        qmp_savevm_end(&errp);
 660 +    }
 661 +
 662 +}
 663 +
 664 +static const QEMUFileOps block_file_ops = {
 665 +    .put_buffer =     block_state_put_buffer,
 666 +    .close =          block_state_close,
 667 +};
 668 +
 669 +
 670 +void qmp_savevm_start(bool has_statefile, const char *statefile, Error **errp)
 671 +{
 672 +    BlockDriver *drv = NULL;
 673 +    Error *local_err = NULL;
 674 +
 675 +    int bdrv_oflags = BDRV_O_RDWR | BDRV_O_RESIZE;
 676 +    int ret;
 677 +
 678 +    if (snap_state.state != SAVE_STATE_DONE) {
 679 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 680 +                  "VM snapshot already started\n");
 681 +        return;
 682 +    }
 683 +
 684 +    /* initialize snapshot info */
 685 +    snap_state.saved_vm_running = runstate_is_running();
 686 +    snap_state.bs_pos = 0;
 687 +    snap_state.total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 688 +    snap_state.blocker = NULL;
 689 +
 690 +    if (snap_state.error) {
 691 +        error_free(snap_state.error);
 692 +        snap_state.error = NULL;
 693 +    }
 694 +
 695 +    if (!has_statefile) {
 696 +        vm_stop(RUN_STATE_SAVE_VM);
 697 +        snap_state.state = SAVE_STATE_COMPLETED;
 698 +        return;
 699 +    }
 700 +
 701 +    if (qemu_savevm_state_blocked(errp)) {
 702 +        return;
 703 +    }
 704 +
 705 +    /* Open the image */
 706 +    snap_state.bs = bdrv_new();
 707 +
 708 +    QDict *options = NULL;
 709 +    options = qdict_new();
 710 +    qdict_put(options, "driver", qstring_from_str("raw"));
 711 +    ret = bdrv_open(&snap_state.bs, statefile, NULL, options, bdrv_oflags, drv, &local_err);
 712 +    if (ret < 0) {
 713 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "failed to open '%s'", statefile);
 714 +        goto restart;
 715 +    }
 716 +
 717 +    snap_state.file = qemu_fopen_ops(&snap_state, &block_file_ops);
 718 +
 719 +    if (!snap_state.file) {
 720 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "failed to open '%s'", statefile);
 721 +        goto restart;
 722 +    }
 723 +
 724 +
 725 +    error_setg(&snap_state.blocker, "block device is in use by savevm");
 726 +    bdrv_op_block_all(snap_state.bs, snap_state.blocker);
 727 +
 728 +    Coroutine *co = qemu_coroutine_create(process_savevm_co);
 729 +    qemu_coroutine_enter(co);
 730 +
 731 +    return;
 732 +
 733 +restart:
 734 +
 735 +    save_snapshot_error("setup failed");
 736 +
 737 +    if (snap_state.saved_vm_running) {
 738 +        vm_start();
 739 +    }
 740 +}
 741 +
 742 +void qmp_savevm_end(Error **errp)
 743 +{
 744 +    if (snap_state.state == SAVE_STATE_DONE) {
 745 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 746 +                  "VM snapshot not started\n");
 747 +        return;
 748 +    }
 749 +
 750 +    if (snap_state.state == SAVE_STATE_ACTIVE) {
 751 +        snap_state.state = SAVE_STATE_CANCELLED;
 752 +        return;
 753 +    }
 754 +
 755 +    if (snap_state.saved_vm_running) {
 756 +        vm_start();
 757 +    }
 758 +
 759 +    snap_state.state = SAVE_STATE_DONE;
 760 +}
 761 +
 762 +void qmp_snapshot_drive(const char *device, const char *name, Error **errp)
 763 +{
 764 +    BlockBackend *blk;
 765 +    BlockDriverState *bs;
 766 +    QEMUSnapshotInfo sn1, *sn = &sn1;
 767 +    int ret;
 768 +#ifdef _WIN32
 769 +    struct _timeb tb;
 770 +#else
 771 +    struct timeval tv;
 772 +#endif
 773 +
 774 +    if (snap_state.state != SAVE_STATE_COMPLETED) {
 775 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 776 +                  "VM snapshot not ready/started\n");
 777 +        return;
 778 +    }
 779 +
 780 +    blk = blk_by_name(device);
 781 +    if (!blk) {
 782 +        error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
 783 +                  "Device '%s' not found", device);
 784 +        return;
 785 +    }
 786 +
 787 +    bs = blk_bs(blk);
 788 +    if (!bdrv_is_inserted(bs)) {
 789 +        error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
 790 +        return;
 791 +    }
 792 +
 793 +    if (bdrv_is_read_only(bs)) {
 794 +        error_setg(errp, "Node '%s' is read only", device);
 795 +        return;
 796 +    }
 797 +
 798 +    if (!bdrv_can_snapshot(bs)) {
 799 +        error_setg(errp, QERR_UNSUPPORTED);
 800 +        return;
 801 +    }
 802 +
 803 +    if (bdrv_snapshot_find(bs, sn, name) >= 0) {
 804 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 805 +                  "snapshot '%s' already exists", name);
 806 +        return;
 807 +    }
 808 +
 809 +    sn = &sn1;
 810 +    memset(sn, 0, sizeof(*sn));
 811 +
 812 +#ifdef _WIN32
 813 +    _ftime(&tb);
 814 +    sn->date_sec = tb.time;
 815 +    sn->date_nsec = tb.millitm * 1000000;
 816 +#else
 817 +    gettimeofday(&tv, NULL);
 818 +    sn->date_sec = tv.tv_sec;
 819 +    sn->date_nsec = tv.tv_usec * 1000;
 820 +#endif
 821 +    sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
 822 +
 823 +    pstrcpy(sn->name, sizeof(sn->name), name);
 824 +
 825 +    sn->vm_state_size = 0; /* do not save state */
 826 +
 827 +    ret = bdrv_snapshot_create(bs, sn);
 828 +    if (ret < 0) {
 829 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 830 +                  "Error while creating snapshot on '%s'\n", device);
 831 +        return;
 832 +    }
 833 +}
 834 +
 835 +void qmp_delete_drive_snapshot(const char *device, const char *name,
 836 +                               Error **errp)
 837 +{
 838 +    BlockBackend *blk;
 839 +    BlockDriverState *bs;
 840 +    QEMUSnapshotInfo sn1, *sn = &sn1;
 841 +    Error *local_err = NULL;
 842 +
 843 +    int ret;
 844 +
 845 +    blk = blk_by_name(device);
 846 +    if (!blk) {
 847 +        error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
 848 +                  "Device '%s' not found", device);
 849 +        return;
 850 +    }
 851 +
 852 +    bs = blk_bs(blk);
 853 +    if (bdrv_is_read_only(bs)) {
 854 +        error_setg(errp, "Node '%s' is read only", device);
 855 +        return;
 856 +    }
 857 +
 858 +    if (!bdrv_can_snapshot(bs)) {
 859 +        error_setg(errp, QERR_UNSUPPORTED);
 860 +        return;
 861 +    }
 862 +
 863 +    if (bdrv_snapshot_find(bs, sn, name) < 0) {
 864 +        /* return success if snapshot does not exists */
 865 +        return;
 866 +    }
 867 +
 868 +    ret = bdrv_snapshot_delete(bs, NULL, name, &local_err);
 869 +    if (ret < 0) {
 870 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 871 +                  "Error while deleting snapshot on '%s'\n", device);
 872 +        return;
 873 +    }
 874 +}
 875 +
 876 +static int loadstate_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
 877 +                                int size)
 878 +{
 879 +    BlockDriverState *bs = (BlockDriverState *)opaque;
 880 +    int64_t maxlen = bdrv_getlength(bs);
 881 +    if (pos > maxlen) {
 882 +        return -EIO;
 883 +    }
 884 +    if ((pos + size) > maxlen) {
 885 +        size = maxlen - pos - 1;
 886 +    }
 887 +    if (size == 0) {
 888 +        return 0;
 889 +    }
 890 +    return bdrv_pread(bs, pos, buf, size);
 891 +}
 892 +
 893 +static const QEMUFileOps loadstate_file_ops = {
 894 +    .get_buffer = loadstate_get_buffer,
 895 +};
 896 +
 897 +int load_state_from_blockdev(const char *filename)
 898 +{
 899 +    BlockDriverState *bs = NULL;
 900 +    BlockDriver *drv = NULL;
 901 +    Error *local_err = NULL;
 902 +    Error *blocker = NULL;
 903 +
 904 +    QEMUFile *f;
 905 +    int ret;
 906 +
 907 +    bs = bdrv_new();
 908 +    ret = bdrv_open(&bs, filename, NULL, NULL, 0, drv, &local_err);
 909 +    error_setg(&blocker, "block device is in use by load state");
 910 +    bdrv_op_block_all(bs, blocker);
 911 +
 912 +    if (ret < 0) {
 913 +        error_report("Could not open VM state file");
 914 +        goto the_end;
 915 +    }
 916 +
 917 +    /* restore the VM state */
 918 +    f = qemu_fopen_ops(bs, &loadstate_file_ops);
 919 +    if (!f) {
 920 +        error_report("Could not open VM state file");
 921 +        ret = -EINVAL;
 922 +        goto the_end;
 923 +    }
 924 +
 925 +    qemu_system_reset(VMRESET_SILENT);
 926 +    ret = qemu_loadvm_state(f);
 927 +
 928 +    qemu_fclose(f);
 929 +    migration_incoming_state_destroy();
 930 +    if (ret < 0) {
 931 +        error_report("Error %d while loading VM state", ret);
 932 +        goto the_end;
 933 +    }
 934 +
 935 +    ret = 0;
 936 +
 937 + the_end:
 938 +    if (bs) {
 939 +        bdrv_op_unblock_all(bs, blocker);
 940 +        error_free(blocker);
 941 +        bdrv_unref(bs);
 942 +    }
 943 +    return ret;
 944 +}
 945 diff --git a/vl.c b/vl.c
 946 index 868c489..19afd47 100644
 947 --- a/vl.c
 948 +++ b/vl.c
 949 @@ -2960,6 +2960,7 @@ int main(int argc, char **argv, char **envp)
 950      int optind;
 951      const char *optarg;
 952      const char *loadvm = NULL;
 953 +    const char *loadstate = NULL;
 954      MachineClass *machine_class;
 955      const char *cpu_model;
 956      const char *vga_model = NULL;
 957 @@ -3631,6 +3632,9 @@ int main(int argc, char **argv, char **envp)
 958              case QEMU_OPTION_loadvm:
 959                  loadvm = optarg;
 960                  break;
 961 +            case QEMU_OPTION_loadstate:
 962 +                loadstate = optarg;
 963 +                break;
 964              case QEMU_OPTION_full_screen:
 965                  full_screen = 1;
 966                  break;
 967 @@ -4689,6 +4693,10 @@ int main(int argc, char **argv, char **envp)
 968          if (load_vmstate(loadvm) < 0) {
 969              autostart = 0;
 970          }
 971 +    } else if (loadstate) {
 972 +        if (load_state_from_blockdev(loadstate) < 0) {
 973 +            autostart = 0;
 974 +        }
 975      }
 976
 977      qdev_prop_check_globals();
 978 --
 979 2.1.4
 980