debian/patches/pve/0017-PVE-add-savevm-async-for-background-state-snapshots.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: Dietmar Maurer <dietmar@proxmox.com>
   3 Date: Mon, 6 Apr 2020 12:16:46 +0200
   4 Subject: [PATCH] PVE: add savevm-async for background state snapshots
   5
   6 Put qemu_savevm_state_{header,setup} into the main loop and the rest
   7 of the iteration into a coroutine. The former need to lock the
   8 iothread (and we can't unlock it in the coroutine), and the latter
   9 can't deal with being in a separate thread, so a coroutine it must
  10 be.
  11
  12 Truncate output file at 1024 boundary.
  13
  14 Do not block the VM and save the state on aborting a snapshot, as the
  15 snapshot will be invalid anyway.
  16
  17 Also, when aborting, wait for the target file to be closed, otherwise a
  18 client might run into race-conditions when trying to remove the file
  19 still opened by QEMU.
  20
  21 Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
  22 Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
  23 Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
  24 [improve aborting]
  25 Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
  26 [FE: further improve aborting
  27      adapt to removal of QEMUFileOps
  28      improve condition for entering final stage
  29      adapt to QAPI and other changes for 8.0]
  30 Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
  31 ---
  32  hmp-commands-info.hx         |  13 +
  33  hmp-commands.hx              |  33 +++
  34  include/migration/snapshot.h |   2 +
  35  include/monitor/hmp.h        |   5 +
  36  migration/meson.build        |   1 +
  37  migration/savevm-async.c     | 535 +++++++++++++++++++++++++++++++++++
  38  monitor/hmp-cmds.c           |  58 ++++
  39  qapi/migration.json          |  34 +++
  40  qapi/misc.json               |  32 +++
  41  qemu-options.hx              |  12 +
  42  softmmu/vl.c                 |  10 +
  43  11 files changed, 735 insertions(+)
  44  create mode 100644 migration/savevm-async.c
  45
  46 diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
  47 index 47d63d26db..a166bff3d5 100644
  48 --- a/hmp-commands-info.hx
  49 +++ b/hmp-commands-info.hx
  50 @@ -540,6 +540,19 @@ SRST
  51      Show current migration parameters.
  52  ERST
  53
  54 +    {
  55 +        .name       = "savevm",
  56 +        .args_type  = "",
  57 +        .params     = "",
  58 +        .help       = "show savevm status",
  59 +        .cmd = hmp_info_savevm,
  60 +    },
  61 +
  62 +SRST
  63 +  ``info savevm``
  64 +    Show savevm status.
  65 +ERST
  66 +
  67      {
  68          .name       = "balloon",
  69          .args_type  = "",
  70 diff --git a/hmp-commands.hx b/hmp-commands.hx
  71 index bb85ee1d26..b66d7fc4ab 100644
  72 --- a/hmp-commands.hx
  73 +++ b/hmp-commands.hx
  74 @@ -1846,3 +1846,36 @@ SRST
  75    List event channels in the guest
  76  ERST
  77  #endif
  78 +
  79 +    {
  80 +        .name       = "savevm-start",
  81 +        .args_type  = "statefile:s?",
  82 +        .params     = "[statefile]",
  83 +        .help       = "Prepare for snapshot and halt VM. Save VM state to statefile.",
  84 +        .cmd = hmp_savevm_start,
  85 +    },
  86 +
  87 +    {
  88 +        .name       = "snapshot-drive",
  89 +        .args_type  = "device:s,name:s",
  90 +        .params     = "device name",
  91 +        .help       = "Create internal snapshot.",
  92 +        .cmd = hmp_snapshot_drive,
  93 +    },
  94 +
  95 +    {
  96 +        .name       = "delete-drive-snapshot",
  97 +        .args_type  = "device:s,name:s",
  98 +        .params     = "device name",
  99 +        .help       = "Delete internal snapshot.",
 100 +        .cmd = hmp_delete_drive_snapshot,
 101 +    },
 102 +
 103 +    {
 104 +        .name       = "savevm-end",
 105 +        .args_type  = "",
 106 +        .params     = "",
 107 +        .help       = "Resume VM after snaphot.",
 108 +        .cmd        = hmp_savevm_end,
 109 +        .coroutine  = true,
 110 +    },
 111 diff --git a/include/migration/snapshot.h b/include/migration/snapshot.h
 112 index e72083b117..c846d37806 100644
 113 --- a/include/migration/snapshot.h
 114 +++ b/include/migration/snapshot.h
 115 @@ -61,4 +61,6 @@ bool delete_snapshot(const char *name,
 116                      bool has_devices, strList *devices,
 117                      Error **errp);
 118
 119 +int load_snapshot_from_blockdev(const char *filename, Error **errp);
 120 +
 121  #endif
 122 diff --git a/include/monitor/hmp.h b/include/monitor/hmp.h
 123 index fdb69b7f9c..c012bad741 100644
 124 --- a/include/monitor/hmp.h
 125 +++ b/include/monitor/hmp.h
 126 @@ -28,6 +28,7 @@ void hmp_info_status(Monitor *mon, const QDict *qdict);
 127  void hmp_info_uuid(Monitor *mon, const QDict *qdict);
 128  void hmp_info_chardev(Monitor *mon, const QDict *qdict);
 129  void hmp_info_mice(Monitor *mon, const QDict *qdict);
 130 +void hmp_info_savevm(Monitor *mon, const QDict *qdict);
 131  void hmp_info_migrate(Monitor *mon, const QDict *qdict);
 132  void hmp_info_migrate_capabilities(Monitor *mon, const QDict *qdict);
 133  void hmp_info_migrate_parameters(Monitor *mon, const QDict *qdict);
 134 @@ -94,6 +95,10 @@ void hmp_closefd(Monitor *mon, const QDict *qdict);
 135  void hmp_mouse_move(Monitor *mon, const QDict *qdict);
 136  void hmp_mouse_button(Monitor *mon, const QDict *qdict);
 137  void hmp_mouse_set(Monitor *mon, const QDict *qdict);
 138 +void hmp_savevm_start(Monitor *mon, const QDict *qdict);
 139 +void hmp_snapshot_drive(Monitor *mon, const QDict *qdict);
 140 +void hmp_delete_drive_snapshot(Monitor *mon, const QDict *qdict);
 141 +void hmp_savevm_end(Monitor *mon, const QDict *qdict);
 142  void hmp_sendkey(Monitor *mon, const QDict *qdict);
 143  void coroutine_fn hmp_screendump(Monitor *mon, const QDict *qdict);
 144  void hmp_chardev_add(Monitor *mon, const QDict *qdict);
 145 diff --git a/migration/meson.build b/migration/meson.build
 146 index 8a142fc7a9..a7824b5266 100644
 147 --- a/migration/meson.build
 148 +++ b/migration/meson.build
 149 @@ -25,6 +25,7 @@ softmmu_ss.add(files(
 150    'multifd-zlib.c',
 151    'postcopy-ram.c',
 152    'savevm.c',
 153 +  'savevm-async.c',
 154    'socket.c',
 155    'tls.c',
 156    'threadinfo.c',
 157 diff --git a/migration/savevm-async.c b/migration/savevm-async.c
 158 new file mode 100644
 159 index 0000000000..24660af014
 160 --- /dev/null
 161 +++ b/migration/savevm-async.c
 162 @@ -0,0 +1,535 @@
 163 +#include "qemu/osdep.h"
 164 +#include "migration/channel-savevm-async.h"
 165 +#include "migration/migration.h"
 166 +#include "migration/savevm.h"
 167 +#include "migration/snapshot.h"
 168 +#include "migration/global_state.h"
 169 +#include "migration/ram.h"
 170 +#include "migration/qemu-file.h"
 171 +#include "sysemu/sysemu.h"
 172 +#include "sysemu/runstate.h"
 173 +#include "block/block.h"
 174 +#include "sysemu/block-backend.h"
 175 +#include "qapi/error.h"
 176 +#include "qapi/qmp/qerror.h"
 177 +#include "qapi/qmp/qdict.h"
 178 +#include "qapi/qapi-commands-migration.h"
 179 +#include "qapi/qapi-commands-misc.h"
 180 +#include "qapi/qapi-commands-block.h"
 181 +#include "qemu/cutils.h"
 182 +#include "qemu/timer.h"
 183 +#include "qemu/main-loop.h"
 184 +#include "qemu/rcu.h"
 185 +
 186 +/* #define DEBUG_SAVEVM_STATE */
 187 +
 188 +#ifdef DEBUG_SAVEVM_STATE
 189 +#define DPRINTF(fmt, ...) \
 190 +    do { printf("savevm-async: " fmt, ## __VA_ARGS__); } while (0)
 191 +#else
 192 +#define DPRINTF(fmt, ...) \
 193 +    do { } while (0)
 194 +#endif
 195 +
 196 +enum {
 197 +    SAVE_STATE_DONE,
 198 +    SAVE_STATE_ERROR,
 199 +    SAVE_STATE_ACTIVE,
 200 +    SAVE_STATE_COMPLETED,
 201 +    SAVE_STATE_CANCELLED
 202 +};
 203 +
 204 +
 205 +static struct SnapshotState {
 206 +    BlockBackend *target;
 207 +    size_t bs_pos;
 208 +    int state;
 209 +    Error *error;
 210 +    Error *blocker;
 211 +    int saved_vm_running;
 212 +    QEMUFile *file;
 213 +    int64_t total_time;
 214 +    QEMUBH *finalize_bh;
 215 +    Coroutine *co;
 216 +    QemuCoSleep target_close_wait;
 217 +} snap_state;
 218 +
 219 +static bool savevm_aborted(void)
 220 +{
 221 +    return snap_state.state == SAVE_STATE_CANCELLED ||
 222 +        snap_state.state == SAVE_STATE_ERROR;
 223 +}
 224 +
 225 +SaveVMInfo *qmp_query_savevm(Error **errp)
 226 +{
 227 +    SaveVMInfo *info = g_malloc0(sizeof(*info));
 228 +    struct SnapshotState *s = &snap_state;
 229 +
 230 +    if (s->state != SAVE_STATE_DONE) {
 231 +        info->has_bytes = true;
 232 +        info->bytes = s->bs_pos;
 233 +        switch (s->state) {
 234 +        case SAVE_STATE_ERROR:
 235 +            info->status = g_strdup("failed");
 236 +            info->has_total_time = true;
 237 +            info->total_time = s->total_time;
 238 +            if (s->error) {
 239 +                info->error = g_strdup(error_get_pretty(s->error));
 240 +            }
 241 +            break;
 242 +        case SAVE_STATE_ACTIVE:
 243 +            info->status = g_strdup("active");
 244 +            info->has_total_time = true;
 245 +            info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 246 +                - s->total_time;
 247 +            break;
 248 +        case SAVE_STATE_COMPLETED:
 249 +            info->status = g_strdup("completed");
 250 +            info->has_total_time = true;
 251 +            info->total_time = s->total_time;
 252 +            break;
 253 +        }
 254 +    }
 255 +
 256 +    return info;
 257 +}
 258 +
 259 +static int save_snapshot_cleanup(void)
 260 +{
 261 +    int ret = 0;
 262 +
 263 +    DPRINTF("save_snapshot_cleanup\n");
 264 +
 265 +    snap_state.total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
 266 +        snap_state.total_time;
 267 +
 268 +    if (snap_state.file) {
 269 +        ret = qemu_fclose(snap_state.file);
 270 +        snap_state.file = NULL;
 271 +    }
 272 +
 273 +    if (snap_state.target) {
 274 +        if (!savevm_aborted()) {
 275 +            /* try to truncate, but ignore errors (will fail on block devices).
 276 +            * note1: bdrv_read() need whole blocks, so we need to round up
 277 +            * note2: PVE requires 1024 (BDRV_SECTOR_SIZE*2) alignment
 278 +            */
 279 +            size_t size = QEMU_ALIGN_UP(snap_state.bs_pos, BDRV_SECTOR_SIZE*2);
 280 +            blk_truncate(snap_state.target, size, false, PREALLOC_MODE_OFF, 0, NULL);
 281 +        }
 282 +        blk_op_unblock_all(snap_state.target, snap_state.blocker);
 283 +        error_free(snap_state.blocker);
 284 +        snap_state.blocker = NULL;
 285 +        blk_unref(snap_state.target);
 286 +        snap_state.target = NULL;
 287 +
 288 +        qemu_co_sleep_wake(&snap_state.target_close_wait);
 289 +    }
 290 +
 291 +    return ret;
 292 +}
 293 +
 294 +static void save_snapshot_error(const char *fmt, ...)
 295 +{
 296 +    va_list ap;
 297 +    char *msg;
 298 +
 299 +    va_start(ap, fmt);
 300 +    msg = g_strdup_vprintf(fmt, ap);
 301 +    va_end(ap);
 302 +
 303 +    DPRINTF("save_snapshot_error: %s\n", msg);
 304 +
 305 +    if (!snap_state.error) {
 306 +        error_set(&snap_state.error, ERROR_CLASS_GENERIC_ERROR, "%s", msg);
 307 +    }
 308 +
 309 +    g_free (msg);
 310 +
 311 +    snap_state.state = SAVE_STATE_ERROR;
 312 +}
 313 +
 314 +static void process_savevm_finalize(void *opaque)
 315 +{
 316 +    int ret;
 317 +    AioContext *iohandler_ctx = iohandler_get_aio_context();
 318 +    MigrationState *ms = migrate_get_current();
 319 +
 320 +    bool aborted = savevm_aborted();
 321 +
 322 +#ifdef DEBUG_SAVEVM_STATE
 323 +    int64_t start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 324 +#endif
 325 +
 326 +    qemu_bh_delete(snap_state.finalize_bh);
 327 +    snap_state.finalize_bh = NULL;
 328 +    snap_state.co = NULL;
 329 +
 330 +    /* We need to own the target bdrv's context for the following functions,
 331 +     * so move it back. It can stay in the main context and live out its live
 332 +     * there, since we're done with it after this method ends anyway.
 333 +     */
 334 +    aio_context_acquire(iohandler_ctx);
 335 +    blk_set_aio_context(snap_state.target, qemu_get_aio_context(), NULL);
 336 +    aio_context_release(iohandler_ctx);
 337 +
 338 +    ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 339 +    if (ret < 0) {
 340 +        save_snapshot_error("vm_stop_force_state error %d", ret);
 341 +    }
 342 +
 343 +    if (!aborted) {
 344 +        /* skip state saving if we aborted, snapshot will be invalid anyway */
 345 +        (void)qemu_savevm_state_complete_precopy(snap_state.file, false, false);
 346 +        ret = qemu_file_get_error(snap_state.file);
 347 +        if (ret < 0) {
 348 +            save_snapshot_error("qemu_savevm_state_complete_precopy error %d", ret);
 349 +        }
 350 +    }
 351 +
 352 +    DPRINTF("state saving complete\n");
 353 +    DPRINTF("timing: process_savevm_finalize (state saving) took %ld ms\n",
 354 +        qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - start_time);
 355 +
 356 +    /* clear migration state */
 357 +    migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP,
 358 +        ret || aborted ? MIGRATION_STATUS_FAILED : MIGRATION_STATUS_COMPLETED);
 359 +    ms->to_dst_file = NULL;
 360 +
 361 +    qemu_savevm_state_cleanup();
 362 +
 363 +    ret = save_snapshot_cleanup();
 364 +    if (ret < 0) {
 365 +        save_snapshot_error("save_snapshot_cleanup error %d", ret);
 366 +    } else if (snap_state.state == SAVE_STATE_ACTIVE) {
 367 +        snap_state.state = SAVE_STATE_COMPLETED;
 368 +    } else if (aborted) {
 369 +        /*
 370 +         * If there was an error, there's no need to set a new one here.
 371 +         * If the snapshot was canceled, leave setting the state to
 372 +         * qmp_savevm_end(), which is waked by save_snapshot_cleanup().
 373 +         */
 374 +    } else {
 375 +        save_snapshot_error("process_savevm_cleanup: invalid state: %d",
 376 +                            snap_state.state);
 377 +    }
 378 +    if (snap_state.saved_vm_running) {
 379 +        vm_start();
 380 +        snap_state.saved_vm_running = false;
 381 +    }
 382 +
 383 +    DPRINTF("timing: process_savevm_finalize (full) took %ld ms\n",
 384 +        qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - start_time);
 385 +}
 386 +
 387 +static void coroutine_fn process_savevm_co(void *opaque)
 388 +{
 389 +    int ret;
 390 +    int64_t maxlen;
 391 +    BdrvNextIterator it;
 392 +    BlockDriverState *bs = NULL;
 393 +
 394 +#ifdef DEBUG_SAVEVM_STATE
 395 +    int64_t start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 396 +#endif
 397 +
 398 +    ret = qemu_file_get_error(snap_state.file);
 399 +    if (ret < 0) {
 400 +        save_snapshot_error("qemu_savevm_state_setup failed");
 401 +        return;
 402 +    }
 403 +
 404 +    while (snap_state.state == SAVE_STATE_ACTIVE) {
 405 +        uint64_t pending_size, pend_precopy, pend_postcopy;
 406 +
 407 +        /* pending is expected to be called without iothread lock */
 408 +        qemu_mutex_unlock_iothread();
 409 +        qemu_savevm_state_pending_exact(&pend_precopy, &pend_postcopy);
 410 +        qemu_mutex_lock_iothread();
 411 +
 412 +        pending_size = pend_precopy + pend_postcopy;
 413 +
 414 +        /*
 415 +         * A guest reaching this cutoff is dirtying lots of RAM. It should be
 416 +         * large enough so that the guest can't dirty this much between the
 417 +         * check and the guest actually being stopped, but it should be small
 418 +         * enough to avoid long downtimes for non-hibernation snapshots.
 419 +         */
 420 +        maxlen = blk_getlength(snap_state.target) - 100*1024*1024;
 421 +
 422 +        /* Note that there is no progress for pend_postcopy when iterating */
 423 +        if (pending_size - pend_postcopy > 400000 && snap_state.bs_pos + pending_size < maxlen) {
 424 +            ret = qemu_savevm_state_iterate(snap_state.file, false);
 425 +            if (ret < 0) {
 426 +                save_snapshot_error("qemu_savevm_state_iterate error %d", ret);
 427 +                break;
 428 +            }
 429 +            DPRINTF("savevm iterate pending size %lu ret %d\n", pending_size, ret);
 430 +        } else {
 431 +            qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
 432 +            ret = global_state_store();
 433 +            if (ret) {
 434 +                save_snapshot_error("global_state_store error %d", ret);
 435 +                break;
 436 +            }
 437 +
 438 +            DPRINTF("savevm iterate complete\n");
 439 +            break;
 440 +        }
 441 +    }
 442 +
 443 +    DPRINTF("timing: process_savevm_co took %ld ms\n",
 444 +        qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - start_time);
 445 +
 446 +#ifdef DEBUG_SAVEVM_STATE
 447 +    int64_t start_time_flush = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 448 +#endif
 449 +    /* If a drive runs in an IOThread we can flush it async, and only
 450 +     * need to sync-flush whatever IO happens between now and
 451 +     * vm_stop_force_state. bdrv_next can only be called from main AioContext,
 452 +     * so move there now and after every flush.
 453 +     */
 454 +    aio_co_reschedule_self(qemu_get_aio_context());
 455 +    for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
 456 +        /* target has BDRV_O_NO_FLUSH, no sense calling bdrv_flush on it */
 457 +        if (bs == blk_bs(snap_state.target)) {
 458 +            continue;
 459 +        }
 460 +
 461 +        AioContext *bs_ctx = bdrv_get_aio_context(bs);
 462 +        if (bs_ctx != qemu_get_aio_context()) {
 463 +            DPRINTF("savevm: async flushing drive %s\n", bs->filename);
 464 +            aio_co_reschedule_self(bs_ctx);
 465 +            bdrv_graph_co_rdlock();
 466 +            bdrv_flush(bs);
 467 +            bdrv_graph_co_rdunlock();
 468 +            aio_co_reschedule_self(qemu_get_aio_context());
 469 +        }
 470 +    }
 471 +
 472 +    DPRINTF("timing: async flushing took %ld ms\n",
 473 +        qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - start_time_flush);
 474 +
 475 +    qemu_bh_schedule(snap_state.finalize_bh);
 476 +}
 477 +
 478 +void qmp_savevm_start(const char *statefile, Error **errp)
 479 +{
 480 +    Error *local_err = NULL;
 481 +    MigrationState *ms = migrate_get_current();
 482 +    AioContext *iohandler_ctx = iohandler_get_aio_context();
 483 +
 484 +    int bdrv_oflags = BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH;
 485 +
 486 +    if (snap_state.state != SAVE_STATE_DONE) {
 487 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 488 +                  "VM snapshot already started\n");
 489 +        return;
 490 +    }
 491 +
 492 +    if (migration_is_running(ms->state)) {
 493 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, QERR_MIGRATION_ACTIVE);
 494 +        return;
 495 +    }
 496 +
 497 +    if (migrate_use_block()) {
 498 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 499 +                  "Block migration and snapshots are incompatible");
 500 +        return;
 501 +    }
 502 +
 503 +    /* initialize snapshot info */
 504 +    snap_state.saved_vm_running = runstate_is_running();
 505 +    snap_state.bs_pos = 0;
 506 +    snap_state.total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 507 +    snap_state.blocker = NULL;
 508 +    snap_state.target_close_wait = (QemuCoSleep){ .to_wake = NULL };
 509 +
 510 +    if (snap_state.error) {
 511 +        error_free(snap_state.error);
 512 +        snap_state.error = NULL;
 513 +    }
 514 +
 515 +    if (!statefile) {
 516 +        vm_stop(RUN_STATE_SAVE_VM);
 517 +        snap_state.state = SAVE_STATE_COMPLETED;
 518 +        return;
 519 +    }
 520 +
 521 +    if (qemu_savevm_state_blocked(errp)) {
 522 +        return;
 523 +    }
 524 +
 525 +    /* Open the image */
 526 +    QDict *options = NULL;
 527 +    options = qdict_new();
 528 +    qdict_put_str(options, "driver", "raw");
 529 +    snap_state.target = blk_new_open(statefile, NULL, options, bdrv_oflags, &local_err);
 530 +    if (!snap_state.target) {
 531 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "failed to open '%s'", statefile);
 532 +        goto restart;
 533 +    }
 534 +
 535 +    QIOChannel *ioc = QIO_CHANNEL(qio_channel_savevm_async_new(snap_state.target,
 536 +                                                               &snap_state.bs_pos));
 537 +    snap_state.file = qemu_file_new_output(ioc);
 538 +
 539 +    if (!snap_state.file) {
 540 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "failed to open '%s'", statefile);
 541 +        goto restart;
 542 +    }
 543 +
 544 +    /*
 545 +     * qemu_savevm_* paths use migration code and expect a migration state.
 546 +     * State is cleared in process_savevm_co, but has to be initialized
 547 +     * here (blocking main thread, from QMP) to avoid race conditions.
 548 +     */
 549 +    migrate_init(ms);
 550 +    memset(&ram_counters, 0, sizeof(ram_counters));
 551 +    ms->to_dst_file = snap_state.file;
 552 +
 553 +    error_setg(&snap_state.blocker, "block device is in use by savevm");
 554 +    blk_op_block_all(snap_state.target, snap_state.blocker);
 555 +
 556 +    snap_state.state = SAVE_STATE_ACTIVE;
 557 +    snap_state.finalize_bh = qemu_bh_new(process_savevm_finalize, &snap_state);
 558 +    snap_state.co = qemu_coroutine_create(&process_savevm_co, NULL);
 559 +    qemu_mutex_unlock_iothread();
 560 +    qemu_savevm_state_header(snap_state.file);
 561 +    qemu_savevm_state_setup(snap_state.file);
 562 +    qemu_mutex_lock_iothread();
 563 +
 564 +    /* Async processing from here on out happens in iohandler context, so let
 565 +     * the target bdrv have its home there.
 566 +     */
 567 +    blk_set_aio_context(snap_state.target, iohandler_ctx, &local_err);
 568 +
 569 +    aio_co_schedule(iohandler_ctx, snap_state.co);
 570 +
 571 +    return;
 572 +
 573 +restart:
 574 +
 575 +    save_snapshot_error("setup failed");
 576 +
 577 +    if (snap_state.saved_vm_running) {
 578 +        vm_start();
 579 +        snap_state.saved_vm_running = false;
 580 +    }
 581 +}
 582 +
 583 +void coroutine_fn qmp_savevm_end(Error **errp)
 584 +{
 585 +    int64_t timeout;
 586 +
 587 +    if (snap_state.state == SAVE_STATE_DONE) {
 588 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 589 +                  "VM snapshot not started\n");
 590 +        return;
 591 +    }
 592 +
 593 +    if (snap_state.state == SAVE_STATE_ACTIVE) {
 594 +        snap_state.state = SAVE_STATE_CANCELLED;
 595 +        goto wait_for_close;
 596 +    }
 597 +
 598 +    if (snap_state.saved_vm_running) {
 599 +        vm_start();
 600 +        snap_state.saved_vm_running = false;
 601 +    }
 602 +
 603 +    snap_state.state = SAVE_STATE_DONE;
 604 +
 605 +wait_for_close:
 606 +    if (!snap_state.target) {
 607 +        DPRINTF("savevm-end: no target file open\n");
 608 +        return;
 609 +    }
 610 +
 611 +    /* wait until cleanup is done before returning, this ensures that after this
 612 +     * call exits the statefile will be closed and can be removed immediately */
 613 +    DPRINTF("savevm-end: waiting for cleanup\n");
 614 +    timeout = 30L * 1000 * 1000 * 1000;
 615 +    qemu_co_sleep_ns_wakeable(&snap_state.target_close_wait,
 616 +                              QEMU_CLOCK_REALTIME, timeout);
 617 +    if (snap_state.target) {
 618 +        save_snapshot_error("timeout waiting for target file close in "
 619 +                            "qmp_savevm_end");
 620 +        /* we cannot assume the snapshot finished in this case, so leave the
 621 +         * state alone - caller has to figure something out */
 622 +        return;
 623 +    }
 624 +
 625 +    // File closed and no other error, so ensure next snapshot can be started.
 626 +    if (snap_state.state != SAVE_STATE_ERROR) {
 627 +        snap_state.state = SAVE_STATE_DONE;
 628 +    }
 629 +
 630 +    DPRINTF("savevm-end: cleanup done\n");
 631 +}
 632 +
 633 +// FIXME: Deprecated
 634 +void qmp_snapshot_drive(const char *device, const char *name, Error **errp)
 635 +{
 636 +    // Compatibility to older qemu-server.
 637 +    qmp_blockdev_snapshot_internal_sync(device, name, errp);
 638 +}
 639 +
 640 +// FIXME: Deprecated
 641 +void qmp_delete_drive_snapshot(const char *device, const char *name,
 642 +                               Error **errp)
 643 +{
 644 +    // Compatibility to older qemu-server.
 645 +    (void)qmp_blockdev_snapshot_delete_internal_sync(device, NULL, name, errp);
 646 +}
 647 +
 648 +int load_snapshot_from_blockdev(const char *filename, Error **errp)
 649 +{
 650 +    BlockBackend *be;
 651 +    Error *local_err = NULL;
 652 +    Error *blocker = NULL;
 653 +
 654 +    QEMUFile *f;
 655 +    size_t bs_pos = 0;
 656 +    int ret = -EINVAL;
 657 +
 658 +    be = blk_new_open(filename, NULL, NULL, 0, &local_err);
 659 +
 660 +    if (!be) {
 661 +        error_setg(errp, "Could not open VM state file");
 662 +        goto the_end;
 663 +    }
 664 +
 665 +    error_setg(&blocker, "block device is in use by load state");
 666 +    blk_op_block_all(be, blocker);
 667 +
 668 +    /* restore the VM state */
 669 +    f = qemu_file_new_input(QIO_CHANNEL(qio_channel_savevm_async_new(be, &bs_pos)));
 670 +    if (!f) {
 671 +        error_setg(errp, "Could not open VM state file");
 672 +        goto the_end;
 673 +    }
 674 +
 675 +    qemu_system_reset(SHUTDOWN_CAUSE_NONE);
 676 +    ret = qemu_loadvm_state(f);
 677 +
 678 +    /* dirty bitmap migration has a special case we need to trigger manually */
 679 +    dirty_bitmap_mig_before_vm_start();
 680 +
 681 +    qemu_fclose(f);
 682 +    migration_incoming_state_destroy();
 683 +    if (ret < 0) {
 684 +        error_setg_errno(errp, -ret, "Error while loading VM state");
 685 +        goto the_end;
 686 +    }
 687 +
 688 +    ret = 0;
 689 +
 690 + the_end:
 691 +    if (be) {
 692 +        blk_op_unblock_all(be, blocker);
 693 +        error_free(blocker);
 694 +        blk_unref(be);
 695 +    }
 696 +    return ret;
 697 +}
 698 diff --git a/monitor/hmp-cmds.c b/monitor/hmp-cmds.c
 699 index 6c559b48c8..435f9334f9 100644
 700 --- a/monitor/hmp-cmds.c
 701 +++ b/monitor/hmp-cmds.c
 702 @@ -22,6 +22,7 @@
 703  #include "monitor/monitor-internal.h"
 704  #include "qapi/error.h"
 705  #include "qapi/qapi-commands-control.h"
 706 +#include "qapi/qapi-commands-migration.h"
 707  #include "qapi/qapi-commands-misc.h"
 708  #include "qapi/qmp/qdict.h"
 709  #include "qapi/qmp/qerror.h"
 710 @@ -443,3 +444,60 @@ void hmp_info_mtree(Monitor *mon, const QDict *qdict)
 711
 712      mtree_info(flatview, dispatch_tree, owner, disabled);
 713  }
 714 +
 715 +void hmp_savevm_start(Monitor *mon, const QDict *qdict)
 716 +{
 717 +    Error *errp = NULL;
 718 +    const char *statefile = qdict_get_try_str(qdict, "statefile");
 719 +
 720 +    qmp_savevm_start(statefile, &errp);
 721 +    hmp_handle_error(mon, errp);
 722 +}
 723 +
 724 +void hmp_snapshot_drive(Monitor *mon, const QDict *qdict)
 725 +{
 726 +    Error *errp = NULL;
 727 +    const char *name = qdict_get_str(qdict, "name");
 728 +    const char *device = qdict_get_str(qdict, "device");
 729 +
 730 +    qmp_snapshot_drive(device, name, &errp);
 731 +    hmp_handle_error(mon, errp);
 732 +}
 733 +
 734 +void hmp_delete_drive_snapshot(Monitor *mon, const QDict *qdict)
 735 +{
 736 +    Error *errp = NULL;
 737 +    const char *name = qdict_get_str(qdict, "name");
 738 +    const char *device = qdict_get_str(qdict, "device");
 739 +
 740 +    qmp_delete_drive_snapshot(device, name, &errp);
 741 +    hmp_handle_error(mon, errp);
 742 +}
 743 +
 744 +void coroutine_fn hmp_savevm_end(Monitor *mon, const QDict *qdict)
 745 +{
 746 +    Error *errp = NULL;
 747 +
 748 +    qmp_savevm_end(&errp);
 749 +    hmp_handle_error(mon, errp);
 750 +}
 751 +
 752 +void hmp_info_savevm(Monitor *mon, const QDict *qdict)
 753 +{
 754 +    SaveVMInfo *info;
 755 +    info = qmp_query_savevm(NULL);
 756 +
 757 +    if (info->status) {
 758 +        monitor_printf(mon, "savevm status: %s\n", info->status);
 759 +        monitor_printf(mon, "total time: %" PRIu64 " milliseconds\n",
 760 +                       info->total_time);
 761 +    } else {
 762 +        monitor_printf(mon, "savevm status: not running\n");
 763 +    }
 764 +    if (info->has_bytes) {
 765 +        monitor_printf(mon, "Bytes saved: %"PRIu64"\n", info->bytes);
 766 +    }
 767 +    if (info->error) {
 768 +        monitor_printf(mon, "Error: %s\n", info->error);
 769 +    }
 770 +}
 771 diff --git a/qapi/migration.json b/qapi/migration.json
 772 index c84fa10e86..1702b92553 100644
 773 --- a/qapi/migration.json
 774 +++ b/qapi/migration.json
 775 @@ -261,6 +261,40 @@
 776             '*compression': 'CompressionStats',
 777             '*socket-address': ['SocketAddress'] } }
 778
 779 +##
 780 +# @SaveVMInfo:
 781 +#
 782 +# Information about current migration process.
 783 +#
 784 +# @status: string describing the current savevm status.
 785 +#          This can be 'active', 'completed', 'failed'.
 786 +#          If this field is not returned, no savevm process
 787 +#          has been initiated
 788 +#
 789 +# @error: string containing error message is status is failed.
 790 +#
 791 +# @total-time: total amount of milliseconds since savevm started.
 792 +#              If savevm has ended, it returns the total save time
 793 +#
 794 +# @bytes: total amount of data transfered
 795 +#
 796 +# Since: 1.3
 797 +##
 798 +{ 'struct': 'SaveVMInfo',
 799 +  'data': {'*status': 'str', '*error': 'str',
 800 +           '*total-time': 'int', '*bytes': 'int'} }
 801 +
 802 +##
 803 +# @query-savevm:
 804 +#
 805 +# Returns information about current savevm process.
 806 +#
 807 +# Returns: @SaveVMInfo
 808 +#
 809 +# Since: 1.3
 810 +##
 811 +{ 'command': 'query-savevm', 'returns': 'SaveVMInfo' }
 812 +
 813  ##
 814  # @query-migrate:
 815  #
 816 diff --git a/qapi/misc.json b/qapi/misc.json
 817 index 6ddd16ea28..098c9bbe93 100644
 818 --- a/qapi/misc.json
 819 +++ b/qapi/misc.json
 820 @@ -469,6 +469,38 @@
 821  ##
 822  { 'command': 'query-fdsets', 'returns': ['FdsetInfo'] }
 823
 824 +##
 825 +# @savevm-start:
 826 +#
 827 +# Prepare for snapshot and halt VM. Save VM state to statefile.
 828 +#
 829 +##
 830 +{ 'command': 'savevm-start', 'data': { '*statefile': 'str' } }
 831 +
 832 +##
 833 +# @snapshot-drive:
 834 +#
 835 +# Create an internal drive snapshot.
 836 +#
 837 +##
 838 +{ 'command': 'snapshot-drive', 'data': { 'device': 'str', 'name': 'str' } }
 839 +
 840 +##
 841 +# @delete-drive-snapshot:
 842 +#
 843 +# Delete a drive snapshot.
 844 +#
 845 +##
 846 +{ 'command': 'delete-drive-snapshot', 'data': { 'device': 'str', 'name': 'str' } }
 847 +
 848 +##
 849 +# @savevm-end:
 850 +#
 851 +# Resume VM after a snapshot.
 852 +#
 853 +##
 854 +{ 'command': 'savevm-end', 'coroutine': true }
 855 +
 856  ##
 857  # @CommandLineParameterType:
 858  #
 859 diff --git a/qemu-options.hx b/qemu-options.hx
 860 index 59bdf67a2c..fc6cb23dd9 100644
 861 --- a/qemu-options.hx
 862 +++ b/qemu-options.hx
 863 @@ -4378,6 +4378,18 @@ SRST
 864      Start right away with a saved state (``loadvm`` in monitor)
 865  ERST
 866
 867 +DEF("loadstate", HAS_ARG, QEMU_OPTION_loadstate, \
 868 +    "-loadstate file\n" \
 869 +    "                start right away with a saved state\n",
 870 +    QEMU_ARCH_ALL)
 871 +SRST
 872 +``-loadstate file``
 873 +  Start right away with a saved state. This option does not rollback
 874 +  disk state like @code{loadvm}, so user must make sure that disk
 875 +  have correct state. @var{file} can be any valid device URL. See the section
 876 +  for "Device URL Syntax" for more information.
 877 +ERST
 878 +
 879  #ifndef _WIN32
 880  DEF("daemonize", 0, QEMU_OPTION_daemonize, \
 881      "-daemonize      daemonize QEMU after initializing\n", QEMU_ARCH_ALL)
 882 diff --git a/softmmu/vl.c b/softmmu/vl.c
 883 index ea20b23e4c..0eabc71b68 100644
 884 --- a/softmmu/vl.c
 885 +++ b/softmmu/vl.c
 886 @@ -164,6 +164,7 @@ static const char *accelerators;
 887  static bool have_custom_ram_size;
 888  static const char *ram_memdev_id;
 889  static QDict *machine_opts_dict;
 890 +static const char *loadstate;
 891  static QTAILQ_HEAD(, ObjectOption) object_opts = QTAILQ_HEAD_INITIALIZER(object_opts);
 892  static QTAILQ_HEAD(, DeviceOption) device_opts = QTAILQ_HEAD_INITIALIZER(device_opts);
 893  static int display_remote;
 894 @@ -2612,6 +2613,12 @@ void qmp_x_exit_preconfig(Error **errp)
 895
 896      if (loadvm) {
 897          load_snapshot(loadvm, NULL, false, NULL, &error_fatal);
 898 +    } else if (loadstate) {
 899 +        Error *local_err = NULL;
 900 +        if (load_snapshot_from_blockdev(loadstate, &local_err) < 0) {
 901 +            error_report_err(local_err);
 902 +            autostart = 0;
 903 +        }
 904      }
 905      if (replay_mode != REPLAY_MODE_NONE) {
 906          replay_vmstate_init();
 907 @@ -3159,6 +3166,9 @@ void qemu_init(int argc, char **argv)
 908              case QEMU_OPTION_loadvm:
 909                  loadvm = optarg;
 910                  break;
 911 +            case QEMU_OPTION_loadstate:
 912 +                loadstate = optarg;
 913 +                break;
 914              case QEMU_OPTION_full_screen:
 915                  dpy.has_full_screen = true;
 916                  dpy.full_screen = true;