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 Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
  29 ---
  30  hmp-commands-info.hx         |  13 +
  31  hmp-commands.hx              |  33 +++
  32  include/migration/snapshot.h |   2 +
  33  include/monitor/hmp.h        |   5 +
  34  migration/meson.build        |   1 +
  35  migration/savevm-async.c     | 531 +++++++++++++++++++++++++++++++++++
  36  monitor/hmp-cmds.c           |  57 ++++
  37  qapi/migration.json          |  34 +++
  38  qapi/misc.json               |  32 +++
  39  qemu-options.hx              |  12 +
  40  softmmu/vl.c                 |  10 +
  41  11 files changed, 730 insertions(+)
  42  create mode 100644 migration/savevm-async.c
  43
  44 diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
  45 index 754b1e8408..489c524e9e 100644
  46 --- a/hmp-commands-info.hx
  47 +++ b/hmp-commands-info.hx
  48 @@ -540,6 +540,19 @@ SRST
  49      Show current migration parameters.
  50  ERST
  51
  52 +    {
  53 +        .name       = "savevm",
  54 +        .args_type  = "",
  55 +        .params     = "",
  56 +        .help       = "show savevm status",
  57 +        .cmd = hmp_info_savevm,
  58 +    },
  59 +
  60 +SRST
  61 +  ``info savevm``
  62 +    Show savevm status.
  63 +ERST
  64 +
  65      {
  66          .name       = "balloon",
  67          .args_type  = "",
  68 diff --git a/hmp-commands.hx b/hmp-commands.hx
  69 index 673e39a697..039be0033d 100644
  70 --- a/hmp-commands.hx
  71 +++ b/hmp-commands.hx
  72 @@ -1815,3 +1815,36 @@ SRST
  73    Dump the FDT in dtb format to *filename*.
  74  ERST
  75  #endif
  76 +
  77 +    {
  78 +        .name       = "savevm-start",
  79 +        .args_type  = "statefile:s?",
  80 +        .params     = "[statefile]",
  81 +        .help       = "Prepare for snapshot and halt VM. Save VM state to statefile.",
  82 +        .cmd = hmp_savevm_start,
  83 +    },
  84 +
  85 +    {
  86 +        .name       = "snapshot-drive",
  87 +        .args_type  = "device:s,name:s",
  88 +        .params     = "device name",
  89 +        .help       = "Create internal snapshot.",
  90 +        .cmd = hmp_snapshot_drive,
  91 +    },
  92 +
  93 +    {
  94 +        .name       = "delete-drive-snapshot",
  95 +        .args_type  = "device:s,name:s",
  96 +        .params     = "device name",
  97 +        .help       = "Delete internal snapshot.",
  98 +        .cmd = hmp_delete_drive_snapshot,
  99 +    },
 100 +
 101 +    {
 102 +        .name       = "savevm-end",
 103 +        .args_type  = "",
 104 +        .params     = "",
 105 +        .help       = "Resume VM after snaphot.",
 106 +        .cmd        = hmp_savevm_end,
 107 +        .coroutine  = true,
 108 +    },
 109 diff --git a/include/migration/snapshot.h b/include/migration/snapshot.h
 110 index e72083b117..c846d37806 100644
 111 --- a/include/migration/snapshot.h
 112 +++ b/include/migration/snapshot.h
 113 @@ -61,4 +61,6 @@ bool delete_snapshot(const char *name,
 114                      bool has_devices, strList *devices,
 115                      Error **errp);
 116
 117 +int load_snapshot_from_blockdev(const char *filename, Error **errp);
 118 +
 119  #endif
 120 diff --git a/include/monitor/hmp.h b/include/monitor/hmp.h
 121 index dfbc0c9a2f..440f86aba8 100644
 122 --- a/include/monitor/hmp.h
 123 +++ b/include/monitor/hmp.h
 124 @@ -27,6 +27,7 @@ void hmp_info_status(Monitor *mon, const QDict *qdict);
 125  void hmp_info_uuid(Monitor *mon, const QDict *qdict);
 126  void hmp_info_chardev(Monitor *mon, const QDict *qdict);
 127  void hmp_info_mice(Monitor *mon, const QDict *qdict);
 128 +void hmp_info_savevm(Monitor *mon, const QDict *qdict);
 129  void hmp_info_migrate(Monitor *mon, const QDict *qdict);
 130  void hmp_info_migrate_capabilities(Monitor *mon, const QDict *qdict);
 131  void hmp_info_migrate_parameters(Monitor *mon, const QDict *qdict);
 132 @@ -81,6 +82,10 @@ void hmp_netdev_add(Monitor *mon, const QDict *qdict);
 133  void hmp_netdev_del(Monitor *mon, const QDict *qdict);
 134  void hmp_getfd(Monitor *mon, const QDict *qdict);
 135  void hmp_closefd(Monitor *mon, const QDict *qdict);
 136 +void hmp_savevm_start(Monitor *mon, const QDict *qdict);
 137 +void hmp_snapshot_drive(Monitor *mon, const QDict *qdict);
 138 +void hmp_delete_drive_snapshot(Monitor *mon, const QDict *qdict);
 139 +void hmp_savevm_end(Monitor *mon, const QDict *qdict);
 140  void hmp_sendkey(Monitor *mon, const QDict *qdict);
 141  void coroutine_fn hmp_screendump(Monitor *mon, const QDict *qdict);
 142  void hmp_chardev_add(Monitor *mon, const QDict *qdict);
 143 diff --git a/migration/meson.build b/migration/meson.build
 144 index 8cac83c06c..0842d00cd2 100644
 145 --- a/migration/meson.build
 146 +++ b/migration/meson.build
 147 @@ -24,6 +24,7 @@ softmmu_ss.add(files(
 148    'multifd-zlib.c',
 149    'postcopy-ram.c',
 150    'savevm.c',
 151 +  'savevm-async.c',
 152    'socket.c',
 153    'tls.c',
 154  ), gnutls)
 155 diff --git a/migration/savevm-async.c b/migration/savevm-async.c
 156 new file mode 100644
 157 index 0000000000..05d394c0e2
 158 --- /dev/null
 159 +++ b/migration/savevm-async.c
 160 @@ -0,0 +1,531 @@
 161 +#include "qemu/osdep.h"
 162 +#include "migration/channel-savevm-async.h"
 163 +#include "migration/migration.h"
 164 +#include "migration/savevm.h"
 165 +#include "migration/snapshot.h"
 166 +#include "migration/global_state.h"
 167 +#include "migration/ram.h"
 168 +#include "migration/qemu-file.h"
 169 +#include "sysemu/sysemu.h"
 170 +#include "sysemu/runstate.h"
 171 +#include "block/block.h"
 172 +#include "sysemu/block-backend.h"
 173 +#include "qapi/error.h"
 174 +#include "qapi/qmp/qerror.h"
 175 +#include "qapi/qmp/qdict.h"
 176 +#include "qapi/qapi-commands-migration.h"
 177 +#include "qapi/qapi-commands-misc.h"
 178 +#include "qapi/qapi-commands-block.h"
 179 +#include "qemu/cutils.h"
 180 +#include "qemu/timer.h"
 181 +#include "qemu/main-loop.h"
 182 +#include "qemu/rcu.h"
 183 +
 184 +/* #define DEBUG_SAVEVM_STATE */
 185 +
 186 +#ifdef DEBUG_SAVEVM_STATE
 187 +#define DPRINTF(fmt, ...) \
 188 +    do { printf("savevm-async: " fmt, ## __VA_ARGS__); } while (0)
 189 +#else
 190 +#define DPRINTF(fmt, ...) \
 191 +    do { } while (0)
 192 +#endif
 193 +
 194 +enum {
 195 +    SAVE_STATE_DONE,
 196 +    SAVE_STATE_ERROR,
 197 +    SAVE_STATE_ACTIVE,
 198 +    SAVE_STATE_COMPLETED,
 199 +    SAVE_STATE_CANCELLED
 200 +};
 201 +
 202 +
 203 +static struct SnapshotState {
 204 +    BlockBackend *target;
 205 +    size_t bs_pos;
 206 +    int state;
 207 +    Error *error;
 208 +    Error *blocker;
 209 +    int saved_vm_running;
 210 +    QEMUFile *file;
 211 +    int64_t total_time;
 212 +    QEMUBH *finalize_bh;
 213 +    Coroutine *co;
 214 +    QemuCoSleep target_close_wait;
 215 +} snap_state;
 216 +
 217 +static bool savevm_aborted(void)
 218 +{
 219 +    return snap_state.state == SAVE_STATE_CANCELLED ||
 220 +        snap_state.state == SAVE_STATE_ERROR;
 221 +}
 222 +
 223 +SaveVMInfo *qmp_query_savevm(Error **errp)
 224 +{
 225 +    SaveVMInfo *info = g_malloc0(sizeof(*info));
 226 +    struct SnapshotState *s = &snap_state;
 227 +
 228 +    if (s->state != SAVE_STATE_DONE) {
 229 +        info->has_bytes = true;
 230 +        info->bytes = s->bs_pos;
 231 +        switch (s->state) {
 232 +        case SAVE_STATE_ERROR:
 233 +            info->has_status = true;
 234 +            info->status = g_strdup("failed");
 235 +            info->has_total_time = true;
 236 +            info->total_time = s->total_time;
 237 +            if (s->error) {
 238 +                info->has_error = true;
 239 +                info->error = g_strdup(error_get_pretty(s->error));
 240 +            }
 241 +            break;
 242 +        case SAVE_STATE_ACTIVE:
 243 +            info->has_status = true;
 244 +            info->status = g_strdup("active");
 245 +            info->has_total_time = true;
 246 +            info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 247 +                - s->total_time;
 248 +            break;
 249 +        case SAVE_STATE_COMPLETED:
 250 +            info->has_status = true;
 251 +            info->status = g_strdup("completed");
 252 +            info->has_total_time = true;
 253 +            info->total_time = s->total_time;
 254 +            break;
 255 +        }
 256 +    }
 257 +
 258 +    return info;
 259 +}
 260 +
 261 +static int save_snapshot_cleanup(void)
 262 +{
 263 +    int ret = 0;
 264 +
 265 +    DPRINTF("save_snapshot_cleanup\n");
 266 +
 267 +    snap_state.total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
 268 +        snap_state.total_time;
 269 +
 270 +    if (snap_state.file) {
 271 +        ret = qemu_fclose(snap_state.file);
 272 +        snap_state.file = NULL;
 273 +    }
 274 +
 275 +    if (snap_state.target) {
 276 +        if (!savevm_aborted()) {
 277 +            /* try to truncate, but ignore errors (will fail on block devices).
 278 +            * note1: bdrv_read() need whole blocks, so we need to round up
 279 +            * note2: PVE requires 1024 (BDRV_SECTOR_SIZE*2) alignment
 280 +            */
 281 +            size_t size = QEMU_ALIGN_UP(snap_state.bs_pos, BDRV_SECTOR_SIZE*2);
 282 +            blk_truncate(snap_state.target, size, false, PREALLOC_MODE_OFF, 0, NULL);
 283 +        }
 284 +        blk_op_unblock_all(snap_state.target, snap_state.blocker);
 285 +        error_free(snap_state.blocker);
 286 +        snap_state.blocker = NULL;
 287 +        blk_unref(snap_state.target);
 288 +        snap_state.target = NULL;
 289 +
 290 +        qemu_co_sleep_wake(&snap_state.target_close_wait);
 291 +    }
 292 +
 293 +    return ret;
 294 +}
 295 +
 296 +static void save_snapshot_error(const char *fmt, ...)
 297 +{
 298 +    va_list ap;
 299 +    char *msg;
 300 +
 301 +    va_start(ap, fmt);
 302 +    msg = g_strdup_vprintf(fmt, ap);
 303 +    va_end(ap);
 304 +
 305 +    DPRINTF("save_snapshot_error: %s\n", msg);
 306 +
 307 +    if (!snap_state.error) {
 308 +        error_set(&snap_state.error, ERROR_CLASS_GENERIC_ERROR, "%s", msg);
 309 +    }
 310 +
 311 +    g_free (msg);
 312 +
 313 +    snap_state.state = SAVE_STATE_ERROR;
 314 +}
 315 +
 316 +static void process_savevm_finalize(void *opaque)
 317 +{
 318 +    int ret;
 319 +    AioContext *iohandler_ctx = iohandler_get_aio_context();
 320 +    MigrationState *ms = migrate_get_current();
 321 +
 322 +    bool aborted = savevm_aborted();
 323 +
 324 +#ifdef DEBUG_SAVEVM_STATE
 325 +    int64_t start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 326 +#endif
 327 +
 328 +    qemu_bh_delete(snap_state.finalize_bh);
 329 +    snap_state.finalize_bh = NULL;
 330 +    snap_state.co = NULL;
 331 +
 332 +    /* We need to own the target bdrv's context for the following functions,
 333 +     * so move it back. It can stay in the main context and live out its live
 334 +     * there, since we're done with it after this method ends anyway.
 335 +     */
 336 +    aio_context_acquire(iohandler_ctx);
 337 +    blk_set_aio_context(snap_state.target, qemu_get_aio_context(), NULL);
 338 +    aio_context_release(iohandler_ctx);
 339 +
 340 +    ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 341 +    if (ret < 0) {
 342 +        save_snapshot_error("vm_stop_force_state error %d", ret);
 343 +    }
 344 +
 345 +    if (!aborted) {
 346 +        /* skip state saving if we aborted, snapshot will be invalid anyway */
 347 +        (void)qemu_savevm_state_complete_precopy(snap_state.file, false, false);
 348 +        ret = qemu_file_get_error(snap_state.file);
 349 +        if (ret < 0) {
 350 +                save_snapshot_error("qemu_savevm_state_iterate error %d", ret);
 351 +        }
 352 +    }
 353 +
 354 +    DPRINTF("state saving complete\n");
 355 +    DPRINTF("timing: process_savevm_finalize (state saving) took %ld ms\n",
 356 +        qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - start_time);
 357 +
 358 +    /* clear migration state */
 359 +    migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP,
 360 +        ret || aborted ? MIGRATION_STATUS_FAILED : MIGRATION_STATUS_COMPLETED);
 361 +    ms->to_dst_file = NULL;
 362 +
 363 +    qemu_savevm_state_cleanup();
 364 +
 365 +    ret = save_snapshot_cleanup();
 366 +    if (ret < 0) {
 367 +        save_snapshot_error("save_snapshot_cleanup error %d", ret);
 368 +    } else if (snap_state.state == SAVE_STATE_ACTIVE) {
 369 +        snap_state.state = SAVE_STATE_COMPLETED;
 370 +    } else if (aborted) {
 371 +        /*
 372 +         * If there was an error, there's no need to set a new one here.
 373 +         * If the snapshot was canceled, leave setting the state to
 374 +         * qmp_savevm_end(), which is waked by save_snapshot_cleanup().
 375 +         */
 376 +    } else {
 377 +        save_snapshot_error("process_savevm_cleanup: invalid state: %d",
 378 +                            snap_state.state);
 379 +    }
 380 +    if (snap_state.saved_vm_running) {
 381 +        vm_start();
 382 +        snap_state.saved_vm_running = false;
 383 +    }
 384 +
 385 +    DPRINTF("timing: process_savevm_finalize (full) took %ld ms\n",
 386 +        qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - start_time);
 387 +}
 388 +
 389 +static void coroutine_fn process_savevm_co(void *opaque)
 390 +{
 391 +    int ret;
 392 +    int64_t maxlen;
 393 +    BdrvNextIterator it;
 394 +    BlockDriverState *bs = NULL;
 395 +
 396 +#ifdef DEBUG_SAVEVM_STATE
 397 +    int64_t start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 398 +#endif
 399 +
 400 +    ret = qemu_file_get_error(snap_state.file);
 401 +    if (ret < 0) {
 402 +        save_snapshot_error("qemu_savevm_state_setup failed");
 403 +        return;
 404 +    }
 405 +
 406 +    while (snap_state.state == SAVE_STATE_ACTIVE) {
 407 +        uint64_t pending_size, pend_precopy, pend_compatible, pend_postcopy;
 408 +
 409 +        /* pending is expected to be called without iothread lock */
 410 +        qemu_mutex_unlock_iothread();
 411 +        qemu_savevm_state_pending(snap_state.file, 0, &pend_precopy, &pend_compatible, &pend_postcopy);
 412 +        qemu_mutex_lock_iothread();
 413 +
 414 +        pending_size = pend_precopy + pend_compatible + pend_postcopy;
 415 +
 416 +        maxlen = blk_getlength(snap_state.target) - 30*1024*1024;
 417 +
 418 +        if (pending_size > 400000 && snap_state.bs_pos + pending_size < maxlen) {
 419 +            ret = qemu_savevm_state_iterate(snap_state.file, false);
 420 +            if (ret < 0) {
 421 +                save_snapshot_error("qemu_savevm_state_iterate error %d", ret);
 422 +                break;
 423 +            }
 424 +            DPRINTF("savevm iterate pending size %lu ret %d\n", pending_size, ret);
 425 +        } else {
 426 +            qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
 427 +            ret = global_state_store();
 428 +            if (ret) {
 429 +                save_snapshot_error("global_state_store error %d", ret);
 430 +                break;
 431 +            }
 432 +
 433 +            DPRINTF("savevm iterate complete\n");
 434 +            break;
 435 +        }
 436 +    }
 437 +
 438 +    DPRINTF("timing: process_savevm_co took %ld ms\n",
 439 +        qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - start_time);
 440 +
 441 +#ifdef DEBUG_SAVEVM_STATE
 442 +    int64_t start_time_flush = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 443 +#endif
 444 +    /* If a drive runs in an IOThread we can flush it async, and only
 445 +     * need to sync-flush whatever IO happens between now and
 446 +     * vm_stop_force_state. bdrv_next can only be called from main AioContext,
 447 +     * so move there now and after every flush.
 448 +     */
 449 +    aio_co_reschedule_self(qemu_get_aio_context());
 450 +    for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
 451 +        /* target has BDRV_O_NO_FLUSH, no sense calling bdrv_flush on it */
 452 +        if (bs == blk_bs(snap_state.target)) {
 453 +            continue;
 454 +        }
 455 +
 456 +        AioContext *bs_ctx = bdrv_get_aio_context(bs);
 457 +        if (bs_ctx != qemu_get_aio_context()) {
 458 +            DPRINTF("savevm: async flushing drive %s\n", bs->filename);
 459 +            aio_co_reschedule_self(bs_ctx);
 460 +            bdrv_flush(bs);
 461 +            aio_co_reschedule_self(qemu_get_aio_context());
 462 +        }
 463 +    }
 464 +
 465 +    DPRINTF("timing: async flushing took %ld ms\n",
 466 +        qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - start_time_flush);
 467 +
 468 +    qemu_bh_schedule(snap_state.finalize_bh);
 469 +}
 470 +
 471 +void qmp_savevm_start(bool has_statefile, const char *statefile, Error **errp)
 472 +{
 473 +    Error *local_err = NULL;
 474 +    MigrationState *ms = migrate_get_current();
 475 +    AioContext *iohandler_ctx = iohandler_get_aio_context();
 476 +
 477 +    int bdrv_oflags = BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH;
 478 +
 479 +    if (snap_state.state != SAVE_STATE_DONE) {
 480 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 481 +                  "VM snapshot already started\n");
 482 +        return;
 483 +    }
 484 +
 485 +    if (migration_is_running(ms->state)) {
 486 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, QERR_MIGRATION_ACTIVE);
 487 +        return;
 488 +    }
 489 +
 490 +    if (migrate_use_block()) {
 491 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 492 +                  "Block migration and snapshots are incompatible");
 493 +        return;
 494 +    }
 495 +
 496 +    /* initialize snapshot info */
 497 +    snap_state.saved_vm_running = runstate_is_running();
 498 +    snap_state.bs_pos = 0;
 499 +    snap_state.total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 500 +    snap_state.blocker = NULL;
 501 +    snap_state.target_close_wait = (QemuCoSleep){ .to_wake = NULL };
 502 +
 503 +    if (snap_state.error) {
 504 +        error_free(snap_state.error);
 505 +        snap_state.error = NULL;
 506 +    }
 507 +
 508 +    if (!has_statefile) {
 509 +        vm_stop(RUN_STATE_SAVE_VM);
 510 +        snap_state.state = SAVE_STATE_COMPLETED;
 511 +        return;
 512 +    }
 513 +
 514 +    if (qemu_savevm_state_blocked(errp)) {
 515 +        return;
 516 +    }
 517 +
 518 +    /* Open the image */
 519 +    QDict *options = NULL;
 520 +    options = qdict_new();
 521 +    qdict_put_str(options, "driver", "raw");
 522 +    snap_state.target = blk_new_open(statefile, NULL, options, bdrv_oflags, &local_err);
 523 +    if (!snap_state.target) {
 524 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "failed to open '%s'", statefile);
 525 +        goto restart;
 526 +    }
 527 +
 528 +    QIOChannel *ioc = QIO_CHANNEL(qio_channel_savevm_async_new(snap_state.target,
 529 +                                                               &snap_state.bs_pos));
 530 +    snap_state.file = qemu_file_new_output(ioc);
 531 +
 532 +    if (!snap_state.file) {
 533 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "failed to open '%s'", statefile);
 534 +        goto restart;
 535 +    }
 536 +
 537 +    /*
 538 +     * qemu_savevm_* paths use migration code and expect a migration state.
 539 +     * State is cleared in process_savevm_co, but has to be initialized
 540 +     * here (blocking main thread, from QMP) to avoid race conditions.
 541 +     */
 542 +    migrate_init(ms);
 543 +    memset(&ram_counters, 0, sizeof(ram_counters));
 544 +    ms->to_dst_file = snap_state.file;
 545 +
 546 +    error_setg(&snap_state.blocker, "block device is in use by savevm");
 547 +    blk_op_block_all(snap_state.target, snap_state.blocker);
 548 +
 549 +    snap_state.state = SAVE_STATE_ACTIVE;
 550 +    snap_state.finalize_bh = qemu_bh_new(process_savevm_finalize, &snap_state);
 551 +    snap_state.co = qemu_coroutine_create(&process_savevm_co, NULL);
 552 +    qemu_mutex_unlock_iothread();
 553 +    qemu_savevm_state_header(snap_state.file);
 554 +    qemu_savevm_state_setup(snap_state.file);
 555 +    qemu_mutex_lock_iothread();
 556 +
 557 +    /* Async processing from here on out happens in iohandler context, so let
 558 +     * the target bdrv have its home there.
 559 +     */
 560 +    blk_set_aio_context(snap_state.target, iohandler_ctx, &local_err);
 561 +
 562 +    aio_co_schedule(iohandler_ctx, snap_state.co);
 563 +
 564 +    return;
 565 +
 566 +restart:
 567 +
 568 +    save_snapshot_error("setup failed");
 569 +
 570 +    if (snap_state.saved_vm_running) {
 571 +        vm_start();
 572 +        snap_state.saved_vm_running = false;
 573 +    }
 574 +}
 575 +
 576 +void coroutine_fn qmp_savevm_end(Error **errp)
 577 +{
 578 +    int64_t timeout;
 579 +
 580 +    if (snap_state.state == SAVE_STATE_DONE) {
 581 +        error_set(errp, ERROR_CLASS_GENERIC_ERROR,
 582 +                  "VM snapshot not started\n");
 583 +        return;
 584 +    }
 585 +
 586 +    if (snap_state.state == SAVE_STATE_ACTIVE) {
 587 +        snap_state.state = SAVE_STATE_CANCELLED;
 588 +        goto wait_for_close;
 589 +    }
 590 +
 591 +    if (snap_state.saved_vm_running) {
 592 +        vm_start();
 593 +        snap_state.saved_vm_running = false;
 594 +    }
 595 +
 596 +    snap_state.state = SAVE_STATE_DONE;
 597 +
 598 +wait_for_close:
 599 +    if (!snap_state.target) {
 600 +        DPRINTF("savevm-end: no target file open\n");
 601 +        return;
 602 +    }
 603 +
 604 +    /* wait until cleanup is done before returning, this ensures that after this
 605 +     * call exits the statefile will be closed and can be removed immediately */
 606 +    DPRINTF("savevm-end: waiting for cleanup\n");
 607 +    timeout = 30L * 1000 * 1000 * 1000;
 608 +    qemu_co_sleep_ns_wakeable(&snap_state.target_close_wait,
 609 +                              QEMU_CLOCK_REALTIME, timeout);
 610 +    if (snap_state.target) {
 611 +        save_snapshot_error("timeout waiting for target file close in "
 612 +                            "qmp_savevm_end");
 613 +        /* we cannot assume the snapshot finished in this case, so leave the
 614 +         * state alone - caller has to figure something out */
 615 +        return;
 616 +    }
 617 +
 618 +    // File closed and no other error, so ensure next snapshot can be started.
 619 +    if (snap_state.state != SAVE_STATE_ERROR) {
 620 +        snap_state.state = SAVE_STATE_DONE;
 621 +    }
 622 +
 623 +    DPRINTF("savevm-end: cleanup done\n");
 624 +}
 625 +
 626 +// FIXME: Deprecated
 627 +void qmp_snapshot_drive(const char *device, const char *name, Error **errp)
 628 +{
 629 +    // Compatibility to older qemu-server.
 630 +    qmp_blockdev_snapshot_internal_sync(device, name, errp);
 631 +}
 632 +
 633 +// FIXME: Deprecated
 634 +void qmp_delete_drive_snapshot(const char *device, const char *name,
 635 +                               Error **errp)
 636 +{
 637 +    // Compatibility to older qemu-server.
 638 +    (void)qmp_blockdev_snapshot_delete_internal_sync(device, false, NULL,
 639 +                                                     true, name, errp);
 640 +}
 641 +
 642 +int load_snapshot_from_blockdev(const char *filename, Error **errp)
 643 +{
 644 +    BlockBackend *be;
 645 +    Error *local_err = NULL;
 646 +    Error *blocker = NULL;
 647 +
 648 +    QEMUFile *f;
 649 +    size_t bs_pos = 0;
 650 +    int ret = -EINVAL;
 651 +
 652 +    be = blk_new_open(filename, NULL, NULL, 0, &local_err);
 653 +
 654 +    if (!be) {
 655 +        error_setg(errp, "Could not open VM state file");
 656 +        goto the_end;
 657 +    }
 658 +
 659 +    error_setg(&blocker, "block device is in use by load state");
 660 +    blk_op_block_all(be, blocker);
 661 +
 662 +    /* restore the VM state */
 663 +    f = qemu_file_new_input(QIO_CHANNEL(qio_channel_savevm_async_new(be, &bs_pos)));
 664 +    if (!f) {
 665 +        error_setg(errp, "Could not open VM state file");
 666 +        goto the_end;
 667 +    }
 668 +
 669 +    qemu_system_reset(SHUTDOWN_CAUSE_NONE);
 670 +    ret = qemu_loadvm_state(f);
 671 +
 672 +    /* dirty bitmap migration has a special case we need to trigger manually */
 673 +    dirty_bitmap_mig_before_vm_start();
 674 +
 675 +    qemu_fclose(f);
 676 +    migration_incoming_state_destroy();
 677 +    if (ret < 0) {
 678 +        error_setg_errno(errp, -ret, "Error while loading VM state");
 679 +        goto the_end;
 680 +    }
 681 +
 682 +    ret = 0;
 683 +
 684 + the_end:
 685 +    if (be) {
 686 +        blk_op_unblock_all(be, blocker);
 687 +        error_free(blocker);
 688 +        blk_unref(be);
 689 +    }
 690 +    return ret;
 691 +}
 692 diff --git a/monitor/hmp-cmds.c b/monitor/hmp-cmds.c
 693 index 480b798963..cfebfd1db5 100644
 694 --- a/monitor/hmp-cmds.c
 695 +++ b/monitor/hmp-cmds.c
 696 @@ -1906,6 +1906,63 @@ void hmp_info_memory_devices(Monitor *mon, const QDict *qdict)
 697      hmp_handle_error(mon, err);
 698  }
 699
 700 +void hmp_savevm_start(Monitor *mon, const QDict *qdict)
 701 +{
 702 +    Error *errp = NULL;
 703 +    const char *statefile = qdict_get_try_str(qdict, "statefile");
 704 +
 705 +    qmp_savevm_start(statefile != NULL, statefile, &errp);
 706 +    hmp_handle_error(mon, errp);
 707 +}
 708 +
 709 +void hmp_snapshot_drive(Monitor *mon, const QDict *qdict)
 710 +{
 711 +    Error *errp = NULL;
 712 +    const char *name = qdict_get_str(qdict, "name");
 713 +    const char *device = qdict_get_str(qdict, "device");
 714 +
 715 +    qmp_snapshot_drive(device, name, &errp);
 716 +    hmp_handle_error(mon, errp);
 717 +}
 718 +
 719 +void hmp_delete_drive_snapshot(Monitor *mon, const QDict *qdict)
 720 +{
 721 +    Error *errp = NULL;
 722 +    const char *name = qdict_get_str(qdict, "name");
 723 +    const char *device = qdict_get_str(qdict, "device");
 724 +
 725 +    qmp_delete_drive_snapshot(device, name, &errp);
 726 +    hmp_handle_error(mon, errp);
 727 +}
 728 +
 729 +void coroutine_fn hmp_savevm_end(Monitor *mon, const QDict *qdict)
 730 +{
 731 +    Error *errp = NULL;
 732 +
 733 +    qmp_savevm_end(&errp);
 734 +    hmp_handle_error(mon, errp);
 735 +}
 736 +
 737 +void hmp_info_savevm(Monitor *mon, const QDict *qdict)
 738 +{
 739 +    SaveVMInfo *info;
 740 +    info = qmp_query_savevm(NULL);
 741 +
 742 +    if (info->has_status) {
 743 +        monitor_printf(mon, "savevm status: %s\n", info->status);
 744 +        monitor_printf(mon, "total time: %" PRIu64 " milliseconds\n",
 745 +                       info->total_time);
 746 +    } else {
 747 +        monitor_printf(mon, "savevm status: not running\n");
 748 +    }
 749 +    if (info->has_bytes) {
 750 +        monitor_printf(mon, "Bytes saved: %"PRIu64"\n", info->bytes);
 751 +    }
 752 +    if (info->has_error) {
 753 +        monitor_printf(mon, "Error: %s\n", info->error);
 754 +    }
 755 +}
 756 +
 757  void hmp_info_iothreads(Monitor *mon, const QDict *qdict)
 758  {
 759      IOThreadInfoList *info_list = qmp_query_iothreads(NULL);
 760 diff --git a/qapi/migration.json b/qapi/migration.json
 761 index 88ecf86ac8..4435866379 100644
 762 --- a/qapi/migration.json
 763 +++ b/qapi/migration.json
 764 @@ -261,6 +261,40 @@
 765             '*compression': 'CompressionStats',
 766             '*socket-address': ['SocketAddress'] } }
 767
 768 +##
 769 +# @SaveVMInfo:
 770 +#
 771 +# Information about current migration process.
 772 +#
 773 +# @status: string describing the current savevm status.
 774 +#          This can be 'active', 'completed', 'failed'.
 775 +#          If this field is not returned, no savevm process
 776 +#          has been initiated
 777 +#
 778 +# @error: string containing error message is status is failed.
 779 +#
 780 +# @total-time: total amount of milliseconds since savevm started.
 781 +#              If savevm has ended, it returns the total save time
 782 +#
 783 +# @bytes: total amount of data transfered
 784 +#
 785 +# Since: 1.3
 786 +##
 787 +{ 'struct': 'SaveVMInfo',
 788 +  'data': {'*status': 'str', '*error': 'str',
 789 +           '*total-time': 'int', '*bytes': 'int'} }
 790 +
 791 +##
 792 +# @query-savevm:
 793 +#
 794 +# Returns information about current savevm process.
 795 +#
 796 +# Returns: @SaveVMInfo
 797 +#
 798 +# Since: 1.3
 799 +##
 800 +{ 'command': 'query-savevm', 'returns': 'SaveVMInfo' }
 801 +
 802  ##
 803  # @query-migrate:
 804  #
 805 diff --git a/qapi/misc.json b/qapi/misc.json
 806 index 27ef5a2b20..b3ce75dcae 100644
 807 --- a/qapi/misc.json
 808 +++ b/qapi/misc.json
 809 @@ -435,6 +435,38 @@
 810  ##
 811  { 'command': 'query-fdsets', 'returns': ['FdsetInfo'] }
 812
 813 +##
 814 +# @savevm-start:
 815 +#
 816 +# Prepare for snapshot and halt VM. Save VM state to statefile.
 817 +#
 818 +##
 819 +{ 'command': 'savevm-start', 'data': { '*statefile': 'str' } }
 820 +
 821 +##
 822 +# @snapshot-drive:
 823 +#
 824 +# Create an internal drive snapshot.
 825 +#
 826 +##
 827 +{ 'command': 'snapshot-drive', 'data': { 'device': 'str', 'name': 'str' } }
 828 +
 829 +##
 830 +# @delete-drive-snapshot:
 831 +#
 832 +# Delete a drive snapshot.
 833 +#
 834 +##
 835 +{ 'command': 'delete-drive-snapshot', 'data': { 'device': 'str', 'name': 'str' } }
 836 +
 837 +##
 838 +# @savevm-end:
 839 +#
 840 +# Resume VM after a snapshot.
 841 +#
 842 +##
 843 +{ 'command': 'savevm-end', 'coroutine': true }
 844 +
 845  ##
 846  # @CommandLineParameterType:
 847  #
 848 diff --git a/qemu-options.hx b/qemu-options.hx
 849 index 7f99d15b23..54efb127c4 100644
 850 --- a/qemu-options.hx
 851 +++ b/qemu-options.hx
 852 @@ -4391,6 +4391,18 @@ SRST
 853      Start right away with a saved state (``loadvm`` in monitor)
 854  ERST
 855
 856 +DEF("loadstate", HAS_ARG, QEMU_OPTION_loadstate, \
 857 +    "-loadstate file\n" \
 858 +    "                start right away with a saved state\n",
 859 +    QEMU_ARCH_ALL)
 860 +SRST
 861 +``-loadstate file``
 862 +  Start right away with a saved state. This option does not rollback
 863 +  disk state like @code{loadvm}, so user must make sure that disk
 864 +  have correct state. @var{file} can be any valid device URL. See the section
 865 +  for "Device URL Syntax" for more information.
 866 +ERST
 867 +
 868  #ifndef _WIN32
 869  DEF("daemonize", 0, QEMU_OPTION_daemonize, \
 870      "-daemonize      daemonize QEMU after initializing\n", QEMU_ARCH_ALL)
 871 diff --git a/softmmu/vl.c b/softmmu/vl.c
 872 index 5f7f6ca981..21f067d115 100644
 873 --- a/softmmu/vl.c
 874 +++ b/softmmu/vl.c
 875 @@ -164,6 +164,7 @@ static const char *accelerators;
 876  static bool have_custom_ram_size;
 877  static const char *ram_memdev_id;
 878  static QDict *machine_opts_dict;
 879 +static const char *loadstate;
 880  static QTAILQ_HEAD(, ObjectOption) object_opts = QTAILQ_HEAD_INITIALIZER(object_opts);
 881  static QTAILQ_HEAD(, DeviceOption) device_opts = QTAILQ_HEAD_INITIALIZER(device_opts);
 882  static int display_remote;
 883 @@ -2607,6 +2608,12 @@ void qmp_x_exit_preconfig(Error **errp)
 884
 885      if (loadvm) {
 886          load_snapshot(loadvm, NULL, false, NULL, &error_fatal);
 887 +    } else if (loadstate) {
 888 +        Error *local_err = NULL;
 889 +        if (load_snapshot_from_blockdev(loadstate, &local_err) < 0) {
 890 +            error_report_err(local_err);
 891 +            autostart = 0;
 892 +        }
 893      }
 894      if (replay_mode != REPLAY_MODE_NONE) {
 895          replay_vmstate_init();
 896 @@ -3151,6 +3158,9 @@ void qemu_init(int argc, char **argv)
 897              case QEMU_OPTION_loadvm:
 898                  loadvm = optarg;
 899                  break;
 900 +            case QEMU_OPTION_loadstate:
 901 +                loadstate = optarg;
 902 +                break;
 903              case QEMU_OPTION_full_screen:
 904                  dpy.has_full_screen = true;
 905                  dpy.full_screen = true;