qemu/blockdev.c
<<
>>
Prefs
   1/*
   2 * QEMU host block devices
   3 *
   4 * Copyright (c) 2003-2008 Fabrice Bellard
   5 *
   6 * This work is licensed under the terms of the GNU GPL, version 2 or
   7 * later.  See the COPYING file in the top-level directory.
   8 *
   9 * This file incorporates work covered by the following copyright and
  10 * permission notice:
  11 *
  12 * Copyright (c) 2003-2008 Fabrice Bellard
  13 *
  14 * Permission is hereby granted, free of charge, to any person obtaining a copy
  15 * of this software and associated documentation files (the "Software"), to deal
  16 * in the Software without restriction, including without limitation the rights
  17 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  18 * copies of the Software, and to permit persons to whom the Software is
  19 * furnished to do so, subject to the following conditions:
  20 *
  21 * The above copyright notice and this permission notice shall be included in
  22 * all copies or substantial portions of the Software.
  23 *
  24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  25 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  26 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  27 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  28 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  29 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  30 * THE SOFTWARE.
  31 */
  32
  33#include "qemu/osdep.h"
  34#include "sysemu/block-backend.h"
  35#include "sysemu/blockdev.h"
  36#include "hw/block/block.h"
  37#include "block/blockjob.h"
  38#include "block/qdict.h"
  39#include "block/throttle-groups.h"
  40#include "monitor/monitor.h"
  41#include "qemu/error-report.h"
  42#include "qemu/option.h"
  43#include "qemu/config-file.h"
  44#include "qapi/qapi-commands-block.h"
  45#include "qapi/qapi-commands-transaction.h"
  46#include "qapi/qapi-visit-block-core.h"
  47#include "qapi/qmp/qdict.h"
  48#include "qapi/qmp/qnum.h"
  49#include "qapi/qmp/qstring.h"
  50#include "qapi/error.h"
  51#include "qapi/qmp/qerror.h"
  52#include "qapi/qmp/qlist.h"
  53#include "qapi/qobject-output-visitor.h"
  54#include "sysemu/sysemu.h"
  55#include "sysemu/iothread.h"
  56#include "block/block_int.h"
  57#include "block/trace.h"
  58#include "sysemu/arch_init.h"
  59#include "sysemu/qtest.h"
  60#include "qemu/cutils.h"
  61#include "qemu/help_option.h"
  62#include "qemu/throttle-options.h"
  63
  64static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
  65    QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
  66
  67static int do_open_tray(const char *blk_name, const char *qdev_id,
  68                        bool force, Error **errp);
  69static void blockdev_remove_medium(bool has_device, const char *device,
  70                                   bool has_id, const char *id, Error **errp);
  71static void blockdev_insert_medium(bool has_device, const char *device,
  72                                   bool has_id, const char *id,
  73                                   const char *node_name, Error **errp);
  74
  75static const char *const if_name[IF_COUNT] = {
  76    [IF_NONE] = "none",
  77    [IF_IDE] = "ide",
  78    [IF_SCSI] = "scsi",
  79    [IF_FLOPPY] = "floppy",
  80    [IF_PFLASH] = "pflash",
  81    [IF_MTD] = "mtd",
  82    [IF_SD] = "sd",
  83    [IF_VIRTIO] = "virtio",
  84    [IF_XEN] = "xen",
  85};
  86
  87static int if_max_devs[IF_COUNT] = {
  88    /*
  89     * Do not change these numbers!  They govern how drive option
  90     * index maps to unit and bus.  That mapping is ABI.
  91     *
  92     * All controllers used to implement if=T drives need to support
  93     * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
  94     * Otherwise, some index values map to "impossible" bus, unit
  95     * values.
  96     *
  97     * For instance, if you change [IF_SCSI] to 255, -drive
  98     * if=scsi,index=12 no longer means bus=1,unit=5, but
  99     * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
 100     * the drive can't be set up.  Regression.
 101     */
 102    [IF_IDE] = 2,
 103    [IF_SCSI] = 7,
 104};
 105
 106/**
 107 * Boards may call this to offer board-by-board overrides
 108 * of the default, global values.
 109 */
 110void override_max_devs(BlockInterfaceType type, int max_devs)
 111{
 112    BlockBackend *blk;
 113    DriveInfo *dinfo;
 114
 115    if (max_devs <= 0) {
 116        return;
 117    }
 118
 119    for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 120        dinfo = blk_legacy_dinfo(blk);
 121        if (dinfo->type == type) {
 122            fprintf(stderr, "Cannot override units-per-bus property of"
 123                    " the %s interface, because a drive of that type has"
 124                    " already been added.\n", if_name[type]);
 125            g_assert_not_reached();
 126        }
 127    }
 128
 129    if_max_devs[type] = max_devs;
 130}
 131
 132/*
 133 * We automatically delete the drive when a device using it gets
 134 * unplugged.  Questionable feature, but we can't just drop it.
 135 * Device models call blockdev_mark_auto_del() to schedule the
 136 * automatic deletion, and generic qdev code calls blockdev_auto_del()
 137 * when deletion is actually safe.
 138 */
 139void blockdev_mark_auto_del(BlockBackend *blk)
 140{
 141    DriveInfo *dinfo = blk_legacy_dinfo(blk);
 142    BlockDriverState *bs = blk_bs(blk);
 143    AioContext *aio_context;
 144
 145    if (!dinfo) {
 146        return;
 147    }
 148
 149    if (bs) {
 150        aio_context = bdrv_get_aio_context(bs);
 151        aio_context_acquire(aio_context);
 152
 153        if (bs->job) {
 154            job_cancel(&bs->job->job, false);
 155        }
 156
 157        aio_context_release(aio_context);
 158    }
 159
 160    dinfo->auto_del = 1;
 161}
 162
 163void blockdev_auto_del(BlockBackend *blk)
 164{
 165    DriveInfo *dinfo = blk_legacy_dinfo(blk);
 166
 167    if (dinfo && dinfo->auto_del) {
 168        monitor_remove_blk(blk);
 169        blk_unref(blk);
 170    }
 171}
 172
 173/**
 174 * Returns the current mapping of how many units per bus
 175 * a particular interface can support.
 176 *
 177 *  A positive integer indicates n units per bus.
 178 *  0 implies the mapping has not been established.
 179 * -1 indicates an invalid BlockInterfaceType was given.
 180 */
 181int drive_get_max_devs(BlockInterfaceType type)
 182{
 183    if (type >= IF_IDE && type < IF_COUNT) {
 184        return if_max_devs[type];
 185    }
 186
 187    return -1;
 188}
 189
 190static int drive_index_to_bus_id(BlockInterfaceType type, int index)
 191{
 192    int max_devs = if_max_devs[type];
 193    return max_devs ? index / max_devs : 0;
 194}
 195
 196static int drive_index_to_unit_id(BlockInterfaceType type, int index)
 197{
 198    int max_devs = if_max_devs[type];
 199    return max_devs ? index % max_devs : index;
 200}
 201
 202QemuOpts *drive_def(const char *optstr)
 203{
 204    return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
 205}
 206
 207QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
 208                    const char *optstr)
 209{
 210    QemuOpts *opts;
 211
 212    opts = drive_def(optstr);
 213    if (!opts) {
 214        return NULL;
 215    }
 216    if (type != IF_DEFAULT) {
 217        qemu_opt_set(opts, "if", if_name[type], &error_abort);
 218    }
 219    if (index >= 0) {
 220        qemu_opt_set_number(opts, "index", index, &error_abort);
 221    }
 222    if (file)
 223        qemu_opt_set(opts, "file", file, &error_abort);
 224    return opts;
 225}
 226
 227DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
 228{
 229    BlockBackend *blk;
 230    DriveInfo *dinfo;
 231
 232    for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 233        dinfo = blk_legacy_dinfo(blk);
 234        if (dinfo && dinfo->type == type
 235            && dinfo->bus == bus && dinfo->unit == unit) {
 236            return dinfo;
 237        }
 238    }
 239
 240    return NULL;
 241}
 242
 243void drive_check_orphaned(void)
 244{
 245    BlockBackend *blk;
 246    DriveInfo *dinfo;
 247    Location loc;
 248    bool orphans = false;
 249
 250    for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 251        dinfo = blk_legacy_dinfo(blk);
 252        if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
 253            dinfo->type != IF_NONE) {
 254            loc_push_none(&loc);
 255            qemu_opts_loc_restore(dinfo->opts);
 256            error_report("machine type does not support"
 257                         " if=%s,bus=%d,unit=%d",
 258                         if_name[dinfo->type], dinfo->bus, dinfo->unit);
 259            loc_pop(&loc);
 260            orphans = true;
 261        }
 262    }
 263
 264    if (orphans) {
 265        exit(1);
 266    }
 267}
 268
 269DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
 270{
 271    return drive_get(type,
 272                     drive_index_to_bus_id(type, index),
 273                     drive_index_to_unit_id(type, index));
 274}
 275
 276int drive_get_max_bus(BlockInterfaceType type)
 277{
 278    int max_bus;
 279    BlockBackend *blk;
 280    DriveInfo *dinfo;
 281
 282    max_bus = -1;
 283    for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 284        dinfo = blk_legacy_dinfo(blk);
 285        if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
 286            max_bus = dinfo->bus;
 287        }
 288    }
 289    return max_bus;
 290}
 291
 292/* Get a block device.  This should only be used for single-drive devices
 293   (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
 294   appropriate bus.  */
 295DriveInfo *drive_get_next(BlockInterfaceType type)
 296{
 297    static int next_block_unit[IF_COUNT];
 298
 299    return drive_get(type, 0, next_block_unit[type]++);
 300}
 301
 302static void bdrv_format_print(void *opaque, const char *name)
 303{
 304    error_printf(" %s", name);
 305}
 306
 307typedef struct {
 308    QEMUBH *bh;
 309    BlockDriverState *bs;
 310} BDRVPutRefBH;
 311
 312static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
 313{
 314    if (!strcmp(buf, "ignore")) {
 315        return BLOCKDEV_ON_ERROR_IGNORE;
 316    } else if (!is_read && !strcmp(buf, "enospc")) {
 317        return BLOCKDEV_ON_ERROR_ENOSPC;
 318    } else if (!strcmp(buf, "stop")) {
 319        return BLOCKDEV_ON_ERROR_STOP;
 320    } else if (!strcmp(buf, "report")) {
 321        return BLOCKDEV_ON_ERROR_REPORT;
 322    } else {
 323        error_setg(errp, "'%s' invalid %s error action",
 324                   buf, is_read ? "read" : "write");
 325        return -1;
 326    }
 327}
 328
 329static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
 330                                  Error **errp)
 331{
 332    const QListEntry *entry;
 333    for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
 334        switch (qobject_type(entry->value)) {
 335
 336        case QTYPE_QSTRING: {
 337            unsigned long long length;
 338            const char *str = qstring_get_str(qobject_to(QString,
 339                                                         entry->value));
 340            if (parse_uint_full(str, &length, 10) == 0 &&
 341                length > 0 && length <= UINT_MAX) {
 342                block_acct_add_interval(stats, (unsigned) length);
 343            } else {
 344                error_setg(errp, "Invalid interval length: %s", str);
 345                return false;
 346            }
 347            break;
 348        }
 349
 350        case QTYPE_QNUM: {
 351            int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
 352
 353            if (length > 0 && length <= UINT_MAX) {
 354                block_acct_add_interval(stats, (unsigned) length);
 355            } else {
 356                error_setg(errp, "Invalid interval length: %" PRId64, length);
 357                return false;
 358            }
 359            break;
 360        }
 361
 362        default:
 363            error_setg(errp, "The specification of stats-intervals is invalid");
 364            return false;
 365        }
 366    }
 367    return true;
 368}
 369
 370typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
 371
 372/* All parameters but @opts are optional and may be set to NULL. */
 373static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
 374    const char **throttling_group, ThrottleConfig *throttle_cfg,
 375    BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
 376{
 377    Error *local_error = NULL;
 378    const char *aio;
 379
 380    if (bdrv_flags) {
 381        if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
 382            *bdrv_flags |= BDRV_O_COPY_ON_READ;
 383        }
 384
 385        if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
 386            if (!strcmp(aio, "native")) {
 387                *bdrv_flags |= BDRV_O_NATIVE_AIO;
 388            } else if (!strcmp(aio, "threads")) {
 389                /* this is the default */
 390            } else {
 391               error_setg(errp, "invalid aio option");
 392               return;
 393            }
 394        }
 395    }
 396
 397    /* disk I/O throttling */
 398    if (throttling_group) {
 399        *throttling_group = qemu_opt_get(opts, "throttling.group");
 400    }
 401
 402    if (throttle_cfg) {
 403        throttle_config_init(throttle_cfg);
 404        throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
 405            qemu_opt_get_number(opts, "throttling.bps-total", 0);
 406        throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
 407            qemu_opt_get_number(opts, "throttling.bps-read", 0);
 408        throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
 409            qemu_opt_get_number(opts, "throttling.bps-write", 0);
 410        throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
 411            qemu_opt_get_number(opts, "throttling.iops-total", 0);
 412        throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
 413            qemu_opt_get_number(opts, "throttling.iops-read", 0);
 414        throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
 415            qemu_opt_get_number(opts, "throttling.iops-write", 0);
 416
 417        throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
 418            qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
 419        throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
 420            qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
 421        throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
 422            qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
 423        throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
 424            qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
 425        throttle_cfg->buckets[THROTTLE_OPS_READ].max =
 426            qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
 427        throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
 428            qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
 429
 430        throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
 431            qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
 432        throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
 433            qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
 434        throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
 435            qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
 436        throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
 437            qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
 438        throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
 439            qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
 440        throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
 441            qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
 442
 443        throttle_cfg->op_size =
 444            qemu_opt_get_number(opts, "throttling.iops-size", 0);
 445
 446        if (!throttle_is_valid(throttle_cfg, errp)) {
 447            return;
 448        }
 449    }
 450
 451    if (detect_zeroes) {
 452        *detect_zeroes =
 453            qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
 454                            qemu_opt_get(opts, "detect-zeroes"),
 455                            BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
 456                            &local_error);
 457        if (local_error) {
 458            error_propagate(errp, local_error);
 459            return;
 460        }
 461    }
 462}
 463
 464/* Takes the ownership of bs_opts */
 465static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
 466                                   Error **errp)
 467{
 468    const char *buf;
 469    int bdrv_flags = 0;
 470    int on_read_error, on_write_error;
 471    bool account_invalid, account_failed;
 472    bool writethrough, read_only;
 473    BlockBackend *blk;
 474    BlockDriverState *bs;
 475    ThrottleConfig cfg;
 476    int snapshot = 0;
 477    Error *error = NULL;
 478    QemuOpts *opts;
 479    QDict *interval_dict = NULL;
 480    QList *interval_list = NULL;
 481    const char *id;
 482    BlockdevDetectZeroesOptions detect_zeroes =
 483        BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
 484    const char *throttling_group = NULL;
 485
 486    /* Check common options by copying from bs_opts to opts, all other options
 487     * stay in bs_opts for processing by bdrv_open(). */
 488    id = qdict_get_try_str(bs_opts, "id");
 489    opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
 490    if (error) {
 491        error_propagate(errp, error);
 492        goto err_no_opts;
 493    }
 494
 495    qemu_opts_absorb_qdict(opts, bs_opts, &error);
 496    if (error) {
 497        error_propagate(errp, error);
 498        goto early_err;
 499    }
 500
 501    if (id) {
 502        qdict_del(bs_opts, "id");
 503    }
 504
 505    /* extract parameters */
 506    snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
 507
 508    account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
 509    account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
 510
 511    writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
 512
 513    id = qemu_opts_id(opts);
 514
 515    qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
 516    qdict_array_split(interval_dict, &interval_list);
 517
 518    if (qdict_size(interval_dict) != 0) {
 519        error_setg(errp, "Invalid option stats-intervals.%s",
 520                   qdict_first(interval_dict)->key);
 521        goto early_err;
 522    }
 523
 524    extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
 525                                    &detect_zeroes, &error);
 526    if (error) {
 527        error_propagate(errp, error);
 528        goto early_err;
 529    }
 530
 531    if ((buf = qemu_opt_get(opts, "format")) != NULL) {
 532        if (is_help_option(buf)) {
 533            error_printf("Supported formats:");
 534            bdrv_iterate_format(bdrv_format_print, NULL);
 535            error_printf("\n");
 536            goto early_err;
 537        }
 538
 539        if (qdict_haskey(bs_opts, "driver")) {
 540            error_setg(errp, "Cannot specify both 'driver' and 'format'");
 541            goto early_err;
 542        }
 543        qdict_put_str(bs_opts, "driver", buf);
 544    }
 545
 546    on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
 547    if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
 548        on_write_error = parse_block_error_action(buf, 0, &error);
 549        if (error) {
 550            error_propagate(errp, error);
 551            goto early_err;
 552        }
 553    }
 554
 555    on_read_error = BLOCKDEV_ON_ERROR_REPORT;
 556    if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
 557        on_read_error = parse_block_error_action(buf, 1, &error);
 558        if (error) {
 559            error_propagate(errp, error);
 560            goto early_err;
 561        }
 562    }
 563
 564    if (snapshot) {
 565        bdrv_flags |= BDRV_O_SNAPSHOT;
 566    }
 567
 568    read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
 569
 570    /* init */
 571    if ((!file || !*file) && !qdict_size(bs_opts)) {
 572        BlockBackendRootState *blk_rs;
 573
 574        blk = blk_new(0, BLK_PERM_ALL);
 575        blk_rs = blk_get_root_state(blk);
 576        blk_rs->open_flags    = bdrv_flags;
 577        blk_rs->read_only     = read_only;
 578        blk_rs->detect_zeroes = detect_zeroes;
 579
 580        qobject_unref(bs_opts);
 581    } else {
 582        if (file && !*file) {
 583            file = NULL;
 584        }
 585
 586        /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
 587         * with other callers) rather than what we want as the real defaults.
 588         * Apply the defaults here instead. */
 589        qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
 590        qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
 591        qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
 592                              read_only ? "on" : "off");
 593        assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
 594
 595        if (runstate_check(RUN_STATE_INMIGRATE)) {
 596            bdrv_flags |= BDRV_O_INACTIVE;
 597        }
 598
 599        blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
 600        if (!blk) {
 601            goto err_no_bs_opts;
 602        }
 603        bs = blk_bs(blk);
 604
 605        bs->detect_zeroes = detect_zeroes;
 606
 607        block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
 608
 609        if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
 610            blk_unref(blk);
 611            blk = NULL;
 612            goto err_no_bs_opts;
 613        }
 614    }
 615
 616    /* disk I/O throttling */
 617    if (throttle_enabled(&cfg)) {
 618        if (!throttling_group) {
 619            throttling_group = id;
 620        }
 621        blk_io_limits_enable(blk, throttling_group);
 622        blk_set_io_limits(blk, &cfg);
 623    }
 624
 625    blk_set_enable_write_cache(blk, !writethrough);
 626    blk_set_on_error(blk, on_read_error, on_write_error);
 627
 628    if (!monitor_add_blk(blk, id, errp)) {
 629        blk_unref(blk);
 630        blk = NULL;
 631        goto err_no_bs_opts;
 632    }
 633
 634err_no_bs_opts:
 635    qemu_opts_del(opts);
 636    qobject_unref(interval_dict);
 637    qobject_unref(interval_list);
 638    return blk;
 639
 640early_err:
 641    qemu_opts_del(opts);
 642    qobject_unref(interval_dict);
 643    qobject_unref(interval_list);
 644err_no_opts:
 645    qobject_unref(bs_opts);
 646    return NULL;
 647}
 648
 649/* Takes the ownership of bs_opts */
 650static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
 651{
 652    int bdrv_flags = 0;
 653
 654    /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
 655     * with other callers) rather than what we want as the real defaults.
 656     * Apply the defaults here instead. */
 657    qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
 658    qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
 659    qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
 660
 661    if (runstate_check(RUN_STATE_INMIGRATE)) {
 662        bdrv_flags |= BDRV_O_INACTIVE;
 663    }
 664
 665    return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
 666}
 667
 668void blockdev_close_all_bdrv_states(void)
 669{
 670    BlockDriverState *bs, *next_bs;
 671
 672    QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
 673        AioContext *ctx = bdrv_get_aio_context(bs);
 674
 675        aio_context_acquire(ctx);
 676        bdrv_unref(bs);
 677        aio_context_release(ctx);
 678    }
 679}
 680
 681/* Iterates over the list of monitor-owned BlockDriverStates */
 682BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
 683{
 684    return bs ? QTAILQ_NEXT(bs, monitor_list)
 685              : QTAILQ_FIRST(&monitor_bdrv_states);
 686}
 687
 688static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
 689                            Error **errp)
 690{
 691    const char *value;
 692
 693    value = qemu_opt_get(opts, from);
 694    if (value) {
 695        if (qemu_opt_find(opts, to)) {
 696            error_setg(errp, "'%s' and its alias '%s' can't be used at the "
 697                       "same time", to, from);
 698            return;
 699        }
 700    }
 701
 702    /* rename all items in opts */
 703    while ((value = qemu_opt_get(opts, from))) {
 704        qemu_opt_set(opts, to, value, &error_abort);
 705        qemu_opt_unset(opts, from);
 706    }
 707}
 708
 709QemuOptsList qemu_legacy_drive_opts = {
 710    .name = "drive",
 711    .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
 712    .desc = {
 713        {
 714            .name = "bus",
 715            .type = QEMU_OPT_NUMBER,
 716            .help = "bus number",
 717        },{
 718            .name = "unit",
 719            .type = QEMU_OPT_NUMBER,
 720            .help = "unit number (i.e. lun for scsi)",
 721        },{
 722            .name = "index",
 723            .type = QEMU_OPT_NUMBER,
 724            .help = "index number",
 725        },{
 726            .name = "media",
 727            .type = QEMU_OPT_STRING,
 728            .help = "media type (disk, cdrom)",
 729        },{
 730            .name = "if",
 731            .type = QEMU_OPT_STRING,
 732            .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
 733        },{
 734            .name = "cyls",
 735            .type = QEMU_OPT_NUMBER,
 736            .help = "number of cylinders (ide disk geometry)",
 737        },{
 738            .name = "heads",
 739            .type = QEMU_OPT_NUMBER,
 740            .help = "number of heads (ide disk geometry)",
 741        },{
 742            .name = "secs",
 743            .type = QEMU_OPT_NUMBER,
 744            .help = "number of sectors (ide disk geometry)",
 745        },{
 746            .name = "trans",
 747            .type = QEMU_OPT_STRING,
 748            .help = "chs translation (auto, lba, none)",
 749        },{
 750            .name = "addr",
 751            .type = QEMU_OPT_STRING,
 752            .help = "pci address (virtio only)",
 753        },{
 754            .name = "serial",
 755            .type = QEMU_OPT_STRING,
 756            .help = "disk serial number",
 757        },{
 758            .name = "file",
 759            .type = QEMU_OPT_STRING,
 760            .help = "file name",
 761        },
 762
 763        /* Options that are passed on, but have special semantics with -drive */
 764        {
 765            .name = BDRV_OPT_READ_ONLY,
 766            .type = QEMU_OPT_BOOL,
 767            .help = "open drive file as read-only",
 768        },{
 769            .name = "rerror",
 770            .type = QEMU_OPT_STRING,
 771            .help = "read error action",
 772        },{
 773            .name = "werror",
 774            .type = QEMU_OPT_STRING,
 775            .help = "write error action",
 776        },{
 777            .name = "copy-on-read",
 778            .type = QEMU_OPT_BOOL,
 779            .help = "copy read data from backing file into image file",
 780        },
 781
 782        { /* end of list */ }
 783    },
 784};
 785
 786DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
 787{
 788    const char *value;
 789    BlockBackend *blk;
 790    DriveInfo *dinfo = NULL;
 791    QDict *bs_opts;
 792    QemuOpts *legacy_opts;
 793    DriveMediaType media = MEDIA_DISK;
 794    BlockInterfaceType type;
 795    int cyls, heads, secs, translation;
 796    int max_devs, bus_id, unit_id, index;
 797    const char *devaddr;
 798    const char *werror, *rerror;
 799    bool read_only = false;
 800    bool copy_on_read;
 801    const char *serial;
 802    const char *filename;
 803    Error *local_err = NULL;
 804    int i;
 805    const char *deprecated[] = {
 806        "serial", "trans", "secs", "heads", "cyls", "addr"
 807    };
 808
 809    /* Change legacy command line options into QMP ones */
 810    static const struct {
 811        const char *from;
 812        const char *to;
 813    } opt_renames[] = {
 814        { "iops",           "throttling.iops-total" },
 815        { "iops_rd",        "throttling.iops-read" },
 816        { "iops_wr",        "throttling.iops-write" },
 817
 818        { "bps",            "throttling.bps-total" },
 819        { "bps_rd",         "throttling.bps-read" },
 820        { "bps_wr",         "throttling.bps-write" },
 821
 822        { "iops_max",       "throttling.iops-total-max" },
 823        { "iops_rd_max",    "throttling.iops-read-max" },
 824        { "iops_wr_max",    "throttling.iops-write-max" },
 825
 826        { "bps_max",        "throttling.bps-total-max" },
 827        { "bps_rd_max",     "throttling.bps-read-max" },
 828        { "bps_wr_max",     "throttling.bps-write-max" },
 829
 830        { "iops_size",      "throttling.iops-size" },
 831
 832        { "group",          "throttling.group" },
 833
 834        { "readonly",       BDRV_OPT_READ_ONLY },
 835    };
 836
 837    for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
 838        qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
 839                        &local_err);
 840        if (local_err) {
 841            error_report_err(local_err);
 842            return NULL;
 843        }
 844    }
 845
 846    value = qemu_opt_get(all_opts, "cache");
 847    if (value) {
 848        int flags = 0;
 849        bool writethrough;
 850
 851        if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
 852            error_report("invalid cache option");
 853            return NULL;
 854        }
 855
 856        /* Specific options take precedence */
 857        if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
 858            qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
 859                              !writethrough, &error_abort);
 860        }
 861        if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
 862            qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
 863                              !!(flags & BDRV_O_NOCACHE), &error_abort);
 864        }
 865        if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
 866            qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
 867                              !!(flags & BDRV_O_NO_FLUSH), &error_abort);
 868        }
 869        qemu_opt_unset(all_opts, "cache");
 870    }
 871
 872    /* Get a QDict for processing the options */
 873    bs_opts = qdict_new();
 874    qemu_opts_to_qdict(all_opts, bs_opts);
 875
 876    legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
 877                                   &error_abort);
 878    qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
 879    if (local_err) {
 880        error_report_err(local_err);
 881        goto fail;
 882    }
 883
 884    /* Other deprecated options */
 885    if (!qtest_enabled()) {
 886        for (i = 0; i < ARRAY_SIZE(deprecated); i++) {
 887            if (qemu_opt_get(legacy_opts, deprecated[i]) != NULL) {
 888                error_report("'%s' is deprecated, please use the corresponding "
 889                             "option of '-device' instead", deprecated[i]);
 890            }
 891        }
 892    }
 893
 894    /* Media type */
 895    value = qemu_opt_get(legacy_opts, "media");
 896    if (value) {
 897        if (!strcmp(value, "disk")) {
 898            media = MEDIA_DISK;
 899        } else if (!strcmp(value, "cdrom")) {
 900            media = MEDIA_CDROM;
 901            read_only = true;
 902        } else {
 903            error_report("'%s' invalid media", value);
 904            goto fail;
 905        }
 906    }
 907
 908    /* copy-on-read is disabled with a warning for read-only devices */
 909    read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
 910    copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
 911
 912    if (read_only && copy_on_read) {
 913        warn_report("disabling copy-on-read on read-only drive");
 914        copy_on_read = false;
 915    }
 916
 917    qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
 918    qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
 919
 920    /* Controller type */
 921    value = qemu_opt_get(legacy_opts, "if");
 922    if (value) {
 923        for (type = 0;
 924             type < IF_COUNT && strcmp(value, if_name[type]);
 925             type++) {
 926        }
 927        if (type == IF_COUNT) {
 928            error_report("unsupported bus type '%s'", value);
 929            goto fail;
 930        }
 931    } else {
 932        type = block_default_type;
 933    }
 934
 935    /* Geometry */
 936    cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
 937    heads = qemu_opt_get_number(legacy_opts, "heads", 0);
 938    secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
 939
 940    if (cyls || heads || secs) {
 941        if (cyls < 1) {
 942            error_report("invalid physical cyls number");
 943            goto fail;
 944        }
 945        if (heads < 1) {
 946            error_report("invalid physical heads number");
 947            goto fail;
 948        }
 949        if (secs < 1) {
 950            error_report("invalid physical secs number");
 951            goto fail;
 952        }
 953    }
 954
 955    translation = BIOS_ATA_TRANSLATION_AUTO;
 956    value = qemu_opt_get(legacy_opts, "trans");
 957    if (value != NULL) {
 958        if (!cyls) {
 959            error_report("'%s' trans must be used with cyls, heads and secs",
 960                         value);
 961            goto fail;
 962        }
 963        if (!strcmp(value, "none")) {
 964            translation = BIOS_ATA_TRANSLATION_NONE;
 965        } else if (!strcmp(value, "lba")) {
 966            translation = BIOS_ATA_TRANSLATION_LBA;
 967        } else if (!strcmp(value, "large")) {
 968            translation = BIOS_ATA_TRANSLATION_LARGE;
 969        } else if (!strcmp(value, "rechs")) {
 970            translation = BIOS_ATA_TRANSLATION_RECHS;
 971        } else if (!strcmp(value, "auto")) {
 972            translation = BIOS_ATA_TRANSLATION_AUTO;
 973        } else {
 974            error_report("'%s' invalid translation type", value);
 975            goto fail;
 976        }
 977    }
 978
 979    if (media == MEDIA_CDROM) {
 980        if (cyls || secs || heads) {
 981            error_report("CHS can't be set with media=cdrom");
 982            goto fail;
 983        }
 984    }
 985
 986    /* Device address specified by bus/unit or index.
 987     * If none was specified, try to find the first free one. */
 988    bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
 989    unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
 990    index   = qemu_opt_get_number(legacy_opts, "index", -1);
 991
 992    max_devs = if_max_devs[type];
 993
 994    if (index != -1) {
 995        if (bus_id != 0 || unit_id != -1) {
 996            error_report("index cannot be used with bus and unit");
 997            goto fail;
 998        }
 999        bus_id = drive_index_to_bus_id(type, index);
1000        unit_id = drive_index_to_unit_id(type, index);
1001    }
1002
1003    if (unit_id == -1) {
1004       unit_id = 0;
1005       while (drive_get(type, bus_id, unit_id) != NULL) {
1006           unit_id++;
1007           if (max_devs && unit_id >= max_devs) {
1008               unit_id -= max_devs;
1009               bus_id++;
1010           }
1011       }
1012    }
1013
1014    if (max_devs && unit_id >= max_devs) {
1015        error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1016        goto fail;
1017    }
1018
1019    if (drive_get(type, bus_id, unit_id) != NULL) {
1020        error_report("drive with bus=%d, unit=%d (index=%d) exists",
1021                     bus_id, unit_id, index);
1022        goto fail;
1023    }
1024
1025    /* Serial number */
1026    serial = qemu_opt_get(legacy_opts, "serial");
1027
1028    /* no id supplied -> create one */
1029    if (qemu_opts_id(all_opts) == NULL) {
1030        char *new_id;
1031        const char *mediastr = "";
1032        if (type == IF_IDE || type == IF_SCSI) {
1033            mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1034        }
1035        if (max_devs) {
1036            new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1037                                     mediastr, unit_id);
1038        } else {
1039            new_id = g_strdup_printf("%s%s%i", if_name[type],
1040                                     mediastr, unit_id);
1041        }
1042        qdict_put_str(bs_opts, "id", new_id);
1043        g_free(new_id);
1044    }
1045
1046    /* Add virtio block device */
1047    devaddr = qemu_opt_get(legacy_opts, "addr");
1048    if (devaddr && type != IF_VIRTIO) {
1049        error_report("addr is not supported by this bus type");
1050        goto fail;
1051    }
1052
1053    if (type == IF_VIRTIO) {
1054        QemuOpts *devopts;
1055        devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1056                                   &error_abort);
1057        if (arch_type == QEMU_ARCH_S390X) {
1058            qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1059        } else {
1060            qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1061        }
1062        qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1063                     &error_abort);
1064        if (devaddr) {
1065            qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1066        }
1067    }
1068
1069    filename = qemu_opt_get(legacy_opts, "file");
1070
1071    /* Check werror/rerror compatibility with if=... */
1072    werror = qemu_opt_get(legacy_opts, "werror");
1073    if (werror != NULL) {
1074        if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1075            type != IF_NONE) {
1076            error_report("werror is not supported by this bus type");
1077            goto fail;
1078        }
1079        qdict_put_str(bs_opts, "werror", werror);
1080    }
1081
1082    rerror = qemu_opt_get(legacy_opts, "rerror");
1083    if (rerror != NULL) {
1084        if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1085            type != IF_NONE) {
1086            error_report("rerror is not supported by this bus type");
1087            goto fail;
1088        }
1089        qdict_put_str(bs_opts, "rerror", rerror);
1090    }
1091
1092    /* Actual block device init: Functionality shared with blockdev-add */
1093    blk = blockdev_init(filename, bs_opts, &local_err);
1094    bs_opts = NULL;
1095    if (!blk) {
1096        if (local_err) {
1097            error_report_err(local_err);
1098        }
1099        goto fail;
1100    } else {
1101        assert(!local_err);
1102    }
1103
1104    /* Create legacy DriveInfo */
1105    dinfo = g_malloc0(sizeof(*dinfo));
1106    dinfo->opts = all_opts;
1107
1108    dinfo->cyls = cyls;
1109    dinfo->heads = heads;
1110    dinfo->secs = secs;
1111    dinfo->trans = translation;
1112
1113    dinfo->type = type;
1114    dinfo->bus = bus_id;
1115    dinfo->unit = unit_id;
1116    dinfo->devaddr = devaddr;
1117    dinfo->serial = g_strdup(serial);
1118
1119    blk_set_legacy_dinfo(blk, dinfo);
1120
1121    switch(type) {
1122    case IF_IDE:
1123    case IF_SCSI:
1124    case IF_XEN:
1125    case IF_NONE:
1126        dinfo->media_cd = media == MEDIA_CDROM;
1127        break;
1128    default:
1129        break;
1130    }
1131
1132fail:
1133    qemu_opts_del(legacy_opts);
1134    qobject_unref(bs_opts);
1135    return dinfo;
1136}
1137
1138static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1139{
1140    BlockDriverState *bs;
1141
1142    bs = bdrv_lookup_bs(name, name, errp);
1143    if (bs == NULL) {
1144        return NULL;
1145    }
1146
1147    if (!bdrv_is_root_node(bs)) {
1148        error_setg(errp, "Need a root block node");
1149        return NULL;
1150    }
1151
1152    if (!bdrv_is_inserted(bs)) {
1153        error_setg(errp, "Device has no medium");
1154        return NULL;
1155    }
1156
1157    return bs;
1158}
1159
1160static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1161                                 Error **errp)
1162{
1163    BlockBackend *blk;
1164
1165    if (!blk_name == !qdev_id) {
1166        error_setg(errp, "Need exactly one of 'device' and 'id'");
1167        return NULL;
1168    }
1169
1170    if (qdev_id) {
1171        blk = blk_by_qdev_id(qdev_id, errp);
1172    } else {
1173        blk = blk_by_name(blk_name);
1174        if (blk == NULL) {
1175            error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1176                      "Device '%s' not found", blk_name);
1177        }
1178    }
1179
1180    return blk;
1181}
1182
1183void hmp_commit(Monitor *mon, const QDict *qdict)
1184{
1185    const char *device = qdict_get_str(qdict, "device");
1186    BlockBackend *blk;
1187    int ret;
1188
1189    if (!strcmp(device, "all")) {
1190        ret = blk_commit_all();
1191    } else {
1192        BlockDriverState *bs;
1193        AioContext *aio_context;
1194
1195        blk = blk_by_name(device);
1196        if (!blk) {
1197            monitor_printf(mon, "Device '%s' not found\n", device);
1198            return;
1199        }
1200        if (!blk_is_available(blk)) {
1201            monitor_printf(mon, "Device '%s' has no medium\n", device);
1202            return;
1203        }
1204
1205        bs = blk_bs(blk);
1206        aio_context = bdrv_get_aio_context(bs);
1207        aio_context_acquire(aio_context);
1208
1209        ret = bdrv_commit(bs);
1210
1211        aio_context_release(aio_context);
1212    }
1213    if (ret < 0) {
1214        monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1215                       strerror(-ret));
1216    }
1217}
1218
1219static void blockdev_do_action(TransactionAction *action, Error **errp)
1220{
1221    TransactionActionList list;
1222
1223    list.value = action;
1224    list.next = NULL;
1225    qmp_transaction(&list, false, NULL, errp);
1226}
1227
1228void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1229                                bool has_node_name, const char *node_name,
1230                                const char *snapshot_file,
1231                                bool has_snapshot_node_name,
1232                                const char *snapshot_node_name,
1233                                bool has_format, const char *format,
1234                                bool has_mode, NewImageMode mode, Error **errp)
1235{
1236    BlockdevSnapshotSync snapshot = {
1237        .has_device = has_device,
1238        .device = (char *) device,
1239        .has_node_name = has_node_name,
1240        .node_name = (char *) node_name,
1241        .snapshot_file = (char *) snapshot_file,
1242        .has_snapshot_node_name = has_snapshot_node_name,
1243        .snapshot_node_name = (char *) snapshot_node_name,
1244        .has_format = has_format,
1245        .format = (char *) format,
1246        .has_mode = has_mode,
1247        .mode = mode,
1248    };
1249    TransactionAction action = {
1250        .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1251        .u.blockdev_snapshot_sync.data = &snapshot,
1252    };
1253    blockdev_do_action(&action, errp);
1254}
1255
1256void qmp_blockdev_snapshot(const char *node, const char *overlay,
1257                           Error **errp)
1258{
1259    BlockdevSnapshot snapshot_data = {
1260        .node = (char *) node,
1261        .overlay = (char *) overlay
1262    };
1263    TransactionAction action = {
1264        .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1265        .u.blockdev_snapshot.data = &snapshot_data,
1266    };
1267    blockdev_do_action(&action, errp);
1268}
1269
1270void qmp_blockdev_snapshot_internal_sync(const char *device,
1271                                         const char *name,
1272                                         Error **errp)
1273{
1274    BlockdevSnapshotInternal snapshot = {
1275        .device = (char *) device,
1276        .name = (char *) name
1277    };
1278    TransactionAction action = {
1279        .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1280        .u.blockdev_snapshot_internal_sync.data = &snapshot,
1281    };
1282    blockdev_do_action(&action, errp);
1283}
1284
1285SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1286                                                         bool has_id,
1287                                                         const char *id,
1288                                                         bool has_name,
1289                                                         const char *name,
1290                                                         Error **errp)
1291{
1292    BlockDriverState *bs;
1293    AioContext *aio_context;
1294    QEMUSnapshotInfo sn;
1295    Error *local_err = NULL;
1296    SnapshotInfo *info = NULL;
1297    int ret;
1298
1299    bs = qmp_get_root_bs(device, errp);
1300    if (!bs) {
1301        return NULL;
1302    }
1303    aio_context = bdrv_get_aio_context(bs);
1304    aio_context_acquire(aio_context);
1305
1306    if (!has_id) {
1307        id = NULL;
1308    }
1309
1310    if (!has_name) {
1311        name = NULL;
1312    }
1313
1314    if (!id && !name) {
1315        error_setg(errp, "Name or id must be provided");
1316        goto out_aio_context;
1317    }
1318
1319    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1320        goto out_aio_context;
1321    }
1322
1323    ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1324    if (local_err) {
1325        error_propagate(errp, local_err);
1326        goto out_aio_context;
1327    }
1328    if (!ret) {
1329        error_setg(errp,
1330                   "Snapshot with id '%s' and name '%s' does not exist on "
1331                   "device '%s'",
1332                   STR_OR_NULL(id), STR_OR_NULL(name), device);
1333        goto out_aio_context;
1334    }
1335
1336    bdrv_snapshot_delete(bs, id, name, &local_err);
1337    if (local_err) {
1338        error_propagate(errp, local_err);
1339        goto out_aio_context;
1340    }
1341
1342    aio_context_release(aio_context);
1343
1344    info = g_new0(SnapshotInfo, 1);
1345    info->id = g_strdup(sn.id_str);
1346    info->name = g_strdup(sn.name);
1347    info->date_nsec = sn.date_nsec;
1348    info->date_sec = sn.date_sec;
1349    info->vm_state_size = sn.vm_state_size;
1350    info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1351    info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1352
1353    return info;
1354
1355out_aio_context:
1356    aio_context_release(aio_context);
1357    return NULL;
1358}
1359
1360/**
1361 * block_dirty_bitmap_lookup:
1362 * Return a dirty bitmap (if present), after validating
1363 * the node reference and bitmap names.
1364 *
1365 * @node: The name of the BDS node to search for bitmaps
1366 * @name: The name of the bitmap to search for
1367 * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1368 * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1369 * @errp: Output pointer for error information. Can be NULL.
1370 *
1371 * @return: A bitmap object on success, or NULL on failure.
1372 */
1373static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1374                                                  const char *name,
1375                                                  BlockDriverState **pbs,
1376                                                  Error **errp)
1377{
1378    BlockDriverState *bs;
1379    BdrvDirtyBitmap *bitmap;
1380
1381    if (!node) {
1382        error_setg(errp, "Node cannot be NULL");
1383        return NULL;
1384    }
1385    if (!name) {
1386        error_setg(errp, "Bitmap name cannot be NULL");
1387        return NULL;
1388    }
1389    bs = bdrv_lookup_bs(node, node, NULL);
1390    if (!bs) {
1391        error_setg(errp, "Node '%s' not found", node);
1392        return NULL;
1393    }
1394
1395    bitmap = bdrv_find_dirty_bitmap(bs, name);
1396    if (!bitmap) {
1397        error_setg(errp, "Dirty bitmap '%s' not found", name);
1398        return NULL;
1399    }
1400
1401    if (pbs) {
1402        *pbs = bs;
1403    }
1404
1405    return bitmap;
1406}
1407
1408/* New and old BlockDriverState structs for atomic group operations */
1409
1410typedef struct BlkActionState BlkActionState;
1411
1412/**
1413 * BlkActionOps:
1414 * Table of operations that define an Action.
1415 *
1416 * @instance_size: Size of state struct, in bytes.
1417 * @prepare: Prepare the work, must NOT be NULL.
1418 * @commit: Commit the changes, can be NULL.
1419 * @abort: Abort the changes on fail, can be NULL.
1420 * @clean: Clean up resources after all transaction actions have called
1421 *         commit() or abort(). Can be NULL.
1422 *
1423 * Only prepare() may fail. In a single transaction, only one of commit() or
1424 * abort() will be called. clean() will always be called if it is present.
1425 */
1426typedef struct BlkActionOps {
1427    size_t instance_size;
1428    void (*prepare)(BlkActionState *common, Error **errp);
1429    void (*commit)(BlkActionState *common);
1430    void (*abort)(BlkActionState *common);
1431    void (*clean)(BlkActionState *common);
1432} BlkActionOps;
1433
1434/**
1435 * BlkActionState:
1436 * Describes one Action's state within a Transaction.
1437 *
1438 * @action: QAPI-defined enum identifying which Action to perform.
1439 * @ops: Table of ActionOps this Action can perform.
1440 * @block_job_txn: Transaction which this action belongs to.
1441 * @entry: List membership for all Actions in this Transaction.
1442 *
1443 * This structure must be arranged as first member in a subclassed type,
1444 * assuming that the compiler will also arrange it to the same offsets as the
1445 * base class.
1446 */
1447struct BlkActionState {
1448    TransactionAction *action;
1449    const BlkActionOps *ops;
1450    JobTxn *block_job_txn;
1451    TransactionProperties *txn_props;
1452    QSIMPLEQ_ENTRY(BlkActionState) entry;
1453};
1454
1455/* internal snapshot private data */
1456typedef struct InternalSnapshotState {
1457    BlkActionState common;
1458    BlockDriverState *bs;
1459    QEMUSnapshotInfo sn;
1460    bool created;
1461} InternalSnapshotState;
1462
1463
1464static int action_check_completion_mode(BlkActionState *s, Error **errp)
1465{
1466    if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1467        error_setg(errp,
1468                   "Action '%s' does not support Transaction property "
1469                   "completion-mode = %s",
1470                   TransactionActionKind_str(s->action->type),
1471                   ActionCompletionMode_str(s->txn_props->completion_mode));
1472        return -1;
1473    }
1474    return 0;
1475}
1476
1477static void internal_snapshot_prepare(BlkActionState *common,
1478                                      Error **errp)
1479{
1480    Error *local_err = NULL;
1481    const char *device;
1482    const char *name;
1483    BlockDriverState *bs;
1484    QEMUSnapshotInfo old_sn, *sn;
1485    bool ret;
1486    qemu_timeval tv;
1487    BlockdevSnapshotInternal *internal;
1488    InternalSnapshotState *state;
1489    AioContext *aio_context;
1490    int ret1;
1491
1492    g_assert(common->action->type ==
1493             TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1494    internal = common->action->u.blockdev_snapshot_internal_sync.data;
1495    state = DO_UPCAST(InternalSnapshotState, common, common);
1496
1497    /* 1. parse input */
1498    device = internal->device;
1499    name = internal->name;
1500
1501    /* 2. check for validation */
1502    if (action_check_completion_mode(common, errp) < 0) {
1503        return;
1504    }
1505
1506    bs = qmp_get_root_bs(device, errp);
1507    if (!bs) {
1508        return;
1509    }
1510
1511    aio_context = bdrv_get_aio_context(bs);
1512    aio_context_acquire(aio_context);
1513
1514    state->bs = bs;
1515
1516    /* Paired with .clean() */
1517    bdrv_drained_begin(bs);
1518
1519    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1520        goto out;
1521    }
1522
1523    if (bdrv_is_read_only(bs)) {
1524        error_setg(errp, "Device '%s' is read only", device);
1525        goto out;
1526    }
1527
1528    if (!bdrv_can_snapshot(bs)) {
1529        error_setg(errp, "Block format '%s' used by device '%s' "
1530                   "does not support internal snapshots",
1531                   bs->drv->format_name, device);
1532        goto out;
1533    }
1534
1535    if (!strlen(name)) {
1536        error_setg(errp, "Name is empty");
1537        goto out;
1538    }
1539
1540    /* check whether a snapshot with name exist */
1541    ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1542                                            &local_err);
1543    if (local_err) {
1544        error_propagate(errp, local_err);
1545        goto out;
1546    } else if (ret) {
1547        error_setg(errp,
1548                   "Snapshot with name '%s' already exists on device '%s'",
1549                   name, device);
1550        goto out;
1551    }
1552
1553    /* 3. take the snapshot */
1554    sn = &state->sn;
1555    pstrcpy(sn->name, sizeof(sn->name), name);
1556    qemu_gettimeofday(&tv);
1557    sn->date_sec = tv.tv_sec;
1558    sn->date_nsec = tv.tv_usec * 1000;
1559    sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1560
1561    ret1 = bdrv_snapshot_create(bs, sn);
1562    if (ret1 < 0) {
1563        error_setg_errno(errp, -ret1,
1564                         "Failed to create snapshot '%s' on device '%s'",
1565                         name, device);
1566        goto out;
1567    }
1568
1569    /* 4. succeed, mark a snapshot is created */
1570    state->created = true;
1571
1572out:
1573    aio_context_release(aio_context);
1574}
1575
1576static void internal_snapshot_abort(BlkActionState *common)
1577{
1578    InternalSnapshotState *state =
1579                             DO_UPCAST(InternalSnapshotState, common, common);
1580    BlockDriverState *bs = state->bs;
1581    QEMUSnapshotInfo *sn = &state->sn;
1582    AioContext *aio_context;
1583    Error *local_error = NULL;
1584
1585    if (!state->created) {
1586        return;
1587    }
1588
1589    aio_context = bdrv_get_aio_context(state->bs);
1590    aio_context_acquire(aio_context);
1591
1592    if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1593        error_reportf_err(local_error,
1594                          "Failed to delete snapshot with id '%s' and "
1595                          "name '%s' on device '%s' in abort: ",
1596                          sn->id_str, sn->name,
1597                          bdrv_get_device_name(bs));
1598    }
1599
1600    aio_context_release(aio_context);
1601}
1602
1603static void internal_snapshot_clean(BlkActionState *common)
1604{
1605    InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1606                                             common, common);
1607    AioContext *aio_context;
1608
1609    if (!state->bs) {
1610        return;
1611    }
1612
1613    aio_context = bdrv_get_aio_context(state->bs);
1614    aio_context_acquire(aio_context);
1615
1616    bdrv_drained_end(state->bs);
1617
1618    aio_context_release(aio_context);
1619}
1620
1621/* external snapshot private data */
1622typedef struct ExternalSnapshotState {
1623    BlkActionState common;
1624    BlockDriverState *old_bs;
1625    BlockDriverState *new_bs;
1626    bool overlay_appended;
1627} ExternalSnapshotState;
1628
1629static void external_snapshot_prepare(BlkActionState *common,
1630                                      Error **errp)
1631{
1632    int flags = 0;
1633    QDict *options = NULL;
1634    Error *local_err = NULL;
1635    /* Device and node name of the image to generate the snapshot from */
1636    const char *device;
1637    const char *node_name;
1638    /* Reference to the new image (for 'blockdev-snapshot') */
1639    const char *snapshot_ref;
1640    /* File name of the new image (for 'blockdev-snapshot-sync') */
1641    const char *new_image_file;
1642    ExternalSnapshotState *state =
1643                             DO_UPCAST(ExternalSnapshotState, common, common);
1644    TransactionAction *action = common->action;
1645    AioContext *aio_context;
1646
1647    /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1648     * purpose but a different set of parameters */
1649    switch (action->type) {
1650    case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1651        {
1652            BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1653            device = s->node;
1654            node_name = s->node;
1655            new_image_file = NULL;
1656            snapshot_ref = s->overlay;
1657        }
1658        break;
1659    case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1660        {
1661            BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1662            device = s->has_device ? s->device : NULL;
1663            node_name = s->has_node_name ? s->node_name : NULL;
1664            new_image_file = s->snapshot_file;
1665            snapshot_ref = NULL;
1666        }
1667        break;
1668    default:
1669        g_assert_not_reached();
1670    }
1671
1672    /* start processing */
1673    if (action_check_completion_mode(common, errp) < 0) {
1674        return;
1675    }
1676
1677    state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1678    if (!state->old_bs) {
1679        return;
1680    }
1681
1682    aio_context = bdrv_get_aio_context(state->old_bs);
1683    aio_context_acquire(aio_context);
1684
1685    /* Paired with .clean() */
1686    bdrv_drained_begin(state->old_bs);
1687
1688    if (!bdrv_is_inserted(state->old_bs)) {
1689        error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1690        goto out;
1691    }
1692
1693    if (bdrv_op_is_blocked(state->old_bs,
1694                           BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1695        goto out;
1696    }
1697
1698    if (!bdrv_is_read_only(state->old_bs)) {
1699        if (bdrv_flush(state->old_bs)) {
1700            error_setg(errp, QERR_IO_ERROR);
1701            goto out;
1702        }
1703    }
1704
1705    if (!bdrv_is_first_non_filter(state->old_bs)) {
1706        error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1707        goto out;
1708    }
1709
1710    if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1711        BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1712        const char *format = s->has_format ? s->format : "qcow2";
1713        enum NewImageMode mode;
1714        const char *snapshot_node_name =
1715            s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1716
1717        if (node_name && !snapshot_node_name) {
1718            error_setg(errp, "New snapshot node name missing");
1719            goto out;
1720        }
1721
1722        if (snapshot_node_name &&
1723            bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1724            error_setg(errp, "New snapshot node name already in use");
1725            goto out;
1726        }
1727
1728        flags = state->old_bs->open_flags;
1729        flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1730        flags |= BDRV_O_NO_BACKING;
1731
1732        /* create new image w/backing file */
1733        mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1734        if (mode != NEW_IMAGE_MODE_EXISTING) {
1735            int64_t size = bdrv_getlength(state->old_bs);
1736            if (size < 0) {
1737                error_setg_errno(errp, -size, "bdrv_getlength failed");
1738                goto out;
1739            }
1740            bdrv_img_create(new_image_file, format,
1741                            state->old_bs->filename,
1742                            state->old_bs->drv->format_name,
1743                            NULL, size, flags, false, &local_err);
1744            if (local_err) {
1745                error_propagate(errp, local_err);
1746                goto out;
1747            }
1748        }
1749
1750        options = qdict_new();
1751        if (s->has_snapshot_node_name) {
1752            qdict_put_str(options, "node-name", snapshot_node_name);
1753        }
1754        qdict_put_str(options, "driver", format);
1755    }
1756
1757    state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1758                              errp);
1759    /* We will manually add the backing_hd field to the bs later */
1760    if (!state->new_bs) {
1761        goto out;
1762    }
1763
1764    if (bdrv_has_blk(state->new_bs)) {
1765        error_setg(errp, "The snapshot is already in use");
1766        goto out;
1767    }
1768
1769    if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1770                           errp)) {
1771        goto out;
1772    }
1773
1774    if (state->new_bs->backing != NULL) {
1775        error_setg(errp, "The snapshot already has a backing image");
1776        goto out;
1777    }
1778
1779    if (!state->new_bs->drv->supports_backing) {
1780        error_setg(errp, "The snapshot does not support backing images");
1781        goto out;
1782    }
1783
1784    bdrv_set_aio_context(state->new_bs, aio_context);
1785
1786    /* This removes our old bs and adds the new bs. This is an operation that
1787     * can fail, so we need to do it in .prepare; undoing it for abort is
1788     * always possible. */
1789    bdrv_ref(state->new_bs);
1790    bdrv_append(state->new_bs, state->old_bs, &local_err);
1791    if (local_err) {
1792        error_propagate(errp, local_err);
1793        goto out;
1794    }
1795    state->overlay_appended = true;
1796
1797out:
1798    aio_context_release(aio_context);
1799}
1800
1801static void external_snapshot_commit(BlkActionState *common)
1802{
1803    ExternalSnapshotState *state =
1804                             DO_UPCAST(ExternalSnapshotState, common, common);
1805    AioContext *aio_context;
1806
1807    aio_context = bdrv_get_aio_context(state->old_bs);
1808    aio_context_acquire(aio_context);
1809
1810    /* We don't need (or want) to use the transactional
1811     * bdrv_reopen_multiple() across all the entries at once, because we
1812     * don't want to abort all of them if one of them fails the reopen */
1813    if (!atomic_read(&state->old_bs->copy_on_read)) {
1814        bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1815                    NULL);
1816    }
1817
1818    aio_context_release(aio_context);
1819}
1820
1821static void external_snapshot_abort(BlkActionState *common)
1822{
1823    ExternalSnapshotState *state =
1824                             DO_UPCAST(ExternalSnapshotState, common, common);
1825    if (state->new_bs) {
1826        if (state->overlay_appended) {
1827            AioContext *aio_context;
1828
1829            aio_context = bdrv_get_aio_context(state->old_bs);
1830            aio_context_acquire(aio_context);
1831
1832            bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1833                                          close state->old_bs; we need it */
1834            bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1835            bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1836            bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1837
1838            aio_context_release(aio_context);
1839        }
1840    }
1841}
1842
1843static void external_snapshot_clean(BlkActionState *common)
1844{
1845    ExternalSnapshotState *state =
1846                             DO_UPCAST(ExternalSnapshotState, common, common);
1847    AioContext *aio_context;
1848
1849    if (!state->old_bs) {
1850        return;
1851    }
1852
1853    aio_context = bdrv_get_aio_context(state->old_bs);
1854    aio_context_acquire(aio_context);
1855
1856    bdrv_drained_end(state->old_bs);
1857    bdrv_unref(state->new_bs);
1858
1859    aio_context_release(aio_context);
1860}
1861
1862typedef struct DriveBackupState {
1863    BlkActionState common;
1864    BlockDriverState *bs;
1865    BlockJob *job;
1866} DriveBackupState;
1867
1868static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
1869                            Error **errp);
1870
1871static void drive_backup_prepare(BlkActionState *common, Error **errp)
1872{
1873    DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1874    BlockDriverState *bs;
1875    DriveBackup *backup;
1876    AioContext *aio_context;
1877    Error *local_err = NULL;
1878
1879    assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1880    backup = common->action->u.drive_backup.data;
1881
1882    bs = qmp_get_root_bs(backup->device, errp);
1883    if (!bs) {
1884        return;
1885    }
1886
1887    aio_context = bdrv_get_aio_context(bs);
1888    aio_context_acquire(aio_context);
1889
1890    /* Paired with .clean() */
1891    bdrv_drained_begin(bs);
1892
1893    state->bs = bs;
1894
1895    state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1896    if (local_err) {
1897        error_propagate(errp, local_err);
1898        goto out;
1899    }
1900
1901out:
1902    aio_context_release(aio_context);
1903}
1904
1905static void drive_backup_commit(BlkActionState *common)
1906{
1907    DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1908    AioContext *aio_context;
1909
1910    aio_context = bdrv_get_aio_context(state->bs);
1911    aio_context_acquire(aio_context);
1912
1913    assert(state->job);
1914    job_start(&state->job->job);
1915
1916    aio_context_release(aio_context);
1917}
1918
1919static void drive_backup_abort(BlkActionState *common)
1920{
1921    DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1922
1923    if (state->job) {
1924        AioContext *aio_context;
1925
1926        aio_context = bdrv_get_aio_context(state->bs);
1927        aio_context_acquire(aio_context);
1928
1929        job_cancel_sync(&state->job->job);
1930
1931        aio_context_release(aio_context);
1932    }
1933}
1934
1935static void drive_backup_clean(BlkActionState *common)
1936{
1937    DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1938    AioContext *aio_context;
1939
1940    if (!state->bs) {
1941        return;
1942    }
1943
1944    aio_context = bdrv_get_aio_context(state->bs);
1945    aio_context_acquire(aio_context);
1946
1947    bdrv_drained_end(state->bs);
1948
1949    aio_context_release(aio_context);
1950}
1951
1952typedef struct BlockdevBackupState {
1953    BlkActionState common;
1954    BlockDriverState *bs;
1955    BlockJob *job;
1956} BlockdevBackupState;
1957
1958static BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
1959                                    Error **errp);
1960
1961static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1962{
1963    BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1964    BlockdevBackup *backup;
1965    BlockDriverState *bs, *target;
1966    AioContext *aio_context;
1967    Error *local_err = NULL;
1968
1969    assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1970    backup = common->action->u.blockdev_backup.data;
1971
1972    bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1973    if (!bs) {
1974        return;
1975    }
1976
1977    target = bdrv_lookup_bs(backup->target, backup->target, errp);
1978    if (!target) {
1979        return;
1980    }
1981
1982    aio_context = bdrv_get_aio_context(bs);
1983    if (aio_context != bdrv_get_aio_context(target)) {
1984        error_setg(errp, "Backup between two IO threads is not implemented");
1985        return;
1986    }
1987    aio_context_acquire(aio_context);
1988    state->bs = bs;
1989
1990    /* Paired with .clean() */
1991    bdrv_drained_begin(state->bs);
1992
1993    state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1994    if (local_err) {
1995        error_propagate(errp, local_err);
1996        goto out;
1997    }
1998
1999out:
2000    aio_context_release(aio_context);
2001}
2002
2003static void blockdev_backup_commit(BlkActionState *common)
2004{
2005    BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2006    AioContext *aio_context;
2007
2008    aio_context = bdrv_get_aio_context(state->bs);
2009    aio_context_acquire(aio_context);
2010
2011    assert(state->job);
2012    job_start(&state->job->job);
2013
2014    aio_context_release(aio_context);
2015}
2016
2017static void blockdev_backup_abort(BlkActionState *common)
2018{
2019    BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2020
2021    if (state->job) {
2022        AioContext *aio_context;
2023
2024        aio_context = bdrv_get_aio_context(state->bs);
2025        aio_context_acquire(aio_context);
2026
2027        job_cancel_sync(&state->job->job);
2028
2029        aio_context_release(aio_context);
2030    }
2031}
2032
2033static void blockdev_backup_clean(BlkActionState *common)
2034{
2035    BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2036    AioContext *aio_context;
2037
2038    if (!state->bs) {
2039        return;
2040    }
2041
2042    aio_context = bdrv_get_aio_context(state->bs);
2043    aio_context_acquire(aio_context);
2044
2045    bdrv_drained_end(state->bs);
2046
2047    aio_context_release(aio_context);
2048}
2049
2050typedef struct BlockDirtyBitmapState {
2051    BlkActionState common;
2052    BdrvDirtyBitmap *bitmap;
2053    BlockDriverState *bs;
2054    HBitmap *backup;
2055    bool prepared;
2056    bool was_enabled;
2057} BlockDirtyBitmapState;
2058
2059static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2060                                           Error **errp)
2061{
2062    Error *local_err = NULL;
2063    BlockDirtyBitmapAdd *action;
2064    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2065                                             common, common);
2066
2067    if (action_check_completion_mode(common, errp) < 0) {
2068        return;
2069    }
2070
2071    action = common->action->u.block_dirty_bitmap_add.data;
2072    /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2073    qmp_block_dirty_bitmap_add(action->node, action->name,
2074                               action->has_granularity, action->granularity,
2075                               action->has_persistent, action->persistent,
2076                               action->has_autoload, action->autoload,
2077                               action->has_x_disabled, action->x_disabled,
2078                               &local_err);
2079
2080    if (!local_err) {
2081        state->prepared = true;
2082    } else {
2083        error_propagate(errp, local_err);
2084    }
2085}
2086
2087static void block_dirty_bitmap_add_abort(BlkActionState *common)
2088{
2089    BlockDirtyBitmapAdd *action;
2090    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2091                                             common, common);
2092
2093    action = common->action->u.block_dirty_bitmap_add.data;
2094    /* Should not be able to fail: IF the bitmap was added via .prepare(),
2095     * then the node reference and bitmap name must have been valid.
2096     */
2097    if (state->prepared) {
2098        qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2099    }
2100}
2101
2102static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2103                                             Error **errp)
2104{
2105    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2106                                             common, common);
2107    BlockDirtyBitmap *action;
2108
2109    if (action_check_completion_mode(common, errp) < 0) {
2110        return;
2111    }
2112
2113    action = common->action->u.block_dirty_bitmap_clear.data;
2114    state->bitmap = block_dirty_bitmap_lookup(action->node,
2115                                              action->name,
2116                                              &state->bs,
2117                                              errp);
2118    if (!state->bitmap) {
2119        return;
2120    }
2121
2122    if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2123        error_setg(errp, "Cannot modify a frozen bitmap");
2124        return;
2125    } else if (bdrv_dirty_bitmap_qmp_locked(state->bitmap)) {
2126        error_setg(errp, "Cannot modify a locked bitmap");
2127        return;
2128    } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2129        error_setg(errp, "Cannot clear a disabled bitmap");
2130        return;
2131    } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2132        error_setg(errp, "Cannot clear a readonly bitmap");
2133        return;
2134    }
2135
2136    bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2137}
2138
2139static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2140{
2141    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2142                                             common, common);
2143
2144    if (state->backup) {
2145        bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2146    }
2147}
2148
2149static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2150{
2151    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2152                                             common, common);
2153
2154    hbitmap_free(state->backup);
2155}
2156
2157static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2158                                              Error **errp)
2159{
2160    BlockDirtyBitmap *action;
2161    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2162                                             common, common);
2163
2164    if (action_check_completion_mode(common, errp) < 0) {
2165        return;
2166    }
2167
2168    action = common->action->u.x_block_dirty_bitmap_enable.data;
2169    state->bitmap = block_dirty_bitmap_lookup(action->node,
2170                                              action->name,
2171                                              NULL,
2172                                              errp);
2173    if (!state->bitmap) {
2174        return;
2175    }
2176
2177    state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2178    bdrv_enable_dirty_bitmap(state->bitmap);
2179}
2180
2181static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2182{
2183    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2184                                             common, common);
2185
2186    if (!state->was_enabled) {
2187        bdrv_disable_dirty_bitmap(state->bitmap);
2188    }
2189}
2190
2191static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2192                                               Error **errp)
2193{
2194    BlockDirtyBitmap *action;
2195    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2196                                             common, common);
2197
2198    if (action_check_completion_mode(common, errp) < 0) {
2199        return;
2200    }
2201
2202    action = common->action->u.x_block_dirty_bitmap_disable.data;
2203    state->bitmap = block_dirty_bitmap_lookup(action->node,
2204                                              action->name,
2205                                              NULL,
2206                                              errp);
2207    if (!state->bitmap) {
2208        return;
2209    }
2210
2211    state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2212    bdrv_disable_dirty_bitmap(state->bitmap);
2213}
2214
2215static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2216{
2217    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2218                                             common, common);
2219
2220    if (state->was_enabled) {
2221        bdrv_enable_dirty_bitmap(state->bitmap);
2222    }
2223}
2224
2225static void abort_prepare(BlkActionState *common, Error **errp)
2226{
2227    error_setg(errp, "Transaction aborted using Abort action");
2228}
2229
2230static void abort_commit(BlkActionState *common)
2231{
2232    g_assert_not_reached(); /* this action never succeeds */
2233}
2234
2235static const BlkActionOps actions[] = {
2236    [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2237        .instance_size = sizeof(ExternalSnapshotState),
2238        .prepare  = external_snapshot_prepare,
2239        .commit   = external_snapshot_commit,
2240        .abort = external_snapshot_abort,
2241        .clean = external_snapshot_clean,
2242    },
2243    [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2244        .instance_size = sizeof(ExternalSnapshotState),
2245        .prepare  = external_snapshot_prepare,
2246        .commit   = external_snapshot_commit,
2247        .abort = external_snapshot_abort,
2248        .clean = external_snapshot_clean,
2249    },
2250    [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2251        .instance_size = sizeof(DriveBackupState),
2252        .prepare = drive_backup_prepare,
2253        .commit = drive_backup_commit,
2254        .abort = drive_backup_abort,
2255        .clean = drive_backup_clean,
2256    },
2257    [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2258        .instance_size = sizeof(BlockdevBackupState),
2259        .prepare = blockdev_backup_prepare,
2260        .commit = blockdev_backup_commit,
2261        .abort = blockdev_backup_abort,
2262        .clean = blockdev_backup_clean,
2263    },
2264    [TRANSACTION_ACTION_KIND_ABORT] = {
2265        .instance_size = sizeof(BlkActionState),
2266        .prepare = abort_prepare,
2267        .commit = abort_commit,
2268    },
2269    [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2270        .instance_size = sizeof(InternalSnapshotState),
2271        .prepare  = internal_snapshot_prepare,
2272        .abort = internal_snapshot_abort,
2273        .clean = internal_snapshot_clean,
2274    },
2275    [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2276        .instance_size = sizeof(BlockDirtyBitmapState),
2277        .prepare = block_dirty_bitmap_add_prepare,
2278        .abort = block_dirty_bitmap_add_abort,
2279    },
2280    [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2281        .instance_size = sizeof(BlockDirtyBitmapState),
2282        .prepare = block_dirty_bitmap_clear_prepare,
2283        .commit = block_dirty_bitmap_clear_commit,
2284        .abort = block_dirty_bitmap_clear_abort,
2285    },
2286    [TRANSACTION_ACTION_KIND_X_BLOCK_DIRTY_BITMAP_ENABLE] = {
2287        .instance_size = sizeof(BlockDirtyBitmapState),
2288        .prepare = block_dirty_bitmap_enable_prepare,
2289        .abort = block_dirty_bitmap_enable_abort,
2290    },
2291    [TRANSACTION_ACTION_KIND_X_BLOCK_DIRTY_BITMAP_DISABLE] = {
2292        .instance_size = sizeof(BlockDirtyBitmapState),
2293        .prepare = block_dirty_bitmap_disable_prepare,
2294        .abort = block_dirty_bitmap_disable_abort,
2295     }
2296};
2297
2298/**
2299 * Allocate a TransactionProperties structure if necessary, and fill
2300 * that structure with desired defaults if they are unset.
2301 */
2302static TransactionProperties *get_transaction_properties(
2303    TransactionProperties *props)
2304{
2305    if (!props) {
2306        props = g_new0(TransactionProperties, 1);
2307    }
2308
2309    if (!props->has_completion_mode) {
2310        props->has_completion_mode = true;
2311        props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2312    }
2313
2314    return props;
2315}
2316
2317/*
2318 * 'Atomic' group operations.  The operations are performed as a set, and if
2319 * any fail then we roll back all operations in the group.
2320 */
2321void qmp_transaction(TransactionActionList *dev_list,
2322                     bool has_props,
2323                     struct TransactionProperties *props,
2324                     Error **errp)
2325{
2326    TransactionActionList *dev_entry = dev_list;
2327    JobTxn *block_job_txn = NULL;
2328    BlkActionState *state, *next;
2329    Error *local_err = NULL;
2330
2331    QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2332    QSIMPLEQ_INIT(&snap_bdrv_states);
2333
2334    /* Does this transaction get canceled as a group on failure?
2335     * If not, we don't really need to make a JobTxn.
2336     */
2337    props = get_transaction_properties(props);
2338    if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2339        block_job_txn = job_txn_new();
2340    }
2341
2342    /* drain all i/o before any operations */
2343    bdrv_drain_all();
2344
2345    /* We don't do anything in this loop that commits us to the operations */
2346    while (NULL != dev_entry) {
2347        TransactionAction *dev_info = NULL;
2348        const BlkActionOps *ops;
2349
2350        dev_info = dev_entry->value;
2351        dev_entry = dev_entry->next;
2352
2353        assert(dev_info->type < ARRAY_SIZE(actions));
2354
2355        ops = &actions[dev_info->type];
2356        assert(ops->instance_size > 0);
2357
2358        state = g_malloc0(ops->instance_size);
2359        state->ops = ops;
2360        state->action = dev_info;
2361        state->block_job_txn = block_job_txn;
2362        state->txn_props = props;
2363        QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2364
2365        state->ops->prepare(state, &local_err);
2366        if (local_err) {
2367            error_propagate(errp, local_err);
2368            goto delete_and_fail;
2369        }
2370    }
2371
2372    QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2373        if (state->ops->commit) {
2374            state->ops->commit(state);
2375        }
2376    }
2377
2378    /* success */
2379    goto exit;
2380
2381delete_and_fail:
2382    /* failure, and it is all-or-none; roll back all operations */
2383    QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2384        if (state->ops->abort) {
2385            state->ops->abort(state);
2386        }
2387    }
2388exit:
2389    QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2390        if (state->ops->clean) {
2391            state->ops->clean(state);
2392        }
2393        g_free(state);
2394    }
2395    if (!has_props) {
2396        qapi_free_TransactionProperties(props);
2397    }
2398    job_txn_unref(block_job_txn);
2399}
2400
2401void qmp_eject(bool has_device, const char *device,
2402               bool has_id, const char *id,
2403               bool has_force, bool force, Error **errp)
2404{
2405    Error *local_err = NULL;
2406    int rc;
2407
2408    if (!has_force) {
2409        force = false;
2410    }
2411
2412    rc = do_open_tray(has_device ? device : NULL,
2413                      has_id ? id : NULL,
2414                      force, &local_err);
2415    if (rc && rc != -ENOSYS) {
2416        error_propagate(errp, local_err);
2417        return;
2418    }
2419    error_free(local_err);
2420
2421    blockdev_remove_medium(has_device, device, has_id, id, errp);
2422}
2423
2424void qmp_block_passwd(bool has_device, const char *device,
2425                      bool has_node_name, const char *node_name,
2426                      const char *password, Error **errp)
2427{
2428    error_setg(errp,
2429               "Setting block passwords directly is no longer supported");
2430}
2431
2432/*
2433 * Attempt to open the tray of @device.
2434 * If @force, ignore its tray lock.
2435 * Else, if the tray is locked, don't open it, but ask the guest to open it.
2436 * On error, store an error through @errp and return -errno.
2437 * If @device does not exist, return -ENODEV.
2438 * If it has no removable media, return -ENOTSUP.
2439 * If it has no tray, return -ENOSYS.
2440 * If the guest was asked to open the tray, return -EINPROGRESS.
2441 * Else, return 0.
2442 */
2443static int do_open_tray(const char *blk_name, const char *qdev_id,
2444                        bool force, Error **errp)
2445{
2446    BlockBackend *blk;
2447    const char *device = qdev_id ?: blk_name;
2448    bool locked;
2449
2450    blk = qmp_get_blk(blk_name, qdev_id, errp);
2451    if (!blk) {
2452        return -ENODEV;
2453    }
2454
2455    if (!blk_dev_has_removable_media(blk)) {
2456        error_setg(errp, "Device '%s' is not removable", device);
2457        return -ENOTSUP;
2458    }
2459
2460    if (!blk_dev_has_tray(blk)) {
2461        error_setg(errp, "Device '%s' does not have a tray", device);
2462        return -ENOSYS;
2463    }
2464
2465    if (blk_dev_is_tray_open(blk)) {
2466        return 0;
2467    }
2468
2469    locked = blk_dev_is_medium_locked(blk);
2470    if (locked) {
2471        blk_dev_eject_request(blk, force);
2472    }
2473
2474    if (!locked || force) {
2475        blk_dev_change_media_cb(blk, false, &error_abort);
2476    }
2477
2478    if (locked && !force) {
2479        error_setg(errp, "Device '%s' is locked and force was not specified, "
2480                   "wait for tray to open and try again", device);
2481        return -EINPROGRESS;
2482    }
2483
2484    return 0;
2485}
2486
2487void qmp_blockdev_open_tray(bool has_device, const char *device,
2488                            bool has_id, const char *id,
2489                            bool has_force, bool force,
2490                            Error **errp)
2491{
2492    Error *local_err = NULL;
2493    int rc;
2494
2495    if (!has_force) {
2496        force = false;
2497    }
2498    rc = do_open_tray(has_device ? device : NULL,
2499                      has_id ? id : NULL,
2500                      force, &local_err);
2501    if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2502        error_propagate(errp, local_err);
2503        return;
2504    }
2505    error_free(local_err);
2506}
2507
2508void qmp_blockdev_close_tray(bool has_device, const char *device,
2509                             bool has_id, const char *id,
2510                             Error **errp)
2511{
2512    BlockBackend *blk;
2513    Error *local_err = NULL;
2514
2515    device = has_device ? device : NULL;
2516    id = has_id ? id : NULL;
2517
2518    blk = qmp_get_blk(device, id, errp);
2519    if (!blk) {
2520        return;
2521    }
2522
2523    if (!blk_dev_has_removable_media(blk)) {
2524        error_setg(errp, "Device '%s' is not removable", device ?: id);
2525        return;
2526    }
2527
2528    if (!blk_dev_has_tray(blk)) {
2529        /* Ignore this command on tray-less devices */
2530        return;
2531    }
2532
2533    if (!blk_dev_is_tray_open(blk)) {
2534        return;
2535    }
2536
2537    blk_dev_change_media_cb(blk, true, &local_err);
2538    if (local_err) {
2539        error_propagate(errp, local_err);
2540        return;
2541    }
2542}
2543
2544static void blockdev_remove_medium(bool has_device, const char *device,
2545                                   bool has_id, const char *id, Error **errp)
2546{
2547    BlockBackend *blk;
2548    BlockDriverState *bs;
2549    AioContext *aio_context;
2550    bool has_attached_device;
2551
2552    device = has_device ? device : NULL;
2553    id = has_id ? id : NULL;
2554
2555    blk = qmp_get_blk(device, id, errp);
2556    if (!blk) {
2557        return;
2558    }
2559
2560    /* For BBs without a device, we can exchange the BDS tree at will */
2561    has_attached_device = blk_get_attached_dev(blk);
2562
2563    if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2564        error_setg(errp, "Device '%s' is not removable", device ?: id);
2565        return;
2566    }
2567
2568    if (has_attached_device && blk_dev_has_tray(blk) &&
2569        !blk_dev_is_tray_open(blk))
2570    {
2571        error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2572        return;
2573    }
2574
2575    bs = blk_bs(blk);
2576    if (!bs) {
2577        return;
2578    }
2579
2580    aio_context = bdrv_get_aio_context(bs);
2581    aio_context_acquire(aio_context);
2582
2583    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2584        goto out;
2585    }
2586
2587    blk_remove_bs(blk);
2588
2589    if (!blk_dev_has_tray(blk)) {
2590        /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2591         * called at all); therefore, the medium needs to be ejected here.
2592         * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2593         * value passed here (i.e. false). */
2594        blk_dev_change_media_cb(blk, false, &error_abort);
2595    }
2596
2597out:
2598    aio_context_release(aio_context);
2599}
2600
2601void qmp_blockdev_remove_medium(const char *id, Error **errp)
2602{
2603    blockdev_remove_medium(false, NULL, true, id, errp);
2604}
2605
2606static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2607                                            BlockDriverState *bs, Error **errp)
2608{
2609    Error *local_err = NULL;
2610    bool has_device;
2611    int ret;
2612
2613    /* For BBs without a device, we can exchange the BDS tree at will */
2614    has_device = blk_get_attached_dev(blk);
2615
2616    if (has_device && !blk_dev_has_removable_media(blk)) {
2617        error_setg(errp, "Device is not removable");
2618        return;
2619    }
2620
2621    if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2622        error_setg(errp, "Tray of the device is not open");
2623        return;
2624    }
2625
2626    if (blk_bs(blk)) {
2627        error_setg(errp, "There already is a medium in the device");
2628        return;
2629    }
2630
2631    ret = blk_insert_bs(blk, bs, errp);
2632    if (ret < 0) {
2633        return;
2634    }
2635
2636    if (!blk_dev_has_tray(blk)) {
2637        /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2638         * called at all); therefore, the medium needs to be pushed into the
2639         * slot here.
2640         * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2641         * value passed here (i.e. true). */
2642        blk_dev_change_media_cb(blk, true, &local_err);
2643        if (local_err) {
2644            error_propagate(errp, local_err);
2645            blk_remove_bs(blk);
2646            return;
2647        }
2648    }
2649}
2650
2651static void blockdev_insert_medium(bool has_device, const char *device,
2652                                   bool has_id, const char *id,
2653                                   const char *node_name, Error **errp)
2654{
2655    BlockBackend *blk;
2656    BlockDriverState *bs;
2657
2658    blk = qmp_get_blk(has_device ? device : NULL,
2659                      has_id ? id : NULL,
2660                      errp);
2661    if (!blk) {
2662        return;
2663    }
2664
2665    bs = bdrv_find_node(node_name);
2666    if (!bs) {
2667        error_setg(errp, "Node '%s' not found", node_name);
2668        return;
2669    }
2670
2671    if (bdrv_has_blk(bs)) {
2672        error_setg(errp, "Node '%s' is already in use", node_name);
2673        return;
2674    }
2675
2676    qmp_blockdev_insert_anon_medium(blk, bs, errp);
2677}
2678
2679void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2680                                Error **errp)
2681{
2682    blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2683}
2684
2685void qmp_blockdev_change_medium(bool has_device, const char *device,
2686                                bool has_id, const char *id,
2687                                const char *filename,
2688                                bool has_format, const char *format,
2689                                bool has_read_only,
2690                                BlockdevChangeReadOnlyMode read_only,
2691                                Error **errp)
2692{
2693    BlockBackend *blk;
2694    BlockDriverState *medium_bs = NULL;
2695    int bdrv_flags;
2696    bool detect_zeroes;
2697    int rc;
2698    QDict *options = NULL;
2699    Error *err = NULL;
2700
2701    blk = qmp_get_blk(has_device ? device : NULL,
2702                      has_id ? id : NULL,
2703                      errp);
2704    if (!blk) {
2705        goto fail;
2706    }
2707
2708    if (blk_bs(blk)) {
2709        blk_update_root_state(blk);
2710    }
2711
2712    bdrv_flags = blk_get_open_flags_from_root_state(blk);
2713    bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2714        BDRV_O_PROTOCOL);
2715
2716    if (!has_read_only) {
2717        read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2718    }
2719
2720    switch (read_only) {
2721    case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2722        break;
2723
2724    case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2725        bdrv_flags &= ~BDRV_O_RDWR;
2726        break;
2727
2728    case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2729        bdrv_flags |= BDRV_O_RDWR;
2730        break;
2731
2732    default:
2733        abort();
2734    }
2735
2736    options = qdict_new();
2737    detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2738    qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2739
2740    if (has_format) {
2741        qdict_put_str(options, "driver", format);
2742    }
2743
2744    medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2745    if (!medium_bs) {
2746        goto fail;
2747    }
2748
2749    rc = do_open_tray(has_device ? device : NULL,
2750                      has_id ? id : NULL,
2751                      false, &err);
2752    if (rc && rc != -ENOSYS) {
2753        error_propagate(errp, err);
2754        goto fail;
2755    }
2756    error_free(err);
2757    err = NULL;
2758
2759    blockdev_remove_medium(has_device, device, has_id, id, &err);
2760    if (err) {
2761        error_propagate(errp, err);
2762        goto fail;
2763    }
2764
2765    qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2766    if (err) {
2767        error_propagate(errp, err);
2768        goto fail;
2769    }
2770
2771    qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2772
2773fail:
2774    /* If the medium has been inserted, the device has its own reference, so
2775     * ours must be relinquished; and if it has not been inserted successfully,
2776     * the reference must be relinquished anyway */
2777    bdrv_unref(medium_bs);
2778}
2779
2780/* throttling disk I/O limits */
2781void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2782{
2783    ThrottleConfig cfg;
2784    BlockDriverState *bs;
2785    BlockBackend *blk;
2786    AioContext *aio_context;
2787
2788    blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2789                      arg->has_id ? arg->id : NULL,
2790                      errp);
2791    if (!blk) {
2792        return;
2793    }
2794
2795    aio_context = blk_get_aio_context(blk);
2796    aio_context_acquire(aio_context);
2797
2798    bs = blk_bs(blk);
2799    if (!bs) {
2800        error_setg(errp, "Device has no medium");
2801        goto out;
2802    }
2803
2804    throttle_config_init(&cfg);
2805    cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2806    cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2807    cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2808
2809    cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2810    cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2811    cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2812
2813    if (arg->has_bps_max) {
2814        cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2815    }
2816    if (arg->has_bps_rd_max) {
2817        cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2818    }
2819    if (arg->has_bps_wr_max) {
2820        cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2821    }
2822    if (arg->has_iops_max) {
2823        cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2824    }
2825    if (arg->has_iops_rd_max) {
2826        cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2827    }
2828    if (arg->has_iops_wr_max) {
2829        cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2830    }
2831
2832    if (arg->has_bps_max_length) {
2833        cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2834    }
2835    if (arg->has_bps_rd_max_length) {
2836        cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2837    }
2838    if (arg->has_bps_wr_max_length) {
2839        cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2840    }
2841    if (arg->has_iops_max_length) {
2842        cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2843    }
2844    if (arg->has_iops_rd_max_length) {
2845        cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2846    }
2847    if (arg->has_iops_wr_max_length) {
2848        cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2849    }
2850
2851    if (arg->has_iops_size) {
2852        cfg.op_size = arg->iops_size;
2853    }
2854
2855    if (!throttle_is_valid(&cfg, errp)) {
2856        goto out;
2857    }
2858
2859    if (throttle_enabled(&cfg)) {
2860        /* Enable I/O limits if they're not enabled yet, otherwise
2861         * just update the throttling group. */
2862        if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2863            blk_io_limits_enable(blk,
2864                                 arg->has_group ? arg->group :
2865                                 arg->has_device ? arg->device :
2866                                 arg->id);
2867        } else if (arg->has_group) {
2868            blk_io_limits_update_group(blk, arg->group);
2869        }
2870        /* Set the new throttling configuration */
2871        blk_set_io_limits(blk, &cfg);
2872    } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2873        /* If all throttling settings are set to 0, disable I/O limits */
2874        blk_io_limits_disable(blk);
2875    }
2876
2877out:
2878    aio_context_release(aio_context);
2879}
2880
2881void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2882                                bool has_granularity, uint32_t granularity,
2883                                bool has_persistent, bool persistent,
2884                                bool has_autoload, bool autoload,
2885                                bool has_disabled, bool disabled,
2886                                Error **errp)
2887{
2888    BlockDriverState *bs;
2889    BdrvDirtyBitmap *bitmap;
2890
2891    if (!name || name[0] == '\0') {
2892        error_setg(errp, "Bitmap name cannot be empty");
2893        return;
2894    }
2895
2896    bs = bdrv_lookup_bs(node, node, errp);
2897    if (!bs) {
2898        return;
2899    }
2900
2901    if (has_granularity) {
2902        if (granularity < 512 || !is_power_of_2(granularity)) {
2903            error_setg(errp, "Granularity must be power of 2 "
2904                             "and at least 512");
2905            return;
2906        }
2907    } else {
2908        /* Default to cluster size, if available: */
2909        granularity = bdrv_get_default_bitmap_granularity(bs);
2910    }
2911
2912    if (!has_persistent) {
2913        persistent = false;
2914    }
2915
2916    if (has_autoload) {
2917        warn_report("Autoload option is deprecated and its value is ignored");
2918    }
2919
2920    if (!has_disabled) {
2921        disabled = false;
2922    }
2923
2924    if (persistent &&
2925        !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2926    {
2927        return;
2928    }
2929
2930    bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2931    if (bitmap == NULL) {
2932        return;
2933    }
2934
2935    if (disabled) {
2936        bdrv_disable_dirty_bitmap(bitmap);
2937    }
2938
2939    bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2940}
2941
2942void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2943                                   Error **errp)
2944{
2945    BlockDriverState *bs;
2946    BdrvDirtyBitmap *bitmap;
2947    Error *local_err = NULL;
2948
2949    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2950    if (!bitmap || !bs) {
2951        return;
2952    }
2953
2954    if (bdrv_dirty_bitmap_frozen(bitmap)) {
2955        error_setg(errp,
2956                   "Bitmap '%s' is currently frozen and cannot be removed",
2957                   name);
2958        return;
2959    } else if (bdrv_dirty_bitmap_qmp_locked(bitmap)) {
2960        error_setg(errp,
2961                   "Bitmap '%s' is currently locked and cannot be removed",
2962                   name);
2963        return;
2964    }
2965
2966    if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2967        bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2968        if (local_err != NULL) {
2969            error_propagate(errp, local_err);
2970            return;
2971        }
2972    }
2973
2974    bdrv_release_dirty_bitmap(bs, bitmap);
2975}
2976
2977/**
2978 * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2979 * immediately after a full backup operation.
2980 */
2981void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2982                                  Error **errp)
2983{
2984    BdrvDirtyBitmap *bitmap;
2985    BlockDriverState *bs;
2986
2987    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2988    if (!bitmap || !bs) {
2989        return;
2990    }
2991
2992    if (bdrv_dirty_bitmap_frozen(bitmap)) {
2993        error_setg(errp,
2994                   "Bitmap '%s' is currently frozen and cannot be modified",
2995                   name);
2996        return;
2997    } else if (bdrv_dirty_bitmap_qmp_locked(bitmap)) {
2998        error_setg(errp,
2999                   "Bitmap '%s' is currently locked and cannot be modified",
3000                   name);
3001        return;
3002    } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
3003        error_setg(errp,
3004                   "Bitmap '%s' is currently disabled and cannot be cleared",
3005                   name);
3006        return;
3007    } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
3008        error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
3009        return;
3010    }
3011
3012    bdrv_clear_dirty_bitmap(bitmap, NULL);
3013}
3014
3015void qmp_x_block_dirty_bitmap_enable(const char *node, const char *name,
3016                                   Error **errp)
3017{
3018    BlockDriverState *bs;
3019    BdrvDirtyBitmap *bitmap;
3020
3021    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3022    if (!bitmap) {
3023        return;
3024    }
3025
3026    if (bdrv_dirty_bitmap_frozen(bitmap)) {
3027        error_setg(errp,
3028                   "Bitmap '%s' is currently frozen and cannot be enabled",
3029                   name);
3030        return;
3031    }
3032
3033    bdrv_enable_dirty_bitmap(bitmap);
3034}
3035
3036void qmp_x_block_dirty_bitmap_disable(const char *node, const char *name,
3037                                    Error **errp)
3038{
3039    BlockDriverState *bs;
3040    BdrvDirtyBitmap *bitmap;
3041
3042    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3043    if (!bitmap) {
3044        return;
3045    }
3046
3047    if (bdrv_dirty_bitmap_frozen(bitmap)) {
3048        error_setg(errp,
3049                   "Bitmap '%s' is currently frozen and cannot be disabled",
3050                   name);
3051        return;
3052    }
3053
3054    bdrv_disable_dirty_bitmap(bitmap);
3055}
3056
3057void qmp_x_block_dirty_bitmap_merge(const char *node, const char *dst_name,
3058                                    const char *src_name, Error **errp)
3059{
3060    BlockDriverState *bs;
3061    BdrvDirtyBitmap *dst, *src;
3062
3063    dst = block_dirty_bitmap_lookup(node, dst_name, &bs, errp);
3064    if (!dst) {
3065        return;
3066    }
3067
3068    if (bdrv_dirty_bitmap_frozen(dst)) {
3069        error_setg(errp, "Bitmap '%s' is frozen and cannot be modified",
3070                   dst_name);
3071        return;
3072    } else if (bdrv_dirty_bitmap_readonly(dst)) {
3073        error_setg(errp, "Bitmap '%s' is readonly and cannot be modified",
3074                   dst_name);
3075        return;
3076    }
3077
3078    src = bdrv_find_dirty_bitmap(bs, src_name);
3079    if (!src) {
3080        error_setg(errp, "Dirty bitmap '%s' not found", src_name);
3081        return;
3082    }
3083
3084    bdrv_merge_dirty_bitmap(dst, src, errp);
3085}
3086
3087BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3088                                                              const char *name,
3089                                                              Error **errp)
3090{
3091    BdrvDirtyBitmap *bitmap;
3092    BlockDriverState *bs;
3093    BlockDirtyBitmapSha256 *ret = NULL;
3094    char *sha256;
3095
3096    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3097    if (!bitmap || !bs) {
3098        return NULL;
3099    }
3100
3101    sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3102    if (sha256 == NULL) {
3103        return NULL;
3104    }
3105
3106    ret = g_new(BlockDirtyBitmapSha256, 1);
3107    ret->sha256 = sha256;
3108
3109    return ret;
3110}
3111
3112void hmp_drive_del(Monitor *mon, const QDict *qdict)
3113{
3114    const char *id = qdict_get_str(qdict, "id");
3115    BlockBackend *blk;
3116    BlockDriverState *bs;
3117    AioContext *aio_context;
3118    Error *local_err = NULL;
3119
3120    bs = bdrv_find_node(id);
3121    if (bs) {
3122        qmp_blockdev_del(id, &local_err);
3123        if (local_err) {
3124            error_report_err(local_err);
3125        }
3126        return;
3127    }
3128
3129    blk = blk_by_name(id);
3130    if (!blk) {
3131        error_report("Device '%s' not found", id);
3132        return;
3133    }
3134
3135    if (!blk_legacy_dinfo(blk)) {
3136        error_report("Deleting device added with blockdev-add"
3137                     " is not supported");
3138        return;
3139    }
3140
3141    aio_context = blk_get_aio_context(blk);
3142    aio_context_acquire(aio_context);
3143
3144    bs = blk_bs(blk);
3145    if (bs) {
3146        if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3147            error_report_err(local_err);
3148            aio_context_release(aio_context);
3149            return;
3150        }
3151
3152        blk_remove_bs(blk);
3153    }
3154
3155    /* Make the BlockBackend and the attached BlockDriverState anonymous */
3156    monitor_remove_blk(blk);
3157
3158    /* If this BlockBackend has a device attached to it, its refcount will be
3159     * decremented when the device is removed; otherwise we have to do so here.
3160     */
3161    if (blk_get_attached_dev(blk)) {
3162        /* Further I/O must not pause the guest */
3163        blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3164                         BLOCKDEV_ON_ERROR_REPORT);
3165    } else {
3166        blk_unref(blk);
3167    }
3168
3169    aio_context_release(aio_context);
3170}
3171
3172void qmp_block_resize(bool has_device, const char *device,
3173                      bool has_node_name, const char *node_name,
3174                      int64_t size, Error **errp)
3175{
3176    Error *local_err = NULL;
3177    BlockBackend *blk = NULL;
3178    BlockDriverState *bs;
3179    AioContext *aio_context;
3180    int ret;
3181
3182    bs = bdrv_lookup_bs(has_device ? device : NULL,
3183                        has_node_name ? node_name : NULL,
3184                        &local_err);
3185    if (local_err) {
3186        error_propagate(errp, local_err);
3187        return;
3188    }
3189
3190    aio_context = bdrv_get_aio_context(bs);
3191    aio_context_acquire(aio_context);
3192
3193    if (!bdrv_is_first_non_filter(bs)) {
3194        error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3195        goto out;
3196    }
3197
3198    if (size < 0) {
3199        error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3200        goto out;
3201    }
3202
3203    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3204        error_setg(errp, QERR_DEVICE_IN_USE, device);
3205        goto out;
3206    }
3207
3208    blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3209    ret = blk_insert_bs(blk, bs, errp);
3210    if (ret < 0) {
3211        goto out;
3212    }
3213
3214    bdrv_drained_begin(bs);
3215    ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3216    bdrv_drained_end(bs);
3217
3218out:
3219    blk_unref(blk);
3220    aio_context_release(aio_context);
3221}
3222
3223void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3224                      bool has_base, const char *base,
3225                      bool has_base_node, const char *base_node,
3226                      bool has_backing_file, const char *backing_file,
3227                      bool has_speed, int64_t speed,
3228                      bool has_on_error, BlockdevOnError on_error,
3229                      Error **errp)
3230{
3231    BlockDriverState *bs, *iter;
3232    BlockDriverState *base_bs = NULL;
3233    AioContext *aio_context;
3234    Error *local_err = NULL;
3235    const char *base_name = NULL;
3236
3237    if (!has_on_error) {
3238        on_error = BLOCKDEV_ON_ERROR_REPORT;
3239    }
3240
3241    bs = bdrv_lookup_bs(device, device, errp);
3242    if (!bs) {
3243        return;
3244    }
3245
3246    aio_context = bdrv_get_aio_context(bs);
3247    aio_context_acquire(aio_context);
3248
3249    if (has_base && has_base_node) {
3250        error_setg(errp, "'base' and 'base-node' cannot be specified "
3251                   "at the same time");
3252        goto out;
3253    }
3254
3255    if (has_base) {
3256        base_bs = bdrv_find_backing_image(bs, base);
3257        if (base_bs == NULL) {
3258            error_setg(errp, QERR_BASE_NOT_FOUND, base);
3259            goto out;
3260        }
3261        assert(bdrv_get_aio_context(base_bs) == aio_context);
3262        base_name = base;
3263    }
3264
3265    if (has_base_node) {
3266        base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3267        if (!base_bs) {
3268            goto out;
3269        }
3270        if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3271            error_setg(errp, "Node '%s' is not a backing image of '%s'",
3272                       base_node, device);
3273            goto out;
3274        }
3275        assert(bdrv_get_aio_context(base_bs) == aio_context);
3276        base_name = base_bs->filename;
3277    }
3278
3279    /* Check for op blockers in the whole chain between bs and base */
3280    for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3281        if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3282            goto out;
3283        }
3284    }
3285
3286    /* if we are streaming the entire chain, the result will have no backing
3287     * file, and specifying one is therefore an error */
3288    if (base_bs == NULL && has_backing_file) {
3289        error_setg(errp, "backing file specified, but streaming the "
3290                         "entire chain");
3291        goto out;
3292    }
3293
3294    /* backing_file string overrides base bs filename */
3295    base_name = has_backing_file ? backing_file : base_name;
3296
3297    stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3298                 has_speed ? speed : 0, on_error, &local_err);
3299    if (local_err) {
3300        error_propagate(errp, local_err);
3301        goto out;
3302    }
3303
3304    trace_qmp_block_stream(bs, bs->job);
3305
3306out:
3307    aio_context_release(aio_context);
3308}
3309
3310void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3311                      bool has_base, const char *base,
3312                      bool has_top, const char *top,
3313                      bool has_backing_file, const char *backing_file,
3314                      bool has_speed, int64_t speed,
3315                      bool has_filter_node_name, const char *filter_node_name,
3316                      Error **errp)
3317{
3318    BlockDriverState *bs;
3319    BlockDriverState *iter;
3320    BlockDriverState *base_bs, *top_bs;
3321    AioContext *aio_context;
3322    Error *local_err = NULL;
3323    /* This will be part of the QMP command, if/when the
3324     * BlockdevOnError change for blkmirror makes it in
3325     */
3326    BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3327
3328    if (!has_speed) {
3329        speed = 0;
3330    }
3331    if (!has_filter_node_name) {
3332        filter_node_name = NULL;
3333    }
3334
3335    /* Important Note:
3336     *  libvirt relies on the DeviceNotFound error class in order to probe for
3337     *  live commit feature versions; for this to work, we must make sure to
3338     *  perform the device lookup before any generic errors that may occur in a
3339     *  scenario in which all optional arguments are omitted. */
3340    bs = qmp_get_root_bs(device, &local_err);
3341    if (!bs) {
3342        bs = bdrv_lookup_bs(device, device, NULL);
3343        if (!bs) {
3344            error_free(local_err);
3345            error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3346                      "Device '%s' not found", device);
3347        } else {
3348            error_propagate(errp, local_err);
3349        }
3350        return;
3351    }
3352
3353    aio_context = bdrv_get_aio_context(bs);
3354    aio_context_acquire(aio_context);
3355
3356    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3357        goto out;
3358    }
3359
3360    /* default top_bs is the active layer */
3361    top_bs = bs;
3362
3363    if (has_top && top) {
3364        if (strcmp(bs->filename, top) != 0) {
3365            top_bs = bdrv_find_backing_image(bs, top);
3366        }
3367    }
3368
3369    if (top_bs == NULL) {
3370        error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3371        goto out;
3372    }
3373
3374    assert(bdrv_get_aio_context(top_bs) == aio_context);
3375
3376    if (has_base && base) {
3377        base_bs = bdrv_find_backing_image(top_bs, base);
3378    } else {
3379        base_bs = bdrv_find_base(top_bs);
3380    }
3381
3382    if (base_bs == NULL) {
3383        error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3384        goto out;
3385    }
3386
3387    assert(bdrv_get_aio_context(base_bs) == aio_context);
3388
3389    for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3390        if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3391            goto out;
3392        }
3393    }
3394
3395    /* Do not allow attempts to commit an image into itself */
3396    if (top_bs == base_bs) {
3397        error_setg(errp, "cannot commit an image into itself");
3398        goto out;
3399    }
3400
3401    if (top_bs == bs) {
3402        if (has_backing_file) {
3403            error_setg(errp, "'backing-file' specified,"
3404                             " but 'top' is the active layer");
3405            goto out;
3406        }
3407        commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3408                            JOB_DEFAULT, speed, on_error,
3409                            filter_node_name, NULL, NULL, false, &local_err);
3410    } else {
3411        BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3412        if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3413            goto out;
3414        }
3415        commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3416                     on_error, has_backing_file ? backing_file : NULL,
3417                     filter_node_name, &local_err);
3418    }
3419    if (local_err != NULL) {
3420        error_propagate(errp, local_err);
3421        goto out;
3422    }
3423
3424out:
3425    aio_context_release(aio_context);
3426}
3427
3428static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
3429                                 Error **errp)
3430{
3431    BlockDriverState *bs;
3432    BlockDriverState *target_bs;
3433    BlockDriverState *source = NULL;
3434    BlockJob *job = NULL;
3435    BdrvDirtyBitmap *bmap = NULL;
3436    AioContext *aio_context;
3437    QDict *options = NULL;
3438    Error *local_err = NULL;
3439    int flags, job_flags = JOB_DEFAULT;
3440    int64_t size;
3441    bool set_backing_hd = false;
3442
3443    if (!backup->has_speed) {
3444        backup->speed = 0;
3445    }
3446    if (!backup->has_on_source_error) {
3447        backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3448    }
3449    if (!backup->has_on_target_error) {
3450        backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3451    }
3452    if (!backup->has_mode) {
3453        backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3454    }
3455    if (!backup->has_job_id) {
3456        backup->job_id = NULL;
3457    }
3458    if (!backup->has_auto_finalize) {
3459        backup->auto_finalize = true;
3460    }
3461    if (!backup->has_auto_dismiss) {
3462        backup->auto_dismiss = true;
3463    }
3464    if (!backup->has_compress) {
3465        backup->compress = false;
3466    }
3467
3468    bs = qmp_get_root_bs(backup->device, errp);
3469    if (!bs) {
3470        return NULL;
3471    }
3472
3473    aio_context = bdrv_get_aio_context(bs);
3474    aio_context_acquire(aio_context);
3475
3476    if (!backup->has_format) {
3477        backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3478                         NULL : (char*) bs->drv->format_name;
3479    }
3480
3481    /* Early check to avoid creating target */
3482    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3483        goto out;
3484    }
3485
3486    flags = bs->open_flags | BDRV_O_RDWR;
3487
3488    /* See if we have a backing HD we can use to create our new image
3489     * on top of. */
3490    if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3491        source = backing_bs(bs);
3492        if (!source) {
3493            backup->sync = MIRROR_SYNC_MODE_FULL;
3494        }
3495    }
3496    if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3497        source = bs;
3498        flags |= BDRV_O_NO_BACKING;
3499        set_backing_hd = true;
3500    }
3501
3502    size = bdrv_getlength(bs);
3503    if (size < 0) {
3504        error_setg_errno(errp, -size, "bdrv_getlength failed");
3505        goto out;
3506    }
3507
3508    if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3509        assert(backup->format);
3510        if (source) {
3511            bdrv_img_create(backup->target, backup->format, source->filename,
3512                            source->drv->format_name, NULL,
3513                            size, flags, false, &local_err);
3514        } else {
3515            bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3516                            size, flags, false, &local_err);
3517        }
3518    }
3519
3520    if (local_err) {
3521        error_propagate(errp, local_err);
3522        goto out;
3523    }
3524
3525    if (backup->format) {
3526        if (!options) {
3527            options = qdict_new();
3528        }
3529        qdict_put_str(options, "driver", backup->format);
3530    }
3531
3532    target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3533    if (!target_bs) {
3534        goto out;
3535    }
3536
3537    bdrv_set_aio_context(target_bs, aio_context);
3538
3539    if (set_backing_hd) {
3540        bdrv_set_backing_hd(target_bs, source, &local_err);
3541        if (local_err) {
3542            bdrv_unref(target_bs);
3543            goto out;
3544        }
3545    }
3546
3547    if (backup->has_bitmap) {
3548        bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3549        if (!bmap) {
3550            error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3551            bdrv_unref(target_bs);
3552            goto out;
3553        }
3554        if (bdrv_dirty_bitmap_qmp_locked(bmap)) {
3555            error_setg(errp,
3556                       "Bitmap '%s' is currently locked and cannot be used for "
3557                       "backup", backup->bitmap);
3558            goto out;
3559        }
3560    }
3561    if (!backup->auto_finalize) {
3562        job_flags |= JOB_MANUAL_FINALIZE;
3563    }
3564    if (!backup->auto_dismiss) {
3565        job_flags |= JOB_MANUAL_DISMISS;
3566    }
3567
3568    job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3569                            backup->sync, bmap, backup->compress,
3570                            backup->on_source_error, backup->on_target_error,
3571                            job_flags, NULL, NULL, txn, &local_err);
3572    bdrv_unref(target_bs);
3573    if (local_err != NULL) {
3574        error_propagate(errp, local_err);
3575        goto out;
3576    }
3577
3578out:
3579    aio_context_release(aio_context);
3580    return job;
3581}
3582
3583void qmp_drive_backup(DriveBackup *arg, Error **errp)
3584{
3585
3586    BlockJob *job;
3587    job = do_drive_backup(arg, NULL, errp);
3588    if (job) {
3589        job_start(&job->job);
3590    }
3591}
3592
3593BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3594{
3595    return bdrv_named_nodes_list(errp);
3596}
3597
3598BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
3599                             Error **errp)
3600{
3601    BlockDriverState *bs;
3602    BlockDriverState *target_bs;
3603    Error *local_err = NULL;
3604    AioContext *aio_context;
3605    BlockJob *job = NULL;
3606    int job_flags = JOB_DEFAULT;
3607
3608    if (!backup->has_speed) {
3609        backup->speed = 0;
3610    }
3611    if (!backup->has_on_source_error) {
3612        backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3613    }
3614    if (!backup->has_on_target_error) {
3615        backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3616    }
3617    if (!backup->has_job_id) {
3618        backup->job_id = NULL;
3619    }
3620    if (!backup->has_auto_finalize) {
3621        backup->auto_finalize = true;
3622    }
3623    if (!backup->has_auto_dismiss) {
3624        backup->auto_dismiss = true;
3625    }
3626    if (!backup->has_compress) {
3627        backup->compress = false;
3628    }
3629
3630    bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3631    if (!bs) {
3632        return NULL;
3633    }
3634
3635    aio_context = bdrv_get_aio_context(bs);
3636    aio_context_acquire(aio_context);
3637
3638    target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3639    if (!target_bs) {
3640        goto out;
3641    }
3642
3643    if (bdrv_get_aio_context(target_bs) != aio_context) {
3644        if (!bdrv_has_blk(target_bs)) {
3645            /* The target BDS is not attached, we can safely move it to another
3646             * AioContext. */
3647            bdrv_set_aio_context(target_bs, aio_context);
3648        } else {
3649            error_setg(errp, "Target is attached to a different thread from "
3650                             "source.");
3651            goto out;
3652        }
3653    }
3654    if (!backup->auto_finalize) {
3655        job_flags |= JOB_MANUAL_FINALIZE;
3656    }
3657    if (!backup->auto_dismiss) {
3658        job_flags |= JOB_MANUAL_DISMISS;
3659    }
3660    job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3661                            backup->sync, NULL, backup->compress,
3662                            backup->on_source_error, backup->on_target_error,
3663                            job_flags, NULL, NULL, txn, &local_err);
3664    if (local_err != NULL) {
3665        error_propagate(errp, local_err);
3666    }
3667out:
3668    aio_context_release(aio_context);
3669    return job;
3670}
3671
3672void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3673{
3674    BlockJob *job;
3675    job = do_blockdev_backup(arg, NULL, errp);
3676    if (job) {
3677        job_start(&job->job);
3678    }
3679}
3680
3681/* Parameter check and block job starting for drive mirroring.
3682 * Caller should hold @device and @target's aio context (must be the same).
3683 **/
3684static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3685                                   BlockDriverState *target,
3686                                   bool has_replaces, const char *replaces,
3687                                   enum MirrorSyncMode sync,
3688                                   BlockMirrorBackingMode backing_mode,
3689                                   bool has_speed, int64_t speed,
3690                                   bool has_granularity, uint32_t granularity,
3691                                   bool has_buf_size, int64_t buf_size,
3692                                   bool has_on_source_error,
3693                                   BlockdevOnError on_source_error,
3694                                   bool has_on_target_error,
3695                                   BlockdevOnError on_target_error,
3696                                   bool has_unmap, bool unmap,
3697                                   bool has_filter_node_name,
3698                                   const char *filter_node_name,
3699                                   bool has_copy_mode, MirrorCopyMode copy_mode,
3700                                   Error **errp)
3701{
3702
3703    if (!has_speed) {
3704        speed = 0;
3705    }
3706    if (!has_on_source_error) {
3707        on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3708    }
3709    if (!has_on_target_error) {
3710        on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3711    }
3712    if (!has_granularity) {
3713        granularity = 0;
3714    }
3715    if (!has_buf_size) {
3716        buf_size = 0;
3717    }
3718    if (!has_unmap) {
3719        unmap = true;
3720    }
3721    if (!has_filter_node_name) {
3722        filter_node_name = NULL;
3723    }
3724    if (!has_copy_mode) {
3725        copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3726    }
3727
3728    if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3729        error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3730                   "a value in range [512B, 64MB]");
3731        return;
3732    }
3733    if (granularity & (granularity - 1)) {
3734        error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3735                   "power of 2");
3736        return;
3737    }
3738
3739    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3740        return;
3741    }
3742    if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3743        return;
3744    }
3745
3746    if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3747        sync = MIRROR_SYNC_MODE_FULL;
3748    }
3749
3750    /* pass the node name to replace to mirror start since it's loose coupling
3751     * and will allow to check whether the node still exist at mirror completion
3752     */
3753    mirror_start(job_id, bs, target,
3754                 has_replaces ? replaces : NULL,
3755                 speed, granularity, buf_size, sync, backing_mode,
3756                 on_source_error, on_target_error, unmap, filter_node_name,
3757                 copy_mode, errp);
3758}
3759
3760void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3761{
3762    BlockDriverState *bs;
3763    BlockDriverState *source, *target_bs;
3764    AioContext *aio_context;
3765    BlockMirrorBackingMode backing_mode;
3766    Error *local_err = NULL;
3767    QDict *options = NULL;
3768    int flags;
3769    int64_t size;
3770    const char *format = arg->format;
3771
3772    bs = qmp_get_root_bs(arg->device, errp);
3773    if (!bs) {
3774        return;
3775    }
3776
3777    /* Early check to avoid creating target */
3778    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3779        return;
3780    }
3781
3782    aio_context = bdrv_get_aio_context(bs);
3783    aio_context_acquire(aio_context);
3784
3785    if (!arg->has_mode) {
3786        arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3787    }
3788
3789    if (!arg->has_format) {
3790        format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3791                  ? NULL : bs->drv->format_name);
3792    }
3793
3794    flags = bs->open_flags | BDRV_O_RDWR;
3795    source = backing_bs(bs);
3796    if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3797        arg->sync = MIRROR_SYNC_MODE_FULL;
3798    }
3799    if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3800        source = bs;
3801    }
3802
3803    size = bdrv_getlength(bs);
3804    if (size < 0) {
3805        error_setg_errno(errp, -size, "bdrv_getlength failed");
3806        goto out;
3807    }
3808
3809    if (arg->has_replaces) {
3810        BlockDriverState *to_replace_bs;
3811        AioContext *replace_aio_context;
3812        int64_t replace_size;
3813
3814        if (!arg->has_node_name) {
3815            error_setg(errp, "a node-name must be provided when replacing a"
3816                             " named node of the graph");
3817            goto out;
3818        }
3819
3820        to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3821
3822        if (!to_replace_bs) {
3823            error_propagate(errp, local_err);
3824            goto out;
3825        }
3826
3827        replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3828        aio_context_acquire(replace_aio_context);
3829        replace_size = bdrv_getlength(to_replace_bs);
3830        aio_context_release(replace_aio_context);
3831
3832        if (size != replace_size) {
3833            error_setg(errp, "cannot replace image with a mirror image of "
3834                             "different size");
3835            goto out;
3836        }
3837    }
3838
3839    if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3840        backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3841    } else {
3842        backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3843    }
3844
3845    /* Don't open backing image in create() */
3846    flags |= BDRV_O_NO_BACKING;
3847
3848    if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3849        && arg->mode != NEW_IMAGE_MODE_EXISTING)
3850    {
3851        /* create new image w/o backing file */
3852        assert(format);
3853        bdrv_img_create(arg->target, format,
3854                        NULL, NULL, NULL, size, flags, false, &local_err);
3855    } else {
3856        switch (arg->mode) {
3857        case NEW_IMAGE_MODE_EXISTING:
3858            break;
3859        case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3860            /* create new image with backing file */
3861            bdrv_img_create(arg->target, format,
3862                            source->filename,
3863                            source->drv->format_name,
3864                            NULL, size, flags, false, &local_err);
3865            break;
3866        default:
3867            abort();
3868        }
3869    }
3870
3871    if (local_err) {
3872        error_propagate(errp, local_err);
3873        goto out;
3874    }
3875
3876    options = qdict_new();
3877    if (arg->has_node_name) {
3878        qdict_put_str(options, "node-name", arg->node_name);
3879    }
3880    if (format) {
3881        qdict_put_str(options, "driver", format);
3882    }
3883
3884    /* Mirroring takes care of copy-on-write using the source's backing
3885     * file.
3886     */
3887    target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3888    if (!target_bs) {
3889        goto out;
3890    }
3891
3892    bdrv_set_aio_context(target_bs, aio_context);
3893
3894    blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3895                           arg->has_replaces, arg->replaces, arg->sync,
3896                           backing_mode, arg->has_speed, arg->speed,
3897                           arg->has_granularity, arg->granularity,
3898                           arg->has_buf_size, arg->buf_size,
3899                           arg->has_on_source_error, arg->on_source_error,
3900                           arg->has_on_target_error, arg->on_target_error,
3901                           arg->has_unmap, arg->unmap,
3902                           false, NULL,
3903                           arg->has_copy_mode, arg->copy_mode,
3904                           &local_err);
3905    bdrv_unref(target_bs);
3906    error_propagate(errp, local_err);
3907out:
3908    aio_context_release(aio_context);
3909}
3910
3911void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3912                         const char *device, const char *target,
3913                         bool has_replaces, const char *replaces,
3914                         MirrorSyncMode sync,
3915                         bool has_speed, int64_t speed,
3916                         bool has_granularity, uint32_t granularity,
3917                         bool has_buf_size, int64_t buf_size,
3918                         bool has_on_source_error,
3919                         BlockdevOnError on_source_error,
3920                         bool has_on_target_error,
3921                         BlockdevOnError on_target_error,
3922                         bool has_filter_node_name,
3923                         const char *filter_node_name,
3924                         bool has_copy_mode, MirrorCopyMode copy_mode,
3925                         Error **errp)
3926{
3927    BlockDriverState *bs;
3928    BlockDriverState *target_bs;
3929    AioContext *aio_context;
3930    BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3931    Error *local_err = NULL;
3932
3933    bs = qmp_get_root_bs(device, errp);
3934    if (!bs) {
3935        return;
3936    }
3937
3938    target_bs = bdrv_lookup_bs(target, target, errp);
3939    if (!target_bs) {
3940        return;
3941    }
3942
3943    aio_context = bdrv_get_aio_context(bs);
3944    aio_context_acquire(aio_context);
3945
3946    bdrv_set_aio_context(target_bs, aio_context);
3947
3948    blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3949                           has_replaces, replaces, sync, backing_mode,
3950                           has_speed, speed,
3951                           has_granularity, granularity,
3952                           has_buf_size, buf_size,
3953                           has_on_source_error, on_source_error,
3954                           has_on_target_error, on_target_error,
3955                           true, true,
3956                           has_filter_node_name, filter_node_name,
3957                           has_copy_mode, copy_mode,
3958                           &local_err);
3959    error_propagate(errp, local_err);
3960
3961    aio_context_release(aio_context);
3962}
3963
3964/* Get a block job using its ID and acquire its AioContext */
3965static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3966                                Error **errp)
3967{
3968    BlockJob *job;
3969
3970    assert(id != NULL);
3971
3972    *aio_context = NULL;
3973
3974    job = block_job_get(id);
3975
3976    if (!job) {
3977        error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3978                  "Block job '%s' not found", id);
3979        return NULL;
3980    }
3981
3982    *aio_context = blk_get_aio_context(job->blk);
3983    aio_context_acquire(*aio_context);
3984
3985    return job;
3986}
3987
3988void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3989{
3990    AioContext *aio_context;
3991    BlockJob *job = find_block_job(device, &aio_context, errp);
3992
3993    if (!job) {
3994        return;
3995    }
3996
3997    block_job_set_speed(job, speed, errp);
3998    aio_context_release(aio_context);
3999}
4000
4001void qmp_block_job_cancel(const char *device,
4002                          bool has_force, bool force, Error **errp)
4003{
4004    AioContext *aio_context;
4005    BlockJob *job = find_block_job(device, &aio_context, errp);
4006
4007    if (!job) {
4008        return;
4009    }
4010
4011    if (!has_force) {
4012        force = false;
4013    }
4014
4015    if (job_user_paused(&job->job) && !force) {
4016        error_setg(errp, "The block job for device '%s' is currently paused",
4017                   device);
4018        goto out;
4019    }
4020
4021    trace_qmp_block_job_cancel(job);
4022    job_user_cancel(&job->job, force, errp);
4023out:
4024    aio_context_release(aio_context);
4025}
4026
4027void qmp_block_job_pause(const char *device, Error **errp)
4028{
4029    AioContext *aio_context;
4030    BlockJob *job = find_block_job(device, &aio_context, errp);
4031
4032    if (!job) {
4033        return;
4034    }
4035
4036    trace_qmp_block_job_pause(job);
4037    job_user_pause(&job->job, errp);
4038    aio_context_release(aio_context);
4039}
4040
4041void qmp_block_job_resume(const char *device, Error **errp)
4042{
4043    AioContext *aio_context;
4044    BlockJob *job = find_block_job(device, &aio_context, errp);
4045
4046    if (!job) {
4047        return;
4048    }
4049
4050    trace_qmp_block_job_resume(job);
4051    job_user_resume(&job->job, errp);
4052    aio_context_release(aio_context);
4053}
4054
4055void qmp_block_job_complete(const char *device, Error **errp)
4056{
4057    AioContext *aio_context;
4058    BlockJob *job = find_block_job(device, &aio_context, errp);
4059
4060    if (!job) {
4061        return;
4062    }
4063
4064    trace_qmp_block_job_complete(job);
4065    job_complete(&job->job, errp);
4066    aio_context_release(aio_context);
4067}
4068
4069void qmp_block_job_finalize(const char *id, Error **errp)
4070{
4071    AioContext *aio_context;
4072    BlockJob *job = find_block_job(id, &aio_context, errp);
4073
4074    if (!job) {
4075        return;
4076    }
4077
4078    trace_qmp_block_job_finalize(job);
4079    job_finalize(&job->job, errp);
4080    aio_context_release(aio_context);
4081}
4082
4083void qmp_block_job_dismiss(const char *id, Error **errp)
4084{
4085    AioContext *aio_context;
4086    BlockJob *bjob = find_block_job(id, &aio_context, errp);
4087    Job *job;
4088
4089    if (!bjob) {
4090        return;
4091    }
4092
4093    trace_qmp_block_job_dismiss(bjob);
4094    job = &bjob->job;
4095    job_dismiss(&job, errp);
4096    aio_context_release(aio_context);
4097}
4098
4099void qmp_change_backing_file(const char *device,
4100                             const char *image_node_name,
4101                             const char *backing_file,
4102                             Error **errp)
4103{
4104    BlockDriverState *bs = NULL;
4105    AioContext *aio_context;
4106    BlockDriverState *image_bs = NULL;
4107    Error *local_err = NULL;
4108    bool ro;
4109    int open_flags;
4110    int ret;
4111
4112    bs = qmp_get_root_bs(device, errp);
4113    if (!bs) {
4114        return;
4115    }
4116
4117    aio_context = bdrv_get_aio_context(bs);
4118    aio_context_acquire(aio_context);
4119
4120    image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4121    if (local_err) {
4122        error_propagate(errp, local_err);
4123        goto out;
4124    }
4125
4126    if (!image_bs) {
4127        error_setg(errp, "image file not found");
4128        goto out;
4129    }
4130
4131    if (bdrv_find_base(image_bs) == image_bs) {
4132        error_setg(errp, "not allowing backing file change on an image "
4133                         "without a backing file");
4134        goto out;
4135    }
4136
4137    /* even though we are not necessarily operating on bs, we need it to
4138     * determine if block ops are currently prohibited on the chain */
4139    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4140        goto out;
4141    }
4142
4143    /* final sanity check */
4144    if (!bdrv_chain_contains(bs, image_bs)) {
4145        error_setg(errp, "'%s' and image file are not in the same chain",
4146                   device);
4147        goto out;
4148    }
4149
4150    /* if not r/w, reopen to make r/w */
4151    open_flags = image_bs->open_flags;
4152    ro = bdrv_is_read_only(image_bs);
4153
4154    if (ro) {
4155        bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
4156        if (local_err) {
4157            error_propagate(errp, local_err);
4158            goto out;
4159        }
4160    }
4161
4162    ret = bdrv_change_backing_file(image_bs, backing_file,
4163                               image_bs->drv ? image_bs->drv->format_name : "");
4164
4165    if (ret < 0) {
4166        error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4167                         backing_file);
4168        /* don't exit here, so we can try to restore open flags if
4169         * appropriate */
4170    }
4171
4172    if (ro) {
4173        bdrv_reopen(image_bs, open_flags, &local_err);
4174        error_propagate(errp, local_err);
4175    }
4176
4177out:
4178    aio_context_release(aio_context);
4179}
4180
4181void hmp_drive_add_node(Monitor *mon, const char *optstr)
4182{
4183    QemuOpts *opts;
4184    QDict *qdict;
4185    Error *local_err = NULL;
4186
4187    opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4188    if (!opts) {
4189        return;
4190    }
4191
4192    qdict = qemu_opts_to_qdict(opts, NULL);
4193
4194    if (!qdict_get_try_str(qdict, "node-name")) {
4195        qobject_unref(qdict);
4196        error_report("'node-name' needs to be specified");
4197        goto out;
4198    }
4199
4200    BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4201    if (!bs) {
4202        error_report_err(local_err);
4203        goto out;
4204    }
4205
4206    QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4207
4208out:
4209    qemu_opts_del(opts);
4210}
4211
4212void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4213{
4214    BlockDriverState *bs;
4215    QObject *obj;
4216    Visitor *v = qobject_output_visitor_new(&obj);
4217    QDict *qdict;
4218    Error *local_err = NULL;
4219
4220    visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4221    if (local_err) {
4222        error_propagate(errp, local_err);
4223        goto fail;
4224    }
4225
4226    visit_complete(v, &obj);
4227    qdict = qobject_to(QDict, obj);
4228
4229    qdict_flatten(qdict);
4230
4231    if (!qdict_get_try_str(qdict, "node-name")) {
4232        error_setg(errp, "'node-name' must be specified for the root node");
4233        goto fail;
4234    }
4235
4236    bs = bds_tree_init(qdict, errp);
4237    if (!bs) {
4238        goto fail;
4239    }
4240
4241    QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4242
4243fail:
4244    visit_free(v);
4245}
4246
4247void qmp_blockdev_del(const char *node_name, Error **errp)
4248{
4249    AioContext *aio_context;
4250    BlockDriverState *bs;
4251
4252    bs = bdrv_find_node(node_name);
4253    if (!bs) {
4254        error_setg(errp, "Cannot find node %s", node_name);
4255        return;
4256    }
4257    if (bdrv_has_blk(bs)) {
4258        error_setg(errp, "Node %s is in use", node_name);
4259        return;
4260    }
4261    aio_context = bdrv_get_aio_context(bs);
4262    aio_context_acquire(aio_context);
4263
4264    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4265        goto out;
4266    }
4267
4268    if (!bs->monitor_list.tqe_prev) {
4269        error_setg(errp, "Node %s is not owned by the monitor",
4270                   bs->node_name);
4271        goto out;
4272    }
4273
4274    if (bs->refcnt > 1) {
4275        error_setg(errp, "Block device %s is in use",
4276                   bdrv_get_device_or_node_name(bs));
4277        goto out;
4278    }
4279
4280    QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4281    bdrv_unref(bs);
4282
4283out:
4284    aio_context_release(aio_context);
4285}
4286
4287static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4288                                  const char *child_name)
4289{
4290    BdrvChild *child;
4291
4292    QLIST_FOREACH(child, &parent_bs->children, next) {
4293        if (strcmp(child->name, child_name) == 0) {
4294            return child;
4295        }
4296    }
4297
4298    return NULL;
4299}
4300
4301void qmp_x_blockdev_change(const char *parent, bool has_child,
4302                           const char *child, bool has_node,
4303                           const char *node, Error **errp)
4304{
4305    BlockDriverState *parent_bs, *new_bs = NULL;
4306    BdrvChild *p_child;
4307
4308    parent_bs = bdrv_lookup_bs(parent, parent, errp);
4309    if (!parent_bs) {
4310        return;
4311    }
4312
4313    if (has_child == has_node) {
4314        if (has_child) {
4315            error_setg(errp, "The parameters child and node are in conflict");
4316        } else {
4317            error_setg(errp, "Either child or node must be specified");
4318        }
4319        return;
4320    }
4321
4322    if (has_child) {
4323        p_child = bdrv_find_child(parent_bs, child);
4324        if (!p_child) {
4325            error_setg(errp, "Node '%s' does not have child '%s'",
4326                       parent, child);
4327            return;
4328        }
4329        bdrv_del_child(parent_bs, p_child, errp);
4330    }
4331
4332    if (has_node) {
4333        new_bs = bdrv_find_node(node);
4334        if (!new_bs) {
4335            error_setg(errp, "Node '%s' not found", node);
4336            return;
4337        }
4338        bdrv_add_child(parent_bs, new_bs, errp);
4339    }
4340}
4341
4342BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4343{
4344    BlockJobInfoList *head = NULL, **p_next = &head;
4345    BlockJob *job;
4346
4347    for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4348        BlockJobInfoList *elem;
4349        AioContext *aio_context;
4350
4351        if (block_job_is_internal(job)) {
4352            continue;
4353        }
4354        elem = g_new0(BlockJobInfoList, 1);
4355        aio_context = blk_get_aio_context(job->blk);
4356        aio_context_acquire(aio_context);
4357        elem->value = block_job_query(job, errp);
4358        aio_context_release(aio_context);
4359        if (!elem->value) {
4360            g_free(elem);
4361            qapi_free_BlockJobInfoList(head);
4362            return NULL;
4363        }
4364        *p_next = elem;
4365        p_next = &elem->next;
4366    }
4367
4368    return head;
4369}
4370
4371void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4372                                 bool has_force, bool force, Error **errp)
4373{
4374    AioContext *old_context;
4375    AioContext *new_context;
4376    BlockDriverState *bs;
4377
4378    bs = bdrv_find_node(node_name);
4379    if (!bs) {
4380        error_setg(errp, "Cannot find node %s", node_name);
4381        return;
4382    }
4383
4384    /* Protects against accidents. */
4385    if (!(has_force && force) && bdrv_has_blk(bs)) {
4386        error_setg(errp, "Node %s is associated with a BlockBackend and could "
4387                         "be in use (use force=true to override this check)",
4388                         node_name);
4389        return;
4390    }
4391
4392    if (iothread->type == QTYPE_QSTRING) {
4393        IOThread *obj = iothread_by_id(iothread->u.s);
4394        if (!obj) {
4395            error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4396            return;
4397        }
4398
4399        new_context = iothread_get_aio_context(obj);
4400    } else {
4401        new_context = qemu_get_aio_context();
4402    }
4403
4404    old_context = bdrv_get_aio_context(bs);
4405    aio_context_acquire(old_context);
4406
4407    bdrv_set_aio_context(bs, new_context);
4408
4409    aio_context_release(old_context);
4410}
4411
4412void qmp_x_block_latency_histogram_set(
4413    const char *device,
4414    bool has_boundaries, uint64List *boundaries,
4415    bool has_boundaries_read, uint64List *boundaries_read,
4416    bool has_boundaries_write, uint64List *boundaries_write,
4417    bool has_boundaries_flush, uint64List *boundaries_flush,
4418    Error **errp)
4419{
4420    BlockBackend *blk = blk_by_name(device);
4421    BlockAcctStats *stats;
4422
4423    if (!blk) {
4424        error_setg(errp, "Device '%s' not found", device);
4425        return;
4426    }
4427    stats = blk_get_stats(blk);
4428
4429    if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4430        !has_boundaries_flush)
4431    {
4432        block_latency_histograms_clear(stats);
4433        return;
4434    }
4435
4436    if (has_boundaries || has_boundaries_read) {
4437        block_latency_histogram_set(
4438            stats, BLOCK_ACCT_READ,
4439            has_boundaries_read ? boundaries_read : boundaries);
4440    }
4441
4442    if (has_boundaries || has_boundaries_write) {
4443        block_latency_histogram_set(
4444            stats, BLOCK_ACCT_WRITE,
4445            has_boundaries_write ? boundaries_write : boundaries);
4446    }
4447
4448    if (has_boundaries || has_boundaries_flush) {
4449        block_latency_histogram_set(
4450            stats, BLOCK_ACCT_FLUSH,
4451            has_boundaries_flush ? boundaries_flush : boundaries);
4452    }
4453}
4454
4455QemuOptsList qemu_common_drive_opts = {
4456    .name = "drive",
4457    .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4458    .desc = {
4459        {
4460            .name = "snapshot",
4461            .type = QEMU_OPT_BOOL,
4462            .help = "enable/disable snapshot mode",
4463        },{
4464            .name = "aio",
4465            .type = QEMU_OPT_STRING,
4466            .help = "host AIO implementation (threads, native)",
4467        },{
4468            .name = BDRV_OPT_CACHE_WB,
4469            .type = QEMU_OPT_BOOL,
4470            .help = "Enable writeback mode",
4471        },{
4472            .name = "format",
4473            .type = QEMU_OPT_STRING,
4474            .help = "disk format (raw, qcow2, ...)",
4475        },{
4476            .name = "rerror",
4477            .type = QEMU_OPT_STRING,
4478            .help = "read error action",
4479        },{
4480            .name = "werror",
4481            .type = QEMU_OPT_STRING,
4482            .help = "write error action",
4483        },{
4484            .name = BDRV_OPT_READ_ONLY,
4485            .type = QEMU_OPT_BOOL,
4486            .help = "open drive file as read-only",
4487        },
4488
4489        THROTTLE_OPTS,
4490
4491        {
4492            .name = "throttling.group",
4493            .type = QEMU_OPT_STRING,
4494            .help = "name of the block throttling group",
4495        },{
4496            .name = "copy-on-read",
4497            .type = QEMU_OPT_BOOL,
4498            .help = "copy read data from backing file into image file",
4499        },{
4500            .name = "detect-zeroes",
4501            .type = QEMU_OPT_STRING,
4502            .help = "try to optimize zero writes (off, on, unmap)",
4503        },{
4504            .name = "stats-account-invalid",
4505            .type = QEMU_OPT_BOOL,
4506            .help = "whether to account for invalid I/O operations "
4507                    "in the statistics",
4508        },{
4509            .name = "stats-account-failed",
4510            .type = QEMU_OPT_BOOL,
4511            .help = "whether to account for failed I/O operations "
4512                    "in the statistics",
4513        },
4514        { /* end of list */ }
4515    },
4516};
4517
4518QemuOptsList qemu_drive_opts = {
4519    .name = "drive",
4520    .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4521    .desc = {
4522        /*
4523         * no elements => accept any params
4524         * validation will happen later
4525         */
4526        { /* end of list */ }
4527    },
4528};
4529