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        qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
 594        assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
 595
 596        if (runstate_check(RUN_STATE_INMIGRATE)) {
 597            bdrv_flags |= BDRV_O_INACTIVE;
 598        }
 599
 600        blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
 601        if (!blk) {
 602            goto err_no_bs_opts;
 603        }
 604        bs = blk_bs(blk);
 605
 606        bs->detect_zeroes = detect_zeroes;
 607
 608        block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
 609
 610        if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
 611            blk_unref(blk);
 612            blk = NULL;
 613            goto err_no_bs_opts;
 614        }
 615    }
 616
 617    /* disk I/O throttling */
 618    if (throttle_enabled(&cfg)) {
 619        if (!throttling_group) {
 620            throttling_group = id;
 621        }
 622        blk_io_limits_enable(blk, throttling_group);
 623        blk_set_io_limits(blk, &cfg);
 624    }
 625
 626    blk_set_enable_write_cache(blk, !writethrough);
 627    blk_set_on_error(blk, on_read_error, on_write_error);
 628
 629    if (!monitor_add_blk(blk, id, errp)) {
 630        blk_unref(blk);
 631        blk = NULL;
 632        goto err_no_bs_opts;
 633    }
 634
 635err_no_bs_opts:
 636    qemu_opts_del(opts);
 637    qobject_unref(interval_dict);
 638    qobject_unref(interval_list);
 639    return blk;
 640
 641early_err:
 642    qemu_opts_del(opts);
 643    qobject_unref(interval_dict);
 644    qobject_unref(interval_list);
 645err_no_opts:
 646    qobject_unref(bs_opts);
 647    return NULL;
 648}
 649
 650/* Takes the ownership of bs_opts */
 651static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
 652{
 653    int bdrv_flags = 0;
 654
 655    /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
 656     * with other callers) rather than what we want as the real defaults.
 657     * Apply the defaults here instead. */
 658    qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
 659    qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
 660    qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
 661
 662    if (runstate_check(RUN_STATE_INMIGRATE)) {
 663        bdrv_flags |= BDRV_O_INACTIVE;
 664    }
 665
 666    return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
 667}
 668
 669void blockdev_close_all_bdrv_states(void)
 670{
 671    BlockDriverState *bs, *next_bs;
 672
 673    QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
 674        AioContext *ctx = bdrv_get_aio_context(bs);
 675
 676        aio_context_acquire(ctx);
 677        bdrv_unref(bs);
 678        aio_context_release(ctx);
 679    }
 680}
 681
 682/* Iterates over the list of monitor-owned BlockDriverStates */
 683BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
 684{
 685    return bs ? QTAILQ_NEXT(bs, monitor_list)
 686              : QTAILQ_FIRST(&monitor_bdrv_states);
 687}
 688
 689static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
 690                            Error **errp)
 691{
 692    const char *value;
 693
 694    value = qemu_opt_get(opts, from);
 695    if (value) {
 696        if (qemu_opt_find(opts, to)) {
 697            error_setg(errp, "'%s' and its alias '%s' can't be used at the "
 698                       "same time", to, from);
 699            return;
 700        }
 701    }
 702
 703    /* rename all items in opts */
 704    while ((value = qemu_opt_get(opts, from))) {
 705        qemu_opt_set(opts, to, value, &error_abort);
 706        qemu_opt_unset(opts, from);
 707    }
 708}
 709
 710QemuOptsList qemu_legacy_drive_opts = {
 711    .name = "drive",
 712    .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
 713    .desc = {
 714        {
 715            .name = "bus",
 716            .type = QEMU_OPT_NUMBER,
 717            .help = "bus number",
 718        },{
 719            .name = "unit",
 720            .type = QEMU_OPT_NUMBER,
 721            .help = "unit number (i.e. lun for scsi)",
 722        },{
 723            .name = "index",
 724            .type = QEMU_OPT_NUMBER,
 725            .help = "index number",
 726        },{
 727            .name = "media",
 728            .type = QEMU_OPT_STRING,
 729            .help = "media type (disk, cdrom)",
 730        },{
 731            .name = "if",
 732            .type = QEMU_OPT_STRING,
 733            .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
 734        },{
 735            .name = "file",
 736            .type = QEMU_OPT_STRING,
 737            .help = "file name",
 738        },
 739
 740        /* Options that are passed on, but have special semantics with -drive */
 741        {
 742            .name = BDRV_OPT_READ_ONLY,
 743            .type = QEMU_OPT_BOOL,
 744            .help = "open drive file as read-only",
 745        },{
 746            .name = "rerror",
 747            .type = QEMU_OPT_STRING,
 748            .help = "read error action",
 749        },{
 750            .name = "werror",
 751            .type = QEMU_OPT_STRING,
 752            .help = "write error action",
 753        },{
 754            .name = "copy-on-read",
 755            .type = QEMU_OPT_BOOL,
 756            .help = "copy read data from backing file into image file",
 757        },
 758
 759        { /* end of list */ }
 760    },
 761};
 762
 763DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
 764                     Error **errp)
 765{
 766    const char *value;
 767    BlockBackend *blk;
 768    DriveInfo *dinfo = NULL;
 769    QDict *bs_opts;
 770    QemuOpts *legacy_opts;
 771    DriveMediaType media = MEDIA_DISK;
 772    BlockInterfaceType type;
 773    int max_devs, bus_id, unit_id, index;
 774    const char *werror, *rerror;
 775    bool read_only = false;
 776    bool copy_on_read;
 777    const char *filename;
 778    Error *local_err = NULL;
 779    int i;
 780
 781    /* Change legacy command line options into QMP ones */
 782    static const struct {
 783        const char *from;
 784        const char *to;
 785    } opt_renames[] = {
 786        { "iops",           "throttling.iops-total" },
 787        { "iops_rd",        "throttling.iops-read" },
 788        { "iops_wr",        "throttling.iops-write" },
 789
 790        { "bps",            "throttling.bps-total" },
 791        { "bps_rd",         "throttling.bps-read" },
 792        { "bps_wr",         "throttling.bps-write" },
 793
 794        { "iops_max",       "throttling.iops-total-max" },
 795        { "iops_rd_max",    "throttling.iops-read-max" },
 796        { "iops_wr_max",    "throttling.iops-write-max" },
 797
 798        { "bps_max",        "throttling.bps-total-max" },
 799        { "bps_rd_max",     "throttling.bps-read-max" },
 800        { "bps_wr_max",     "throttling.bps-write-max" },
 801
 802        { "iops_size",      "throttling.iops-size" },
 803
 804        { "group",          "throttling.group" },
 805
 806        { "readonly",       BDRV_OPT_READ_ONLY },
 807    };
 808
 809    for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
 810        qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
 811                        &local_err);
 812        if (local_err) {
 813            error_propagate(errp, local_err);
 814            return NULL;
 815        }
 816    }
 817
 818    value = qemu_opt_get(all_opts, "cache");
 819    if (value) {
 820        int flags = 0;
 821        bool writethrough;
 822
 823        if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
 824            error_setg(errp, "invalid cache option");
 825            return NULL;
 826        }
 827
 828        /* Specific options take precedence */
 829        if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
 830            qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
 831                              !writethrough, &error_abort);
 832        }
 833        if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
 834            qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
 835                              !!(flags & BDRV_O_NOCACHE), &error_abort);
 836        }
 837        if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
 838            qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
 839                              !!(flags & BDRV_O_NO_FLUSH), &error_abort);
 840        }
 841        qemu_opt_unset(all_opts, "cache");
 842    }
 843
 844    /* Get a QDict for processing the options */
 845    bs_opts = qdict_new();
 846    qemu_opts_to_qdict(all_opts, bs_opts);
 847
 848    legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
 849                                   &error_abort);
 850    qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
 851    if (local_err) {
 852        error_propagate(errp, local_err);
 853        goto fail;
 854    }
 855
 856    /* Media type */
 857    value = qemu_opt_get(legacy_opts, "media");
 858    if (value) {
 859        if (!strcmp(value, "disk")) {
 860            media = MEDIA_DISK;
 861        } else if (!strcmp(value, "cdrom")) {
 862            media = MEDIA_CDROM;
 863            read_only = true;
 864        } else {
 865            error_setg(errp, "'%s' invalid media", value);
 866            goto fail;
 867        }
 868    }
 869
 870    /* copy-on-read is disabled with a warning for read-only devices */
 871    read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
 872    copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
 873
 874    if (read_only && copy_on_read) {
 875        warn_report("disabling copy-on-read on read-only drive");
 876        copy_on_read = false;
 877    }
 878
 879    qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
 880    qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
 881
 882    /* Controller type */
 883    value = qemu_opt_get(legacy_opts, "if");
 884    if (value) {
 885        for (type = 0;
 886             type < IF_COUNT && strcmp(value, if_name[type]);
 887             type++) {
 888        }
 889        if (type == IF_COUNT) {
 890            error_setg(errp, "unsupported bus type '%s'", value);
 891            goto fail;
 892        }
 893    } else {
 894        type = block_default_type;
 895    }
 896
 897    /* Device address specified by bus/unit or index.
 898     * If none was specified, try to find the first free one. */
 899    bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
 900    unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
 901    index   = qemu_opt_get_number(legacy_opts, "index", -1);
 902
 903    max_devs = if_max_devs[type];
 904
 905    if (index != -1) {
 906        if (bus_id != 0 || unit_id != -1) {
 907            error_setg(errp, "index cannot be used with bus and unit");
 908            goto fail;
 909        }
 910        bus_id = drive_index_to_bus_id(type, index);
 911        unit_id = drive_index_to_unit_id(type, index);
 912    }
 913
 914    if (unit_id == -1) {
 915       unit_id = 0;
 916       while (drive_get(type, bus_id, unit_id) != NULL) {
 917           unit_id++;
 918           if (max_devs && unit_id >= max_devs) {
 919               unit_id -= max_devs;
 920               bus_id++;
 921           }
 922       }
 923    }
 924
 925    if (max_devs && unit_id >= max_devs) {
 926        error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
 927        goto fail;
 928    }
 929
 930    if (drive_get(type, bus_id, unit_id) != NULL) {
 931        error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
 932                   bus_id, unit_id, index);
 933        goto fail;
 934    }
 935
 936    /* no id supplied -> create one */
 937    if (qemu_opts_id(all_opts) == NULL) {
 938        char *new_id;
 939        const char *mediastr = "";
 940        if (type == IF_IDE || type == IF_SCSI) {
 941            mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
 942        }
 943        if (max_devs) {
 944            new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
 945                                     mediastr, unit_id);
 946        } else {
 947            new_id = g_strdup_printf("%s%s%i", if_name[type],
 948                                     mediastr, unit_id);
 949        }
 950        qdict_put_str(bs_opts, "id", new_id);
 951        g_free(new_id);
 952    }
 953
 954    /* Add virtio block device */
 955    if (type == IF_VIRTIO) {
 956        QemuOpts *devopts;
 957        devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
 958                                   &error_abort);
 959        if (arch_type == QEMU_ARCH_S390X) {
 960            qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
 961        } else {
 962            qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
 963        }
 964        qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
 965                     &error_abort);
 966    }
 967
 968    filename = qemu_opt_get(legacy_opts, "file");
 969
 970    /* Check werror/rerror compatibility with if=... */
 971    werror = qemu_opt_get(legacy_opts, "werror");
 972    if (werror != NULL) {
 973        if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
 974            type != IF_NONE) {
 975            error_setg(errp, "werror is not supported by this bus type");
 976            goto fail;
 977        }
 978        qdict_put_str(bs_opts, "werror", werror);
 979    }
 980
 981    rerror = qemu_opt_get(legacy_opts, "rerror");
 982    if (rerror != NULL) {
 983        if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
 984            type != IF_NONE) {
 985            error_setg(errp, "rerror is not supported by this bus type");
 986            goto fail;
 987        }
 988        qdict_put_str(bs_opts, "rerror", rerror);
 989    }
 990
 991    /* Actual block device init: Functionality shared with blockdev-add */
 992    blk = blockdev_init(filename, bs_opts, &local_err);
 993    bs_opts = NULL;
 994    if (!blk) {
 995        if (local_err) {
 996            error_propagate(errp, local_err);
 997        }
 998        goto fail;
 999    } else {
1000        assert(!local_err);
1001    }
1002
1003    /* Create legacy DriveInfo */
1004    dinfo = g_malloc0(sizeof(*dinfo));
1005    dinfo->opts = all_opts;
1006
1007    dinfo->type = type;
1008    dinfo->bus = bus_id;
1009    dinfo->unit = unit_id;
1010
1011    blk_set_legacy_dinfo(blk, dinfo);
1012
1013    switch(type) {
1014    case IF_IDE:
1015    case IF_SCSI:
1016    case IF_XEN:
1017    case IF_NONE:
1018        dinfo->media_cd = media == MEDIA_CDROM;
1019        break;
1020    default:
1021        break;
1022    }
1023
1024fail:
1025    qemu_opts_del(legacy_opts);
1026    qobject_unref(bs_opts);
1027    return dinfo;
1028}
1029
1030static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1031{
1032    BlockDriverState *bs;
1033
1034    bs = bdrv_lookup_bs(name, name, errp);
1035    if (bs == NULL) {
1036        return NULL;
1037    }
1038
1039    if (!bdrv_is_root_node(bs)) {
1040        error_setg(errp, "Need a root block node");
1041        return NULL;
1042    }
1043
1044    if (!bdrv_is_inserted(bs)) {
1045        error_setg(errp, "Device has no medium");
1046        return NULL;
1047    }
1048
1049    return bs;
1050}
1051
1052static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1053                                 Error **errp)
1054{
1055    BlockBackend *blk;
1056
1057    if (!blk_name == !qdev_id) {
1058        error_setg(errp, "Need exactly one of 'device' and 'id'");
1059        return NULL;
1060    }
1061
1062    if (qdev_id) {
1063        blk = blk_by_qdev_id(qdev_id, errp);
1064    } else {
1065        blk = blk_by_name(blk_name);
1066        if (blk == NULL) {
1067            error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1068                      "Device '%s' not found", blk_name);
1069        }
1070    }
1071
1072    return blk;
1073}
1074
1075void hmp_commit(Monitor *mon, const QDict *qdict)
1076{
1077    const char *device = qdict_get_str(qdict, "device");
1078    BlockBackend *blk;
1079    int ret;
1080
1081    if (!strcmp(device, "all")) {
1082        ret = blk_commit_all();
1083    } else {
1084        BlockDriverState *bs;
1085        AioContext *aio_context;
1086
1087        blk = blk_by_name(device);
1088        if (!blk) {
1089            monitor_printf(mon, "Device '%s' not found\n", device);
1090            return;
1091        }
1092        if (!blk_is_available(blk)) {
1093            monitor_printf(mon, "Device '%s' has no medium\n", device);
1094            return;
1095        }
1096
1097        bs = blk_bs(blk);
1098        aio_context = bdrv_get_aio_context(bs);
1099        aio_context_acquire(aio_context);
1100
1101        ret = bdrv_commit(bs);
1102
1103        aio_context_release(aio_context);
1104    }
1105    if (ret < 0) {
1106        monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1107                       strerror(-ret));
1108    }
1109}
1110
1111static void blockdev_do_action(TransactionAction *action, Error **errp)
1112{
1113    TransactionActionList list;
1114
1115    list.value = action;
1116    list.next = NULL;
1117    qmp_transaction(&list, false, NULL, errp);
1118}
1119
1120void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1121                                bool has_node_name, const char *node_name,
1122                                const char *snapshot_file,
1123                                bool has_snapshot_node_name,
1124                                const char *snapshot_node_name,
1125                                bool has_format, const char *format,
1126                                bool has_mode, NewImageMode mode, Error **errp)
1127{
1128    BlockdevSnapshotSync snapshot = {
1129        .has_device = has_device,
1130        .device = (char *) device,
1131        .has_node_name = has_node_name,
1132        .node_name = (char *) node_name,
1133        .snapshot_file = (char *) snapshot_file,
1134        .has_snapshot_node_name = has_snapshot_node_name,
1135        .snapshot_node_name = (char *) snapshot_node_name,
1136        .has_format = has_format,
1137        .format = (char *) format,
1138        .has_mode = has_mode,
1139        .mode = mode,
1140    };
1141    TransactionAction action = {
1142        .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1143        .u.blockdev_snapshot_sync.data = &snapshot,
1144    };
1145    blockdev_do_action(&action, errp);
1146}
1147
1148void qmp_blockdev_snapshot(const char *node, const char *overlay,
1149                           Error **errp)
1150{
1151    BlockdevSnapshot snapshot_data = {
1152        .node = (char *) node,
1153        .overlay = (char *) overlay
1154    };
1155    TransactionAction action = {
1156        .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1157        .u.blockdev_snapshot.data = &snapshot_data,
1158    };
1159    blockdev_do_action(&action, errp);
1160}
1161
1162void qmp_blockdev_snapshot_internal_sync(const char *device,
1163                                         const char *name,
1164                                         Error **errp)
1165{
1166    BlockdevSnapshotInternal snapshot = {
1167        .device = (char *) device,
1168        .name = (char *) name
1169    };
1170    TransactionAction action = {
1171        .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1172        .u.blockdev_snapshot_internal_sync.data = &snapshot,
1173    };
1174    blockdev_do_action(&action, errp);
1175}
1176
1177SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1178                                                         bool has_id,
1179                                                         const char *id,
1180                                                         bool has_name,
1181                                                         const char *name,
1182                                                         Error **errp)
1183{
1184    BlockDriverState *bs;
1185    AioContext *aio_context;
1186    QEMUSnapshotInfo sn;
1187    Error *local_err = NULL;
1188    SnapshotInfo *info = NULL;
1189    int ret;
1190
1191    bs = qmp_get_root_bs(device, errp);
1192    if (!bs) {
1193        return NULL;
1194    }
1195    aio_context = bdrv_get_aio_context(bs);
1196    aio_context_acquire(aio_context);
1197
1198    if (!has_id) {
1199        id = NULL;
1200    }
1201
1202    if (!has_name) {
1203        name = NULL;
1204    }
1205
1206    if (!id && !name) {
1207        error_setg(errp, "Name or id must be provided");
1208        goto out_aio_context;
1209    }
1210
1211    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1212        goto out_aio_context;
1213    }
1214
1215    ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1216    if (local_err) {
1217        error_propagate(errp, local_err);
1218        goto out_aio_context;
1219    }
1220    if (!ret) {
1221        error_setg(errp,
1222                   "Snapshot with id '%s' and name '%s' does not exist on "
1223                   "device '%s'",
1224                   STR_OR_NULL(id), STR_OR_NULL(name), device);
1225        goto out_aio_context;
1226    }
1227
1228    bdrv_snapshot_delete(bs, id, name, &local_err);
1229    if (local_err) {
1230        error_propagate(errp, local_err);
1231        goto out_aio_context;
1232    }
1233
1234    aio_context_release(aio_context);
1235
1236    info = g_new0(SnapshotInfo, 1);
1237    info->id = g_strdup(sn.id_str);
1238    info->name = g_strdup(sn.name);
1239    info->date_nsec = sn.date_nsec;
1240    info->date_sec = sn.date_sec;
1241    info->vm_state_size = sn.vm_state_size;
1242    info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1243    info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1244
1245    return info;
1246
1247out_aio_context:
1248    aio_context_release(aio_context);
1249    return NULL;
1250}
1251
1252/**
1253 * block_dirty_bitmap_lookup:
1254 * Return a dirty bitmap (if present), after validating
1255 * the node reference and bitmap names.
1256 *
1257 * @node: The name of the BDS node to search for bitmaps
1258 * @name: The name of the bitmap to search for
1259 * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1260 * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1261 * @errp: Output pointer for error information. Can be NULL.
1262 *
1263 * @return: A bitmap object on success, or NULL on failure.
1264 */
1265static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1266                                                  const char *name,
1267                                                  BlockDriverState **pbs,
1268                                                  Error **errp)
1269{
1270    BlockDriverState *bs;
1271    BdrvDirtyBitmap *bitmap;
1272
1273    if (!node) {
1274        error_setg(errp, "Node cannot be NULL");
1275        return NULL;
1276    }
1277    if (!name) {
1278        error_setg(errp, "Bitmap name cannot be NULL");
1279        return NULL;
1280    }
1281    bs = bdrv_lookup_bs(node, node, NULL);
1282    if (!bs) {
1283        error_setg(errp, "Node '%s' not found", node);
1284        return NULL;
1285    }
1286
1287    bitmap = bdrv_find_dirty_bitmap(bs, name);
1288    if (!bitmap) {
1289        error_setg(errp, "Dirty bitmap '%s' not found", name);
1290        return NULL;
1291    }
1292
1293    if (pbs) {
1294        *pbs = bs;
1295    }
1296
1297    return bitmap;
1298}
1299
1300/* New and old BlockDriverState structs for atomic group operations */
1301
1302typedef struct BlkActionState BlkActionState;
1303
1304/**
1305 * BlkActionOps:
1306 * Table of operations that define an Action.
1307 *
1308 * @instance_size: Size of state struct, in bytes.
1309 * @prepare: Prepare the work, must NOT be NULL.
1310 * @commit: Commit the changes, can be NULL.
1311 * @abort: Abort the changes on fail, can be NULL.
1312 * @clean: Clean up resources after all transaction actions have called
1313 *         commit() or abort(). Can be NULL.
1314 *
1315 * Only prepare() may fail. In a single transaction, only one of commit() or
1316 * abort() will be called. clean() will always be called if it is present.
1317 */
1318typedef struct BlkActionOps {
1319    size_t instance_size;
1320    void (*prepare)(BlkActionState *common, Error **errp);
1321    void (*commit)(BlkActionState *common);
1322    void (*abort)(BlkActionState *common);
1323    void (*clean)(BlkActionState *common);
1324} BlkActionOps;
1325
1326/**
1327 * BlkActionState:
1328 * Describes one Action's state within a Transaction.
1329 *
1330 * @action: QAPI-defined enum identifying which Action to perform.
1331 * @ops: Table of ActionOps this Action can perform.
1332 * @block_job_txn: Transaction which this action belongs to.
1333 * @entry: List membership for all Actions in this Transaction.
1334 *
1335 * This structure must be arranged as first member in a subclassed type,
1336 * assuming that the compiler will also arrange it to the same offsets as the
1337 * base class.
1338 */
1339struct BlkActionState {
1340    TransactionAction *action;
1341    const BlkActionOps *ops;
1342    JobTxn *block_job_txn;
1343    TransactionProperties *txn_props;
1344    QSIMPLEQ_ENTRY(BlkActionState) entry;
1345};
1346
1347/* internal snapshot private data */
1348typedef struct InternalSnapshotState {
1349    BlkActionState common;
1350    BlockDriverState *bs;
1351    QEMUSnapshotInfo sn;
1352    bool created;
1353} InternalSnapshotState;
1354
1355
1356static int action_check_completion_mode(BlkActionState *s, Error **errp)
1357{
1358    if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1359        error_setg(errp,
1360                   "Action '%s' does not support Transaction property "
1361                   "completion-mode = %s",
1362                   TransactionActionKind_str(s->action->type),
1363                   ActionCompletionMode_str(s->txn_props->completion_mode));
1364        return -1;
1365    }
1366    return 0;
1367}
1368
1369static void internal_snapshot_prepare(BlkActionState *common,
1370                                      Error **errp)
1371{
1372    Error *local_err = NULL;
1373    const char *device;
1374    const char *name;
1375    BlockDriverState *bs;
1376    QEMUSnapshotInfo old_sn, *sn;
1377    bool ret;
1378    qemu_timeval tv;
1379    BlockdevSnapshotInternal *internal;
1380    InternalSnapshotState *state;
1381    AioContext *aio_context;
1382    int ret1;
1383
1384    g_assert(common->action->type ==
1385             TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1386    internal = common->action->u.blockdev_snapshot_internal_sync.data;
1387    state = DO_UPCAST(InternalSnapshotState, common, common);
1388
1389    /* 1. parse input */
1390    device = internal->device;
1391    name = internal->name;
1392
1393    /* 2. check for validation */
1394    if (action_check_completion_mode(common, errp) < 0) {
1395        return;
1396    }
1397
1398    bs = qmp_get_root_bs(device, errp);
1399    if (!bs) {
1400        return;
1401    }
1402
1403    aio_context = bdrv_get_aio_context(bs);
1404    aio_context_acquire(aio_context);
1405
1406    state->bs = bs;
1407
1408    /* Paired with .clean() */
1409    bdrv_drained_begin(bs);
1410
1411    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1412        goto out;
1413    }
1414
1415    if (bdrv_is_read_only(bs)) {
1416        error_setg(errp, "Device '%s' is read only", device);
1417        goto out;
1418    }
1419
1420    if (!bdrv_can_snapshot(bs)) {
1421        error_setg(errp, "Block format '%s' used by device '%s' "
1422                   "does not support internal snapshots",
1423                   bs->drv->format_name, device);
1424        goto out;
1425    }
1426
1427    if (!strlen(name)) {
1428        error_setg(errp, "Name is empty");
1429        goto out;
1430    }
1431
1432    /* check whether a snapshot with name exist */
1433    ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1434                                            &local_err);
1435    if (local_err) {
1436        error_propagate(errp, local_err);
1437        goto out;
1438    } else if (ret) {
1439        error_setg(errp,
1440                   "Snapshot with name '%s' already exists on device '%s'",
1441                   name, device);
1442        goto out;
1443    }
1444
1445    /* 3. take the snapshot */
1446    sn = &state->sn;
1447    pstrcpy(sn->name, sizeof(sn->name), name);
1448    qemu_gettimeofday(&tv);
1449    sn->date_sec = tv.tv_sec;
1450    sn->date_nsec = tv.tv_usec * 1000;
1451    sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1452
1453    ret1 = bdrv_snapshot_create(bs, sn);
1454    if (ret1 < 0) {
1455        error_setg_errno(errp, -ret1,
1456                         "Failed to create snapshot '%s' on device '%s'",
1457                         name, device);
1458        goto out;
1459    }
1460
1461    /* 4. succeed, mark a snapshot is created */
1462    state->created = true;
1463
1464out:
1465    aio_context_release(aio_context);
1466}
1467
1468static void internal_snapshot_abort(BlkActionState *common)
1469{
1470    InternalSnapshotState *state =
1471                             DO_UPCAST(InternalSnapshotState, common, common);
1472    BlockDriverState *bs = state->bs;
1473    QEMUSnapshotInfo *sn = &state->sn;
1474    AioContext *aio_context;
1475    Error *local_error = NULL;
1476
1477    if (!state->created) {
1478        return;
1479    }
1480
1481    aio_context = bdrv_get_aio_context(state->bs);
1482    aio_context_acquire(aio_context);
1483
1484    if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1485        error_reportf_err(local_error,
1486                          "Failed to delete snapshot with id '%s' and "
1487                          "name '%s' on device '%s' in abort: ",
1488                          sn->id_str, sn->name,
1489                          bdrv_get_device_name(bs));
1490    }
1491
1492    aio_context_release(aio_context);
1493}
1494
1495static void internal_snapshot_clean(BlkActionState *common)
1496{
1497    InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1498                                             common, common);
1499    AioContext *aio_context;
1500
1501    if (!state->bs) {
1502        return;
1503    }
1504
1505    aio_context = bdrv_get_aio_context(state->bs);
1506    aio_context_acquire(aio_context);
1507
1508    bdrv_drained_end(state->bs);
1509
1510    aio_context_release(aio_context);
1511}
1512
1513/* external snapshot private data */
1514typedef struct ExternalSnapshotState {
1515    BlkActionState common;
1516    BlockDriverState *old_bs;
1517    BlockDriverState *new_bs;
1518    bool overlay_appended;
1519} ExternalSnapshotState;
1520
1521static void external_snapshot_prepare(BlkActionState *common,
1522                                      Error **errp)
1523{
1524    int flags = 0;
1525    QDict *options = NULL;
1526    Error *local_err = NULL;
1527    /* Device and node name of the image to generate the snapshot from */
1528    const char *device;
1529    const char *node_name;
1530    /* Reference to the new image (for 'blockdev-snapshot') */
1531    const char *snapshot_ref;
1532    /* File name of the new image (for 'blockdev-snapshot-sync') */
1533    const char *new_image_file;
1534    ExternalSnapshotState *state =
1535                             DO_UPCAST(ExternalSnapshotState, common, common);
1536    TransactionAction *action = common->action;
1537    AioContext *aio_context;
1538
1539    /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1540     * purpose but a different set of parameters */
1541    switch (action->type) {
1542    case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1543        {
1544            BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1545            device = s->node;
1546            node_name = s->node;
1547            new_image_file = NULL;
1548            snapshot_ref = s->overlay;
1549        }
1550        break;
1551    case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1552        {
1553            BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1554            device = s->has_device ? s->device : NULL;
1555            node_name = s->has_node_name ? s->node_name : NULL;
1556            new_image_file = s->snapshot_file;
1557            snapshot_ref = NULL;
1558        }
1559        break;
1560    default:
1561        g_assert_not_reached();
1562    }
1563
1564    /* start processing */
1565    if (action_check_completion_mode(common, errp) < 0) {
1566        return;
1567    }
1568
1569    state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1570    if (!state->old_bs) {
1571        return;
1572    }
1573
1574    aio_context = bdrv_get_aio_context(state->old_bs);
1575    aio_context_acquire(aio_context);
1576
1577    /* Paired with .clean() */
1578    bdrv_drained_begin(state->old_bs);
1579
1580    if (!bdrv_is_inserted(state->old_bs)) {
1581        error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1582        goto out;
1583    }
1584
1585    if (bdrv_op_is_blocked(state->old_bs,
1586                           BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1587        goto out;
1588    }
1589
1590    if (!bdrv_is_read_only(state->old_bs)) {
1591        if (bdrv_flush(state->old_bs)) {
1592            error_setg(errp, QERR_IO_ERROR);
1593            goto out;
1594        }
1595    }
1596
1597    if (!bdrv_is_first_non_filter(state->old_bs)) {
1598        error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1599        goto out;
1600    }
1601
1602    if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1603        BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1604        const char *format = s->has_format ? s->format : "qcow2";
1605        enum NewImageMode mode;
1606        const char *snapshot_node_name =
1607            s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1608
1609        if (node_name && !snapshot_node_name) {
1610            error_setg(errp, "New snapshot node name missing");
1611            goto out;
1612        }
1613
1614        if (snapshot_node_name &&
1615            bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1616            error_setg(errp, "New snapshot node name already in use");
1617            goto out;
1618        }
1619
1620        flags = state->old_bs->open_flags;
1621        flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1622        flags |= BDRV_O_NO_BACKING;
1623
1624        /* create new image w/backing file */
1625        mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1626        if (mode != NEW_IMAGE_MODE_EXISTING) {
1627            int64_t size = bdrv_getlength(state->old_bs);
1628            if (size < 0) {
1629                error_setg_errno(errp, -size, "bdrv_getlength failed");
1630                goto out;
1631            }
1632            bdrv_img_create(new_image_file, format,
1633                            state->old_bs->filename,
1634                            state->old_bs->drv->format_name,
1635                            NULL, size, flags, false, &local_err);
1636            if (local_err) {
1637                error_propagate(errp, local_err);
1638                goto out;
1639            }
1640        }
1641
1642        options = qdict_new();
1643        if (snapshot_node_name) {
1644            qdict_put_str(options, "node-name", snapshot_node_name);
1645        }
1646        qdict_put_str(options, "driver", format);
1647    }
1648
1649    state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1650                              errp);
1651    /* We will manually add the backing_hd field to the bs later */
1652    if (!state->new_bs) {
1653        goto out;
1654    }
1655
1656    if (bdrv_has_blk(state->new_bs)) {
1657        error_setg(errp, "The snapshot is already in use");
1658        goto out;
1659    }
1660
1661    if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1662                           errp)) {
1663        goto out;
1664    }
1665
1666    if (state->new_bs->backing != NULL) {
1667        error_setg(errp, "The snapshot already has a backing image");
1668        goto out;
1669    }
1670
1671    if (!state->new_bs->drv->supports_backing) {
1672        error_setg(errp, "The snapshot does not support backing images");
1673        goto out;
1674    }
1675
1676    bdrv_set_aio_context(state->new_bs, aio_context);
1677
1678    /* This removes our old bs and adds the new bs. This is an operation that
1679     * can fail, so we need to do it in .prepare; undoing it for abort is
1680     * always possible. */
1681    bdrv_ref(state->new_bs);
1682    bdrv_append(state->new_bs, state->old_bs, &local_err);
1683    if (local_err) {
1684        error_propagate(errp, local_err);
1685        goto out;
1686    }
1687    state->overlay_appended = true;
1688
1689out:
1690    aio_context_release(aio_context);
1691}
1692
1693static void external_snapshot_commit(BlkActionState *common)
1694{
1695    ExternalSnapshotState *state =
1696                             DO_UPCAST(ExternalSnapshotState, common, common);
1697    AioContext *aio_context;
1698
1699    aio_context = bdrv_get_aio_context(state->old_bs);
1700    aio_context_acquire(aio_context);
1701
1702    /* We don't need (or want) to use the transactional
1703     * bdrv_reopen_multiple() across all the entries at once, because we
1704     * don't want to abort all of them if one of them fails the reopen */
1705    if (!atomic_read(&state->old_bs->copy_on_read)) {
1706        bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1707                    NULL);
1708    }
1709
1710    aio_context_release(aio_context);
1711}
1712
1713static void external_snapshot_abort(BlkActionState *common)
1714{
1715    ExternalSnapshotState *state =
1716                             DO_UPCAST(ExternalSnapshotState, common, common);
1717    if (state->new_bs) {
1718        if (state->overlay_appended) {
1719            AioContext *aio_context;
1720
1721            aio_context = bdrv_get_aio_context(state->old_bs);
1722            aio_context_acquire(aio_context);
1723
1724            bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1725                                          close state->old_bs; we need it */
1726            bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1727            bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1728            bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1729
1730            aio_context_release(aio_context);
1731        }
1732    }
1733}
1734
1735static void external_snapshot_clean(BlkActionState *common)
1736{
1737    ExternalSnapshotState *state =
1738                             DO_UPCAST(ExternalSnapshotState, common, common);
1739    AioContext *aio_context;
1740
1741    if (!state->old_bs) {
1742        return;
1743    }
1744
1745    aio_context = bdrv_get_aio_context(state->old_bs);
1746    aio_context_acquire(aio_context);
1747
1748    bdrv_drained_end(state->old_bs);
1749    bdrv_unref(state->new_bs);
1750
1751    aio_context_release(aio_context);
1752}
1753
1754typedef struct DriveBackupState {
1755    BlkActionState common;
1756    BlockDriverState *bs;
1757    BlockJob *job;
1758} DriveBackupState;
1759
1760static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
1761                            Error **errp);
1762
1763static void drive_backup_prepare(BlkActionState *common, Error **errp)
1764{
1765    DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1766    BlockDriverState *bs;
1767    DriveBackup *backup;
1768    AioContext *aio_context;
1769    Error *local_err = NULL;
1770
1771    assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1772    backup = common->action->u.drive_backup.data;
1773
1774    bs = qmp_get_root_bs(backup->device, errp);
1775    if (!bs) {
1776        return;
1777    }
1778
1779    aio_context = bdrv_get_aio_context(bs);
1780    aio_context_acquire(aio_context);
1781
1782    /* Paired with .clean() */
1783    bdrv_drained_begin(bs);
1784
1785    state->bs = bs;
1786
1787    state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1788    if (local_err) {
1789        error_propagate(errp, local_err);
1790        goto out;
1791    }
1792
1793out:
1794    aio_context_release(aio_context);
1795}
1796
1797static void drive_backup_commit(BlkActionState *common)
1798{
1799    DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1800    AioContext *aio_context;
1801
1802    aio_context = bdrv_get_aio_context(state->bs);
1803    aio_context_acquire(aio_context);
1804
1805    assert(state->job);
1806    job_start(&state->job->job);
1807
1808    aio_context_release(aio_context);
1809}
1810
1811static void drive_backup_abort(BlkActionState *common)
1812{
1813    DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1814
1815    if (state->job) {
1816        AioContext *aio_context;
1817
1818        aio_context = bdrv_get_aio_context(state->bs);
1819        aio_context_acquire(aio_context);
1820
1821        job_cancel_sync(&state->job->job);
1822
1823        aio_context_release(aio_context);
1824    }
1825}
1826
1827static void drive_backup_clean(BlkActionState *common)
1828{
1829    DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1830    AioContext *aio_context;
1831
1832    if (!state->bs) {
1833        return;
1834    }
1835
1836    aio_context = bdrv_get_aio_context(state->bs);
1837    aio_context_acquire(aio_context);
1838
1839    bdrv_drained_end(state->bs);
1840
1841    aio_context_release(aio_context);
1842}
1843
1844typedef struct BlockdevBackupState {
1845    BlkActionState common;
1846    BlockDriverState *bs;
1847    BlockJob *job;
1848} BlockdevBackupState;
1849
1850static BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
1851                                    Error **errp);
1852
1853static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1854{
1855    BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1856    BlockdevBackup *backup;
1857    BlockDriverState *bs, *target;
1858    AioContext *aio_context;
1859    Error *local_err = NULL;
1860
1861    assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1862    backup = common->action->u.blockdev_backup.data;
1863
1864    bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1865    if (!bs) {
1866        return;
1867    }
1868
1869    target = bdrv_lookup_bs(backup->target, backup->target, errp);
1870    if (!target) {
1871        return;
1872    }
1873
1874    aio_context = bdrv_get_aio_context(bs);
1875    if (aio_context != bdrv_get_aio_context(target)) {
1876        error_setg(errp, "Backup between two IO threads is not implemented");
1877        return;
1878    }
1879    aio_context_acquire(aio_context);
1880    state->bs = bs;
1881
1882    /* Paired with .clean() */
1883    bdrv_drained_begin(state->bs);
1884
1885    state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1886    if (local_err) {
1887        error_propagate(errp, local_err);
1888        goto out;
1889    }
1890
1891out:
1892    aio_context_release(aio_context);
1893}
1894
1895static void blockdev_backup_commit(BlkActionState *common)
1896{
1897    BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1898    AioContext *aio_context;
1899
1900    aio_context = bdrv_get_aio_context(state->bs);
1901    aio_context_acquire(aio_context);
1902
1903    assert(state->job);
1904    job_start(&state->job->job);
1905
1906    aio_context_release(aio_context);
1907}
1908
1909static void blockdev_backup_abort(BlkActionState *common)
1910{
1911    BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1912
1913    if (state->job) {
1914        AioContext *aio_context;
1915
1916        aio_context = bdrv_get_aio_context(state->bs);
1917        aio_context_acquire(aio_context);
1918
1919        job_cancel_sync(&state->job->job);
1920
1921        aio_context_release(aio_context);
1922    }
1923}
1924
1925static void blockdev_backup_clean(BlkActionState *common)
1926{
1927    BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1928    AioContext *aio_context;
1929
1930    if (!state->bs) {
1931        return;
1932    }
1933
1934    aio_context = bdrv_get_aio_context(state->bs);
1935    aio_context_acquire(aio_context);
1936
1937    bdrv_drained_end(state->bs);
1938
1939    aio_context_release(aio_context);
1940}
1941
1942typedef struct BlockDirtyBitmapState {
1943    BlkActionState common;
1944    BdrvDirtyBitmap *bitmap;
1945    BlockDriverState *bs;
1946    HBitmap *backup;
1947    bool prepared;
1948    bool was_enabled;
1949} BlockDirtyBitmapState;
1950
1951static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1952                                           Error **errp)
1953{
1954    Error *local_err = NULL;
1955    BlockDirtyBitmapAdd *action;
1956    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1957                                             common, common);
1958
1959    if (action_check_completion_mode(common, errp) < 0) {
1960        return;
1961    }
1962
1963    action = common->action->u.block_dirty_bitmap_add.data;
1964    /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1965    qmp_block_dirty_bitmap_add(action->node, action->name,
1966                               action->has_granularity, action->granularity,
1967                               action->has_persistent, action->persistent,
1968                               action->has_autoload, action->autoload,
1969                               action->has_x_disabled, action->x_disabled,
1970                               &local_err);
1971
1972    if (!local_err) {
1973        state->prepared = true;
1974    } else {
1975        error_propagate(errp, local_err);
1976    }
1977}
1978
1979static void block_dirty_bitmap_add_abort(BlkActionState *common)
1980{
1981    BlockDirtyBitmapAdd *action;
1982    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1983                                             common, common);
1984
1985    action = common->action->u.block_dirty_bitmap_add.data;
1986    /* Should not be able to fail: IF the bitmap was added via .prepare(),
1987     * then the node reference and bitmap name must have been valid.
1988     */
1989    if (state->prepared) {
1990        qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1991    }
1992}
1993
1994static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1995                                             Error **errp)
1996{
1997    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1998                                             common, common);
1999    BlockDirtyBitmap *action;
2000
2001    if (action_check_completion_mode(common, errp) < 0) {
2002        return;
2003    }
2004
2005    action = common->action->u.block_dirty_bitmap_clear.data;
2006    state->bitmap = block_dirty_bitmap_lookup(action->node,
2007                                              action->name,
2008                                              &state->bs,
2009                                              errp);
2010    if (!state->bitmap) {
2011        return;
2012    }
2013
2014    if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2015        error_setg(errp, "Cannot modify a bitmap in use by another operation");
2016        return;
2017    } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2018        error_setg(errp, "Cannot clear a readonly bitmap");
2019        return;
2020    }
2021
2022    bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2023}
2024
2025static void block_dirty_bitmap_restore(BlkActionState *common)
2026{
2027    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2028                                             common, common);
2029
2030    if (state->backup) {
2031        bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2032    }
2033}
2034
2035static void block_dirty_bitmap_free_backup(BlkActionState *common)
2036{
2037    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2038                                             common, common);
2039
2040    hbitmap_free(state->backup);
2041}
2042
2043static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2044                                              Error **errp)
2045{
2046    BlockDirtyBitmap *action;
2047    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2048                                             common, common);
2049
2050    if (action_check_completion_mode(common, errp) < 0) {
2051        return;
2052    }
2053
2054    action = common->action->u.x_block_dirty_bitmap_enable.data;
2055    state->bitmap = block_dirty_bitmap_lookup(action->node,
2056                                              action->name,
2057                                              NULL,
2058                                              errp);
2059    if (!state->bitmap) {
2060        return;
2061    }
2062
2063    if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2064        error_setg(errp,
2065                   "Bitmap '%s' is currently in use by another operation"
2066                   " and cannot be enabled", action->name);
2067        return;
2068    }
2069
2070    state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2071    bdrv_enable_dirty_bitmap(state->bitmap);
2072}
2073
2074static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2075{
2076    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2077                                             common, common);
2078
2079    if (!state->was_enabled) {
2080        bdrv_disable_dirty_bitmap(state->bitmap);
2081    }
2082}
2083
2084static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2085                                               Error **errp)
2086{
2087    BlockDirtyBitmap *action;
2088    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2089                                             common, common);
2090
2091    if (action_check_completion_mode(common, errp) < 0) {
2092        return;
2093    }
2094
2095    action = common->action->u.x_block_dirty_bitmap_disable.data;
2096    state->bitmap = block_dirty_bitmap_lookup(action->node,
2097                                              action->name,
2098                                              NULL,
2099                                              errp);
2100    if (!state->bitmap) {
2101        return;
2102    }
2103
2104    if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2105        error_setg(errp,
2106                   "Bitmap '%s' is currently in use by another operation"
2107                   " and cannot be disabled", action->name);
2108        return;
2109    }
2110
2111    state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2112    bdrv_disable_dirty_bitmap(state->bitmap);
2113}
2114
2115static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2116{
2117    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2118                                             common, common);
2119
2120    if (state->was_enabled) {
2121        bdrv_enable_dirty_bitmap(state->bitmap);
2122    }
2123}
2124
2125static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2126                                             Error **errp)
2127{
2128    BlockDirtyBitmapMerge *action;
2129    BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2130                                             common, common);
2131    BdrvDirtyBitmap *merge_source;
2132
2133    if (action_check_completion_mode(common, errp) < 0) {
2134        return;
2135    }
2136
2137    action = common->action->u.x_block_dirty_bitmap_merge.data;
2138    state->bitmap = block_dirty_bitmap_lookup(action->node,
2139                                              action->dst_name,
2140                                              &state->bs,
2141                                              errp);
2142    if (!state->bitmap) {
2143        return;
2144    }
2145
2146    merge_source = bdrv_find_dirty_bitmap(state->bs, action->src_name);
2147    if (!merge_source) {
2148        return;
2149    }
2150
2151    bdrv_merge_dirty_bitmap(state->bitmap, merge_source, &state->backup, errp);
2152}
2153
2154static void abort_prepare(BlkActionState *common, Error **errp)
2155{
2156    error_setg(errp, "Transaction aborted using Abort action");
2157}
2158
2159static void abort_commit(BlkActionState *common)
2160{
2161    g_assert_not_reached(); /* this action never succeeds */
2162}
2163
2164static const BlkActionOps actions[] = {
2165    [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2166        .instance_size = sizeof(ExternalSnapshotState),
2167        .prepare  = external_snapshot_prepare,
2168        .commit   = external_snapshot_commit,
2169        .abort = external_snapshot_abort,
2170        .clean = external_snapshot_clean,
2171    },
2172    [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2173        .instance_size = sizeof(ExternalSnapshotState),
2174        .prepare  = external_snapshot_prepare,
2175        .commit   = external_snapshot_commit,
2176        .abort = external_snapshot_abort,
2177        .clean = external_snapshot_clean,
2178    },
2179    [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2180        .instance_size = sizeof(DriveBackupState),
2181        .prepare = drive_backup_prepare,
2182        .commit = drive_backup_commit,
2183        .abort = drive_backup_abort,
2184        .clean = drive_backup_clean,
2185    },
2186    [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2187        .instance_size = sizeof(BlockdevBackupState),
2188        .prepare = blockdev_backup_prepare,
2189        .commit = blockdev_backup_commit,
2190        .abort = blockdev_backup_abort,
2191        .clean = blockdev_backup_clean,
2192    },
2193    [TRANSACTION_ACTION_KIND_ABORT] = {
2194        .instance_size = sizeof(BlkActionState),
2195        .prepare = abort_prepare,
2196        .commit = abort_commit,
2197    },
2198    [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2199        .instance_size = sizeof(InternalSnapshotState),
2200        .prepare  = internal_snapshot_prepare,
2201        .abort = internal_snapshot_abort,
2202        .clean = internal_snapshot_clean,
2203    },
2204    [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2205        .instance_size = sizeof(BlockDirtyBitmapState),
2206        .prepare = block_dirty_bitmap_add_prepare,
2207        .abort = block_dirty_bitmap_add_abort,
2208    },
2209    [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2210        .instance_size = sizeof(BlockDirtyBitmapState),
2211        .prepare = block_dirty_bitmap_clear_prepare,
2212        .commit = block_dirty_bitmap_free_backup,
2213        .abort = block_dirty_bitmap_restore,
2214    },
2215    [TRANSACTION_ACTION_KIND_X_BLOCK_DIRTY_BITMAP_ENABLE] = {
2216        .instance_size = sizeof(BlockDirtyBitmapState),
2217        .prepare = block_dirty_bitmap_enable_prepare,
2218        .abort = block_dirty_bitmap_enable_abort,
2219    },
2220    [TRANSACTION_ACTION_KIND_X_BLOCK_DIRTY_BITMAP_DISABLE] = {
2221        .instance_size = sizeof(BlockDirtyBitmapState),
2222        .prepare = block_dirty_bitmap_disable_prepare,
2223        .abort = block_dirty_bitmap_disable_abort,
2224    },
2225    [TRANSACTION_ACTION_KIND_X_BLOCK_DIRTY_BITMAP_MERGE] = {
2226        .instance_size = sizeof(BlockDirtyBitmapState),
2227        .prepare = block_dirty_bitmap_merge_prepare,
2228        .commit = block_dirty_bitmap_free_backup,
2229        .abort = block_dirty_bitmap_restore,
2230    },
2231    /* Where are transactions for MIRROR, COMMIT and STREAM?
2232     * Although these blockjobs use transaction callbacks like the backup job,
2233     * these jobs do not necessarily adhere to transaction semantics.
2234     * These jobs may not fully undo all of their actions on abort, nor do they
2235     * necessarily work in transactions with more than one job in them.
2236     */
2237};
2238
2239/**
2240 * Allocate a TransactionProperties structure if necessary, and fill
2241 * that structure with desired defaults if they are unset.
2242 */
2243static TransactionProperties *get_transaction_properties(
2244    TransactionProperties *props)
2245{
2246    if (!props) {
2247        props = g_new0(TransactionProperties, 1);
2248    }
2249
2250    if (!props->has_completion_mode) {
2251        props->has_completion_mode = true;
2252        props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2253    }
2254
2255    return props;
2256}
2257
2258/*
2259 * 'Atomic' group operations.  The operations are performed as a set, and if
2260 * any fail then we roll back all operations in the group.
2261 */
2262void qmp_transaction(TransactionActionList *dev_list,
2263                     bool has_props,
2264                     struct TransactionProperties *props,
2265                     Error **errp)
2266{
2267    TransactionActionList *dev_entry = dev_list;
2268    JobTxn *block_job_txn = NULL;
2269    BlkActionState *state, *next;
2270    Error *local_err = NULL;
2271
2272    QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2273    QSIMPLEQ_INIT(&snap_bdrv_states);
2274
2275    /* Does this transaction get canceled as a group on failure?
2276     * If not, we don't really need to make a JobTxn.
2277     */
2278    props = get_transaction_properties(props);
2279    if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2280        block_job_txn = job_txn_new();
2281    }
2282
2283    /* drain all i/o before any operations */
2284    bdrv_drain_all();
2285
2286    /* We don't do anything in this loop that commits us to the operations */
2287    while (NULL != dev_entry) {
2288        TransactionAction *dev_info = NULL;
2289        const BlkActionOps *ops;
2290
2291        dev_info = dev_entry->value;
2292        dev_entry = dev_entry->next;
2293
2294        assert(dev_info->type < ARRAY_SIZE(actions));
2295
2296        ops = &actions[dev_info->type];
2297        assert(ops->instance_size > 0);
2298
2299        state = g_malloc0(ops->instance_size);
2300        state->ops = ops;
2301        state->action = dev_info;
2302        state->block_job_txn = block_job_txn;
2303        state->txn_props = props;
2304        QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2305
2306        state->ops->prepare(state, &local_err);
2307        if (local_err) {
2308            error_propagate(errp, local_err);
2309            goto delete_and_fail;
2310        }
2311    }
2312
2313    QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2314        if (state->ops->commit) {
2315            state->ops->commit(state);
2316        }
2317    }
2318
2319    /* success */
2320    goto exit;
2321
2322delete_and_fail:
2323    /* failure, and it is all-or-none; roll back all operations */
2324    QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2325        if (state->ops->abort) {
2326            state->ops->abort(state);
2327        }
2328    }
2329exit:
2330    QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2331        if (state->ops->clean) {
2332            state->ops->clean(state);
2333        }
2334        g_free(state);
2335    }
2336    if (!has_props) {
2337        qapi_free_TransactionProperties(props);
2338    }
2339    job_txn_unref(block_job_txn);
2340}
2341
2342void qmp_eject(bool has_device, const char *device,
2343               bool has_id, const char *id,
2344               bool has_force, bool force, Error **errp)
2345{
2346    Error *local_err = NULL;
2347    int rc;
2348
2349    if (!has_force) {
2350        force = false;
2351    }
2352
2353    rc = do_open_tray(has_device ? device : NULL,
2354                      has_id ? id : NULL,
2355                      force, &local_err);
2356    if (rc && rc != -ENOSYS) {
2357        error_propagate(errp, local_err);
2358        return;
2359    }
2360    error_free(local_err);
2361
2362    blockdev_remove_medium(has_device, device, has_id, id, errp);
2363}
2364
2365void qmp_block_passwd(bool has_device, const char *device,
2366                      bool has_node_name, const char *node_name,
2367                      const char *password, Error **errp)
2368{
2369    error_setg(errp,
2370               "Setting block passwords directly is no longer supported");
2371}
2372
2373/*
2374 * Attempt to open the tray of @device.
2375 * If @force, ignore its tray lock.
2376 * Else, if the tray is locked, don't open it, but ask the guest to open it.
2377 * On error, store an error through @errp and return -errno.
2378 * If @device does not exist, return -ENODEV.
2379 * If it has no removable media, return -ENOTSUP.
2380 * If it has no tray, return -ENOSYS.
2381 * If the guest was asked to open the tray, return -EINPROGRESS.
2382 * Else, return 0.
2383 */
2384static int do_open_tray(const char *blk_name, const char *qdev_id,
2385                        bool force, Error **errp)
2386{
2387    BlockBackend *blk;
2388    const char *device = qdev_id ?: blk_name;
2389    bool locked;
2390
2391    blk = qmp_get_blk(blk_name, qdev_id, errp);
2392    if (!blk) {
2393        return -ENODEV;
2394    }
2395
2396    if (!blk_dev_has_removable_media(blk)) {
2397        error_setg(errp, "Device '%s' is not removable", device);
2398        return -ENOTSUP;
2399    }
2400
2401    if (!blk_dev_has_tray(blk)) {
2402        error_setg(errp, "Device '%s' does not have a tray", device);
2403        return -ENOSYS;
2404    }
2405
2406    if (blk_dev_is_tray_open(blk)) {
2407        return 0;
2408    }
2409
2410    locked = blk_dev_is_medium_locked(blk);
2411    if (locked) {
2412        blk_dev_eject_request(blk, force);
2413    }
2414
2415    if (!locked || force) {
2416        blk_dev_change_media_cb(blk, false, &error_abort);
2417    }
2418
2419    if (locked && !force) {
2420        error_setg(errp, "Device '%s' is locked and force was not specified, "
2421                   "wait for tray to open and try again", device);
2422        return -EINPROGRESS;
2423    }
2424
2425    return 0;
2426}
2427
2428void qmp_blockdev_open_tray(bool has_device, const char *device,
2429                            bool has_id, const char *id,
2430                            bool has_force, bool force,
2431                            Error **errp)
2432{
2433    Error *local_err = NULL;
2434    int rc;
2435
2436    if (!has_force) {
2437        force = false;
2438    }
2439    rc = do_open_tray(has_device ? device : NULL,
2440                      has_id ? id : NULL,
2441                      force, &local_err);
2442    if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2443        error_propagate(errp, local_err);
2444        return;
2445    }
2446    error_free(local_err);
2447}
2448
2449void qmp_blockdev_close_tray(bool has_device, const char *device,
2450                             bool has_id, const char *id,
2451                             Error **errp)
2452{
2453    BlockBackend *blk;
2454    Error *local_err = NULL;
2455
2456    device = has_device ? device : NULL;
2457    id = has_id ? id : NULL;
2458
2459    blk = qmp_get_blk(device, id, errp);
2460    if (!blk) {
2461        return;
2462    }
2463
2464    if (!blk_dev_has_removable_media(blk)) {
2465        error_setg(errp, "Device '%s' is not removable", device ?: id);
2466        return;
2467    }
2468
2469    if (!blk_dev_has_tray(blk)) {
2470        /* Ignore this command on tray-less devices */
2471        return;
2472    }
2473
2474    if (!blk_dev_is_tray_open(blk)) {
2475        return;
2476    }
2477
2478    blk_dev_change_media_cb(blk, true, &local_err);
2479    if (local_err) {
2480        error_propagate(errp, local_err);
2481        return;
2482    }
2483}
2484
2485static void blockdev_remove_medium(bool has_device, const char *device,
2486                                   bool has_id, const char *id, Error **errp)
2487{
2488    BlockBackend *blk;
2489    BlockDriverState *bs;
2490    AioContext *aio_context;
2491    bool has_attached_device;
2492
2493    device = has_device ? device : NULL;
2494    id = has_id ? id : NULL;
2495
2496    blk = qmp_get_blk(device, id, errp);
2497    if (!blk) {
2498        return;
2499    }
2500
2501    /* For BBs without a device, we can exchange the BDS tree at will */
2502    has_attached_device = blk_get_attached_dev(blk);
2503
2504    if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2505        error_setg(errp, "Device '%s' is not removable", device ?: id);
2506        return;
2507    }
2508
2509    if (has_attached_device && blk_dev_has_tray(blk) &&
2510        !blk_dev_is_tray_open(blk))
2511    {
2512        error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2513        return;
2514    }
2515
2516    bs = blk_bs(blk);
2517    if (!bs) {
2518        return;
2519    }
2520
2521    aio_context = bdrv_get_aio_context(bs);
2522    aio_context_acquire(aio_context);
2523
2524    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2525        goto out;
2526    }
2527
2528    blk_remove_bs(blk);
2529
2530    if (!blk_dev_has_tray(blk)) {
2531        /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2532         * called at all); therefore, the medium needs to be ejected here.
2533         * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2534         * value passed here (i.e. false). */
2535        blk_dev_change_media_cb(blk, false, &error_abort);
2536    }
2537
2538out:
2539    aio_context_release(aio_context);
2540}
2541
2542void qmp_blockdev_remove_medium(const char *id, Error **errp)
2543{
2544    blockdev_remove_medium(false, NULL, true, id, errp);
2545}
2546
2547static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2548                                            BlockDriverState *bs, Error **errp)
2549{
2550    Error *local_err = NULL;
2551    bool has_device;
2552    int ret;
2553
2554    /* For BBs without a device, we can exchange the BDS tree at will */
2555    has_device = blk_get_attached_dev(blk);
2556
2557    if (has_device && !blk_dev_has_removable_media(blk)) {
2558        error_setg(errp, "Device is not removable");
2559        return;
2560    }
2561
2562    if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2563        error_setg(errp, "Tray of the device is not open");
2564        return;
2565    }
2566
2567    if (blk_bs(blk)) {
2568        error_setg(errp, "There already is a medium in the device");
2569        return;
2570    }
2571
2572    ret = blk_insert_bs(blk, bs, errp);
2573    if (ret < 0) {
2574        return;
2575    }
2576
2577    if (!blk_dev_has_tray(blk)) {
2578        /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2579         * called at all); therefore, the medium needs to be pushed into the
2580         * slot here.
2581         * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2582         * value passed here (i.e. true). */
2583        blk_dev_change_media_cb(blk, true, &local_err);
2584        if (local_err) {
2585            error_propagate(errp, local_err);
2586            blk_remove_bs(blk);
2587            return;
2588        }
2589    }
2590}
2591
2592static void blockdev_insert_medium(bool has_device, const char *device,
2593                                   bool has_id, const char *id,
2594                                   const char *node_name, Error **errp)
2595{
2596    BlockBackend *blk;
2597    BlockDriverState *bs;
2598
2599    blk = qmp_get_blk(has_device ? device : NULL,
2600                      has_id ? id : NULL,
2601                      errp);
2602    if (!blk) {
2603        return;
2604    }
2605
2606    bs = bdrv_find_node(node_name);
2607    if (!bs) {
2608        error_setg(errp, "Node '%s' not found", node_name);
2609        return;
2610    }
2611
2612    if (bdrv_has_blk(bs)) {
2613        error_setg(errp, "Node '%s' is already in use", node_name);
2614        return;
2615    }
2616
2617    qmp_blockdev_insert_anon_medium(blk, bs, errp);
2618}
2619
2620void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2621                                Error **errp)
2622{
2623    blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2624}
2625
2626void qmp_blockdev_change_medium(bool has_device, const char *device,
2627                                bool has_id, const char *id,
2628                                const char *filename,
2629                                bool has_format, const char *format,
2630                                bool has_read_only,
2631                                BlockdevChangeReadOnlyMode read_only,
2632                                Error **errp)
2633{
2634    BlockBackend *blk;
2635    BlockDriverState *medium_bs = NULL;
2636    int bdrv_flags;
2637    bool detect_zeroes;
2638    int rc;
2639    QDict *options = NULL;
2640    Error *err = NULL;
2641
2642    blk = qmp_get_blk(has_device ? device : NULL,
2643                      has_id ? id : NULL,
2644                      errp);
2645    if (!blk) {
2646        goto fail;
2647    }
2648
2649    if (blk_bs(blk)) {
2650        blk_update_root_state(blk);
2651    }
2652
2653    bdrv_flags = blk_get_open_flags_from_root_state(blk);
2654    bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2655        BDRV_O_PROTOCOL | BDRV_O_AUTO_RDONLY);
2656
2657    if (!has_read_only) {
2658        read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2659    }
2660
2661    switch (read_only) {
2662    case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2663        break;
2664
2665    case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2666        bdrv_flags &= ~BDRV_O_RDWR;
2667        break;
2668
2669    case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2670        bdrv_flags |= BDRV_O_RDWR;
2671        break;
2672
2673    default:
2674        abort();
2675    }
2676
2677    options = qdict_new();
2678    detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2679    qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2680
2681    if (has_format) {
2682        qdict_put_str(options, "driver", format);
2683    }
2684
2685    medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2686    if (!medium_bs) {
2687        goto fail;
2688    }
2689
2690    rc = do_open_tray(has_device ? device : NULL,
2691                      has_id ? id : NULL,
2692                      false, &err);
2693    if (rc && rc != -ENOSYS) {
2694        error_propagate(errp, err);
2695        goto fail;
2696    }
2697    error_free(err);
2698    err = NULL;
2699
2700    blockdev_remove_medium(has_device, device, has_id, id, &err);
2701    if (err) {
2702        error_propagate(errp, err);
2703        goto fail;
2704    }
2705
2706    qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2707    if (err) {
2708        error_propagate(errp, err);
2709        goto fail;
2710    }
2711
2712    qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2713
2714fail:
2715    /* If the medium has been inserted, the device has its own reference, so
2716     * ours must be relinquished; and if it has not been inserted successfully,
2717     * the reference must be relinquished anyway */
2718    bdrv_unref(medium_bs);
2719}
2720
2721/* throttling disk I/O limits */
2722void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2723{
2724    ThrottleConfig cfg;
2725    BlockDriverState *bs;
2726    BlockBackend *blk;
2727    AioContext *aio_context;
2728
2729    blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2730                      arg->has_id ? arg->id : NULL,
2731                      errp);
2732    if (!blk) {
2733        return;
2734    }
2735
2736    aio_context = blk_get_aio_context(blk);
2737    aio_context_acquire(aio_context);
2738
2739    bs = blk_bs(blk);
2740    if (!bs) {
2741        error_setg(errp, "Device has no medium");
2742        goto out;
2743    }
2744
2745    throttle_config_init(&cfg);
2746    cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2747    cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2748    cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2749
2750    cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2751    cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2752    cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2753
2754    if (arg->has_bps_max) {
2755        cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2756    }
2757    if (arg->has_bps_rd_max) {
2758        cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2759    }
2760    if (arg->has_bps_wr_max) {
2761        cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2762    }
2763    if (arg->has_iops_max) {
2764        cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2765    }
2766    if (arg->has_iops_rd_max) {
2767        cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2768    }
2769    if (arg->has_iops_wr_max) {
2770        cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2771    }
2772
2773    if (arg->has_bps_max_length) {
2774        cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2775    }
2776    if (arg->has_bps_rd_max_length) {
2777        cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2778    }
2779    if (arg->has_bps_wr_max_length) {
2780        cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2781    }
2782    if (arg->has_iops_max_length) {
2783        cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2784    }
2785    if (arg->has_iops_rd_max_length) {
2786        cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2787    }
2788    if (arg->has_iops_wr_max_length) {
2789        cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2790    }
2791
2792    if (arg->has_iops_size) {
2793        cfg.op_size = arg->iops_size;
2794    }
2795
2796    if (!throttle_is_valid(&cfg, errp)) {
2797        goto out;
2798    }
2799
2800    if (throttle_enabled(&cfg)) {
2801        /* Enable I/O limits if they're not enabled yet, otherwise
2802         * just update the throttling group. */
2803        if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2804            blk_io_limits_enable(blk,
2805                                 arg->has_group ? arg->group :
2806                                 arg->has_device ? arg->device :
2807                                 arg->id);
2808        } else if (arg->has_group) {
2809            blk_io_limits_update_group(blk, arg->group);
2810        }
2811        /* Set the new throttling configuration */
2812        blk_set_io_limits(blk, &cfg);
2813    } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2814        /* If all throttling settings are set to 0, disable I/O limits */
2815        blk_io_limits_disable(blk);
2816    }
2817
2818out:
2819    aio_context_release(aio_context);
2820}
2821
2822void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2823                                bool has_granularity, uint32_t granularity,
2824                                bool has_persistent, bool persistent,
2825                                bool has_autoload, bool autoload,
2826                                bool has_disabled, bool disabled,
2827                                Error **errp)
2828{
2829    BlockDriverState *bs;
2830    BdrvDirtyBitmap *bitmap;
2831
2832    if (!name || name[0] == '\0') {
2833        error_setg(errp, "Bitmap name cannot be empty");
2834        return;
2835    }
2836
2837    bs = bdrv_lookup_bs(node, node, errp);
2838    if (!bs) {
2839        return;
2840    }
2841
2842    if (has_granularity) {
2843        if (granularity < 512 || !is_power_of_2(granularity)) {
2844            error_setg(errp, "Granularity must be power of 2 "
2845                             "and at least 512");
2846            return;
2847        }
2848    } else {
2849        /* Default to cluster size, if available: */
2850        granularity = bdrv_get_default_bitmap_granularity(bs);
2851    }
2852
2853    if (!has_persistent) {
2854        persistent = false;
2855    }
2856
2857    if (has_autoload) {
2858        warn_report("Autoload option is deprecated and its value is ignored");
2859    }
2860
2861    if (!has_disabled) {
2862        disabled = false;
2863    }
2864
2865    if (persistent &&
2866        !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2867    {
2868        return;
2869    }
2870
2871    bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2872    if (bitmap == NULL) {
2873        return;
2874    }
2875
2876    if (disabled) {
2877        bdrv_disable_dirty_bitmap(bitmap);
2878    }
2879
2880    bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2881}
2882
2883void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2884                                   Error **errp)
2885{
2886    BlockDriverState *bs;
2887    BdrvDirtyBitmap *bitmap;
2888    Error *local_err = NULL;
2889
2890    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2891    if (!bitmap || !bs) {
2892        return;
2893    }
2894
2895    if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2896        error_setg(errp,
2897                   "Bitmap '%s' is currently in use by another operation and"
2898                   " cannot be removed", name);
2899        return;
2900    }
2901
2902    if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2903        bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2904        if (local_err != NULL) {
2905            error_propagate(errp, local_err);
2906            return;
2907        }
2908    }
2909
2910    bdrv_release_dirty_bitmap(bs, bitmap);
2911}
2912
2913/**
2914 * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2915 * immediately after a full backup operation.
2916 */
2917void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2918                                  Error **errp)
2919{
2920    BdrvDirtyBitmap *bitmap;
2921    BlockDriverState *bs;
2922
2923    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2924    if (!bitmap || !bs) {
2925        return;
2926    }
2927
2928    if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2929        error_setg(errp,
2930                   "Bitmap '%s' is currently in use by another operation"
2931                   " and cannot be cleared", name);
2932        return;
2933    } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2934        error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2935        return;
2936    }
2937
2938    bdrv_clear_dirty_bitmap(bitmap, NULL);
2939}
2940
2941void qmp_x_block_dirty_bitmap_enable(const char *node, const char *name,
2942                                   Error **errp)
2943{
2944    BlockDriverState *bs;
2945    BdrvDirtyBitmap *bitmap;
2946
2947    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2948    if (!bitmap) {
2949        return;
2950    }
2951
2952    if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2953        error_setg(errp,
2954                   "Bitmap '%s' is currently in use by another operation"
2955                   " and cannot be enabled", name);
2956        return;
2957    }
2958
2959    bdrv_enable_dirty_bitmap(bitmap);
2960}
2961
2962void qmp_x_block_dirty_bitmap_disable(const char *node, const char *name,
2963                                    Error **errp)
2964{
2965    BlockDriverState *bs;
2966    BdrvDirtyBitmap *bitmap;
2967
2968    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2969    if (!bitmap) {
2970        return;
2971    }
2972
2973    if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2974        error_setg(errp,
2975                   "Bitmap '%s' is currently in use by another operation"
2976                   " and cannot be disabled", name);
2977        return;
2978    }
2979
2980    bdrv_disable_dirty_bitmap(bitmap);
2981}
2982
2983void qmp_x_block_dirty_bitmap_merge(const char *node, const char *dst_name,
2984                                    const char *src_name, Error **errp)
2985{
2986    BlockDriverState *bs;
2987    BdrvDirtyBitmap *dst, *src;
2988
2989    dst = block_dirty_bitmap_lookup(node, dst_name, &bs, errp);
2990    if (!dst) {
2991        return;
2992    }
2993
2994    src = bdrv_find_dirty_bitmap(bs, src_name);
2995    if (!src) {
2996        error_setg(errp, "Dirty bitmap '%s' not found", src_name);
2997        return;
2998    }
2999
3000    bdrv_merge_dirty_bitmap(dst, src, NULL, errp);
3001}
3002
3003BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3004                                                              const char *name,
3005                                                              Error **errp)
3006{
3007    BdrvDirtyBitmap *bitmap;
3008    BlockDriverState *bs;
3009    BlockDirtyBitmapSha256 *ret = NULL;
3010    char *sha256;
3011
3012    bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3013    if (!bitmap || !bs) {
3014        return NULL;
3015    }
3016
3017    sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3018    if (sha256 == NULL) {
3019        return NULL;
3020    }
3021
3022    ret = g_new(BlockDirtyBitmapSha256, 1);
3023    ret->sha256 = sha256;
3024
3025    return ret;
3026}
3027
3028void hmp_drive_del(Monitor *mon, const QDict *qdict)
3029{
3030    const char *id = qdict_get_str(qdict, "id");
3031    BlockBackend *blk;
3032    BlockDriverState *bs;
3033    AioContext *aio_context;
3034    Error *local_err = NULL;
3035
3036    bs = bdrv_find_node(id);
3037    if (bs) {
3038        qmp_blockdev_del(id, &local_err);
3039        if (local_err) {
3040            error_report_err(local_err);
3041        }
3042        return;
3043    }
3044
3045    blk = blk_by_name(id);
3046    if (!blk) {
3047        error_report("Device '%s' not found", id);
3048        return;
3049    }
3050
3051    if (!blk_legacy_dinfo(blk)) {
3052        error_report("Deleting device added with blockdev-add"
3053                     " is not supported");
3054        return;
3055    }
3056
3057    aio_context = blk_get_aio_context(blk);
3058    aio_context_acquire(aio_context);
3059
3060    bs = blk_bs(blk);
3061    if (bs) {
3062        if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3063            error_report_err(local_err);
3064            aio_context_release(aio_context);
3065            return;
3066        }
3067
3068        blk_remove_bs(blk);
3069    }
3070
3071    /* Make the BlockBackend and the attached BlockDriverState anonymous */
3072    monitor_remove_blk(blk);
3073
3074    /* If this BlockBackend has a device attached to it, its refcount will be
3075     * decremented when the device is removed; otherwise we have to do so here.
3076     */
3077    if (blk_get_attached_dev(blk)) {
3078        /* Further I/O must not pause the guest */
3079        blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3080                         BLOCKDEV_ON_ERROR_REPORT);
3081    } else {
3082        blk_unref(blk);
3083    }
3084
3085    aio_context_release(aio_context);
3086}
3087
3088void qmp_block_resize(bool has_device, const char *device,
3089                      bool has_node_name, const char *node_name,
3090                      int64_t size, Error **errp)
3091{
3092    Error *local_err = NULL;
3093    BlockBackend *blk = NULL;
3094    BlockDriverState *bs;
3095    AioContext *aio_context;
3096    int ret;
3097
3098    bs = bdrv_lookup_bs(has_device ? device : NULL,
3099                        has_node_name ? node_name : NULL,
3100                        &local_err);
3101    if (local_err) {
3102        error_propagate(errp, local_err);
3103        return;
3104    }
3105
3106    aio_context = bdrv_get_aio_context(bs);
3107    aio_context_acquire(aio_context);
3108
3109    if (!bdrv_is_first_non_filter(bs)) {
3110        error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3111        goto out;
3112    }
3113
3114    if (size < 0) {
3115        error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3116        goto out;
3117    }
3118
3119    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3120        error_setg(errp, QERR_DEVICE_IN_USE, device);
3121        goto out;
3122    }
3123
3124    blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3125    ret = blk_insert_bs(blk, bs, errp);
3126    if (ret < 0) {
3127        goto out;
3128    }
3129
3130    bdrv_drained_begin(bs);
3131    ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3132    bdrv_drained_end(bs);
3133
3134out:
3135    blk_unref(blk);
3136    aio_context_release(aio_context);
3137}
3138
3139void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3140                      bool has_base, const char *base,
3141                      bool has_base_node, const char *base_node,
3142                      bool has_backing_file, const char *backing_file,
3143                      bool has_speed, int64_t speed,
3144                      bool has_on_error, BlockdevOnError on_error,
3145                      bool has_auto_finalize, bool auto_finalize,
3146                      bool has_auto_dismiss, bool auto_dismiss,
3147                      Error **errp)
3148{
3149    BlockDriverState *bs, *iter;
3150    BlockDriverState *base_bs = NULL;
3151    AioContext *aio_context;
3152    Error *local_err = NULL;
3153    const char *base_name = NULL;
3154    int job_flags = JOB_DEFAULT;
3155
3156    if (!has_on_error) {
3157        on_error = BLOCKDEV_ON_ERROR_REPORT;
3158    }
3159
3160    bs = bdrv_lookup_bs(device, device, errp);
3161    if (!bs) {
3162        return;
3163    }
3164
3165    aio_context = bdrv_get_aio_context(bs);
3166    aio_context_acquire(aio_context);
3167
3168    if (has_base && has_base_node) {
3169        error_setg(errp, "'base' and 'base-node' cannot be specified "
3170                   "at the same time");
3171        goto out;
3172    }
3173
3174    if (has_base) {
3175        base_bs = bdrv_find_backing_image(bs, base);
3176        if (base_bs == NULL) {
3177            error_setg(errp, QERR_BASE_NOT_FOUND, base);
3178            goto out;
3179        }
3180        assert(bdrv_get_aio_context(base_bs) == aio_context);
3181        base_name = base;
3182    }
3183
3184    if (has_base_node) {
3185        base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3186        if (!base_bs) {
3187            goto out;
3188        }
3189        if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3190            error_setg(errp, "Node '%s' is not a backing image of '%s'",
3191                       base_node, device);
3192            goto out;
3193        }
3194        assert(bdrv_get_aio_context(base_bs) == aio_context);
3195        base_name = base_bs->filename;
3196    }
3197
3198    /* Check for op blockers in the whole chain between bs and base */
3199    for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3200        if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3201            goto out;
3202        }
3203    }
3204
3205    /* if we are streaming the entire chain, the result will have no backing
3206     * file, and specifying one is therefore an error */
3207    if (base_bs == NULL && has_backing_file) {
3208        error_setg(errp, "backing file specified, but streaming the "
3209                         "entire chain");
3210        goto out;
3211    }
3212
3213    /* backing_file string overrides base bs filename */
3214    base_name = has_backing_file ? backing_file : base_name;
3215
3216    if (has_auto_finalize && !auto_finalize) {
3217        job_flags |= JOB_MANUAL_FINALIZE;
3218    }
3219    if (has_auto_dismiss && !auto_dismiss) {
3220        job_flags |= JOB_MANUAL_DISMISS;
3221    }
3222
3223    stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3224                 job_flags, has_speed ? speed : 0, on_error, &local_err);
3225    if (local_err) {
3226        error_propagate(errp, local_err);
3227        goto out;
3228    }
3229
3230    trace_qmp_block_stream(bs, bs->job);
3231
3232out:
3233    aio_context_release(aio_context);
3234}
3235
3236void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3237                      bool has_base_node, const char *base_node,
3238                      bool has_base, const char *base,
3239                      bool has_top_node, const char *top_node,
3240                      bool has_top, const char *top,
3241                      bool has_backing_file, const char *backing_file,
3242                      bool has_speed, int64_t speed,
3243                      bool has_filter_node_name, const char *filter_node_name,
3244                      bool has_auto_finalize, bool auto_finalize,
3245                      bool has_auto_dismiss, bool auto_dismiss,
3246                      Error **errp)
3247{
3248    BlockDriverState *bs;
3249    BlockDriverState *iter;
3250    BlockDriverState *base_bs, *top_bs;
3251    AioContext *aio_context;
3252    Error *local_err = NULL;
3253    /* This will be part of the QMP command, if/when the
3254     * BlockdevOnError change for blkmirror makes it in
3255     */
3256    BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3257    int job_flags = JOB_DEFAULT;
3258
3259    if (!has_speed) {
3260        speed = 0;
3261    }
3262    if (!has_filter_node_name) {
3263        filter_node_name = NULL;
3264    }
3265    if (has_auto_finalize && !auto_finalize) {
3266        job_flags |= JOB_MANUAL_FINALIZE;
3267    }
3268    if (has_auto_dismiss && !auto_dismiss) {
3269        job_flags |= JOB_MANUAL_DISMISS;
3270    }
3271
3272    /* Important Note:
3273     *  libvirt relies on the DeviceNotFound error class in order to probe for
3274     *  live commit feature versions; for this to work, we must make sure to
3275     *  perform the device lookup before any generic errors that may occur in a
3276     *  scenario in which all optional arguments are omitted. */
3277    bs = qmp_get_root_bs(device, &local_err);
3278    if (!bs) {
3279        bs = bdrv_lookup_bs(device, device, NULL);
3280        if (!bs) {
3281            error_free(local_err);
3282            error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3283                      "Device '%s' not found", device);
3284        } else {
3285            error_propagate(errp, local_err);
3286        }
3287        return;
3288    }
3289
3290    aio_context = bdrv_get_aio_context(bs);
3291    aio_context_acquire(aio_context);
3292
3293    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3294        goto out;
3295    }
3296
3297    /* default top_bs is the active layer */
3298    top_bs = bs;
3299
3300    if (has_top_node && has_top) {
3301        error_setg(errp, "'top-node' and 'top' are mutually exclusive");
3302        goto out;
3303    } else if (has_top_node) {
3304        top_bs = bdrv_lookup_bs(NULL, top_node, errp);
3305        if (top_bs == NULL) {
3306            goto out;
3307        }
3308        if (!bdrv_chain_contains(bs, top_bs)) {
3309            error_setg(errp, "'%s' is not in this backing file chain",
3310                       top_node);
3311            goto out;
3312        }
3313    } else if (has_top && top) {
3314        if (strcmp(bs->filename, top) != 0) {
3315            top_bs = bdrv_find_backing_image(bs, top);
3316        }
3317    }
3318
3319    if (top_bs == NULL) {
3320        error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3321        goto out;
3322    }
3323
3324    assert(bdrv_get_aio_context(top_bs) == aio_context);
3325
3326    if (has_base_node && has_base) {
3327        error_setg(errp, "'base-node' and 'base' are mutually exclusive");
3328        goto out;
3329    } else if (has_base_node) {
3330        base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3331        if (base_bs == NULL) {
3332            goto out;
3333        }
3334        if (!bdrv_chain_contains(top_bs, base_bs)) {
3335            error_setg(errp, "'%s' is not in this backing file chain",
3336                       base_node);
3337            goto out;
3338        }
3339    } else if (has_base && base) {
3340        base_bs = bdrv_find_backing_image(top_bs, base);
3341    } else {
3342        base_bs = bdrv_find_base(top_bs);
3343    }
3344
3345    if (base_bs == NULL) {
3346        error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3347        goto out;
3348    }
3349
3350    assert(bdrv_get_aio_context(base_bs) == aio_context);
3351
3352    for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3353        if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3354            goto out;
3355        }
3356    }
3357
3358    /* Do not allow attempts to commit an image into itself */
3359    if (top_bs == base_bs) {
3360        error_setg(errp, "cannot commit an image into itself");
3361        goto out;
3362    }
3363
3364    if (top_bs == bs) {
3365        if (has_backing_file) {
3366            error_setg(errp, "'backing-file' specified,"
3367                             " but 'top' is the active layer");
3368            goto out;
3369        }
3370        commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3371                            job_flags, speed, on_error,
3372                            filter_node_name, NULL, NULL, false, &local_err);
3373    } else {
3374        BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3375        if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3376            goto out;
3377        }
3378        commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
3379                     speed, on_error, has_backing_file ? backing_file : NULL,
3380                     filter_node_name, &local_err);
3381    }
3382    if (local_err != NULL) {
3383        error_propagate(errp, local_err);
3384        goto out;
3385    }
3386
3387out:
3388    aio_context_release(aio_context);
3389}
3390
3391static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
3392                                 Error **errp)
3393{
3394    BlockDriverState *bs;
3395    BlockDriverState *target_bs;
3396    BlockDriverState *source = NULL;
3397    BlockJob *job = NULL;
3398    BdrvDirtyBitmap *bmap = NULL;
3399    AioContext *aio_context;
3400    QDict *options = NULL;
3401    Error *local_err = NULL;
3402    int flags, job_flags = JOB_DEFAULT;
3403    int64_t size;
3404    bool set_backing_hd = false;
3405
3406    if (!backup->has_speed) {
3407        backup->speed = 0;
3408    }
3409    if (!backup->has_on_source_error) {
3410        backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3411    }
3412    if (!backup->has_on_target_error) {
3413        backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3414    }
3415    if (!backup->has_mode) {
3416        backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3417    }
3418    if (!backup->has_job_id) {
3419        backup->job_id = NULL;
3420    }
3421    if (!backup->has_auto_finalize) {
3422        backup->auto_finalize = true;
3423    }
3424    if (!backup->has_auto_dismiss) {
3425        backup->auto_dismiss = true;
3426    }
3427    if (!backup->has_compress) {
3428        backup->compress = false;
3429    }
3430
3431    bs = qmp_get_root_bs(backup->device, errp);
3432    if (!bs) {
3433        return NULL;
3434    }
3435
3436    aio_context = bdrv_get_aio_context(bs);
3437    aio_context_acquire(aio_context);
3438
3439    if (!backup->has_format) {
3440        backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3441                         NULL : (char*) bs->drv->format_name;
3442    }
3443
3444    /* Early check to avoid creating target */
3445    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3446        goto out;
3447    }
3448
3449    flags = bs->open_flags | BDRV_O_RDWR;
3450
3451    /* See if we have a backing HD we can use to create our new image
3452     * on top of. */
3453    if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3454        source = backing_bs(bs);
3455        if (!source) {
3456            backup->sync = MIRROR_SYNC_MODE_FULL;
3457        }
3458    }
3459    if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3460        source = bs;
3461        flags |= BDRV_O_NO_BACKING;
3462        set_backing_hd = true;
3463    }
3464
3465    size = bdrv_getlength(bs);
3466    if (size < 0) {
3467        error_setg_errno(errp, -size, "bdrv_getlength failed");
3468        goto out;
3469    }
3470
3471    if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3472        assert(backup->format);
3473        if (source) {
3474            bdrv_img_create(backup->target, backup->format, source->filename,
3475                            source->drv->format_name, NULL,
3476                            size, flags, false, &local_err);
3477        } else {
3478            bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3479                            size, flags, false, &local_err);
3480        }
3481    }
3482
3483    if (local_err) {
3484        error_propagate(errp, local_err);
3485        goto out;
3486    }
3487
3488    if (backup->format) {
3489        if (!options) {
3490            options = qdict_new();
3491        }
3492        qdict_put_str(options, "driver", backup->format);
3493    }
3494
3495    target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3496    if (!target_bs) {
3497        goto out;
3498    }
3499
3500    bdrv_set_aio_context(target_bs, aio_context);
3501
3502    if (set_backing_hd) {
3503        bdrv_set_backing_hd(target_bs, source, &local_err);
3504        if (local_err) {
3505            bdrv_unref(target_bs);
3506            goto out;
3507        }
3508    }
3509
3510    if (backup->has_bitmap) {
3511        bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3512        if (!bmap) {
3513            error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3514            bdrv_unref(target_bs);
3515            goto out;
3516        }
3517        if (bdrv_dirty_bitmap_user_locked(bmap)) {
3518            error_setg(errp,
3519                       "Bitmap '%s' is currently in use by another operation"
3520                       " and cannot be used for backup", backup->bitmap);
3521            goto out;
3522        }
3523    }
3524    if (!backup->auto_finalize) {
3525        job_flags |= JOB_MANUAL_FINALIZE;
3526    }
3527    if (!backup->auto_dismiss) {
3528        job_flags |= JOB_MANUAL_DISMISS;
3529    }
3530
3531    job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3532                            backup->sync, bmap, backup->compress,
3533                            backup->on_source_error, backup->on_target_error,
3534                            job_flags, NULL, NULL, txn, &local_err);
3535    bdrv_unref(target_bs);
3536    if (local_err != NULL) {
3537        error_propagate(errp, local_err);
3538        goto out;
3539    }
3540
3541out:
3542    aio_context_release(aio_context);
3543    return job;
3544}
3545
3546void qmp_drive_backup(DriveBackup *arg, Error **errp)
3547{
3548
3549    BlockJob *job;
3550    job = do_drive_backup(arg, NULL, errp);
3551    if (job) {
3552        job_start(&job->job);
3553    }
3554}
3555
3556BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3557{
3558    return bdrv_named_nodes_list(errp);
3559}
3560
3561BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
3562                             Error **errp)
3563{
3564    BlockDriverState *bs;
3565    BlockDriverState *target_bs;
3566    Error *local_err = NULL;
3567    BdrvDirtyBitmap *bmap = NULL;
3568    AioContext *aio_context;
3569    BlockJob *job = NULL;
3570    int job_flags = JOB_DEFAULT;
3571
3572    if (!backup->has_speed) {
3573        backup->speed = 0;
3574    }
3575    if (!backup->has_on_source_error) {
3576        backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3577    }
3578    if (!backup->has_on_target_error) {
3579        backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3580    }
3581    if (!backup->has_job_id) {
3582        backup->job_id = NULL;
3583    }
3584    if (!backup->has_auto_finalize) {
3585        backup->auto_finalize = true;
3586    }
3587    if (!backup->has_auto_dismiss) {
3588        backup->auto_dismiss = true;
3589    }
3590    if (!backup->has_compress) {
3591        backup->compress = false;
3592    }
3593
3594    bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3595    if (!bs) {
3596        return NULL;
3597    }
3598
3599    aio_context = bdrv_get_aio_context(bs);
3600    aio_context_acquire(aio_context);
3601
3602    target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3603    if (!target_bs) {
3604        goto out;
3605    }
3606
3607    if (bdrv_get_aio_context(target_bs) != aio_context) {
3608        if (!bdrv_has_blk(target_bs)) {
3609            /* The target BDS is not attached, we can safely move it to another
3610             * AioContext. */
3611            bdrv_set_aio_context(target_bs, aio_context);
3612        } else {
3613            error_setg(errp, "Target is attached to a different thread from "
3614                             "source.");
3615            goto out;
3616        }
3617    }
3618
3619    if (backup->has_bitmap) {
3620        bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3621        if (!bmap) {
3622            error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3623            goto out;
3624        }
3625        if (bdrv_dirty_bitmap_user_locked(bmap)) {
3626            error_setg(errp,
3627                       "Bitmap '%s' is currently in use by another operation"
3628                       " and cannot be used for backup", backup->bitmap);
3629            goto out;
3630        }
3631    }
3632
3633    if (!backup->auto_finalize) {
3634        job_flags |= JOB_MANUAL_FINALIZE;
3635    }
3636    if (!backup->auto_dismiss) {
3637        job_flags |= JOB_MANUAL_DISMISS;
3638    }
3639    job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3640                            backup->sync, bmap, backup->compress,
3641                            backup->on_source_error, backup->on_target_error,
3642                            job_flags, NULL, NULL, txn, &local_err);
3643    if (local_err != NULL) {
3644        error_propagate(errp, local_err);
3645    }
3646out:
3647    aio_context_release(aio_context);
3648    return job;
3649}
3650
3651void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3652{
3653    BlockJob *job;
3654    job = do_blockdev_backup(arg, NULL, errp);
3655    if (job) {
3656        job_start(&job->job);
3657    }
3658}
3659
3660/* Parameter check and block job starting for drive mirroring.
3661 * Caller should hold @device and @target's aio context (must be the same).
3662 **/
3663static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3664                                   BlockDriverState *target,
3665                                   bool has_replaces, const char *replaces,
3666                                   enum MirrorSyncMode sync,
3667                                   BlockMirrorBackingMode backing_mode,
3668                                   bool has_speed, int64_t speed,
3669                                   bool has_granularity, uint32_t granularity,
3670                                   bool has_buf_size, int64_t buf_size,
3671                                   bool has_on_source_error,
3672                                   BlockdevOnError on_source_error,
3673                                   bool has_on_target_error,
3674                                   BlockdevOnError on_target_error,
3675                                   bool has_unmap, bool unmap,
3676                                   bool has_filter_node_name,
3677                                   const char *filter_node_name,
3678                                   bool has_copy_mode, MirrorCopyMode copy_mode,
3679                                   bool has_auto_finalize, bool auto_finalize,
3680                                   bool has_auto_dismiss, bool auto_dismiss,
3681                                   Error **errp)
3682{
3683    int job_flags = JOB_DEFAULT;
3684
3685    if (!has_speed) {
3686        speed = 0;
3687    }
3688    if (!has_on_source_error) {
3689        on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3690    }
3691    if (!has_on_target_error) {
3692        on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3693    }
3694    if (!has_granularity) {
3695        granularity = 0;
3696    }
3697    if (!has_buf_size) {
3698        buf_size = 0;
3699    }
3700    if (!has_unmap) {
3701        unmap = true;
3702    }
3703    if (!has_filter_node_name) {
3704        filter_node_name = NULL;
3705    }
3706    if (!has_copy_mode) {
3707        copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3708    }
3709    if (has_auto_finalize && !auto_finalize) {
3710        job_flags |= JOB_MANUAL_FINALIZE;
3711    }
3712    if (has_auto_dismiss && !auto_dismiss) {
3713        job_flags |= JOB_MANUAL_DISMISS;
3714    }
3715
3716    if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3717        error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3718                   "a value in range [512B, 64MB]");
3719        return;
3720    }
3721    if (granularity & (granularity - 1)) {
3722        error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3723                   "power of 2");
3724        return;
3725    }
3726
3727    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3728        return;
3729    }
3730    if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3731        return;
3732    }
3733
3734    if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3735        sync = MIRROR_SYNC_MODE_FULL;
3736    }
3737
3738    /* pass the node name to replace to mirror start since it's loose coupling
3739     * and will allow to check whether the node still exist at mirror completion
3740     */
3741    mirror_start(job_id, bs, target,
3742                 has_replaces ? replaces : NULL, job_flags,
3743                 speed, granularity, buf_size, sync, backing_mode,
3744                 on_source_error, on_target_error, unmap, filter_node_name,
3745                 copy_mode, errp);
3746}
3747
3748void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3749{
3750    BlockDriverState *bs;
3751    BlockDriverState *source, *target_bs;
3752    AioContext *aio_context;
3753    BlockMirrorBackingMode backing_mode;
3754    Error *local_err = NULL;
3755    QDict *options = NULL;
3756    int flags;
3757    int64_t size;
3758    const char *format = arg->format;
3759
3760    bs = qmp_get_root_bs(arg->device, errp);
3761    if (!bs) {
3762        return;
3763    }
3764
3765    /* Early check to avoid creating target */
3766    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3767        return;
3768    }
3769
3770    aio_context = bdrv_get_aio_context(bs);
3771    aio_context_acquire(aio_context);
3772
3773    if (!arg->has_mode) {
3774        arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3775    }
3776
3777    if (!arg->has_format) {
3778        format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3779                  ? NULL : bs->drv->format_name);
3780    }
3781
3782    flags = bs->open_flags | BDRV_O_RDWR;
3783    source = backing_bs(bs);
3784    if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3785        arg->sync = MIRROR_SYNC_MODE_FULL;
3786    }
3787    if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3788        source = bs;
3789    }
3790
3791    size = bdrv_getlength(bs);
3792    if (size < 0) {
3793        error_setg_errno(errp, -size, "bdrv_getlength failed");
3794        goto out;
3795    }
3796
3797    if (arg->has_replaces) {
3798        BlockDriverState *to_replace_bs;
3799        AioContext *replace_aio_context;
3800        int64_t replace_size;
3801
3802        if (!arg->has_node_name) {
3803            error_setg(errp, "a node-name must be provided when replacing a"
3804                             " named node of the graph");
3805            goto out;
3806        }
3807
3808        to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3809
3810        if (!to_replace_bs) {
3811            error_propagate(errp, local_err);
3812            goto out;
3813        }
3814
3815        replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3816        aio_context_acquire(replace_aio_context);
3817        replace_size = bdrv_getlength(to_replace_bs);
3818        aio_context_release(replace_aio_context);
3819
3820        if (size != replace_size) {
3821            error_setg(errp, "cannot replace image with a mirror image of "
3822                             "different size");
3823            goto out;
3824        }
3825    }
3826
3827    if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3828        backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3829    } else {
3830        backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3831    }
3832
3833    /* Don't open backing image in create() */
3834    flags |= BDRV_O_NO_BACKING;
3835
3836    if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3837        && arg->mode != NEW_IMAGE_MODE_EXISTING)
3838    {
3839        /* create new image w/o backing file */
3840        assert(format);
3841        bdrv_img_create(arg->target, format,
3842                        NULL, NULL, NULL, size, flags, false, &local_err);
3843    } else {
3844        switch (arg->mode) {
3845        case NEW_IMAGE_MODE_EXISTING:
3846            break;
3847        case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3848            /* create new image with backing file */
3849            bdrv_img_create(arg->target, format,
3850                            source->filename,
3851                            source->drv->format_name,
3852                            NULL, size, flags, false, &local_err);
3853            break;
3854        default:
3855            abort();
3856        }
3857    }
3858
3859    if (local_err) {
3860        error_propagate(errp, local_err);
3861        goto out;
3862    }
3863
3864    options = qdict_new();
3865    if (arg->has_node_name) {
3866        qdict_put_str(options, "node-name", arg->node_name);
3867    }
3868    if (format) {
3869        qdict_put_str(options, "driver", format);
3870    }
3871
3872    /* Mirroring takes care of copy-on-write using the source's backing
3873     * file.
3874     */
3875    target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3876    if (!target_bs) {
3877        goto out;
3878    }
3879
3880    bdrv_set_aio_context(target_bs, aio_context);
3881
3882    blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3883                           arg->has_replaces, arg->replaces, arg->sync,
3884                           backing_mode, arg->has_speed, arg->speed,
3885                           arg->has_granularity, arg->granularity,
3886                           arg->has_buf_size, arg->buf_size,
3887                           arg->has_on_source_error, arg->on_source_error,
3888                           arg->has_on_target_error, arg->on_target_error,
3889                           arg->has_unmap, arg->unmap,
3890                           false, NULL,
3891                           arg->has_copy_mode, arg->copy_mode,
3892                           arg->has_auto_finalize, arg->auto_finalize,
3893                           arg->has_auto_dismiss, arg->auto_dismiss,
3894                           &local_err);
3895    bdrv_unref(target_bs);
3896    error_propagate(errp, local_err);
3897out:
3898    aio_context_release(aio_context);
3899}
3900
3901void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3902                         const char *device, const char *target,
3903                         bool has_replaces, const char *replaces,
3904                         MirrorSyncMode sync,
3905                         bool has_speed, int64_t speed,
3906                         bool has_granularity, uint32_t granularity,
3907                         bool has_buf_size, int64_t buf_size,
3908                         bool has_on_source_error,
3909                         BlockdevOnError on_source_error,
3910                         bool has_on_target_error,
3911                         BlockdevOnError on_target_error,
3912                         bool has_filter_node_name,
3913                         const char *filter_node_name,
3914                         bool has_copy_mode, MirrorCopyMode copy_mode,
3915                         bool has_auto_finalize, bool auto_finalize,
3916                         bool has_auto_dismiss, bool auto_dismiss,
3917                         Error **errp)
3918{
3919    BlockDriverState *bs;
3920    BlockDriverState *target_bs;
3921    AioContext *aio_context;
3922    BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3923    Error *local_err = NULL;
3924
3925    bs = qmp_get_root_bs(device, errp);
3926    if (!bs) {
3927        return;
3928    }
3929
3930    target_bs = bdrv_lookup_bs(target, target, errp);
3931    if (!target_bs) {
3932        return;
3933    }
3934
3935    aio_context = bdrv_get_aio_context(bs);
3936    aio_context_acquire(aio_context);
3937
3938    bdrv_set_aio_context(target_bs, aio_context);
3939
3940    blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3941                           has_replaces, replaces, sync, backing_mode,
3942                           has_speed, speed,
3943                           has_granularity, granularity,
3944                           has_buf_size, buf_size,
3945                           has_on_source_error, on_source_error,
3946                           has_on_target_error, on_target_error,
3947                           true, true,
3948                           has_filter_node_name, filter_node_name,
3949                           has_copy_mode, copy_mode,
3950                           has_auto_finalize, auto_finalize,
3951                           has_auto_dismiss, auto_dismiss,
3952                           &local_err);
3953    error_propagate(errp, local_err);
3954
3955    aio_context_release(aio_context);
3956}
3957
3958/* Get a block job using its ID and acquire its AioContext */
3959static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3960                                Error **errp)
3961{
3962    BlockJob *job;
3963
3964    assert(id != NULL);
3965
3966    *aio_context = NULL;
3967
3968    job = block_job_get(id);
3969
3970    if (!job) {
3971        error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3972                  "Block job '%s' not found", id);
3973        return NULL;
3974    }
3975
3976    *aio_context = blk_get_aio_context(job->blk);
3977    aio_context_acquire(*aio_context);
3978
3979    return job;
3980}
3981
3982void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3983{
3984    AioContext *aio_context;
3985    BlockJob *job = find_block_job(device, &aio_context, errp);
3986
3987    if (!job) {
3988        return;
3989    }
3990
3991    block_job_set_speed(job, speed, errp);
3992    aio_context_release(aio_context);
3993}
3994
3995void qmp_block_job_cancel(const char *device,
3996                          bool has_force, bool force, Error **errp)
3997{
3998    AioContext *aio_context;
3999    BlockJob *job = find_block_job(device, &aio_context, errp);
4000
4001    if (!job) {
4002        return;
4003    }
4004
4005    if (!has_force) {
4006        force = false;
4007    }
4008
4009    if (job_user_paused(&job->job) && !force) {
4010        error_setg(errp, "The block job for device '%s' is currently paused",
4011                   device);
4012        goto out;
4013    }
4014
4015    trace_qmp_block_job_cancel(job);
4016    job_user_cancel(&job->job, force, errp);
4017out:
4018    aio_context_release(aio_context);
4019}
4020
4021void qmp_block_job_pause(const char *device, Error **errp)
4022{
4023    AioContext *aio_context;
4024    BlockJob *job = find_block_job(device, &aio_context, errp);
4025
4026    if (!job) {
4027        return;
4028    }
4029
4030    trace_qmp_block_job_pause(job);
4031    job_user_pause(&job->job, errp);
4032    aio_context_release(aio_context);
4033}
4034
4035void qmp_block_job_resume(const char *device, Error **errp)
4036{
4037    AioContext *aio_context;
4038    BlockJob *job = find_block_job(device, &aio_context, errp);
4039
4040    if (!job) {
4041        return;
4042    }
4043
4044    trace_qmp_block_job_resume(job);
4045    job_user_resume(&job->job, errp);
4046    aio_context_release(aio_context);
4047}
4048
4049void qmp_block_job_complete(const char *device, Error **errp)
4050{
4051    AioContext *aio_context;
4052    BlockJob *job = find_block_job(device, &aio_context, errp);
4053
4054    if (!job) {
4055        return;
4056    }
4057
4058    trace_qmp_block_job_complete(job);
4059    job_complete(&job->job, errp);
4060    aio_context_release(aio_context);
4061}
4062
4063void qmp_block_job_finalize(const char *id, Error **errp)
4064{
4065    AioContext *aio_context;
4066    BlockJob *job = find_block_job(id, &aio_context, errp);
4067
4068    if (!job) {
4069        return;
4070    }
4071
4072    trace_qmp_block_job_finalize(job);
4073    job_finalize(&job->job, errp);
4074    aio_context_release(aio_context);
4075}
4076
4077void qmp_block_job_dismiss(const char *id, Error **errp)
4078{
4079    AioContext *aio_context;
4080    BlockJob *bjob = find_block_job(id, &aio_context, errp);
4081    Job *job;
4082
4083    if (!bjob) {
4084        return;
4085    }
4086
4087    trace_qmp_block_job_dismiss(bjob);
4088    job = &bjob->job;
4089    job_dismiss(&job, errp);
4090    aio_context_release(aio_context);
4091}
4092
4093void qmp_change_backing_file(const char *device,
4094                             const char *image_node_name,
4095                             const char *backing_file,
4096                             Error **errp)
4097{
4098    BlockDriverState *bs = NULL;
4099    AioContext *aio_context;
4100    BlockDriverState *image_bs = NULL;
4101    Error *local_err = NULL;
4102    bool ro;
4103    int open_flags;
4104    int ret;
4105
4106    bs = qmp_get_root_bs(device, errp);
4107    if (!bs) {
4108        return;
4109    }
4110
4111    aio_context = bdrv_get_aio_context(bs);
4112    aio_context_acquire(aio_context);
4113
4114    image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4115    if (local_err) {
4116        error_propagate(errp, local_err);
4117        goto out;
4118    }
4119
4120    if (!image_bs) {
4121        error_setg(errp, "image file not found");
4122        goto out;
4123    }
4124
4125    if (bdrv_find_base(image_bs) == image_bs) {
4126        error_setg(errp, "not allowing backing file change on an image "
4127                         "without a backing file");
4128        goto out;
4129    }
4130
4131    /* even though we are not necessarily operating on bs, we need it to
4132     * determine if block ops are currently prohibited on the chain */
4133    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4134        goto out;
4135    }
4136
4137    /* final sanity check */
4138    if (!bdrv_chain_contains(bs, image_bs)) {
4139        error_setg(errp, "'%s' and image file are not in the same chain",
4140                   device);
4141        goto out;
4142    }
4143
4144    /* if not r/w, reopen to make r/w */
4145    open_flags = image_bs->open_flags;
4146    ro = bdrv_is_read_only(image_bs);
4147
4148    if (ro) {
4149        bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
4150        if (local_err) {
4151            error_propagate(errp, local_err);
4152            goto out;
4153        }
4154    }
4155
4156    ret = bdrv_change_backing_file(image_bs, backing_file,
4157                               image_bs->drv ? image_bs->drv->format_name : "");
4158
4159    if (ret < 0) {
4160        error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4161                         backing_file);
4162        /* don't exit here, so we can try to restore open flags if
4163         * appropriate */
4164    }
4165
4166    if (ro) {
4167        bdrv_reopen(image_bs, open_flags, &local_err);
4168        error_propagate(errp, local_err);
4169    }
4170
4171out:
4172    aio_context_release(aio_context);
4173}
4174
4175void hmp_drive_add_node(Monitor *mon, const char *optstr)
4176{
4177    QemuOpts *opts;
4178    QDict *qdict;
4179    Error *local_err = NULL;
4180
4181    opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4182    if (!opts) {
4183        return;
4184    }
4185
4186    qdict = qemu_opts_to_qdict(opts, NULL);
4187
4188    if (!qdict_get_try_str(qdict, "node-name")) {
4189        qobject_unref(qdict);
4190        error_report("'node-name' needs to be specified");
4191        goto out;
4192    }
4193
4194    BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4195    if (!bs) {
4196        error_report_err(local_err);
4197        goto out;
4198    }
4199
4200    QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4201
4202out:
4203    qemu_opts_del(opts);
4204}
4205
4206void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4207{
4208    BlockDriverState *bs;
4209    QObject *obj;
4210    Visitor *v = qobject_output_visitor_new(&obj);
4211    QDict *qdict;
4212    Error *local_err = NULL;
4213
4214    visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4215    if (local_err) {
4216        error_propagate(errp, local_err);
4217        goto fail;
4218    }
4219
4220    visit_complete(v, &obj);
4221    qdict = qobject_to(QDict, obj);
4222
4223    qdict_flatten(qdict);
4224
4225    if (!qdict_get_try_str(qdict, "node-name")) {
4226        error_setg(errp, "'node-name' must be specified for the root node");
4227        goto fail;
4228    }
4229
4230    bs = bds_tree_init(qdict, errp);
4231    if (!bs) {
4232        goto fail;
4233    }
4234
4235    QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4236
4237fail:
4238    visit_free(v);
4239}
4240
4241void qmp_blockdev_del(const char *node_name, Error **errp)
4242{
4243    AioContext *aio_context;
4244    BlockDriverState *bs;
4245
4246    bs = bdrv_find_node(node_name);
4247    if (!bs) {
4248        error_setg(errp, "Cannot find node %s", node_name);
4249        return;
4250    }
4251    if (bdrv_has_blk(bs)) {
4252        error_setg(errp, "Node %s is in use", node_name);
4253        return;
4254    }
4255    aio_context = bdrv_get_aio_context(bs);
4256    aio_context_acquire(aio_context);
4257
4258    if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4259        goto out;
4260    }
4261
4262    if (!bs->monitor_list.tqe_prev) {
4263        error_setg(errp, "Node %s is not owned by the monitor",
4264                   bs->node_name);
4265        goto out;
4266    }
4267
4268    if (bs->refcnt > 1) {
4269        error_setg(errp, "Block device %s is in use",
4270                   bdrv_get_device_or_node_name(bs));
4271        goto out;
4272    }
4273
4274    QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4275    bdrv_unref(bs);
4276
4277out:
4278    aio_context_release(aio_context);
4279}
4280
4281static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4282                                  const char *child_name)
4283{
4284    BdrvChild *child;
4285
4286    QLIST_FOREACH(child, &parent_bs->children, next) {
4287        if (strcmp(child->name, child_name) == 0) {
4288            return child;
4289        }
4290    }
4291
4292    return NULL;
4293}
4294
4295void qmp_x_blockdev_change(const char *parent, bool has_child,
4296                           const char *child, bool has_node,
4297                           const char *node, Error **errp)
4298{
4299    BlockDriverState *parent_bs, *new_bs = NULL;
4300    BdrvChild *p_child;
4301
4302    parent_bs = bdrv_lookup_bs(parent, parent, errp);
4303    if (!parent_bs) {
4304        return;
4305    }
4306
4307    if (has_child == has_node) {
4308        if (has_child) {
4309            error_setg(errp, "The parameters child and node are in conflict");
4310        } else {
4311            error_setg(errp, "Either child or node must be specified");
4312        }
4313        return;
4314    }
4315
4316    if (has_child) {
4317        p_child = bdrv_find_child(parent_bs, child);
4318        if (!p_child) {
4319            error_setg(errp, "Node '%s' does not have child '%s'",
4320                       parent, child);
4321            return;
4322        }
4323        bdrv_del_child(parent_bs, p_child, errp);
4324    }
4325
4326    if (has_node) {
4327        new_bs = bdrv_find_node(node);
4328        if (!new_bs) {
4329            error_setg(errp, "Node '%s' not found", node);
4330            return;
4331        }
4332        bdrv_add_child(parent_bs, new_bs, errp);
4333    }
4334}
4335
4336BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4337{
4338    BlockJobInfoList *head = NULL, **p_next = &head;
4339    BlockJob *job;
4340
4341    for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4342        BlockJobInfoList *elem;
4343        AioContext *aio_context;
4344
4345        if (block_job_is_internal(job)) {
4346            continue;
4347        }
4348        elem = g_new0(BlockJobInfoList, 1);
4349        aio_context = blk_get_aio_context(job->blk);
4350        aio_context_acquire(aio_context);
4351        elem->value = block_job_query(job, errp);
4352        aio_context_release(aio_context);
4353        if (!elem->value) {
4354            g_free(elem);
4355            qapi_free_BlockJobInfoList(head);
4356            return NULL;
4357        }
4358        *p_next = elem;
4359        p_next = &elem->next;
4360    }
4361
4362    return head;
4363}
4364
4365void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4366                                 bool has_force, bool force, Error **errp)
4367{
4368    AioContext *old_context;
4369    AioContext *new_context;
4370    BlockDriverState *bs;
4371
4372    bs = bdrv_find_node(node_name);
4373    if (!bs) {
4374        error_setg(errp, "Cannot find node %s", node_name);
4375        return;
4376    }
4377
4378    /* Protects against accidents. */
4379    if (!(has_force && force) && bdrv_has_blk(bs)) {
4380        error_setg(errp, "Node %s is associated with a BlockBackend and could "
4381                         "be in use (use force=true to override this check)",
4382                         node_name);
4383        return;
4384    }
4385
4386    if (iothread->type == QTYPE_QSTRING) {
4387        IOThread *obj = iothread_by_id(iothread->u.s);
4388        if (!obj) {
4389            error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4390            return;
4391        }
4392
4393        new_context = iothread_get_aio_context(obj);
4394    } else {
4395        new_context = qemu_get_aio_context();
4396    }
4397
4398    old_context = bdrv_get_aio_context(bs);
4399    aio_context_acquire(old_context);
4400
4401    bdrv_set_aio_context(bs, new_context);
4402
4403    aio_context_release(old_context);
4404}
4405
4406void qmp_x_block_latency_histogram_set(
4407    const char *device,
4408    bool has_boundaries, uint64List *boundaries,
4409    bool has_boundaries_read, uint64List *boundaries_read,
4410    bool has_boundaries_write, uint64List *boundaries_write,
4411    bool has_boundaries_flush, uint64List *boundaries_flush,
4412    Error **errp)
4413{
4414    BlockBackend *blk = blk_by_name(device);
4415    BlockAcctStats *stats;
4416    int ret;
4417
4418    if (!blk) {
4419        error_setg(errp, "Device '%s' not found", device);
4420        return;
4421    }
4422    stats = blk_get_stats(blk);
4423
4424    if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4425        !has_boundaries_flush)
4426    {
4427        block_latency_histograms_clear(stats);
4428        return;
4429    }
4430
4431    if (has_boundaries || has_boundaries_read) {
4432        ret = block_latency_histogram_set(
4433            stats, BLOCK_ACCT_READ,
4434            has_boundaries_read ? boundaries_read : boundaries);
4435        if (ret) {
4436            error_setg(errp, "Device '%s' set read boundaries fail", device);
4437            return;
4438        }
4439    }
4440
4441    if (has_boundaries || has_boundaries_write) {
4442        ret = block_latency_histogram_set(
4443            stats, BLOCK_ACCT_WRITE,
4444            has_boundaries_write ? boundaries_write : boundaries);
4445        if (ret) {
4446            error_setg(errp, "Device '%s' set write boundaries fail", device);
4447            return;
4448        }
4449    }
4450
4451    if (has_boundaries || has_boundaries_flush) {
4452        ret = block_latency_histogram_set(
4453            stats, BLOCK_ACCT_FLUSH,
4454            has_boundaries_flush ? boundaries_flush : boundaries);
4455        if (ret) {
4456            error_setg(errp, "Device '%s' set flush boundaries fail", device);
4457            return;
4458        }
4459    }
4460}
4461
4462QemuOptsList qemu_common_drive_opts = {
4463    .name = "drive",
4464    .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4465    .desc = {
4466        {
4467            .name = "snapshot",
4468            .type = QEMU_OPT_BOOL,
4469            .help = "enable/disable snapshot mode",
4470        },{
4471            .name = "aio",
4472            .type = QEMU_OPT_STRING,
4473            .help = "host AIO implementation (threads, native)",
4474        },{
4475            .name = BDRV_OPT_CACHE_WB,
4476            .type = QEMU_OPT_BOOL,
4477            .help = "Enable writeback mode",
4478        },{
4479            .name = "format",
4480            .type = QEMU_OPT_STRING,
4481            .help = "disk format (raw, qcow2, ...)",
4482        },{
4483            .name = "rerror",
4484            .type = QEMU_OPT_STRING,
4485            .help = "read error action",
4486        },{
4487            .name = "werror",
4488            .type = QEMU_OPT_STRING,
4489            .help = "write error action",
4490        },{
4491            .name = BDRV_OPT_READ_ONLY,
4492            .type = QEMU_OPT_BOOL,
4493            .help = "open drive file as read-only",
4494        },
4495
4496        THROTTLE_OPTS,
4497
4498        {
4499            .name = "throttling.group",
4500            .type = QEMU_OPT_STRING,
4501            .help = "name of the block throttling group",
4502        },{
4503            .name = "copy-on-read",
4504            .type = QEMU_OPT_BOOL,
4505            .help = "copy read data from backing file into image file",
4506        },{
4507            .name = "detect-zeroes",
4508            .type = QEMU_OPT_STRING,
4509            .help = "try to optimize zero writes (off, on, unmap)",
4510        },{
4511            .name = "stats-account-invalid",
4512            .type = QEMU_OPT_BOOL,
4513            .help = "whether to account for invalid I/O operations "
4514                    "in the statistics",
4515        },{
4516            .name = "stats-account-failed",
4517            .type = QEMU_OPT_BOOL,
4518            .help = "whether to account for failed I/O operations "
4519                    "in the statistics",
4520        },
4521        { /* end of list */ }
4522    },
4523};
4524
4525QemuOptsList qemu_drive_opts = {
4526    .name = "drive",
4527    .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4528    .desc = {
4529        /*
4530         * no elements => accept any params
4531         * validation will happen later
4532         */
4533        { /* end of list */ }
4534    },
4535};
4536