blockdev.c

   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
  10 #include "blockdev.h"
  11 #include "hw/block-common.h"
  12 #include "monitor.h"
  13 #include "qerror.h"
  14 #include "qemu-option.h"
  15 #include "qemu-config.h"
  16 #include "qemu-objects.h"
  17 #include "sysemu.h"
  18 #include "block_int.h"
  19 #include "qmp-commands.h"
  20 #include "trace.h"
  21 #include "arch_init.h"
  22
  23 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
  24
  25 static const char *const if_name[IF_COUNT] = {
  26     [IF_NONE] = "none",
  27     [IF_IDE] = "ide",
  28     [IF_SCSI] = "scsi",
  29     [IF_FLOPPY] = "floppy",
  30     [IF_PFLASH] = "pflash",
  31     [IF_MTD] = "mtd",
  32     [IF_SD] = "sd",
  33     [IF_VIRTIO] = "virtio",
  34     [IF_XEN] = "xen",
  35 };
  36
  37 static const int if_max_devs[IF_COUNT] = {
  38     /*
  39      * Do not change these numbers!  They govern how drive option
  40      * index maps to unit and bus.  That mapping is ABI.
  41      *
  42      * All controllers used to imlement if=T drives need to support
  43      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
  44      * Otherwise, some index values map to "impossible" bus, unit
  45      * values.
  46      *
  47      * For instance, if you change [IF_SCSI] to 255, -drive
  48      * if=scsi,index=12 no longer means bus=1,unit=5, but
  49      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
  50      * the drive can't be set up.  Regression.
  51      */
  52     [IF_IDE] = 2,
  53     [IF_SCSI] = 7,
  54 };
  55
  56 /*
  57  * We automatically delete the drive when a device using it gets
  58  * unplugged.  Questionable feature, but we can't just drop it.
  59  * Device models call blockdev_mark_auto_del() to schedule the
  60  * automatic deletion, and generic qdev code calls blockdev_auto_del()
  61  * when deletion is actually safe.
  62  */
  63 void blockdev_mark_auto_del(BlockDriverState *bs)
  64 {
  65     DriveInfo *dinfo = drive_get_by_blockdev(bs);
  66
  67     if (bs->job) {
  68         block_job_cancel(bs->job);
  69     }
  70     if (dinfo) {
  71         dinfo->auto_del = 1;
  72     }
  73 }
  74
  75 void blockdev_auto_del(BlockDriverState *bs)
  76 {
  77     DriveInfo *dinfo = drive_get_by_blockdev(bs);
  78
  79     if (dinfo && dinfo->auto_del) {
  80         drive_put_ref(dinfo);
  81     }
  82 }
  83
  84 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
  85 {
  86     int max_devs = if_max_devs[type];
  87     return max_devs ? index / max_devs : 0;
  88 }
  89
  90 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
  91 {
  92     int max_devs = if_max_devs[type];
  93     return max_devs ? index % max_devs : index;
  94 }
  95
  96 QemuOpts *drive_def(const char *optstr)
  97 {
  98     return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
  99 }
 100
 101 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
 102                     const char *optstr)
 103 {
 104     QemuOpts *opts;
 105     char buf[32];
 106
 107     opts = drive_def(optstr);
 108     if (!opts) {
 109         return NULL;
 110     }
 111     if (type != IF_DEFAULT) {
 112         qemu_opt_set(opts, "if", if_name[type]);
 113     }
 114     if (index >= 0) {
 115         snprintf(buf, sizeof(buf), "%d", index);
 116         qemu_opt_set(opts, "index", buf);
 117     }
 118     if (file)
 119         qemu_opt_set(opts, "file", file);
 120     return opts;
 121 }
 122
 123 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
 124 {
 125     DriveInfo *dinfo;
 126
 127     /* seek interface, bus and unit */
 128
 129     QTAILQ_FOREACH(dinfo, &drives, next) {
 130         if (dinfo->type == type &&
 131             dinfo->bus == bus &&
 132             dinfo->unit == unit)
 133             return dinfo;
 134     }
 135
 136     return NULL;
 137 }
 138
 139 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
 140 {
 141     return drive_get(type,
 142                      drive_index_to_bus_id(type, index),
 143                      drive_index_to_unit_id(type, index));
 144 }
 145
 146 int drive_get_max_bus(BlockInterfaceType type)
 147 {
 148     int max_bus;
 149     DriveInfo *dinfo;
 150
 151     max_bus = -1;
 152     QTAILQ_FOREACH(dinfo, &drives, next) {
 153         if(dinfo->type == type &&
 154            dinfo->bus > max_bus)
 155             max_bus = dinfo->bus;
 156     }
 157     return max_bus;
 158 }
 159
 160 /* Get a block device.  This should only be used for single-drive devices
 161    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
 162    appropriate bus.  */
 163 DriveInfo *drive_get_next(BlockInterfaceType type)
 164 {
 165     static int next_block_unit[IF_COUNT];
 166
 167     return drive_get(type, 0, next_block_unit[type]++);
 168 }
 169
 170 DriveInfo *drive_get_by_blockdev(BlockDriverState *bs)
 171 {
 172     DriveInfo *dinfo;
 173
 174     QTAILQ_FOREACH(dinfo, &drives, next) {
 175         if (dinfo->bdrv == bs) {
 176             return dinfo;
 177         }
 178     }
 179     return NULL;
 180 }
 181
 182 static void bdrv_format_print(void *opaque, const char *name)
 183 {
 184     error_printf(" %s", name);
 185 }
 186
 187 static void drive_uninit(DriveInfo *dinfo)
 188 {
 189     qemu_opts_del(dinfo->opts);
 190     bdrv_delete(dinfo->bdrv);
 191     g_free(dinfo->id);
 192     QTAILQ_REMOVE(&drives, dinfo, next);
 193     g_free(dinfo);
 194 }
 195
 196 void drive_put_ref(DriveInfo *dinfo)
 197 {
 198     assert(dinfo->refcount);
 199     if (--dinfo->refcount == 0) {
 200         drive_uninit(dinfo);
 201     }
 202 }
 203
 204 void drive_get_ref(DriveInfo *dinfo)
 205 {
 206     dinfo->refcount++;
 207 }
 208
 209 typedef struct {
 210     QEMUBH *bh;
 211     DriveInfo *dinfo;
 212 } DrivePutRefBH;
 213
 214 static void drive_put_ref_bh(void *opaque)
 215 {
 216     DrivePutRefBH *s = opaque;
 217
 218     drive_put_ref(s->dinfo);
 219     qemu_bh_delete(s->bh);
 220     g_free(s);
 221 }
 222
 223 /*
 224  * Release a drive reference in a BH
 225  *
 226  * It is not possible to use drive_put_ref() from a callback function when the
 227  * callers still need the drive.  In such cases we schedule a BH to release the
 228  * reference.
 229  */
 230 static void drive_put_ref_bh_schedule(DriveInfo *dinfo)
 231 {
 232     DrivePutRefBH *s;
 233
 234     s = g_new(DrivePutRefBH, 1);
 235     s->bh = qemu_bh_new(drive_put_ref_bh, s);
 236     s->dinfo = dinfo;
 237     qemu_bh_schedule(s->bh);
 238 }
 239
 240 static int parse_block_error_action(const char *buf, int is_read)
 241 {
 242     if (!strcmp(buf, "ignore")) {
 243         return BLOCK_ERR_IGNORE;
 244     } else if (!is_read && !strcmp(buf, "enospc")) {
 245         return BLOCK_ERR_STOP_ENOSPC;
 246     } else if (!strcmp(buf, "stop")) {
 247         return BLOCK_ERR_STOP_ANY;
 248     } else if (!strcmp(buf, "report")) {
 249         return BLOCK_ERR_REPORT;
 250     } else {
 251         error_report("'%s' invalid %s error action",
 252                      buf, is_read ? "read" : "write");
 253         return -1;
 254     }
 255 }
 256
 257 static bool do_check_io_limits(BlockIOLimit *io_limits)
 258 {
 259     bool bps_flag;
 260     bool iops_flag;
 261
 262     assert(io_limits);
 263
 264     bps_flag  = (io_limits->bps[BLOCK_IO_LIMIT_TOTAL] != 0)
 265                  && ((io_limits->bps[BLOCK_IO_LIMIT_READ] != 0)
 266                  || (io_limits->bps[BLOCK_IO_LIMIT_WRITE] != 0));
 267     iops_flag = (io_limits->iops[BLOCK_IO_LIMIT_TOTAL] != 0)
 268                  && ((io_limits->iops[BLOCK_IO_LIMIT_READ] != 0)
 269                  || (io_limits->iops[BLOCK_IO_LIMIT_WRITE] != 0));
 270     if (bps_flag || iops_flag) {
 271         return false;
 272     }
 273
 274     return true;
 275 }
 276
 277 DriveInfo *drive_init(QemuOpts *opts, int default_to_scsi)
 278 {
 279     const char *buf;
 280     const char *file = NULL;
 281     const char *serial;
 282     const char *mediastr = "";
 283     BlockInterfaceType type;
 284     enum { MEDIA_DISK, MEDIA_CDROM } media;
 285     int bus_id, unit_id;
 286     int cyls, heads, secs, translation;
 287     BlockDriver *drv = NULL;
 288     int max_devs;
 289     int index;
 290     int ro = 0;
 291     int bdrv_flags = 0;
 292     int on_read_error, on_write_error;
 293     const char *devaddr;
 294     DriveInfo *dinfo;
 295     BlockIOLimit io_limits;
 296     int snapshot = 0;
 297     bool copy_on_read;
 298     int ret;
 299
 300     translation = BIOS_ATA_TRANSLATION_AUTO;
 301     media = MEDIA_DISK;
 302
 303     /* extract parameters */
 304     bus_id  = qemu_opt_get_number(opts, "bus", 0);
 305     unit_id = qemu_opt_get_number(opts, "unit", -1);
 306     index   = qemu_opt_get_number(opts, "index", -1);
 307
 308     cyls  = qemu_opt_get_number(opts, "cyls", 0);
 309     heads = qemu_opt_get_number(opts, "heads", 0);
 310     secs  = qemu_opt_get_number(opts, "secs", 0);
 311
 312     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
 313     ro = qemu_opt_get_bool(opts, "readonly", 0);
 314     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
 315
 316     file = qemu_opt_get(opts, "file");
 317     serial = qemu_opt_get(opts, "serial");
 318
 319     if ((buf = qemu_opt_get(opts, "if")) != NULL) {
 320         for (type = 0; type < IF_COUNT && strcmp(buf, if_name[type]); type++)
 321             ;
 322         if (type == IF_COUNT) {
 323             error_report("unsupported bus type '%s'", buf);
 324             return NULL;
 325         }
 326     } else {
 327         type = default_to_scsi ? IF_SCSI : IF_IDE;
 328     }
 329
 330     max_devs = if_max_devs[type];
 331
 332     if (cyls || heads || secs) {
 333         if (cyls < 1) {
 334             error_report("invalid physical cyls number");
 335             return NULL;
 336         }
 337         if (heads < 1) {
 338             error_report("invalid physical heads number");
 339             return NULL;
 340         }
 341         if (secs < 1) {
 342             error_report("invalid physical secs number");
 343             return NULL;
 344         }
 345     }
 346
 347     if ((buf = qemu_opt_get(opts, "trans")) != NULL) {
 348         if (!cyls) {
 349             error_report("'%s' trans must be used with cyls, heads and secs",
 350                          buf);
 351             return NULL;
 352         }
 353         if (!strcmp(buf, "none"))
 354             translation = BIOS_ATA_TRANSLATION_NONE;
 355         else if (!strcmp(buf, "lba"))
 356             translation = BIOS_ATA_TRANSLATION_LBA;
 357         else if (!strcmp(buf, "auto"))
 358             translation = BIOS_ATA_TRANSLATION_AUTO;
 359         else {
 360             error_report("'%s' invalid translation type", buf);
 361             return NULL;
 362         }
 363     }
 364
 365     if ((buf = qemu_opt_get(opts, "media")) != NULL) {
 366         if (!strcmp(buf, "disk")) {
 367             media = MEDIA_DISK;
 368         } else if (!strcmp(buf, "cdrom")) {
 369             if (cyls || secs || heads) {
 370                 error_report("CHS can't be set with media=%s", buf);
 371                 return NULL;
 372             }
 373             media = MEDIA_CDROM;
 374         } else {
 375             error_report("'%s' invalid media", buf);
 376             return NULL;
 377         }
 378     }
 379
 380     bdrv_flags |= BDRV_O_CACHE_WB;
 381     if ((buf = qemu_opt_get(opts, "cache")) != NULL) {
 382         if (bdrv_parse_cache_flags(buf, &bdrv_flags) != 0) {
 383             error_report("invalid cache option");
 384             return NULL;
 385         }
 386     }
 387
 388 #ifdef CONFIG_LINUX_AIO
 389     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
 390         if (!strcmp(buf, "native")) {
 391             bdrv_flags |= BDRV_O_NATIVE_AIO;
 392         } else if (!strcmp(buf, "threads")) {
 393             /* this is the default */
 394         } else {
 395            error_report("invalid aio option");
 396            return NULL;
 397         }
 398     }
 399 #endif
 400
 401     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
 402         if (is_help_option(buf)) {
 403             error_printf("Supported formats:");
 404             bdrv_iterate_format(bdrv_format_print, NULL);
 405             error_printf("\n");
 406             return NULL;
 407         }
 408         drv = bdrv_find_whitelisted_format(buf);
 409         if (!drv) {
 410             error_report("'%s' invalid format", buf);
 411             return NULL;
 412         }
 413     }
 414
 415     /* disk I/O throttling */
 416     io_limits.bps[BLOCK_IO_LIMIT_TOTAL]  =
 417                            qemu_opt_get_number(opts, "bps", 0);
 418     io_limits.bps[BLOCK_IO_LIMIT_READ]   =
 419                            qemu_opt_get_number(opts, "bps_rd", 0);
 420     io_limits.bps[BLOCK_IO_LIMIT_WRITE]  =
 421                            qemu_opt_get_number(opts, "bps_wr", 0);
 422     io_limits.iops[BLOCK_IO_LIMIT_TOTAL] =
 423                            qemu_opt_get_number(opts, "iops", 0);
 424     io_limits.iops[BLOCK_IO_LIMIT_READ]  =
 425                            qemu_opt_get_number(opts, "iops_rd", 0);
 426     io_limits.iops[BLOCK_IO_LIMIT_WRITE] =
 427                            qemu_opt_get_number(opts, "iops_wr", 0);
 428
 429     if (!do_check_io_limits(&io_limits)) {
 430         error_report("bps(iops) and bps_rd/bps_wr(iops_rd/iops_wr) "
 431                      "cannot be used at the same time");
 432         return NULL;
 433     }
 434
 435     if (qemu_opt_get(opts, "boot") != NULL) {
 436         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
 437                 "ignored. Future versions will reject this parameter. Please "
 438                 "update your scripts.\n");
 439     }
 440
 441     on_write_error = BLOCK_ERR_STOP_ENOSPC;
 442     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
 443         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) {
 444             error_report("werror is not supported by this bus type");
 445             return NULL;
 446         }
 447
 448         on_write_error = parse_block_error_action(buf, 0);
 449         if (on_write_error < 0) {
 450             return NULL;
 451         }
 452     }
 453
 454     on_read_error = BLOCK_ERR_REPORT;
 455     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
 456         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) {
 457             error_report("rerror is not supported by this bus type");
 458             return NULL;
 459         }
 460
 461         on_read_error = parse_block_error_action(buf, 1);
 462         if (on_read_error < 0) {
 463             return NULL;
 464         }
 465     }
 466
 467     if ((devaddr = qemu_opt_get(opts, "addr")) != NULL) {
 468         if (type != IF_VIRTIO) {
 469             error_report("addr is not supported by this bus type");
 470             return NULL;
 471         }
 472     }
 473
 474     /* compute bus and unit according index */
 475
 476     if (index != -1) {
 477         if (bus_id != 0 || unit_id != -1) {
 478             error_report("index cannot be used with bus and unit");
 479             return NULL;
 480         }
 481         bus_id = drive_index_to_bus_id(type, index);
 482         unit_id = drive_index_to_unit_id(type, index);
 483     }
 484
 485     /* if user doesn't specify a unit_id,
 486      * try to find the first free
 487      */
 488
 489     if (unit_id == -1) {
 490        unit_id = 0;
 491        while (drive_get(type, bus_id, unit_id) != NULL) {
 492            unit_id++;
 493            if (max_devs && unit_id >= max_devs) {
 494                unit_id -= max_devs;
 495                bus_id++;
 496            }
 497        }
 498     }
 499
 500     /* check unit id */
 501
 502     if (max_devs && unit_id >= max_devs) {
 503         error_report("unit %d too big (max is %d)",
 504                      unit_id, max_devs - 1);
 505         return NULL;
 506     }
 507
 508     /*
 509      * catch multiple definitions
 510      */
 511
 512     if (drive_get(type, bus_id, unit_id) != NULL) {
 513         error_report("drive with bus=%d, unit=%d (index=%d) exists",
 514                      bus_id, unit_id, index);
 515         return NULL;
 516     }
 517
 518     /* init */
 519
 520     dinfo = g_malloc0(sizeof(*dinfo));
 521     if ((buf = qemu_opts_id(opts)) != NULL) {
 522         dinfo->id = g_strdup(buf);
 523     } else {
 524         /* no id supplied -> create one */
 525         dinfo->id = g_malloc0(32);
 526         if (type == IF_IDE || type == IF_SCSI)
 527             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
 528         if (max_devs)
 529             snprintf(dinfo->id, 32, "%s%i%s%i",
 530                      if_name[type], bus_id, mediastr, unit_id);
 531         else
 532             snprintf(dinfo->id, 32, "%s%s%i",
 533                      if_name[type], mediastr, unit_id);
 534     }
 535     dinfo->bdrv = bdrv_new(dinfo->id);
 536     dinfo->bdrv->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
 537     dinfo->bdrv->read_only = ro;
 538     dinfo->devaddr = devaddr;
 539     dinfo->type = type;
 540     dinfo->bus = bus_id;
 541     dinfo->unit = unit_id;
 542     dinfo->cyls = cyls;
 543     dinfo->heads = heads;
 544     dinfo->secs = secs;
 545     dinfo->trans = translation;
 546     dinfo->opts = opts;
 547     dinfo->refcount = 1;
 548     dinfo->serial = serial;
 549     QTAILQ_INSERT_TAIL(&drives, dinfo, next);
 550
 551     bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
 552
 553     /* disk I/O throttling */
 554     bdrv_set_io_limits(dinfo->bdrv, &io_limits);
 555
 556     switch(type) {
 557     case IF_IDE:
 558     case IF_SCSI:
 559     case IF_XEN:
 560     case IF_NONE:
 561         dinfo->media_cd = media == MEDIA_CDROM;
 562         break;
 563     case IF_SD:
 564     case IF_FLOPPY:
 565     case IF_PFLASH:
 566     case IF_MTD:
 567         break;
 568     case IF_VIRTIO:
 569         /* add virtio block device */
 570         opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0, NULL);
 571         if (arch_type == QEMU_ARCH_S390X) {
 572             qemu_opt_set(opts, "driver", "virtio-blk-s390");
 573         } else {
 574             qemu_opt_set(opts, "driver", "virtio-blk-pci");
 575         }
 576         qemu_opt_set(opts, "drive", dinfo->id);
 577         if (devaddr)
 578             qemu_opt_set(opts, "addr", devaddr);
 579         break;
 580     default:
 581         abort();
 582     }
 583     if (!file || !*file) {
 584         return dinfo;
 585     }
 586     if (snapshot) {
 587         /* always use cache=unsafe with snapshot */
 588         bdrv_flags &= ~BDRV_O_CACHE_MASK;
 589         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
 590     }
 591
 592     if (copy_on_read) {
 593         bdrv_flags |= BDRV_O_COPY_ON_READ;
 594     }
 595
 596     if (runstate_check(RUN_STATE_INMIGRATE)) {
 597         bdrv_flags |= BDRV_O_INCOMING;
 598     }
 599
 600     if (media == MEDIA_CDROM) {
 601         /* CDROM is fine for any interface, don't check.  */
 602         ro = 1;
 603     } else if (ro == 1) {
 604         if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY &&
 605             type != IF_NONE && type != IF_PFLASH) {
 606             error_report("readonly not supported by this bus type");
 607             goto err;
 608         }
 609     }
 610
 611     bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
 612
 613     if (ro && copy_on_read) {
 614         error_report("warning: disabling copy_on_read on readonly drive");
 615     }
 616
 617     ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv);
 618     if (ret < 0) {
 619         error_report("could not open disk image %s: %s",
 620                      file, strerror(-ret));
 621         goto err;
 622     }
 623
 624     if (bdrv_key_required(dinfo->bdrv))
 625         autostart = 0;
 626     return dinfo;
 627
 628 err:
 629     bdrv_delete(dinfo->bdrv);
 630     g_free(dinfo->id);
 631     QTAILQ_REMOVE(&drives, dinfo, next);
 632     g_free(dinfo);
 633     return NULL;
 634 }
 635
 636 void do_commit(Monitor *mon, const QDict *qdict)
 637 {
 638     const char *device = qdict_get_str(qdict, "device");
 639     BlockDriverState *bs;
 640     int ret;
 641
 642     if (!strcmp(device, "all")) {
 643         ret = bdrv_commit_all();
 644         if (ret == -EBUSY) {
 645             qerror_report(QERR_DEVICE_IN_USE, device);
 646             return;
 647         }
 648     } else {
 649         bs = bdrv_find(device);
 650         if (!bs) {
 651             qerror_report(QERR_DEVICE_NOT_FOUND, device);
 652             return;
 653         }
 654         ret = bdrv_commit(bs);
 655         if (ret == -EBUSY) {
 656             qerror_report(QERR_DEVICE_IN_USE, device);
 657             return;
 658         }
 659     }
 660 }
 661
 662 static void blockdev_do_action(int kind, void *data, Error **errp)
 663 {
 664     BlockdevAction action;
 665     BlockdevActionList list;
 666
 667     action.kind = kind;
 668     action.data = data;
 669     list.value = &action;
 670     list.next = NULL;
 671     qmp_transaction(&list, errp);
 672 }
 673
 674 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
 675                                 bool has_format, const char *format,
 676                                 bool has_mode, enum NewImageMode mode,
 677                                 Error **errp)
 678 {
 679     BlockdevSnapshot snapshot = {
 680         .device = (char *) device,
 681         .snapshot_file = (char *) snapshot_file,
 682         .has_format = has_format,
 683         .format = (char *) format,
 684         .has_mode = has_mode,
 685         .mode = mode,
 686     };
 687     blockdev_do_action(BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC, &snapshot,
 688                        errp);
 689 }
 690
 691
 692 /* New and old BlockDriverState structs for group snapshots */
 693 typedef struct BlkTransactionStates {
 694     BlockDriverState *old_bs;
 695     BlockDriverState *new_bs;
 696     QSIMPLEQ_ENTRY(BlkTransactionStates) entry;
 697 } BlkTransactionStates;
 698
 699 /*
 700  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
 701  *  then we do not pivot any of the devices in the group, and abandon the
 702  *  snapshots
 703  */
 704 void qmp_transaction(BlockdevActionList *dev_list, Error **errp)
 705 {
 706     int ret = 0;
 707     BlockdevActionList *dev_entry = dev_list;
 708     BlkTransactionStates *states, *next;
 709
 710     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionStates) snap_bdrv_states;
 711     QSIMPLEQ_INIT(&snap_bdrv_states);
 712
 713     /* drain all i/o before any snapshots */
 714     bdrv_drain_all();
 715
 716     /* We don't do anything in this loop that commits us to the snapshot */
 717     while (NULL != dev_entry) {
 718         BlockdevAction *dev_info = NULL;
 719         BlockDriver *proto_drv;
 720         BlockDriver *drv;
 721         int flags;
 722         enum NewImageMode mode;
 723         const char *new_image_file;
 724         const char *device;
 725         const char *format = "qcow2";
 726
 727         dev_info = dev_entry->value;
 728         dev_entry = dev_entry->next;
 729
 730         states = g_malloc0(sizeof(BlkTransactionStates));
 731         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, states, entry);
 732
 733         switch (dev_info->kind) {
 734         case BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
 735             device = dev_info->blockdev_snapshot_sync->device;
 736             if (!dev_info->blockdev_snapshot_sync->has_mode) {
 737                 dev_info->blockdev_snapshot_sync->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
 738             }
 739             new_image_file = dev_info->blockdev_snapshot_sync->snapshot_file;
 740             if (dev_info->blockdev_snapshot_sync->has_format) {
 741                 format = dev_info->blockdev_snapshot_sync->format;
 742             }
 743             mode = dev_info->blockdev_snapshot_sync->mode;
 744             break;
 745         default:
 746             abort();
 747         }
 748
 749         drv = bdrv_find_format(format);
 750         if (!drv) {
 751             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 752             goto delete_and_fail;
 753         }
 754
 755         states->old_bs = bdrv_find(device);
 756         if (!states->old_bs) {
 757             error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 758             goto delete_and_fail;
 759         }
 760
 761         if (!bdrv_is_inserted(states->old_bs)) {
 762             error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
 763             goto delete_and_fail;
 764         }
 765
 766         if (bdrv_in_use(states->old_bs)) {
 767             error_set(errp, QERR_DEVICE_IN_USE, device);
 768             goto delete_and_fail;
 769         }
 770
 771         if (!bdrv_is_read_only(states->old_bs)) {
 772             if (bdrv_flush(states->old_bs)) {
 773                 error_set(errp, QERR_IO_ERROR);
 774                 goto delete_and_fail;
 775             }
 776         }
 777
 778         flags = states->old_bs->open_flags;
 779
 780         proto_drv = bdrv_find_protocol(new_image_file);
 781         if (!proto_drv) {
 782             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 783             goto delete_and_fail;
 784         }
 785
 786         /* create new image w/backing file */
 787         if (mode != NEW_IMAGE_MODE_EXISTING) {
 788             ret = bdrv_img_create(new_image_file, format,
 789                                   states->old_bs->filename,
 790                                   states->old_bs->drv->format_name,
 791                                   NULL, -1, flags);
 792             if (ret) {
 793                 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 794                 goto delete_and_fail;
 795             }
 796         }
 797
 798         /* We will manually add the backing_hd field to the bs later */
 799         states->new_bs = bdrv_new("");
 800         ret = bdrv_open(states->new_bs, new_image_file,
 801                         flags | BDRV_O_NO_BACKING, drv);
 802         if (ret != 0) {
 803             error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 804             goto delete_and_fail;
 805         }
 806     }
 807
 808
 809     /* Now we are going to do the actual pivot.  Everything up to this point
 810      * is reversible, but we are committed at this point */
 811     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 812         /* This removes our old bs from the bdrv_states, and adds the new bs */
 813         bdrv_append(states->new_bs, states->old_bs);
 814     }
 815
 816     /* success */
 817     goto exit;
 818
 819 delete_and_fail:
 820     /*
 821     * failure, and it is all-or-none; abandon each new bs, and keep using
 822     * the original bs for all images
 823     */
 824     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 825         if (states->new_bs) {
 826              bdrv_delete(states->new_bs);
 827         }
 828     }
 829 exit:
 830     QSIMPLEQ_FOREACH_SAFE(states, &snap_bdrv_states, entry, next) {
 831         g_free(states);
 832     }
 833     return;
 834 }
 835
 836
 837 static void eject_device(BlockDriverState *bs, int force, Error **errp)
 838 {
 839     if (bdrv_in_use(bs)) {
 840         error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 841         return;
 842     }
 843     if (!bdrv_dev_has_removable_media(bs)) {
 844         error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs));
 845         return;
 846     }
 847
 848     if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) {
 849         bdrv_dev_eject_request(bs, force);
 850         if (!force) {
 851             error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
 852             return;
 853         }
 854     }
 855
 856     bdrv_close(bs);
 857 }
 858
 859 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
 860 {
 861     BlockDriverState *bs;
 862
 863     bs = bdrv_find(device);
 864     if (!bs) {
 865         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 866         return;
 867     }
 868
 869     eject_device(bs, force, errp);
 870 }
 871
 872 void qmp_block_passwd(const char *device, const char *password, Error **errp)
 873 {
 874     BlockDriverState *bs;
 875     int err;
 876
 877     bs = bdrv_find(device);
 878     if (!bs) {
 879         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 880         return;
 881     }
 882
 883     err = bdrv_set_key(bs, password);
 884     if (err == -EINVAL) {
 885         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
 886         return;
 887     } else if (err < 0) {
 888         error_set(errp, QERR_INVALID_PASSWORD);
 889         return;
 890     }
 891 }
 892
 893 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
 894                                     int bdrv_flags, BlockDriver *drv,
 895                                     const char *password, Error **errp)
 896 {
 897     if (bdrv_open(bs, filename, bdrv_flags, drv) < 0) {
 898         error_set(errp, QERR_OPEN_FILE_FAILED, filename);
 899         return;
 900     }
 901
 902     if (bdrv_key_required(bs)) {
 903         if (password) {
 904             if (bdrv_set_key(bs, password) < 0) {
 905                 error_set(errp, QERR_INVALID_PASSWORD);
 906             }
 907         } else {
 908             error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs),
 909                       bdrv_get_encrypted_filename(bs));
 910         }
 911     } else if (password) {
 912         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
 913     }
 914 }
 915
 916 void qmp_change_blockdev(const char *device, const char *filename,
 917                          bool has_format, const char *format, Error **errp)
 918 {
 919     BlockDriverState *bs;
 920     BlockDriver *drv = NULL;
 921     int bdrv_flags;
 922     Error *err = NULL;
 923
 924     bs = bdrv_find(device);
 925     if (!bs) {
 926         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 927         return;
 928     }
 929
 930     if (format) {
 931         drv = bdrv_find_whitelisted_format(format);
 932         if (!drv) {
 933             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 934             return;
 935         }
 936     }
 937
 938     eject_device(bs, 0, &err);
 939     if (error_is_set(&err)) {
 940         error_propagate(errp, err);
 941         return;
 942     }
 943
 944     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
 945     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
 946
 947     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
 948 }
 949
 950 /* throttling disk I/O limits */
 951 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
 952                                int64_t bps_wr, int64_t iops, int64_t iops_rd,
 953                                int64_t iops_wr, Error **errp)
 954 {
 955     BlockIOLimit io_limits;
 956     BlockDriverState *bs;
 957
 958     bs = bdrv_find(device);
 959     if (!bs) {
 960         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 961         return;
 962     }
 963
 964     io_limits.bps[BLOCK_IO_LIMIT_TOTAL] = bps;
 965     io_limits.bps[BLOCK_IO_LIMIT_READ]  = bps_rd;
 966     io_limits.bps[BLOCK_IO_LIMIT_WRITE] = bps_wr;
 967     io_limits.iops[BLOCK_IO_LIMIT_TOTAL]= iops;
 968     io_limits.iops[BLOCK_IO_LIMIT_READ] = iops_rd;
 969     io_limits.iops[BLOCK_IO_LIMIT_WRITE]= iops_wr;
 970
 971     if (!do_check_io_limits(&io_limits)) {
 972         error_set(errp, QERR_INVALID_PARAMETER_COMBINATION);
 973         return;
 974     }
 975
 976     bs->io_limits = io_limits;
 977     bs->slice_time = BLOCK_IO_SLICE_TIME;
 978
 979     if (!bs->io_limits_enabled && bdrv_io_limits_enabled(bs)) {
 980         bdrv_io_limits_enable(bs);
 981     } else if (bs->io_limits_enabled && !bdrv_io_limits_enabled(bs)) {
 982         bdrv_io_limits_disable(bs);
 983     } else {
 984         if (bs->block_timer) {
 985             qemu_mod_timer(bs->block_timer, qemu_get_clock_ns(vm_clock));
 986         }
 987     }
 988 }
 989
 990 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
 991 {
 992     const char *id = qdict_get_str(qdict, "id");
 993     BlockDriverState *bs;
 994
 995     bs = bdrv_find(id);
 996     if (!bs) {
 997         qerror_report(QERR_DEVICE_NOT_FOUND, id);
 998         return -1;
 999     }
1000     if (bdrv_in_use(bs)) {
1001         qerror_report(QERR_DEVICE_IN_USE, id);
1002         return -1;
1003     }
1004
1005     /* quiesce block driver; prevent further io */
1006     bdrv_drain_all();
1007     bdrv_flush(bs);
1008     bdrv_close(bs);
1009
1010     /* if we have a device attached to this BlockDriverState
1011      * then we need to make the drive anonymous until the device
1012      * can be removed.  If this is a drive with no device backing
1013      * then we can just get rid of the block driver state right here.
1014      */
1015     if (bdrv_get_attached_dev(bs)) {
1016         bdrv_make_anon(bs);
1017     } else {
1018         drive_uninit(drive_get_by_blockdev(bs));
1019     }
1020
1021     return 0;
1022 }
1023
1024 void qmp_block_resize(const char *device, int64_t size, Error **errp)
1025 {
1026     BlockDriverState *bs;
1027
1028     bs = bdrv_find(device);
1029     if (!bs) {
1030         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1031         return;
1032     }
1033
1034     if (size < 0) {
1035         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
1036         return;
1037     }
1038
1039     switch (bdrv_truncate(bs, size)) {
1040     case 0:
1041         break;
1042     case -ENOMEDIUM:
1043         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1044         break;
1045     case -ENOTSUP:
1046         error_set(errp, QERR_UNSUPPORTED);
1047         break;
1048     case -EACCES:
1049         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1050         break;
1051     case -EBUSY:
1052         error_set(errp, QERR_DEVICE_IN_USE, device);
1053         break;
1054     default:
1055         error_set(errp, QERR_UNDEFINED_ERROR);
1056         break;
1057     }
1058 }
1059
1060 static QObject *qobject_from_block_job(BlockJob *job)
1061 {
1062     return qobject_from_jsonf("{ 'type': %s,"
1063                               "'device': %s,"
1064                               "'len': %" PRId64 ","
1065                               "'offset': %" PRId64 ","
1066                               "'speed': %" PRId64 " }",
1067                               job->job_type->job_type,
1068                               bdrv_get_device_name(job->bs),
1069                               job->len,
1070                               job->offset,
1071                               job->speed);
1072 }
1073
1074 static void block_stream_cb(void *opaque, int ret)
1075 {
1076     BlockDriverState *bs = opaque;
1077     QObject *obj;
1078
1079     trace_block_stream_cb(bs, bs->job, ret);
1080
1081     assert(bs->job);
1082     obj = qobject_from_block_job(bs->job);
1083     if (ret < 0) {
1084         QDict *dict = qobject_to_qdict(obj);
1085         qdict_put(dict, "error", qstring_from_str(strerror(-ret)));
1086     }
1087
1088     if (block_job_is_cancelled(bs->job)) {
1089         monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj);
1090     } else {
1091         monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj);
1092     }
1093     qobject_decref(obj);
1094
1095     drive_put_ref_bh_schedule(drive_get_by_blockdev(bs));
1096 }
1097
1098 void qmp_block_stream(const char *device, bool has_base,
1099                       const char *base, bool has_speed,
1100                       int64_t speed, Error **errp)
1101 {
1102     BlockDriverState *bs;
1103     BlockDriverState *base_bs = NULL;
1104     Error *local_err = NULL;
1105
1106     bs = bdrv_find(device);
1107     if (!bs) {
1108         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1109         return;
1110     }
1111
1112     if (base) {
1113         base_bs = bdrv_find_backing_image(bs, base);
1114         if (base_bs == NULL) {
1115             error_set(errp, QERR_BASE_NOT_FOUND, base);
1116             return;
1117         }
1118     }
1119
1120     stream_start(bs, base_bs, base, has_speed ? speed : 0,
1121                  block_stream_cb, bs, &local_err);
1122     if (error_is_set(&local_err)) {
1123         error_propagate(errp, local_err);
1124         return;
1125     }
1126
1127     /* Grab a reference so hotplug does not delete the BlockDriverState from
1128      * underneath us.
1129      */
1130     drive_get_ref(drive_get_by_blockdev(bs));
1131
1132     trace_qmp_block_stream(bs, bs->job);
1133 }
1134
1135 static BlockJob *find_block_job(const char *device)
1136 {
1137     BlockDriverState *bs;
1138
1139     bs = bdrv_find(device);
1140     if (!bs || !bs->job) {
1141         return NULL;
1142     }
1143     return bs->job;
1144 }
1145
1146 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
1147 {
1148     BlockJob *job = find_block_job(device);
1149
1150     if (!job) {
1151         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1152         return;
1153     }
1154
1155     block_job_set_speed(job, speed, errp);
1156 }
1157
1158 void qmp_block_job_cancel(const char *device, Error **errp)
1159 {
1160     BlockJob *job = find_block_job(device);
1161
1162     if (!job) {
1163         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1164         return;
1165     }
1166
1167     trace_qmp_block_job_cancel(job);
1168     block_job_cancel(job);
1169 }
1170
1171 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
1172 {
1173     BlockJobInfoList **prev = opaque;
1174     BlockJob *job = bs->job;
1175
1176     if (job) {
1177         BlockJobInfoList *elem;
1178         BlockJobInfo *info = g_new(BlockJobInfo, 1);
1179         *info = (BlockJobInfo){
1180             .type   = g_strdup(job->job_type->job_type),
1181             .device = g_strdup(bdrv_get_device_name(bs)),
1182             .len    = job->len,
1183             .offset = job->offset,
1184             .speed  = job->speed,
1185         };
1186
1187         elem = g_new0(BlockJobInfoList, 1);
1188         elem->value = info;
1189
1190         (*prev)->next = elem;
1191         *prev = elem;
1192     }
1193 }
1194
1195 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
1196 {
1197     /* Dummy is a fake list element for holding the head pointer */
1198     BlockJobInfoList dummy = {};
1199     BlockJobInfoList *prev = &dummy;
1200     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
1201     return dummy.next;
1202 }