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 "block.h"
  11 #include "blockdev.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 || (type == IF_IDE && cyls > 16383)) {
 334             error_report("invalid physical cyls number");
 335             return NULL;
 336         }
 337         if (heads < 1 || (type == IF_IDE && heads > 16)) {
 338             error_report("invalid physical heads number");
 339             return NULL;
 340         }
 341         if (secs < 1 || (type == IF_IDE && secs > 63)) {
 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     if ((buf = qemu_opt_get(opts, "cache")) != NULL) {
 381         if (bdrv_parse_cache_flags(buf, &bdrv_flags) != 0) {
 382             error_report("invalid cache option");
 383             return NULL;
 384         }
 385     }
 386
 387 #ifdef CONFIG_LINUX_AIO
 388     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
 389         if (!strcmp(buf, "native")) {
 390             bdrv_flags |= BDRV_O_NATIVE_AIO;
 391         } else if (!strcmp(buf, "threads")) {
 392             /* this is the default */
 393         } else {
 394            error_report("invalid aio option");
 395            return NULL;
 396         }
 397     }
 398 #endif
 399
 400     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
 401        if (strcmp(buf, "?") == 0) {
 402            error_printf("Supported formats:");
 403            bdrv_iterate_format(bdrv_format_print, NULL);
 404            error_printf("\n");
 405            return NULL;
 406         }
 407         drv = bdrv_find_whitelisted_format(buf);
 408         if (!drv) {
 409             error_report("'%s' invalid format", buf);
 410             return NULL;
 411         }
 412     }
 413
 414     /* disk I/O throttling */
 415     io_limits.bps[BLOCK_IO_LIMIT_TOTAL]  =
 416                            qemu_opt_get_number(opts, "bps", 0);
 417     io_limits.bps[BLOCK_IO_LIMIT_READ]   =
 418                            qemu_opt_get_number(opts, "bps_rd", 0);
 419     io_limits.bps[BLOCK_IO_LIMIT_WRITE]  =
 420                            qemu_opt_get_number(opts, "bps_wr", 0);
 421     io_limits.iops[BLOCK_IO_LIMIT_TOTAL] =
 422                            qemu_opt_get_number(opts, "iops", 0);
 423     io_limits.iops[BLOCK_IO_LIMIT_READ]  =
 424                            qemu_opt_get_number(opts, "iops_rd", 0);
 425     io_limits.iops[BLOCK_IO_LIMIT_WRITE] =
 426                            qemu_opt_get_number(opts, "iops_wr", 0);
 427
 428     if (!do_check_io_limits(&io_limits)) {
 429         error_report("bps(iops) and bps_rd/bps_wr(iops_rd/iops_wr) "
 430                      "cannot be used at the same time");
 431         return NULL;
 432     }
 433
 434     on_write_error = BLOCK_ERR_STOP_ENOSPC;
 435     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
 436         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) {
 437             error_report("werror is not supported by this bus type");
 438             return NULL;
 439         }
 440
 441         on_write_error = parse_block_error_action(buf, 0);
 442         if (on_write_error < 0) {
 443             return NULL;
 444         }
 445     }
 446
 447     on_read_error = BLOCK_ERR_REPORT;
 448     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
 449         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) {
 450             error_report("rerror is not supported by this bus type");
 451             return NULL;
 452         }
 453
 454         on_read_error = parse_block_error_action(buf, 1);
 455         if (on_read_error < 0) {
 456             return NULL;
 457         }
 458     }
 459
 460     if ((devaddr = qemu_opt_get(opts, "addr")) != NULL) {
 461         if (type != IF_VIRTIO) {
 462             error_report("addr is not supported by this bus type");
 463             return NULL;
 464         }
 465     }
 466
 467     /* compute bus and unit according index */
 468
 469     if (index != -1) {
 470         if (bus_id != 0 || unit_id != -1) {
 471             error_report("index cannot be used with bus and unit");
 472             return NULL;
 473         }
 474         bus_id = drive_index_to_bus_id(type, index);
 475         unit_id = drive_index_to_unit_id(type, index);
 476     }
 477
 478     /* if user doesn't specify a unit_id,
 479      * try to find the first free
 480      */
 481
 482     if (unit_id == -1) {
 483        unit_id = 0;
 484        while (drive_get(type, bus_id, unit_id) != NULL) {
 485            unit_id++;
 486            if (max_devs && unit_id >= max_devs) {
 487                unit_id -= max_devs;
 488                bus_id++;
 489            }
 490        }
 491     }
 492
 493     /* check unit id */
 494
 495     if (max_devs && unit_id >= max_devs) {
 496         error_report("unit %d too big (max is %d)",
 497                      unit_id, max_devs - 1);
 498         return NULL;
 499     }
 500
 501     /*
 502      * catch multiple definitions
 503      */
 504
 505     if (drive_get(type, bus_id, unit_id) != NULL) {
 506         error_report("drive with bus=%d, unit=%d (index=%d) exists",
 507                      bus_id, unit_id, index);
 508         return NULL;
 509     }
 510
 511     /* init */
 512
 513     dinfo = g_malloc0(sizeof(*dinfo));
 514     if ((buf = qemu_opts_id(opts)) != NULL) {
 515         dinfo->id = g_strdup(buf);
 516     } else {
 517         /* no id supplied -> create one */
 518         dinfo->id = g_malloc0(32);
 519         if (type == IF_IDE || type == IF_SCSI)
 520             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
 521         if (max_devs)
 522             snprintf(dinfo->id, 32, "%s%i%s%i",
 523                      if_name[type], bus_id, mediastr, unit_id);
 524         else
 525             snprintf(dinfo->id, 32, "%s%s%i",
 526                      if_name[type], mediastr, unit_id);
 527     }
 528     dinfo->bdrv = bdrv_new(dinfo->id);
 529     dinfo->devaddr = devaddr;
 530     dinfo->type = type;
 531     dinfo->bus = bus_id;
 532     dinfo->unit = unit_id;
 533     dinfo->cyls = cyls;
 534     dinfo->heads = heads;
 535     dinfo->secs = secs;
 536     dinfo->trans = translation;
 537     dinfo->opts = opts;
 538     dinfo->refcount = 1;
 539     if (serial) {
 540         pstrcpy(dinfo->serial, sizeof(dinfo->serial), serial);
 541     }
 542     QTAILQ_INSERT_TAIL(&drives, dinfo, next);
 543
 544     bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
 545
 546     /* disk I/O throttling */
 547     bdrv_set_io_limits(dinfo->bdrv, &io_limits);
 548
 549     switch(type) {
 550     case IF_IDE:
 551     case IF_SCSI:
 552     case IF_XEN:
 553     case IF_NONE:
 554         switch(media) {
 555         case MEDIA_DISK:
 556             if (cyls != 0) {
 557                 bdrv_set_geometry_hint(dinfo->bdrv, cyls, heads, secs);
 558                 bdrv_set_translation_hint(dinfo->bdrv, translation);
 559             }
 560             break;
 561         case MEDIA_CDROM:
 562             dinfo->media_cd = 1;
 563             break;
 564         }
 565         break;
 566     case IF_SD:
 567     case IF_FLOPPY:
 568     case IF_PFLASH:
 569     case IF_MTD:
 570         break;
 571     case IF_VIRTIO:
 572         /* add virtio block device */
 573         opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0, NULL);
 574         if (arch_type == QEMU_ARCH_S390X) {
 575             qemu_opt_set(opts, "driver", "virtio-blk-s390");
 576         } else {
 577             qemu_opt_set(opts, "driver", "virtio-blk-pci");
 578         }
 579         qemu_opt_set(opts, "drive", dinfo->id);
 580         if (devaddr)
 581             qemu_opt_set(opts, "addr", devaddr);
 582         break;
 583     default:
 584         abort();
 585     }
 586     if (!file || !*file) {
 587         return dinfo;
 588     }
 589     if (snapshot) {
 590         /* always use cache=unsafe with snapshot */
 591         bdrv_flags &= ~BDRV_O_CACHE_MASK;
 592         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
 593     }
 594
 595     if (copy_on_read) {
 596         bdrv_flags |= BDRV_O_COPY_ON_READ;
 597     }
 598
 599     if (runstate_check(RUN_STATE_INMIGRATE)) {
 600         bdrv_flags |= BDRV_O_INCOMING;
 601     }
 602
 603     if (media == MEDIA_CDROM) {
 604         /* CDROM is fine for any interface, don't check.  */
 605         ro = 1;
 606     } else if (ro == 1) {
 607         if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY &&
 608             type != IF_NONE && type != IF_PFLASH) {
 609             error_report("readonly not supported by this bus type");
 610             goto err;
 611         }
 612     }
 613
 614     bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
 615
 616     if (ro && copy_on_read) {
 617         error_report("warning: disabling copy_on_read on readonly drive");
 618     }
 619
 620     ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv);
 621     if (ret < 0) {
 622         error_report("could not open disk image %s: %s",
 623                      file, strerror(-ret));
 624         goto err;
 625     }
 626
 627     if (bdrv_key_required(dinfo->bdrv))
 628         autostart = 0;
 629     return dinfo;
 630
 631 err:
 632     bdrv_delete(dinfo->bdrv);
 633     g_free(dinfo->id);
 634     QTAILQ_REMOVE(&drives, dinfo, next);
 635     g_free(dinfo);
 636     return NULL;
 637 }
 638
 639 void do_commit(Monitor *mon, const QDict *qdict)
 640 {
 641     const char *device = qdict_get_str(qdict, "device");
 642     BlockDriverState *bs;
 643     int ret;
 644
 645     if (!strcmp(device, "all")) {
 646         ret = bdrv_commit_all();
 647         if (ret == -EBUSY) {
 648             qerror_report(QERR_DEVICE_IN_USE, device);
 649             return;
 650         }
 651     } else {
 652         bs = bdrv_find(device);
 653         if (!bs) {
 654             qerror_report(QERR_DEVICE_NOT_FOUND, device);
 655             return;
 656         }
 657         ret = bdrv_commit(bs);
 658         if (ret == -EBUSY) {
 659             qerror_report(QERR_DEVICE_IN_USE, device);
 660             return;
 661         }
 662     }
 663 }
 664
 665 static void blockdev_do_action(int kind, void *data, Error **errp)
 666 {
 667     BlockdevAction action;
 668     BlockdevActionList list;
 669
 670     action.kind = kind;
 671     action.data = data;
 672     list.value = &action;
 673     list.next = NULL;
 674     qmp_transaction(&list, errp);
 675 }
 676
 677 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
 678                                 bool has_format, const char *format,
 679                                 bool has_mode, enum NewImageMode mode,
 680                                 Error **errp)
 681 {
 682     BlockdevSnapshot snapshot = {
 683         .device = (char *) device,
 684         .snapshot_file = (char *) snapshot_file,
 685         .has_format = has_format,
 686         .format = (char *) format,
 687         .has_mode = has_mode,
 688         .mode = mode,
 689     };
 690     blockdev_do_action(BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC, &snapshot,
 691                        errp);
 692 }
 693
 694
 695 /* New and old BlockDriverState structs for group snapshots */
 696 typedef struct BlkTransactionStates {
 697     BlockDriverState *old_bs;
 698     BlockDriverState *new_bs;
 699     QSIMPLEQ_ENTRY(BlkTransactionStates) entry;
 700 } BlkTransactionStates;
 701
 702 /*
 703  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
 704  *  then we do not pivot any of the devices in the group, and abandon the
 705  *  snapshots
 706  */
 707 void qmp_transaction(BlockdevActionList *dev_list, Error **errp)
 708 {
 709     int ret = 0;
 710     BlockdevActionList *dev_entry = dev_list;
 711     BlkTransactionStates *states, *next;
 712
 713     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionStates) snap_bdrv_states;
 714     QSIMPLEQ_INIT(&snap_bdrv_states);
 715
 716     /* drain all i/o before any snapshots */
 717     bdrv_drain_all();
 718
 719     /* We don't do anything in this loop that commits us to the snapshot */
 720     while (NULL != dev_entry) {
 721         BlockdevAction *dev_info = NULL;
 722         BlockDriver *proto_drv;
 723         BlockDriver *drv;
 724         int flags;
 725         enum NewImageMode mode;
 726         const char *new_image_file;
 727         const char *device;
 728         const char *format = "qcow2";
 729
 730         dev_info = dev_entry->value;
 731         dev_entry = dev_entry->next;
 732
 733         states = g_malloc0(sizeof(BlkTransactionStates));
 734         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, states, entry);
 735
 736         switch (dev_info->kind) {
 737         case BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
 738             device = dev_info->blockdev_snapshot_sync->device;
 739             if (!dev_info->blockdev_snapshot_sync->has_mode) {
 740                 dev_info->blockdev_snapshot_sync->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
 741             }
 742             new_image_file = dev_info->blockdev_snapshot_sync->snapshot_file;
 743             if (dev_info->blockdev_snapshot_sync->has_format) {
 744                 format = dev_info->blockdev_snapshot_sync->format;
 745             }
 746             mode = dev_info->blockdev_snapshot_sync->mode;
 747             break;
 748         default:
 749             abort();
 750         }
 751
 752         drv = bdrv_find_format(format);
 753         if (!drv) {
 754             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 755             goto delete_and_fail;
 756         }
 757
 758         states->old_bs = bdrv_find(device);
 759         if (!states->old_bs) {
 760             error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 761             goto delete_and_fail;
 762         }
 763
 764         if (!bdrv_is_inserted(states->old_bs)) {
 765             error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
 766             goto delete_and_fail;
 767         }
 768
 769         if (bdrv_in_use(states->old_bs)) {
 770             error_set(errp, QERR_DEVICE_IN_USE, device);
 771             goto delete_and_fail;
 772         }
 773
 774         if (!bdrv_is_read_only(states->old_bs)) {
 775             if (bdrv_flush(states->old_bs)) {
 776                 error_set(errp, QERR_IO_ERROR);
 777                 goto delete_and_fail;
 778             }
 779         }
 780
 781         flags = states->old_bs->open_flags;
 782
 783         proto_drv = bdrv_find_protocol(new_image_file);
 784         if (!proto_drv) {
 785             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 786             goto delete_and_fail;
 787         }
 788
 789         /* create new image w/backing file */
 790         if (mode != NEW_IMAGE_MODE_EXISTING) {
 791             ret = bdrv_img_create(new_image_file, format,
 792                                   states->old_bs->filename,
 793                                   states->old_bs->drv->format_name,
 794                                   NULL, -1, flags);
 795             if (ret) {
 796                 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 797                 goto delete_and_fail;
 798             }
 799         }
 800
 801         /* We will manually add the backing_hd field to the bs later */
 802         states->new_bs = bdrv_new("");
 803         ret = bdrv_open(states->new_bs, new_image_file,
 804                         flags | BDRV_O_NO_BACKING, drv);
 805         if (ret != 0) {
 806             error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 807             goto delete_and_fail;
 808         }
 809     }
 810
 811
 812     /* Now we are going to do the actual pivot.  Everything up to this point
 813      * is reversible, but we are committed at this point */
 814     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 815         /* This removes our old bs from the bdrv_states, and adds the new bs */
 816         bdrv_append(states->new_bs, states->old_bs);
 817     }
 818
 819     /* success */
 820     goto exit;
 821
 822 delete_and_fail:
 823     /*
 824     * failure, and it is all-or-none; abandon each new bs, and keep using
 825     * the original bs for all images
 826     */
 827     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 828         if (states->new_bs) {
 829              bdrv_delete(states->new_bs);
 830         }
 831     }
 832 exit:
 833     QSIMPLEQ_FOREACH_SAFE(states, &snap_bdrv_states, entry, next) {
 834         g_free(states);
 835     }
 836     return;
 837 }
 838
 839
 840 static void eject_device(BlockDriverState *bs, int force, Error **errp)
 841 {
 842     if (bdrv_in_use(bs)) {
 843         error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 844         return;
 845     }
 846     if (!bdrv_dev_has_removable_media(bs)) {
 847         error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs));
 848         return;
 849     }
 850
 851     if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) {
 852         bdrv_dev_eject_request(bs, force);
 853         if (!force) {
 854             error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
 855             return;
 856         }
 857     }
 858
 859     bdrv_close(bs);
 860 }
 861
 862 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
 863 {
 864     BlockDriverState *bs;
 865
 866     bs = bdrv_find(device);
 867     if (!bs) {
 868         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 869         return;
 870     }
 871
 872     eject_device(bs, force, errp);
 873 }
 874
 875 void qmp_block_passwd(const char *device, const char *password, Error **errp)
 876 {
 877     BlockDriverState *bs;
 878     int err;
 879
 880     bs = bdrv_find(device);
 881     if (!bs) {
 882         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 883         return;
 884     }
 885
 886     err = bdrv_set_key(bs, password);
 887     if (err == -EINVAL) {
 888         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
 889         return;
 890     } else if (err < 0) {
 891         error_set(errp, QERR_INVALID_PASSWORD);
 892         return;
 893     }
 894 }
 895
 896 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
 897                                     int bdrv_flags, BlockDriver *drv,
 898                                     const char *password, Error **errp)
 899 {
 900     if (bdrv_open(bs, filename, bdrv_flags, drv) < 0) {
 901         error_set(errp, QERR_OPEN_FILE_FAILED, filename);
 902         return;
 903     }
 904
 905     if (bdrv_key_required(bs)) {
 906         if (password) {
 907             if (bdrv_set_key(bs, password) < 0) {
 908                 error_set(errp, QERR_INVALID_PASSWORD);
 909             }
 910         } else {
 911             error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs),
 912                       bdrv_get_encrypted_filename(bs));
 913         }
 914     } else if (password) {
 915         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
 916     }
 917 }
 918
 919 void qmp_change_blockdev(const char *device, const char *filename,
 920                          bool has_format, const char *format, Error **errp)
 921 {
 922     BlockDriverState *bs;
 923     BlockDriver *drv = NULL;
 924     int bdrv_flags;
 925     Error *err = NULL;
 926
 927     bs = bdrv_find(device);
 928     if (!bs) {
 929         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 930         return;
 931     }
 932
 933     if (format) {
 934         drv = bdrv_find_whitelisted_format(format);
 935         if (!drv) {
 936             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 937             return;
 938         }
 939     }
 940
 941     eject_device(bs, 0, &err);
 942     if (error_is_set(&err)) {
 943         error_propagate(errp, err);
 944         return;
 945     }
 946
 947     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
 948     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
 949
 950     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
 951 }
 952
 953 /* throttling disk I/O limits */
 954 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
 955                                int64_t bps_wr, int64_t iops, int64_t iops_rd,
 956                                int64_t iops_wr, Error **errp)
 957 {
 958     BlockIOLimit io_limits;
 959     BlockDriverState *bs;
 960
 961     bs = bdrv_find(device);
 962     if (!bs) {
 963         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 964         return;
 965     }
 966
 967     io_limits.bps[BLOCK_IO_LIMIT_TOTAL] = bps;
 968     io_limits.bps[BLOCK_IO_LIMIT_READ]  = bps_rd;
 969     io_limits.bps[BLOCK_IO_LIMIT_WRITE] = bps_wr;
 970     io_limits.iops[BLOCK_IO_LIMIT_TOTAL]= iops;
 971     io_limits.iops[BLOCK_IO_LIMIT_READ] = iops_rd;
 972     io_limits.iops[BLOCK_IO_LIMIT_WRITE]= iops_wr;
 973
 974     if (!do_check_io_limits(&io_limits)) {
 975         error_set(errp, QERR_INVALID_PARAMETER_COMBINATION);
 976         return;
 977     }
 978
 979     bs->io_limits = io_limits;
 980     bs->slice_time = BLOCK_IO_SLICE_TIME;
 981
 982     if (!bs->io_limits_enabled && bdrv_io_limits_enabled(bs)) {
 983         bdrv_io_limits_enable(bs);
 984     } else if (bs->io_limits_enabled && !bdrv_io_limits_enabled(bs)) {
 985         bdrv_io_limits_disable(bs);
 986     } else {
 987         if (bs->block_timer) {
 988             qemu_mod_timer(bs->block_timer, qemu_get_clock_ns(vm_clock));
 989         }
 990     }
 991 }
 992
 993 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
 994 {
 995     const char *id = qdict_get_str(qdict, "id");
 996     BlockDriverState *bs;
 997
 998     bs = bdrv_find(id);
 999     if (!bs) {
1000         qerror_report(QERR_DEVICE_NOT_FOUND, id);
1001         return -1;
1002     }
1003     if (bdrv_in_use(bs)) {
1004         qerror_report(QERR_DEVICE_IN_USE, id);
1005         return -1;
1006     }
1007
1008     /* quiesce block driver; prevent further io */
1009     bdrv_drain_all();
1010     bdrv_flush(bs);
1011     bdrv_close(bs);
1012
1013     /* if we have a device attached to this BlockDriverState
1014      * then we need to make the drive anonymous until the device
1015      * can be removed.  If this is a drive with no device backing
1016      * then we can just get rid of the block driver state right here.
1017      */
1018     if (bdrv_get_attached_dev(bs)) {
1019         bdrv_make_anon(bs);
1020     } else {
1021         drive_uninit(drive_get_by_blockdev(bs));
1022     }
1023
1024     return 0;
1025 }
1026
1027 void qmp_block_resize(const char *device, int64_t size, Error **errp)
1028 {
1029     BlockDriverState *bs;
1030
1031     bs = bdrv_find(device);
1032     if (!bs) {
1033         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1034         return;
1035     }
1036
1037     if (size < 0) {
1038         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
1039         return;
1040     }
1041
1042     switch (bdrv_truncate(bs, size)) {
1043     case 0:
1044         break;
1045     case -ENOMEDIUM:
1046         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1047         break;
1048     case -ENOTSUP:
1049         error_set(errp, QERR_UNSUPPORTED);
1050         break;
1051     case -EACCES:
1052         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1053         break;
1054     case -EBUSY:
1055         error_set(errp, QERR_DEVICE_IN_USE, device);
1056         break;
1057     default:
1058         error_set(errp, QERR_UNDEFINED_ERROR);
1059         break;
1060     }
1061 }
1062
1063 static QObject *qobject_from_block_job(BlockJob *job)
1064 {
1065     return qobject_from_jsonf("{ 'type': %s,"
1066                               "'device': %s,"
1067                               "'len': %" PRId64 ","
1068                               "'offset': %" PRId64 ","
1069                               "'speed': %" PRId64 " }",
1070                               job->job_type->job_type,
1071                               bdrv_get_device_name(job->bs),
1072                               job->len,
1073                               job->offset,
1074                               job->speed);
1075 }
1076
1077 static void block_stream_cb(void *opaque, int ret)
1078 {
1079     BlockDriverState *bs = opaque;
1080     QObject *obj;
1081
1082     trace_block_stream_cb(bs, bs->job, ret);
1083
1084     assert(bs->job);
1085     obj = qobject_from_block_job(bs->job);
1086     if (ret < 0) {
1087         QDict *dict = qobject_to_qdict(obj);
1088         qdict_put(dict, "error", qstring_from_str(strerror(-ret)));
1089     }
1090
1091     if (block_job_is_cancelled(bs->job)) {
1092         monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj);
1093     } else {
1094         monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj);
1095     }
1096     qobject_decref(obj);
1097
1098     drive_put_ref_bh_schedule(drive_get_by_blockdev(bs));
1099 }
1100
1101 void qmp_block_stream(const char *device, bool has_base,
1102                       const char *base, bool has_speed,
1103                       int64_t speed, Error **errp)
1104 {
1105     BlockDriverState *bs;
1106     BlockDriverState *base_bs = NULL;
1107     Error *local_err = NULL;
1108
1109     bs = bdrv_find(device);
1110     if (!bs) {
1111         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1112         return;
1113     }
1114
1115     if (base) {
1116         base_bs = bdrv_find_backing_image(bs, base);
1117         if (base_bs == NULL) {
1118             error_set(errp, QERR_BASE_NOT_FOUND, base);
1119             return;
1120         }
1121     }
1122
1123     stream_start(bs, base_bs, base, has_speed ? speed : 0,
1124                  block_stream_cb, bs, &local_err);
1125     if (error_is_set(&local_err)) {
1126         error_propagate(errp, local_err);
1127         return;
1128     }
1129
1130     /* Grab a reference so hotplug does not delete the BlockDriverState from
1131      * underneath us.
1132      */
1133     drive_get_ref(drive_get_by_blockdev(bs));
1134
1135     trace_qmp_block_stream(bs, bs->job);
1136 }
1137
1138 static BlockJob *find_block_job(const char *device)
1139 {
1140     BlockDriverState *bs;
1141
1142     bs = bdrv_find(device);
1143     if (!bs || !bs->job) {
1144         return NULL;
1145     }
1146     return bs->job;
1147 }
1148
1149 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
1150 {
1151     BlockJob *job = find_block_job(device);
1152
1153     if (!job) {
1154         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1155         return;
1156     }
1157
1158     block_job_set_speed(job, speed, errp);
1159 }
1160
1161 void qmp_block_job_cancel(const char *device, Error **errp)
1162 {
1163     BlockJob *job = find_block_job(device);
1164
1165     if (!job) {
1166         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1167         return;
1168     }
1169
1170     trace_qmp_block_job_cancel(job);
1171     block_job_cancel(job);
1172 }
1173
1174 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
1175 {
1176     BlockJobInfoList **prev = opaque;
1177     BlockJob *job = bs->job;
1178
1179     if (job) {
1180         BlockJobInfoList *elem;
1181         BlockJobInfo *info = g_new(BlockJobInfo, 1);
1182         *info = (BlockJobInfo){
1183             .type   = g_strdup(job->job_type->job_type),
1184             .device = g_strdup(bdrv_get_device_name(bs)),
1185             .len    = job->len,
1186             .offset = job->offset,
1187             .speed  = job->speed,
1188         };
1189
1190         elem = g_new0(BlockJobInfoList, 1);
1191         elem->value = info;
1192
1193         (*prev)->next = elem;
1194         *prev = elem;
1195     }
1196 }
1197
1198 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
1199 {
1200     /* Dummy is a fake list element for holding the head pointer */
1201     BlockJobInfoList dummy = {};
1202     BlockJobInfoList *prev = &dummy;
1203     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
1204     return dummy.next;
1205 }