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 "blockjob.h"
  13 #include "monitor.h"
  14 #include "qerror.h"
  15 #include "qemu-option.h"
  16 #include "qemu-config.h"
  17 #include "qemu-objects.h"
  18 #include "sysemu.h"
  19 #include "block_int.h"
  20 #include "qmp-commands.h"
  21 #include "trace.h"
  22 #include "arch_init.h"
  23
  24 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
  25
  26 static const char *const if_name[IF_COUNT] = {
  27     [IF_NONE] = "none",
  28     [IF_IDE] = "ide",
  29     [IF_SCSI] = "scsi",
  30     [IF_FLOPPY] = "floppy",
  31     [IF_PFLASH] = "pflash",
  32     [IF_MTD] = "mtd",
  33     [IF_SD] = "sd",
  34     [IF_VIRTIO] = "virtio",
  35     [IF_XEN] = "xen",
  36 };
  37
  38 static const int if_max_devs[IF_COUNT] = {
  39     /*
  40      * Do not change these numbers!  They govern how drive option
  41      * index maps to unit and bus.  That mapping is ABI.
  42      *
  43      * All controllers used to imlement if=T drives need to support
  44      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
  45      * Otherwise, some index values map to "impossible" bus, unit
  46      * values.
  47      *
  48      * For instance, if you change [IF_SCSI] to 255, -drive
  49      * if=scsi,index=12 no longer means bus=1,unit=5, but
  50      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
  51      * the drive can't be set up.  Regression.
  52      */
  53     [IF_IDE] = 2,
  54     [IF_SCSI] = 7,
  55 };
  56
  57 /*
  58  * We automatically delete the drive when a device using it gets
  59  * unplugged.  Questionable feature, but we can't just drop it.
  60  * Device models call blockdev_mark_auto_del() to schedule the
  61  * automatic deletion, and generic qdev code calls blockdev_auto_del()
  62  * when deletion is actually safe.
  63  */
  64 void blockdev_mark_auto_del(BlockDriverState *bs)
  65 {
  66     DriveInfo *dinfo = drive_get_by_blockdev(bs);
  67
  68     if (bs->job) {
  69         block_job_cancel(bs->job);
  70     }
  71     if (dinfo) {
  72         dinfo->auto_del = 1;
  73     }
  74 }
  75
  76 void blockdev_auto_del(BlockDriverState *bs)
  77 {
  78     DriveInfo *dinfo = drive_get_by_blockdev(bs);
  79
  80     if (dinfo && dinfo->auto_del) {
  81         drive_put_ref(dinfo);
  82     }
  83 }
  84
  85 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
  86 {
  87     int max_devs = if_max_devs[type];
  88     return max_devs ? index / max_devs : 0;
  89 }
  90
  91 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
  92 {
  93     int max_devs = if_max_devs[type];
  94     return max_devs ? index % max_devs : index;
  95 }
  96
  97 QemuOpts *drive_def(const char *optstr)
  98 {
  99     return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
 100 }
 101
 102 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
 103                     const char *optstr)
 104 {
 105     QemuOpts *opts;
 106     char buf[32];
 107
 108     opts = drive_def(optstr);
 109     if (!opts) {
 110         return NULL;
 111     }
 112     if (type != IF_DEFAULT) {
 113         qemu_opt_set(opts, "if", if_name[type]);
 114     }
 115     if (index >= 0) {
 116         snprintf(buf, sizeof(buf), "%d", index);
 117         qemu_opt_set(opts, "index", buf);
 118     }
 119     if (file)
 120         qemu_opt_set(opts, "file", file);
 121     return opts;
 122 }
 123
 124 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
 125 {
 126     DriveInfo *dinfo;
 127
 128     /* seek interface, bus and unit */
 129
 130     QTAILQ_FOREACH(dinfo, &drives, next) {
 131         if (dinfo->type == type &&
 132             dinfo->bus == bus &&
 133             dinfo->unit == unit)
 134             return dinfo;
 135     }
 136
 137     return NULL;
 138 }
 139
 140 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
 141 {
 142     return drive_get(type,
 143                      drive_index_to_bus_id(type, index),
 144                      drive_index_to_unit_id(type, index));
 145 }
 146
 147 int drive_get_max_bus(BlockInterfaceType type)
 148 {
 149     int max_bus;
 150     DriveInfo *dinfo;
 151
 152     max_bus = -1;
 153     QTAILQ_FOREACH(dinfo, &drives, next) {
 154         if(dinfo->type == type &&
 155            dinfo->bus > max_bus)
 156             max_bus = dinfo->bus;
 157     }
 158     return max_bus;
 159 }
 160
 161 /* Get a block device.  This should only be used for single-drive devices
 162    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
 163    appropriate bus.  */
 164 DriveInfo *drive_get_next(BlockInterfaceType type)
 165 {
 166     static int next_block_unit[IF_COUNT];
 167
 168     return drive_get(type, 0, next_block_unit[type]++);
 169 }
 170
 171 DriveInfo *drive_get_by_blockdev(BlockDriverState *bs)
 172 {
 173     DriveInfo *dinfo;
 174
 175     QTAILQ_FOREACH(dinfo, &drives, next) {
 176         if (dinfo->bdrv == bs) {
 177             return dinfo;
 178         }
 179     }
 180     return NULL;
 181 }
 182
 183 static void bdrv_format_print(void *opaque, const char *name)
 184 {
 185     error_printf(" %s", name);
 186 }
 187
 188 static void drive_uninit(DriveInfo *dinfo)
 189 {
 190     qemu_opts_del(dinfo->opts);
 191     bdrv_delete(dinfo->bdrv);
 192     g_free(dinfo->id);
 193     QTAILQ_REMOVE(&drives, dinfo, next);
 194     g_free(dinfo);
 195 }
 196
 197 void drive_put_ref(DriveInfo *dinfo)
 198 {
 199     assert(dinfo->refcount);
 200     if (--dinfo->refcount == 0) {
 201         drive_uninit(dinfo);
 202     }
 203 }
 204
 205 void drive_get_ref(DriveInfo *dinfo)
 206 {
 207     dinfo->refcount++;
 208 }
 209
 210 typedef struct {
 211     QEMUBH *bh;
 212     DriveInfo *dinfo;
 213 } DrivePutRefBH;
 214
 215 static void drive_put_ref_bh(void *opaque)
 216 {
 217     DrivePutRefBH *s = opaque;
 218
 219     drive_put_ref(s->dinfo);
 220     qemu_bh_delete(s->bh);
 221     g_free(s);
 222 }
 223
 224 /*
 225  * Release a drive reference in a BH
 226  *
 227  * It is not possible to use drive_put_ref() from a callback function when the
 228  * callers still need the drive.  In such cases we schedule a BH to release the
 229  * reference.
 230  */
 231 static void drive_put_ref_bh_schedule(DriveInfo *dinfo)
 232 {
 233     DrivePutRefBH *s;
 234
 235     s = g_new(DrivePutRefBH, 1);
 236     s->bh = qemu_bh_new(drive_put_ref_bh, s);
 237     s->dinfo = dinfo;
 238     qemu_bh_schedule(s->bh);
 239 }
 240
 241 static int parse_block_error_action(const char *buf, int is_read)
 242 {
 243     if (!strcmp(buf, "ignore")) {
 244         return BLOCK_ERR_IGNORE;
 245     } else if (!is_read && !strcmp(buf, "enospc")) {
 246         return BLOCK_ERR_STOP_ENOSPC;
 247     } else if (!strcmp(buf, "stop")) {
 248         return BLOCK_ERR_STOP_ANY;
 249     } else if (!strcmp(buf, "report")) {
 250         return BLOCK_ERR_REPORT;
 251     } else {
 252         error_report("'%s' invalid %s error action",
 253                      buf, is_read ? "read" : "write");
 254         return -1;
 255     }
 256 }
 257
 258 static bool do_check_io_limits(BlockIOLimit *io_limits)
 259 {
 260     bool bps_flag;
 261     bool iops_flag;
 262
 263     assert(io_limits);
 264
 265     bps_flag  = (io_limits->bps[BLOCK_IO_LIMIT_TOTAL] != 0)
 266                  && ((io_limits->bps[BLOCK_IO_LIMIT_READ] != 0)
 267                  || (io_limits->bps[BLOCK_IO_LIMIT_WRITE] != 0));
 268     iops_flag = (io_limits->iops[BLOCK_IO_LIMIT_TOTAL] != 0)
 269                  && ((io_limits->iops[BLOCK_IO_LIMIT_READ] != 0)
 270                  || (io_limits->iops[BLOCK_IO_LIMIT_WRITE] != 0));
 271     if (bps_flag || iops_flag) {
 272         return false;
 273     }
 274
 275     return true;
 276 }
 277
 278 DriveInfo *drive_init(QemuOpts *opts, int default_to_scsi)
 279 {
 280     const char *buf;
 281     const char *file = NULL;
 282     const char *serial;
 283     const char *mediastr = "";
 284     BlockInterfaceType type;
 285     enum { MEDIA_DISK, MEDIA_CDROM } media;
 286     int bus_id, unit_id;
 287     int cyls, heads, secs, translation;
 288     BlockDriver *drv = NULL;
 289     int max_devs;
 290     int index;
 291     int ro = 0;
 292     int bdrv_flags = 0;
 293     int on_read_error, on_write_error;
 294     const char *devaddr;
 295     DriveInfo *dinfo;
 296     BlockIOLimit io_limits;
 297     int snapshot = 0;
 298     bool copy_on_read;
 299     int ret;
 300
 301     translation = BIOS_ATA_TRANSLATION_AUTO;
 302     media = MEDIA_DISK;
 303
 304     /* extract parameters */
 305     bus_id  = qemu_opt_get_number(opts, "bus", 0);
 306     unit_id = qemu_opt_get_number(opts, "unit", -1);
 307     index   = qemu_opt_get_number(opts, "index", -1);
 308
 309     cyls  = qemu_opt_get_number(opts, "cyls", 0);
 310     heads = qemu_opt_get_number(opts, "heads", 0);
 311     secs  = qemu_opt_get_number(opts, "secs", 0);
 312
 313     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
 314     ro = qemu_opt_get_bool(opts, "readonly", 0);
 315     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
 316
 317     file = qemu_opt_get(opts, "file");
 318     serial = qemu_opt_get(opts, "serial");
 319
 320     if ((buf = qemu_opt_get(opts, "if")) != NULL) {
 321         for (type = 0; type < IF_COUNT && strcmp(buf, if_name[type]); type++)
 322             ;
 323         if (type == IF_COUNT) {
 324             error_report("unsupported bus type '%s'", buf);
 325             return NULL;
 326         }
 327     } else {
 328         type = default_to_scsi ? IF_SCSI : IF_IDE;
 329     }
 330
 331     max_devs = if_max_devs[type];
 332
 333     if (cyls || heads || secs) {
 334         if (cyls < 1) {
 335             error_report("invalid physical cyls number");
 336             return NULL;
 337         }
 338         if (heads < 1) {
 339             error_report("invalid physical heads number");
 340             return NULL;
 341         }
 342         if (secs < 1) {
 343             error_report("invalid physical secs number");
 344             return NULL;
 345         }
 346     }
 347
 348     if ((buf = qemu_opt_get(opts, "trans")) != NULL) {
 349         if (!cyls) {
 350             error_report("'%s' trans must be used with cyls, heads and secs",
 351                          buf);
 352             return NULL;
 353         }
 354         if (!strcmp(buf, "none"))
 355             translation = BIOS_ATA_TRANSLATION_NONE;
 356         else if (!strcmp(buf, "lba"))
 357             translation = BIOS_ATA_TRANSLATION_LBA;
 358         else if (!strcmp(buf, "auto"))
 359             translation = BIOS_ATA_TRANSLATION_AUTO;
 360         else {
 361             error_report("'%s' invalid translation type", buf);
 362             return NULL;
 363         }
 364     }
 365
 366     if ((buf = qemu_opt_get(opts, "media")) != NULL) {
 367         if (!strcmp(buf, "disk")) {
 368             media = MEDIA_DISK;
 369         } else if (!strcmp(buf, "cdrom")) {
 370             if (cyls || secs || heads) {
 371                 error_report("CHS can't be set with media=%s", buf);
 372                 return NULL;
 373             }
 374             media = MEDIA_CDROM;
 375         } else {
 376             error_report("'%s' invalid media", buf);
 377             return NULL;
 378         }
 379     }
 380
 381     bdrv_flags |= BDRV_O_CACHE_WB;
 382     if ((buf = qemu_opt_get(opts, "cache")) != NULL) {
 383         if (bdrv_parse_cache_flags(buf, &bdrv_flags) != 0) {
 384             error_report("invalid cache option");
 385             return NULL;
 386         }
 387     }
 388
 389 #ifdef CONFIG_LINUX_AIO
 390     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
 391         if (!strcmp(buf, "native")) {
 392             bdrv_flags |= BDRV_O_NATIVE_AIO;
 393         } else if (!strcmp(buf, "threads")) {
 394             /* this is the default */
 395         } else {
 396            error_report("invalid aio option");
 397            return NULL;
 398         }
 399     }
 400 #endif
 401
 402     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
 403         if (is_help_option(buf)) {
 404             error_printf("Supported formats:");
 405             bdrv_iterate_format(bdrv_format_print, NULL);
 406             error_printf("\n");
 407             return NULL;
 408         }
 409         drv = bdrv_find_whitelisted_format(buf);
 410         if (!drv) {
 411             error_report("'%s' invalid format", buf);
 412             return NULL;
 413         }
 414     }
 415
 416     /* disk I/O throttling */
 417     io_limits.bps[BLOCK_IO_LIMIT_TOTAL]  =
 418                            qemu_opt_get_number(opts, "bps", 0);
 419     io_limits.bps[BLOCK_IO_LIMIT_READ]   =
 420                            qemu_opt_get_number(opts, "bps_rd", 0);
 421     io_limits.bps[BLOCK_IO_LIMIT_WRITE]  =
 422                            qemu_opt_get_number(opts, "bps_wr", 0);
 423     io_limits.iops[BLOCK_IO_LIMIT_TOTAL] =
 424                            qemu_opt_get_number(opts, "iops", 0);
 425     io_limits.iops[BLOCK_IO_LIMIT_READ]  =
 426                            qemu_opt_get_number(opts, "iops_rd", 0);
 427     io_limits.iops[BLOCK_IO_LIMIT_WRITE] =
 428                            qemu_opt_get_number(opts, "iops_wr", 0);
 429
 430     if (!do_check_io_limits(&io_limits)) {
 431         error_report("bps(iops) and bps_rd/bps_wr(iops_rd/iops_wr) "
 432                      "cannot be used at the same time");
 433         return NULL;
 434     }
 435
 436     on_write_error = BLOCK_ERR_STOP_ENOSPC;
 437     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
 438         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) {
 439             error_report("werror is not supported by this bus type");
 440             return NULL;
 441         }
 442
 443         on_write_error = parse_block_error_action(buf, 0);
 444         if (on_write_error < 0) {
 445             return NULL;
 446         }
 447     }
 448
 449     on_read_error = BLOCK_ERR_REPORT;
 450     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
 451         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) {
 452             error_report("rerror is not supported by this bus type");
 453             return NULL;
 454         }
 455
 456         on_read_error = parse_block_error_action(buf, 1);
 457         if (on_read_error < 0) {
 458             return NULL;
 459         }
 460     }
 461
 462     if ((devaddr = qemu_opt_get(opts, "addr")) != NULL) {
 463         if (type != IF_VIRTIO) {
 464             error_report("addr is not supported by this bus type");
 465             return NULL;
 466         }
 467     }
 468
 469     /* compute bus and unit according index */
 470
 471     if (index != -1) {
 472         if (bus_id != 0 || unit_id != -1) {
 473             error_report("index cannot be used with bus and unit");
 474             return NULL;
 475         }
 476         bus_id = drive_index_to_bus_id(type, index);
 477         unit_id = drive_index_to_unit_id(type, index);
 478     }
 479
 480     /* if user doesn't specify a unit_id,
 481      * try to find the first free
 482      */
 483
 484     if (unit_id == -1) {
 485        unit_id = 0;
 486        while (drive_get(type, bus_id, unit_id) != NULL) {
 487            unit_id++;
 488            if (max_devs && unit_id >= max_devs) {
 489                unit_id -= max_devs;
 490                bus_id++;
 491            }
 492        }
 493     }
 494
 495     /* check unit id */
 496
 497     if (max_devs && unit_id >= max_devs) {
 498         error_report("unit %d too big (max is %d)",
 499                      unit_id, max_devs - 1);
 500         return NULL;
 501     }
 502
 503     /*
 504      * catch multiple definitions
 505      */
 506
 507     if (drive_get(type, bus_id, unit_id) != NULL) {
 508         error_report("drive with bus=%d, unit=%d (index=%d) exists",
 509                      bus_id, unit_id, index);
 510         return NULL;
 511     }
 512
 513     /* init */
 514
 515     dinfo = g_malloc0(sizeof(*dinfo));
 516     if ((buf = qemu_opts_id(opts)) != NULL) {
 517         dinfo->id = g_strdup(buf);
 518     } else {
 519         /* no id supplied -> create one */
 520         dinfo->id = g_malloc0(32);
 521         if (type == IF_IDE || type == IF_SCSI)
 522             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
 523         if (max_devs)
 524             snprintf(dinfo->id, 32, "%s%i%s%i",
 525                      if_name[type], bus_id, mediastr, unit_id);
 526         else
 527             snprintf(dinfo->id, 32, "%s%s%i",
 528                      if_name[type], mediastr, unit_id);
 529     }
 530     dinfo->bdrv = bdrv_new(dinfo->id);
 531     dinfo->bdrv->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
 532     dinfo->bdrv->read_only = ro;
 533     dinfo->devaddr = devaddr;
 534     dinfo->type = type;
 535     dinfo->bus = bus_id;
 536     dinfo->unit = unit_id;
 537     dinfo->cyls = cyls;
 538     dinfo->heads = heads;
 539     dinfo->secs = secs;
 540     dinfo->trans = translation;
 541     dinfo->opts = opts;
 542     dinfo->refcount = 1;
 543     dinfo->serial = serial;
 544     QTAILQ_INSERT_TAIL(&drives, dinfo, next);
 545
 546     bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
 547
 548     /* disk I/O throttling */
 549     bdrv_set_io_limits(dinfo->bdrv, &io_limits);
 550
 551     switch(type) {
 552     case IF_IDE:
 553     case IF_SCSI:
 554     case IF_XEN:
 555     case IF_NONE:
 556         dinfo->media_cd = media == MEDIA_CDROM;
 557         break;
 558     case IF_SD:
 559     case IF_FLOPPY:
 560     case IF_PFLASH:
 561     case IF_MTD:
 562         break;
 563     case IF_VIRTIO:
 564         /* add virtio block device */
 565         opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0, NULL);
 566         if (arch_type == QEMU_ARCH_S390X) {
 567             qemu_opt_set(opts, "driver", "virtio-blk-s390");
 568         } else {
 569             qemu_opt_set(opts, "driver", "virtio-blk-pci");
 570         }
 571         qemu_opt_set(opts, "drive", dinfo->id);
 572         if (devaddr)
 573             qemu_opt_set(opts, "addr", devaddr);
 574         break;
 575     default:
 576         abort();
 577     }
 578     if (!file || !*file) {
 579         return dinfo;
 580     }
 581     if (snapshot) {
 582         /* always use cache=unsafe with snapshot */
 583         bdrv_flags &= ~BDRV_O_CACHE_MASK;
 584         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
 585     }
 586
 587     if (copy_on_read) {
 588         bdrv_flags |= BDRV_O_COPY_ON_READ;
 589     }
 590
 591     if (runstate_check(RUN_STATE_INMIGRATE)) {
 592         bdrv_flags |= BDRV_O_INCOMING;
 593     }
 594
 595     if (media == MEDIA_CDROM) {
 596         /* CDROM is fine for any interface, don't check.  */
 597         ro = 1;
 598     } else if (ro == 1) {
 599         if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY &&
 600             type != IF_NONE && type != IF_PFLASH) {
 601             error_report("readonly not supported by this bus type");
 602             goto err;
 603         }
 604     }
 605
 606     bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
 607
 608     if (ro && copy_on_read) {
 609         error_report("warning: disabling copy_on_read on readonly drive");
 610     }
 611
 612     ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv);
 613     if (ret < 0) {
 614         error_report("could not open disk image %s: %s",
 615                      file, strerror(-ret));
 616         goto err;
 617     }
 618
 619     if (bdrv_key_required(dinfo->bdrv))
 620         autostart = 0;
 621     return dinfo;
 622
 623 err:
 624     bdrv_delete(dinfo->bdrv);
 625     g_free(dinfo->id);
 626     QTAILQ_REMOVE(&drives, dinfo, next);
 627     g_free(dinfo);
 628     return NULL;
 629 }
 630
 631 void do_commit(Monitor *mon, const QDict *qdict)
 632 {
 633     const char *device = qdict_get_str(qdict, "device");
 634     BlockDriverState *bs;
 635     int ret;
 636
 637     if (!strcmp(device, "all")) {
 638         ret = bdrv_commit_all();
 639         if (ret == -EBUSY) {
 640             qerror_report(QERR_DEVICE_IN_USE, device);
 641             return;
 642         }
 643     } else {
 644         bs = bdrv_find(device);
 645         if (!bs) {
 646             qerror_report(QERR_DEVICE_NOT_FOUND, device);
 647             return;
 648         }
 649         ret = bdrv_commit(bs);
 650         if (ret == -EBUSY) {
 651             qerror_report(QERR_DEVICE_IN_USE, device);
 652             return;
 653         }
 654     }
 655 }
 656
 657 static void blockdev_do_action(int kind, void *data, Error **errp)
 658 {
 659     BlockdevAction action;
 660     BlockdevActionList list;
 661
 662     action.kind = kind;
 663     action.data = data;
 664     list.value = &action;
 665     list.next = NULL;
 666     qmp_transaction(&list, errp);
 667 }
 668
 669 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
 670                                 bool has_format, const char *format,
 671                                 bool has_mode, enum NewImageMode mode,
 672                                 Error **errp)
 673 {
 674     BlockdevSnapshot snapshot = {
 675         .device = (char *) device,
 676         .snapshot_file = (char *) snapshot_file,
 677         .has_format = has_format,
 678         .format = (char *) format,
 679         .has_mode = has_mode,
 680         .mode = mode,
 681     };
 682     blockdev_do_action(BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC, &snapshot,
 683                        errp);
 684 }
 685
 686
 687 /* New and old BlockDriverState structs for group snapshots */
 688 typedef struct BlkTransactionStates {
 689     BlockDriverState *old_bs;
 690     BlockDriverState *new_bs;
 691     QSIMPLEQ_ENTRY(BlkTransactionStates) entry;
 692 } BlkTransactionStates;
 693
 694 /*
 695  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
 696  *  then we do not pivot any of the devices in the group, and abandon the
 697  *  snapshots
 698  */
 699 void qmp_transaction(BlockdevActionList *dev_list, Error **errp)
 700 {
 701     int ret = 0;
 702     BlockdevActionList *dev_entry = dev_list;
 703     BlkTransactionStates *states, *next;
 704
 705     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionStates) snap_bdrv_states;
 706     QSIMPLEQ_INIT(&snap_bdrv_states);
 707
 708     /* drain all i/o before any snapshots */
 709     bdrv_drain_all();
 710
 711     /* We don't do anything in this loop that commits us to the snapshot */
 712     while (NULL != dev_entry) {
 713         BlockdevAction *dev_info = NULL;
 714         BlockDriver *proto_drv;
 715         BlockDriver *drv;
 716         int flags;
 717         enum NewImageMode mode;
 718         const char *new_image_file;
 719         const char *device;
 720         const char *format = "qcow2";
 721
 722         dev_info = dev_entry->value;
 723         dev_entry = dev_entry->next;
 724
 725         states = g_malloc0(sizeof(BlkTransactionStates));
 726         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, states, entry);
 727
 728         switch (dev_info->kind) {
 729         case BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
 730             device = dev_info->blockdev_snapshot_sync->device;
 731             if (!dev_info->blockdev_snapshot_sync->has_mode) {
 732                 dev_info->blockdev_snapshot_sync->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
 733             }
 734             new_image_file = dev_info->blockdev_snapshot_sync->snapshot_file;
 735             if (dev_info->blockdev_snapshot_sync->has_format) {
 736                 format = dev_info->blockdev_snapshot_sync->format;
 737             }
 738             mode = dev_info->blockdev_snapshot_sync->mode;
 739             break;
 740         default:
 741             abort();
 742         }
 743
 744         drv = bdrv_find_format(format);
 745         if (!drv) {
 746             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 747             goto delete_and_fail;
 748         }
 749
 750         states->old_bs = bdrv_find(device);
 751         if (!states->old_bs) {
 752             error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 753             goto delete_and_fail;
 754         }
 755
 756         if (!bdrv_is_inserted(states->old_bs)) {
 757             error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
 758             goto delete_and_fail;
 759         }
 760
 761         if (bdrv_in_use(states->old_bs)) {
 762             error_set(errp, QERR_DEVICE_IN_USE, device);
 763             goto delete_and_fail;
 764         }
 765
 766         if (!bdrv_is_read_only(states->old_bs)) {
 767             if (bdrv_flush(states->old_bs)) {
 768                 error_set(errp, QERR_IO_ERROR);
 769                 goto delete_and_fail;
 770             }
 771         }
 772
 773         flags = states->old_bs->open_flags;
 774
 775         proto_drv = bdrv_find_protocol(new_image_file);
 776         if (!proto_drv) {
 777             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 778             goto delete_and_fail;
 779         }
 780
 781         /* create new image w/backing file */
 782         if (mode != NEW_IMAGE_MODE_EXISTING) {
 783             ret = bdrv_img_create(new_image_file, format,
 784                                   states->old_bs->filename,
 785                                   states->old_bs->drv->format_name,
 786                                   NULL, -1, flags);
 787             if (ret) {
 788                 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 789                 goto delete_and_fail;
 790             }
 791         }
 792
 793         /* We will manually add the backing_hd field to the bs later */
 794         states->new_bs = bdrv_new("");
 795         ret = bdrv_open(states->new_bs, new_image_file,
 796                         flags | BDRV_O_NO_BACKING, drv);
 797         if (ret != 0) {
 798             error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 799             goto delete_and_fail;
 800         }
 801     }
 802
 803
 804     /* Now we are going to do the actual pivot.  Everything up to this point
 805      * is reversible, but we are committed at this point */
 806     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 807         /* This removes our old bs from the bdrv_states, and adds the new bs */
 808         bdrv_append(states->new_bs, states->old_bs);
 809         /* We don't need (or want) to use the transactional
 810          * bdrv_reopen_multiple() across all the entries at once, because we
 811          * don't want to abort all of them if one of them fails the reopen */
 812         bdrv_reopen(states->new_bs, states->new_bs->open_flags & ~BDRV_O_RDWR,
 813                     NULL);
 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_job_cb(void *opaque, int ret)
1075 {
1076     BlockDriverState *bs = opaque;
1077     QObject *obj;
1078
1079     trace_block_job_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_job_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 void qmp_block_commit(const char *device,
1136                       bool has_base, const char *base, const char *top,
1137                       bool has_speed, int64_t speed,
1138                       Error **errp)
1139 {
1140     BlockDriverState *bs;
1141     BlockDriverState *base_bs, *top_bs;
1142     Error *local_err = NULL;
1143     /* This will be part of the QMP command, if/when the
1144      * BlockdevOnError change for blkmirror makes it in
1145      */
1146     BlockErrorAction on_error = BLOCK_ERR_REPORT;
1147
1148     /* drain all i/o before commits */
1149     bdrv_drain_all();
1150
1151     bs = bdrv_find(device);
1152     if (!bs) {
1153         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1154         return;
1155     }
1156     if (base && has_base) {
1157         base_bs = bdrv_find_backing_image(bs, base);
1158     } else {
1159         base_bs = bdrv_find_base(bs);
1160     }
1161
1162     if (base_bs == NULL) {
1163         error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
1164         return;
1165     }
1166
1167     /* default top_bs is the active layer */
1168     top_bs = bs;
1169
1170     if (top) {
1171         if (strcmp(bs->filename, top) != 0) {
1172             top_bs = bdrv_find_backing_image(bs, top);
1173         }
1174     }
1175
1176     if (top_bs == NULL) {
1177         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
1178         return;
1179     }
1180
1181     commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
1182                 &local_err);
1183     if (local_err != NULL) {
1184         error_propagate(errp, local_err);
1185         return;
1186     }
1187     /* Grab a reference so hotplug does not delete the BlockDriverState from
1188      * underneath us.
1189      */
1190     drive_get_ref(drive_get_by_blockdev(bs));
1191 }
1192
1193 static BlockJob *find_block_job(const char *device)
1194 {
1195     BlockDriverState *bs;
1196
1197     bs = bdrv_find(device);
1198     if (!bs || !bs->job) {
1199         return NULL;
1200     }
1201     return bs->job;
1202 }
1203
1204 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
1205 {
1206     BlockJob *job = find_block_job(device);
1207
1208     if (!job) {
1209         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
1210         return;
1211     }
1212
1213     block_job_set_speed(job, speed, errp);
1214 }
1215
1216 void qmp_block_job_cancel(const char *device, Error **errp)
1217 {
1218     BlockJob *job = find_block_job(device);
1219
1220     if (!job) {
1221         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
1222         return;
1223     }
1224     if (job->paused) {
1225         error_set(errp, QERR_BLOCK_JOB_PAUSED, device);
1226         return;
1227     }
1228
1229     trace_qmp_block_job_cancel(job);
1230     block_job_cancel(job);
1231 }
1232
1233 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
1234 {
1235     BlockJobInfoList **prev = opaque;
1236     BlockJob *job = bs->job;
1237
1238     if (job) {
1239         BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
1240         elem->value = block_job_query(bs->job);
1241         (*prev)->next = elem;
1242         *prev = elem;
1243     }
1244 }
1245
1246 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
1247 {
1248     /* Dummy is a fake list element for holding the head pointer */
1249     BlockJobInfoList dummy = {};
1250     BlockJobInfoList *prev = &dummy;
1251     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
1252     return dummy.next;
1253 }