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     on_write_error = BLOCK_ERR_STOP_ENOSPC;
 436     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
 437         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) {
 438             error_report("werror is not supported by this bus type");
 439             return NULL;
 440         }
 441
 442         on_write_error = parse_block_error_action(buf, 0);
 443         if (on_write_error < 0) {
 444             return NULL;
 445         }
 446     }
 447
 448     on_read_error = BLOCK_ERR_REPORT;
 449     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
 450         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) {
 451             error_report("rerror is not supported by this bus type");
 452             return NULL;
 453         }
 454
 455         on_read_error = parse_block_error_action(buf, 1);
 456         if (on_read_error < 0) {
 457             return NULL;
 458         }
 459     }
 460
 461     if ((devaddr = qemu_opt_get(opts, "addr")) != NULL) {
 462         if (type != IF_VIRTIO) {
 463             error_report("addr is not supported by this bus type");
 464             return NULL;
 465         }
 466     }
 467
 468     /* compute bus and unit according index */
 469
 470     if (index != -1) {
 471         if (bus_id != 0 || unit_id != -1) {
 472             error_report("index cannot be used with bus and unit");
 473             return NULL;
 474         }
 475         bus_id = drive_index_to_bus_id(type, index);
 476         unit_id = drive_index_to_unit_id(type, index);
 477     }
 478
 479     /* if user doesn't specify a unit_id,
 480      * try to find the first free
 481      */
 482
 483     if (unit_id == -1) {
 484        unit_id = 0;
 485        while (drive_get(type, bus_id, unit_id) != NULL) {
 486            unit_id++;
 487            if (max_devs && unit_id >= max_devs) {
 488                unit_id -= max_devs;
 489                bus_id++;
 490            }
 491        }
 492     }
 493
 494     /* check unit id */
 495
 496     if (max_devs && unit_id >= max_devs) {
 497         error_report("unit %d too big (max is %d)",
 498                      unit_id, max_devs - 1);
 499         return NULL;
 500     }
 501
 502     /*
 503      * catch multiple definitions
 504      */
 505
 506     if (drive_get(type, bus_id, unit_id) != NULL) {
 507         error_report("drive with bus=%d, unit=%d (index=%d) exists",
 508                      bus_id, unit_id, index);
 509         return NULL;
 510     }
 511
 512     /* init */
 513
 514     dinfo = g_malloc0(sizeof(*dinfo));
 515     if ((buf = qemu_opts_id(opts)) != NULL) {
 516         dinfo->id = g_strdup(buf);
 517     } else {
 518         /* no id supplied -> create one */
 519         dinfo->id = g_malloc0(32);
 520         if (type == IF_IDE || type == IF_SCSI)
 521             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
 522         if (max_devs)
 523             snprintf(dinfo->id, 32, "%s%i%s%i",
 524                      if_name[type], bus_id, mediastr, unit_id);
 525         else
 526             snprintf(dinfo->id, 32, "%s%s%i",
 527                      if_name[type], mediastr, unit_id);
 528     }
 529     dinfo->bdrv = bdrv_new(dinfo->id);
 530     dinfo->bdrv->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
 531     dinfo->bdrv->read_only = ro;
 532     dinfo->devaddr = devaddr;
 533     dinfo->type = type;
 534     dinfo->bus = bus_id;
 535     dinfo->unit = unit_id;
 536     dinfo->cyls = cyls;
 537     dinfo->heads = heads;
 538     dinfo->secs = secs;
 539     dinfo->trans = translation;
 540     dinfo->opts = opts;
 541     dinfo->refcount = 1;
 542     dinfo->serial = serial;
 543     QTAILQ_INSERT_TAIL(&drives, dinfo, next);
 544
 545     bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
 546
 547     /* disk I/O throttling */
 548     bdrv_set_io_limits(dinfo->bdrv, &io_limits);
 549
 550     switch(type) {
 551     case IF_IDE:
 552     case IF_SCSI:
 553     case IF_XEN:
 554     case IF_NONE:
 555         dinfo->media_cd = media == MEDIA_CDROM;
 556         break;
 557     case IF_SD:
 558     case IF_FLOPPY:
 559     case IF_PFLASH:
 560     case IF_MTD:
 561         break;
 562     case IF_VIRTIO:
 563         /* add virtio block device */
 564         opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0, NULL);
 565         if (arch_type == QEMU_ARCH_S390X) {
 566             qemu_opt_set(opts, "driver", "virtio-blk-s390");
 567         } else {
 568             qemu_opt_set(opts, "driver", "virtio-blk-pci");
 569         }
 570         qemu_opt_set(opts, "drive", dinfo->id);
 571         if (devaddr)
 572             qemu_opt_set(opts, "addr", devaddr);
 573         break;
 574     default:
 575         abort();
 576     }
 577     if (!file || !*file) {
 578         return dinfo;
 579     }
 580     if (snapshot) {
 581         /* always use cache=unsafe with snapshot */
 582         bdrv_flags &= ~BDRV_O_CACHE_MASK;
 583         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
 584     }
 585
 586     if (copy_on_read) {
 587         bdrv_flags |= BDRV_O_COPY_ON_READ;
 588     }
 589
 590     if (runstate_check(RUN_STATE_INMIGRATE)) {
 591         bdrv_flags |= BDRV_O_INCOMING;
 592     }
 593
 594     if (media == MEDIA_CDROM) {
 595         /* CDROM is fine for any interface, don't check.  */
 596         ro = 1;
 597     } else if (ro == 1) {
 598         if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY &&
 599             type != IF_NONE && type != IF_PFLASH) {
 600             error_report("readonly not supported by this bus type");
 601             goto err;
 602         }
 603     }
 604
 605     bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
 606
 607     if (ro && copy_on_read) {
 608         error_report("warning: disabling copy_on_read on readonly drive");
 609     }
 610
 611     ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv);
 612     if (ret < 0) {
 613         error_report("could not open disk image %s: %s",
 614                      file, strerror(-ret));
 615         goto err;
 616     }
 617
 618     if (bdrv_key_required(dinfo->bdrv))
 619         autostart = 0;
 620     return dinfo;
 621
 622 err:
 623     bdrv_delete(dinfo->bdrv);
 624     g_free(dinfo->id);
 625     QTAILQ_REMOVE(&drives, dinfo, next);
 626     g_free(dinfo);
 627     return NULL;
 628 }
 629
 630 void do_commit(Monitor *mon, const QDict *qdict)
 631 {
 632     const char *device = qdict_get_str(qdict, "device");
 633     BlockDriverState *bs;
 634     int ret;
 635
 636     if (!strcmp(device, "all")) {
 637         ret = bdrv_commit_all();
 638         if (ret == -EBUSY) {
 639             qerror_report(QERR_DEVICE_IN_USE, device);
 640             return;
 641         }
 642     } else {
 643         bs = bdrv_find(device);
 644         if (!bs) {
 645             qerror_report(QERR_DEVICE_NOT_FOUND, device);
 646             return;
 647         }
 648         ret = bdrv_commit(bs);
 649         if (ret == -EBUSY) {
 650             qerror_report(QERR_DEVICE_IN_USE, device);
 651             return;
 652         }
 653     }
 654 }
 655
 656 static void blockdev_do_action(int kind, void *data, Error **errp)
 657 {
 658     BlockdevAction action;
 659     BlockdevActionList list;
 660
 661     action.kind = kind;
 662     action.data = data;
 663     list.value = &action;
 664     list.next = NULL;
 665     qmp_transaction(&list, errp);
 666 }
 667
 668 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
 669                                 bool has_format, const char *format,
 670                                 bool has_mode, enum NewImageMode mode,
 671                                 Error **errp)
 672 {
 673     BlockdevSnapshot snapshot = {
 674         .device = (char *) device,
 675         .snapshot_file = (char *) snapshot_file,
 676         .has_format = has_format,
 677         .format = (char *) format,
 678         .has_mode = has_mode,
 679         .mode = mode,
 680     };
 681     blockdev_do_action(BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC, &snapshot,
 682                        errp);
 683 }
 684
 685
 686 /* New and old BlockDriverState structs for group snapshots */
 687 typedef struct BlkTransactionStates {
 688     BlockDriverState *old_bs;
 689     BlockDriverState *new_bs;
 690     QSIMPLEQ_ENTRY(BlkTransactionStates) entry;
 691 } BlkTransactionStates;
 692
 693 /*
 694  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
 695  *  then we do not pivot any of the devices in the group, and abandon the
 696  *  snapshots
 697  */
 698 void qmp_transaction(BlockdevActionList *dev_list, Error **errp)
 699 {
 700     int ret = 0;
 701     BlockdevActionList *dev_entry = dev_list;
 702     BlkTransactionStates *states, *next;
 703
 704     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionStates) snap_bdrv_states;
 705     QSIMPLEQ_INIT(&snap_bdrv_states);
 706
 707     /* drain all i/o before any snapshots */
 708     bdrv_drain_all();
 709
 710     /* We don't do anything in this loop that commits us to the snapshot */
 711     while (NULL != dev_entry) {
 712         BlockdevAction *dev_info = NULL;
 713         BlockDriver *proto_drv;
 714         BlockDriver *drv;
 715         int flags;
 716         enum NewImageMode mode;
 717         const char *new_image_file;
 718         const char *device;
 719         const char *format = "qcow2";
 720
 721         dev_info = dev_entry->value;
 722         dev_entry = dev_entry->next;
 723
 724         states = g_malloc0(sizeof(BlkTransactionStates));
 725         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, states, entry);
 726
 727         switch (dev_info->kind) {
 728         case BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
 729             device = dev_info->blockdev_snapshot_sync->device;
 730             if (!dev_info->blockdev_snapshot_sync->has_mode) {
 731                 dev_info->blockdev_snapshot_sync->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
 732             }
 733             new_image_file = dev_info->blockdev_snapshot_sync->snapshot_file;
 734             if (dev_info->blockdev_snapshot_sync->has_format) {
 735                 format = dev_info->blockdev_snapshot_sync->format;
 736             }
 737             mode = dev_info->blockdev_snapshot_sync->mode;
 738             break;
 739         default:
 740             abort();
 741         }
 742
 743         drv = bdrv_find_format(format);
 744         if (!drv) {
 745             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 746             goto delete_and_fail;
 747         }
 748
 749         states->old_bs = bdrv_find(device);
 750         if (!states->old_bs) {
 751             error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 752             goto delete_and_fail;
 753         }
 754
 755         if (!bdrv_is_inserted(states->old_bs)) {
 756             error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
 757             goto delete_and_fail;
 758         }
 759
 760         if (bdrv_in_use(states->old_bs)) {
 761             error_set(errp, QERR_DEVICE_IN_USE, device);
 762             goto delete_and_fail;
 763         }
 764
 765         if (!bdrv_is_read_only(states->old_bs)) {
 766             if (bdrv_flush(states->old_bs)) {
 767                 error_set(errp, QERR_IO_ERROR);
 768                 goto delete_and_fail;
 769             }
 770         }
 771
 772         flags = states->old_bs->open_flags;
 773
 774         proto_drv = bdrv_find_protocol(new_image_file);
 775         if (!proto_drv) {
 776             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 777             goto delete_and_fail;
 778         }
 779
 780         /* create new image w/backing file */
 781         if (mode != NEW_IMAGE_MODE_EXISTING) {
 782             ret = bdrv_img_create(new_image_file, format,
 783                                   states->old_bs->filename,
 784                                   states->old_bs->drv->format_name,
 785                                   NULL, -1, flags);
 786             if (ret) {
 787                 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 788                 goto delete_and_fail;
 789             }
 790         }
 791
 792         /* We will manually add the backing_hd field to the bs later */
 793         states->new_bs = bdrv_new("");
 794         ret = bdrv_open(states->new_bs, new_image_file,
 795                         flags | BDRV_O_NO_BACKING, drv);
 796         if (ret != 0) {
 797             error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 798             goto delete_and_fail;
 799         }
 800     }
 801
 802
 803     /* Now we are going to do the actual pivot.  Everything up to this point
 804      * is reversible, but we are committed at this point */
 805     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 806         /* This removes our old bs from the bdrv_states, and adds the new bs */
 807         bdrv_append(states->new_bs, states->old_bs);
 808         /* We don't need (or want) to use the transactional
 809          * bdrv_reopen_multiple() across all the entries at once, because we
 810          * don't want to abort all of them if one of them fails the reopen */
 811         bdrv_reopen(states->new_bs, states->new_bs->open_flags & ~BDRV_O_RDWR,
 812                     NULL);
 813     }
 814
 815     /* success */
 816     goto exit;
 817
 818 delete_and_fail:
 819     /*
 820     * failure, and it is all-or-none; abandon each new bs, and keep using
 821     * the original bs for all images
 822     */
 823     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 824         if (states->new_bs) {
 825              bdrv_delete(states->new_bs);
 826         }
 827     }
 828 exit:
 829     QSIMPLEQ_FOREACH_SAFE(states, &snap_bdrv_states, entry, next) {
 830         g_free(states);
 831     }
 832     return;
 833 }
 834
 835
 836 static void eject_device(BlockDriverState *bs, int force, Error **errp)
 837 {
 838     if (bdrv_in_use(bs)) {
 839         error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 840         return;
 841     }
 842     if (!bdrv_dev_has_removable_media(bs)) {
 843         error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs));
 844         return;
 845     }
 846
 847     if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) {
 848         bdrv_dev_eject_request(bs, force);
 849         if (!force) {
 850             error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
 851             return;
 852         }
 853     }
 854
 855     bdrv_close(bs);
 856 }
 857
 858 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
 859 {
 860     BlockDriverState *bs;
 861
 862     bs = bdrv_find(device);
 863     if (!bs) {
 864         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 865         return;
 866     }
 867
 868     eject_device(bs, force, errp);
 869 }
 870
 871 void qmp_block_passwd(const char *device, const char *password, Error **errp)
 872 {
 873     BlockDriverState *bs;
 874     int err;
 875
 876     bs = bdrv_find(device);
 877     if (!bs) {
 878         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 879         return;
 880     }
 881
 882     err = bdrv_set_key(bs, password);
 883     if (err == -EINVAL) {
 884         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
 885         return;
 886     } else if (err < 0) {
 887         error_set(errp, QERR_INVALID_PASSWORD);
 888         return;
 889     }
 890 }
 891
 892 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
 893                                     int bdrv_flags, BlockDriver *drv,
 894                                     const char *password, Error **errp)
 895 {
 896     if (bdrv_open(bs, filename, bdrv_flags, drv) < 0) {
 897         error_set(errp, QERR_OPEN_FILE_FAILED, filename);
 898         return;
 899     }
 900
 901     if (bdrv_key_required(bs)) {
 902         if (password) {
 903             if (bdrv_set_key(bs, password) < 0) {
 904                 error_set(errp, QERR_INVALID_PASSWORD);
 905             }
 906         } else {
 907             error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs),
 908                       bdrv_get_encrypted_filename(bs));
 909         }
 910     } else if (password) {
 911         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
 912     }
 913 }
 914
 915 void qmp_change_blockdev(const char *device, const char *filename,
 916                          bool has_format, const char *format, Error **errp)
 917 {
 918     BlockDriverState *bs;
 919     BlockDriver *drv = NULL;
 920     int bdrv_flags;
 921     Error *err = NULL;
 922
 923     bs = bdrv_find(device);
 924     if (!bs) {
 925         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 926         return;
 927     }
 928
 929     if (format) {
 930         drv = bdrv_find_whitelisted_format(format);
 931         if (!drv) {
 932             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 933             return;
 934         }
 935     }
 936
 937     eject_device(bs, 0, &err);
 938     if (error_is_set(&err)) {
 939         error_propagate(errp, err);
 940         return;
 941     }
 942
 943     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
 944     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
 945
 946     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
 947 }
 948
 949 /* throttling disk I/O limits */
 950 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
 951                                int64_t bps_wr, int64_t iops, int64_t iops_rd,
 952                                int64_t iops_wr, Error **errp)
 953 {
 954     BlockIOLimit io_limits;
 955     BlockDriverState *bs;
 956
 957     bs = bdrv_find(device);
 958     if (!bs) {
 959         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 960         return;
 961     }
 962
 963     io_limits.bps[BLOCK_IO_LIMIT_TOTAL] = bps;
 964     io_limits.bps[BLOCK_IO_LIMIT_READ]  = bps_rd;
 965     io_limits.bps[BLOCK_IO_LIMIT_WRITE] = bps_wr;
 966     io_limits.iops[BLOCK_IO_LIMIT_TOTAL]= iops;
 967     io_limits.iops[BLOCK_IO_LIMIT_READ] = iops_rd;
 968     io_limits.iops[BLOCK_IO_LIMIT_WRITE]= iops_wr;
 969
 970     if (!do_check_io_limits(&io_limits)) {
 971         error_set(errp, QERR_INVALID_PARAMETER_COMBINATION);
 972         return;
 973     }
 974
 975     bs->io_limits = io_limits;
 976     bs->slice_time = BLOCK_IO_SLICE_TIME;
 977
 978     if (!bs->io_limits_enabled && bdrv_io_limits_enabled(bs)) {
 979         bdrv_io_limits_enable(bs);
 980     } else if (bs->io_limits_enabled && !bdrv_io_limits_enabled(bs)) {
 981         bdrv_io_limits_disable(bs);
 982     } else {
 983         if (bs->block_timer) {
 984             qemu_mod_timer(bs->block_timer, qemu_get_clock_ns(vm_clock));
 985         }
 986     }
 987 }
 988
 989 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
 990 {
 991     const char *id = qdict_get_str(qdict, "id");
 992     BlockDriverState *bs;
 993
 994     bs = bdrv_find(id);
 995     if (!bs) {
 996         qerror_report(QERR_DEVICE_NOT_FOUND, id);
 997         return -1;
 998     }
 999     if (bdrv_in_use(bs)) {
1000         qerror_report(QERR_DEVICE_IN_USE, id);
1001         return -1;
1002     }
1003
1004     /* quiesce block driver; prevent further io */
1005     bdrv_drain_all();
1006     bdrv_flush(bs);
1007     bdrv_close(bs);
1008
1009     /* if we have a device attached to this BlockDriverState
1010      * then we need to make the drive anonymous until the device
1011      * can be removed.  If this is a drive with no device backing
1012      * then we can just get rid of the block driver state right here.
1013      */
1014     if (bdrv_get_attached_dev(bs)) {
1015         bdrv_make_anon(bs);
1016     } else {
1017         drive_uninit(drive_get_by_blockdev(bs));
1018     }
1019
1020     return 0;
1021 }
1022
1023 void qmp_block_resize(const char *device, int64_t size, Error **errp)
1024 {
1025     BlockDriverState *bs;
1026
1027     bs = bdrv_find(device);
1028     if (!bs) {
1029         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1030         return;
1031     }
1032
1033     if (size < 0) {
1034         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
1035         return;
1036     }
1037
1038     switch (bdrv_truncate(bs, size)) {
1039     case 0:
1040         break;
1041     case -ENOMEDIUM:
1042         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1043         break;
1044     case -ENOTSUP:
1045         error_set(errp, QERR_UNSUPPORTED);
1046         break;
1047     case -EACCES:
1048         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1049         break;
1050     case -EBUSY:
1051         error_set(errp, QERR_DEVICE_IN_USE, device);
1052         break;
1053     default:
1054         error_set(errp, QERR_UNDEFINED_ERROR);
1055         break;
1056     }
1057 }
1058
1059 static QObject *qobject_from_block_job(BlockJob *job)
1060 {
1061     return qobject_from_jsonf("{ 'type': %s,"
1062                               "'device': %s,"
1063                               "'len': %" PRId64 ","
1064                               "'offset': %" PRId64 ","
1065                               "'speed': %" PRId64 " }",
1066                               job->job_type->job_type,
1067                               bdrv_get_device_name(job->bs),
1068                               job->len,
1069                               job->offset,
1070                               job->speed);
1071 }
1072
1073 static void block_job_cb(void *opaque, int ret)
1074 {
1075     BlockDriverState *bs = opaque;
1076     QObject *obj;
1077
1078     trace_block_job_cb(bs, bs->job, ret);
1079
1080     assert(bs->job);
1081     obj = qobject_from_block_job(bs->job);
1082     if (ret < 0) {
1083         QDict *dict = qobject_to_qdict(obj);
1084         qdict_put(dict, "error", qstring_from_str(strerror(-ret)));
1085     }
1086
1087     if (block_job_is_cancelled(bs->job)) {
1088         monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj);
1089     } else {
1090         monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj);
1091     }
1092     qobject_decref(obj);
1093
1094     drive_put_ref_bh_schedule(drive_get_by_blockdev(bs));
1095 }
1096
1097 void qmp_block_stream(const char *device, bool has_base,
1098                       const char *base, bool has_speed,
1099                       int64_t speed, Error **errp)
1100 {
1101     BlockDriverState *bs;
1102     BlockDriverState *base_bs = NULL;
1103     Error *local_err = NULL;
1104
1105     bs = bdrv_find(device);
1106     if (!bs) {
1107         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1108         return;
1109     }
1110
1111     if (base) {
1112         base_bs = bdrv_find_backing_image(bs, base);
1113         if (base_bs == NULL) {
1114             error_set(errp, QERR_BASE_NOT_FOUND, base);
1115             return;
1116         }
1117     }
1118
1119     stream_start(bs, base_bs, base, has_speed ? speed : 0,
1120                  block_job_cb, bs, &local_err);
1121     if (error_is_set(&local_err)) {
1122         error_propagate(errp, local_err);
1123         return;
1124     }
1125
1126     /* Grab a reference so hotplug does not delete the BlockDriverState from
1127      * underneath us.
1128      */
1129     drive_get_ref(drive_get_by_blockdev(bs));
1130
1131     trace_qmp_block_stream(bs, bs->job);
1132 }
1133
1134 void qmp_block_commit(const char *device,
1135                       bool has_base, const char *base, const char *top,
1136                       bool has_speed, int64_t speed,
1137                       Error **errp)
1138 {
1139     BlockDriverState *bs;
1140     BlockDriverState *base_bs, *top_bs;
1141     Error *local_err = NULL;
1142     /* This will be part of the QMP command, if/when the
1143      * BlockdevOnError change for blkmirror makes it in
1144      */
1145     BlockErrorAction on_error = BLOCK_ERR_REPORT;
1146
1147     /* drain all i/o before commits */
1148     bdrv_drain_all();
1149
1150     bs = bdrv_find(device);
1151     if (!bs) {
1152         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1153         return;
1154     }
1155     if (base && has_base) {
1156         base_bs = bdrv_find_backing_image(bs, base);
1157     } else {
1158         base_bs = bdrv_find_base(bs);
1159     }
1160
1161     if (base_bs == NULL) {
1162         error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
1163         return;
1164     }
1165
1166     /* default top_bs is the active layer */
1167     top_bs = bs;
1168
1169     if (top) {
1170         if (strcmp(bs->filename, top) != 0) {
1171             top_bs = bdrv_find_backing_image(bs, top);
1172         }
1173     }
1174
1175     if (top_bs == NULL) {
1176         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
1177         return;
1178     }
1179
1180     commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
1181                 &local_err);
1182     if (local_err != NULL) {
1183         error_propagate(errp, local_err);
1184         return;
1185     }
1186     /* Grab a reference so hotplug does not delete the BlockDriverState from
1187      * underneath us.
1188      */
1189     drive_get_ref(drive_get_by_blockdev(bs));
1190 }
1191
1192 static BlockJob *find_block_job(const char *device)
1193 {
1194     BlockDriverState *bs;
1195
1196     bs = bdrv_find(device);
1197     if (!bs || !bs->job) {
1198         return NULL;
1199     }
1200     return bs->job;
1201 }
1202
1203 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
1204 {
1205     BlockJob *job = find_block_job(device);
1206
1207     if (!job) {
1208         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1209         return;
1210     }
1211
1212     block_job_set_speed(job, speed, errp);
1213 }
1214
1215 void qmp_block_job_cancel(const char *device, Error **errp)
1216 {
1217     BlockJob *job = find_block_job(device);
1218
1219     if (!job) {
1220         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1221         return;
1222     }
1223
1224     trace_qmp_block_job_cancel(job);
1225     block_job_cancel(job);
1226 }
1227
1228 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
1229 {
1230     BlockJobInfoList **prev = opaque;
1231     BlockJob *job = bs->job;
1232
1233     if (job) {
1234         BlockJobInfoList *elem;
1235         BlockJobInfo *info = g_new(BlockJobInfo, 1);
1236         *info = (BlockJobInfo){
1237             .type   = g_strdup(job->job_type->job_type),
1238             .device = g_strdup(bdrv_get_device_name(bs)),
1239             .len    = job->len,
1240             .offset = job->offset,
1241             .speed  = job->speed,
1242         };
1243
1244         elem = g_new0(BlockJobInfoList, 1);
1245         elem->value = info;
1246
1247         (*prev)->next = elem;
1248         *prev = elem;
1249     }
1250 }
1251
1252 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
1253 {
1254     /* Dummy is a fake list element for holding the head pointer */
1255     BlockJobInfoList dummy = {};
1256     BlockJobInfoList *prev = &dummy;
1257     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
1258     return dummy.next;
1259 }