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 (dinfo) {
  68         dinfo->auto_del = 1;
  69     }
  70 }
  71
  72 void blockdev_auto_del(BlockDriverState *bs)
  73 {
  74     DriveInfo *dinfo = drive_get_by_blockdev(bs);
  75
  76     if (dinfo && dinfo->auto_del) {
  77         drive_put_ref(dinfo);
  78     }
  79 }
  80
  81 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
  82 {
  83     int max_devs = if_max_devs[type];
  84     return max_devs ? index / max_devs : 0;
  85 }
  86
  87 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
  88 {
  89     int max_devs = if_max_devs[type];
  90     return max_devs ? index % max_devs : index;
  91 }
  92
  93 QemuOpts *drive_def(const char *optstr)
  94 {
  95     return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
  96 }
  97
  98 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
  99                     const char *optstr)
 100 {
 101     QemuOpts *opts;
 102     char buf[32];
 103
 104     opts = drive_def(optstr);
 105     if (!opts) {
 106         return NULL;
 107     }
 108     if (type != IF_DEFAULT) {
 109         qemu_opt_set(opts, "if", if_name[type]);
 110     }
 111     if (index >= 0) {
 112         snprintf(buf, sizeof(buf), "%d", index);
 113         qemu_opt_set(opts, "index", buf);
 114     }
 115     if (file)
 116         qemu_opt_set(opts, "file", file);
 117     return opts;
 118 }
 119
 120 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
 121 {
 122     DriveInfo *dinfo;
 123
 124     /* seek interface, bus and unit */
 125
 126     QTAILQ_FOREACH(dinfo, &drives, next) {
 127         if (dinfo->type == type &&
 128             dinfo->bus == bus &&
 129             dinfo->unit == unit)
 130             return dinfo;
 131     }
 132
 133     return NULL;
 134 }
 135
 136 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
 137 {
 138     return drive_get(type,
 139                      drive_index_to_bus_id(type, index),
 140                      drive_index_to_unit_id(type, index));
 141 }
 142
 143 int drive_get_max_bus(BlockInterfaceType type)
 144 {
 145     int max_bus;
 146     DriveInfo *dinfo;
 147
 148     max_bus = -1;
 149     QTAILQ_FOREACH(dinfo, &drives, next) {
 150         if(dinfo->type == type &&
 151            dinfo->bus > max_bus)
 152             max_bus = dinfo->bus;
 153     }
 154     return max_bus;
 155 }
 156
 157 /* Get a block device.  This should only be used for single-drive devices
 158    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
 159    appropriate bus.  */
 160 DriveInfo *drive_get_next(BlockInterfaceType type)
 161 {
 162     static int next_block_unit[IF_COUNT];
 163
 164     return drive_get(type, 0, next_block_unit[type]++);
 165 }
 166
 167 DriveInfo *drive_get_by_blockdev(BlockDriverState *bs)
 168 {
 169     DriveInfo *dinfo;
 170
 171     QTAILQ_FOREACH(dinfo, &drives, next) {
 172         if (dinfo->bdrv == bs) {
 173             return dinfo;
 174         }
 175     }
 176     return NULL;
 177 }
 178
 179 static void bdrv_format_print(void *opaque, const char *name)
 180 {
 181     error_printf(" %s", name);
 182 }
 183
 184 static void drive_uninit(DriveInfo *dinfo)
 185 {
 186     qemu_opts_del(dinfo->opts);
 187     bdrv_delete(dinfo->bdrv);
 188     g_free(dinfo->id);
 189     QTAILQ_REMOVE(&drives, dinfo, next);
 190     g_free(dinfo);
 191 }
 192
 193 void drive_put_ref(DriveInfo *dinfo)
 194 {
 195     assert(dinfo->refcount);
 196     if (--dinfo->refcount == 0) {
 197         drive_uninit(dinfo);
 198     }
 199 }
 200
 201 void drive_get_ref(DriveInfo *dinfo)
 202 {
 203     dinfo->refcount++;
 204 }
 205
 206 typedef struct {
 207     QEMUBH *bh;
 208     DriveInfo *dinfo;
 209 } DrivePutRefBH;
 210
 211 static void drive_put_ref_bh(void *opaque)
 212 {
 213     DrivePutRefBH *s = opaque;
 214
 215     drive_put_ref(s->dinfo);
 216     qemu_bh_delete(s->bh);
 217     g_free(s);
 218 }
 219
 220 /*
 221  * Release a drive reference in a BH
 222  *
 223  * It is not possible to use drive_put_ref() from a callback function when the
 224  * callers still need the drive.  In such cases we schedule a BH to release the
 225  * reference.
 226  */
 227 static void drive_put_ref_bh_schedule(DriveInfo *dinfo)
 228 {
 229     DrivePutRefBH *s;
 230
 231     s = g_new(DrivePutRefBH, 1);
 232     s->bh = qemu_bh_new(drive_put_ref_bh, s);
 233     s->dinfo = dinfo;
 234     qemu_bh_schedule(s->bh);
 235 }
 236
 237 static int parse_block_error_action(const char *buf, int is_read)
 238 {
 239     if (!strcmp(buf, "ignore")) {
 240         return BLOCK_ERR_IGNORE;
 241     } else if (!is_read && !strcmp(buf, "enospc")) {
 242         return BLOCK_ERR_STOP_ENOSPC;
 243     } else if (!strcmp(buf, "stop")) {
 244         return BLOCK_ERR_STOP_ANY;
 245     } else if (!strcmp(buf, "report")) {
 246         return BLOCK_ERR_REPORT;
 247     } else {
 248         error_report("'%s' invalid %s error action",
 249                      buf, is_read ? "read" : "write");
 250         return -1;
 251     }
 252 }
 253
 254 static bool do_check_io_limits(BlockIOLimit *io_limits)
 255 {
 256     bool bps_flag;
 257     bool iops_flag;
 258
 259     assert(io_limits);
 260
 261     bps_flag  = (io_limits->bps[BLOCK_IO_LIMIT_TOTAL] != 0)
 262                  && ((io_limits->bps[BLOCK_IO_LIMIT_READ] != 0)
 263                  || (io_limits->bps[BLOCK_IO_LIMIT_WRITE] != 0));
 264     iops_flag = (io_limits->iops[BLOCK_IO_LIMIT_TOTAL] != 0)
 265                  && ((io_limits->iops[BLOCK_IO_LIMIT_READ] != 0)
 266                  || (io_limits->iops[BLOCK_IO_LIMIT_WRITE] != 0));
 267     if (bps_flag || iops_flag) {
 268         return false;
 269     }
 270
 271     return true;
 272 }
 273
 274 DriveInfo *drive_init(QemuOpts *opts, int default_to_scsi)
 275 {
 276     const char *buf;
 277     const char *file = NULL;
 278     char devname[128];
 279     const char *serial;
 280     const char *mediastr = "";
 281     BlockInterfaceType type;
 282     enum { MEDIA_DISK, MEDIA_CDROM } media;
 283     int bus_id, unit_id;
 284     int cyls, heads, secs, translation;
 285     BlockDriver *drv = NULL;
 286     int max_devs;
 287     int index;
 288     int ro = 0;
 289     int bdrv_flags = 0;
 290     int on_read_error, on_write_error;
 291     const char *devaddr;
 292     DriveInfo *dinfo;
 293     BlockIOLimit io_limits;
 294     int snapshot = 0;
 295     bool copy_on_read;
 296     int ret;
 297
 298     translation = BIOS_ATA_TRANSLATION_AUTO;
 299     media = MEDIA_DISK;
 300
 301     /* extract parameters */
 302     bus_id  = qemu_opt_get_number(opts, "bus", 0);
 303     unit_id = qemu_opt_get_number(opts, "unit", -1);
 304     index   = qemu_opt_get_number(opts, "index", -1);
 305
 306     cyls  = qemu_opt_get_number(opts, "cyls", 0);
 307     heads = qemu_opt_get_number(opts, "heads", 0);
 308     secs  = qemu_opt_get_number(opts, "secs", 0);
 309
 310     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
 311     ro = qemu_opt_get_bool(opts, "readonly", 0);
 312     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
 313
 314     file = qemu_opt_get(opts, "file");
 315     serial = qemu_opt_get(opts, "serial");
 316
 317     if ((buf = qemu_opt_get(opts, "if")) != NULL) {
 318         pstrcpy(devname, sizeof(devname), buf);
 319         for (type = 0; type < IF_COUNT && strcmp(buf, if_name[type]); type++)
 320             ;
 321         if (type == IF_COUNT) {
 322             error_report("unsupported bus type '%s'", buf);
 323             return NULL;
 324         }
 325     } else {
 326         type = default_to_scsi ? IF_SCSI : IF_IDE;
 327         pstrcpy(devname, sizeof(devname), if_name[type]);
 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                      devname, bus_id, mediastr, unit_id);
 524         else
 525             snprintf(dinfo->id, 32, "%s%s%i",
 526                      devname, 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->opts = opts;
 534     dinfo->refcount = 1;
 535     if (serial)
 536         strncpy(dinfo->serial, serial, sizeof(dinfo->serial) - 1);
 537     QTAILQ_INSERT_TAIL(&drives, dinfo, next);
 538
 539     bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
 540
 541     /* disk I/O throttling */
 542     bdrv_set_io_limits(dinfo->bdrv, &io_limits);
 543
 544     switch(type) {
 545     case IF_IDE:
 546     case IF_SCSI:
 547     case IF_XEN:
 548     case IF_NONE:
 549         switch(media) {
 550         case MEDIA_DISK:
 551             if (cyls != 0) {
 552                 bdrv_set_geometry_hint(dinfo->bdrv, cyls, heads, secs);
 553                 bdrv_set_translation_hint(dinfo->bdrv, translation);
 554             }
 555             break;
 556         case MEDIA_CDROM:
 557             dinfo->media_cd = 1;
 558             break;
 559         }
 560         break;
 561     case IF_SD:
 562     case IF_FLOPPY:
 563     case IF_PFLASH:
 564     case IF_MTD:
 565         break;
 566     case IF_VIRTIO:
 567         /* add virtio block device */
 568         opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0);
 569         if (arch_type == QEMU_ARCH_S390X) {
 570             qemu_opt_set(opts, "driver", "virtio-blk-s390");
 571         } else {
 572             qemu_opt_set(opts, "driver", "virtio-blk-pci");
 573         }
 574         qemu_opt_set(opts, "drive", dinfo->id);
 575         if (devaddr)
 576             qemu_opt_set(opts, "addr", devaddr);
 577         break;
 578     default:
 579         abort();
 580     }
 581     if (!file || !*file) {
 582         return dinfo;
 583     }
 584     if (snapshot) {
 585         /* always use cache=unsafe with snapshot */
 586         bdrv_flags &= ~BDRV_O_CACHE_MASK;
 587         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
 588     }
 589
 590     if (copy_on_read) {
 591         bdrv_flags |= BDRV_O_COPY_ON_READ;
 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     ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv);
 608     if (ret < 0) {
 609         error_report("could not open disk image %s: %s",
 610                      file, strerror(-ret));
 611         goto err;
 612     }
 613
 614     if (bdrv_key_required(dinfo->bdrv))
 615         autostart = 0;
 616     return dinfo;
 617
 618 err:
 619     bdrv_delete(dinfo->bdrv);
 620     g_free(dinfo->id);
 621     QTAILQ_REMOVE(&drives, dinfo, next);
 622     g_free(dinfo);
 623     return NULL;
 624 }
 625
 626 void do_commit(Monitor *mon, const QDict *qdict)
 627 {
 628     const char *device = qdict_get_str(qdict, "device");
 629     BlockDriverState *bs;
 630     int ret;
 631
 632     if (!strcmp(device, "all")) {
 633         ret = bdrv_commit_all();
 634         if (ret == -EBUSY) {
 635             qerror_report(QERR_DEVICE_IN_USE, device);
 636             return;
 637         }
 638     } else {
 639         bs = bdrv_find(device);
 640         if (!bs) {
 641             qerror_report(QERR_DEVICE_NOT_FOUND, device);
 642             return;
 643         }
 644         ret = bdrv_commit(bs);
 645         if (ret == -EBUSY) {
 646             qerror_report(QERR_DEVICE_IN_USE, device);
 647             return;
 648         }
 649     }
 650 }
 651
 652 static void blockdev_do_action(int kind, void *data, Error **errp)
 653 {
 654     BlockdevAction action;
 655     BlockdevActionList list;
 656
 657     action.kind = kind;
 658     action.data = data;
 659     list.value = &action;
 660     list.next = NULL;
 661     qmp_transaction(&list, errp);
 662 }
 663
 664 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
 665                                 bool has_format, const char *format,
 666                                 bool has_mode, enum NewImageMode mode,
 667                                 Error **errp)
 668 {
 669     BlockdevSnapshot snapshot = {
 670         .device = (char *) device,
 671         .snapshot_file = (char *) snapshot_file,
 672         .has_format = has_format,
 673         .format = (char *) format,
 674         .has_mode = has_mode,
 675         .mode = mode,
 676     };
 677     blockdev_do_action(BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC, &snapshot,
 678                        errp);
 679 }
 680
 681
 682 /* New and old BlockDriverState structs for group snapshots */
 683 typedef struct BlkTransactionStates {
 684     BlockDriverState *old_bs;
 685     BlockDriverState *new_bs;
 686     QSIMPLEQ_ENTRY(BlkTransactionStates) entry;
 687 } BlkTransactionStates;
 688
 689 /*
 690  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
 691  *  then we do not pivot any of the devices in the group, and abandon the
 692  *  snapshots
 693  */
 694 void qmp_transaction(BlockdevActionList *dev_list, Error **errp)
 695 {
 696     int ret = 0;
 697     BlockdevActionList *dev_entry = dev_list;
 698     BlkTransactionStates *states, *next;
 699
 700     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionStates) snap_bdrv_states;
 701     QSIMPLEQ_INIT(&snap_bdrv_states);
 702
 703     /* drain all i/o before any snapshots */
 704     bdrv_drain_all();
 705
 706     /* We don't do anything in this loop that commits us to the snapshot */
 707     while (NULL != dev_entry) {
 708         BlockdevAction *dev_info = NULL;
 709         BlockDriver *proto_drv;
 710         BlockDriver *drv;
 711         int flags;
 712         enum NewImageMode mode;
 713         const char *new_image_file;
 714         const char *device;
 715         const char *format = "qcow2";
 716
 717         dev_info = dev_entry->value;
 718         dev_entry = dev_entry->next;
 719
 720         states = g_malloc0(sizeof(BlkTransactionStates));
 721         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, states, entry);
 722
 723         switch (dev_info->kind) {
 724         case BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
 725             device = dev_info->blockdev_snapshot_sync->device;
 726             if (!dev_info->blockdev_snapshot_sync->has_mode) {
 727                 dev_info->blockdev_snapshot_sync->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
 728             }
 729             new_image_file = dev_info->blockdev_snapshot_sync->snapshot_file;
 730             if (dev_info->blockdev_snapshot_sync->has_format) {
 731                 format = dev_info->blockdev_snapshot_sync->format;
 732             }
 733             mode = dev_info->blockdev_snapshot_sync->mode;
 734             break;
 735         default:
 736             abort();
 737         }
 738
 739         drv = bdrv_find_format(format);
 740         if (!drv) {
 741             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 742             goto delete_and_fail;
 743         }
 744
 745         states->old_bs = bdrv_find(device);
 746         if (!states->old_bs) {
 747             error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 748             goto delete_and_fail;
 749         }
 750
 751         if (bdrv_in_use(states->old_bs)) {
 752             error_set(errp, QERR_DEVICE_IN_USE, device);
 753             goto delete_and_fail;
 754         }
 755
 756         if (!bdrv_is_read_only(states->old_bs) &&
 757              bdrv_is_inserted(states->old_bs)) {
 758
 759             if (bdrv_flush(states->old_bs)) {
 760                 error_set(errp, QERR_IO_ERROR);
 761                 goto delete_and_fail;
 762             }
 763         }
 764
 765         flags = states->old_bs->open_flags;
 766
 767         proto_drv = bdrv_find_protocol(new_image_file);
 768         if (!proto_drv) {
 769             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 770             goto delete_and_fail;
 771         }
 772
 773         /* create new image w/backing file */
 774         if (mode != NEW_IMAGE_MODE_EXISTING) {
 775             ret = bdrv_img_create(new_image_file, format,
 776                                   states->old_bs->filename,
 777                                   states->old_bs->drv->format_name,
 778                                   NULL, -1, flags);
 779             if (ret) {
 780                 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 781                 goto delete_and_fail;
 782             }
 783         }
 784
 785         /* We will manually add the backing_hd field to the bs later */
 786         states->new_bs = bdrv_new("");
 787         ret = bdrv_open(states->new_bs, new_image_file,
 788                         flags | BDRV_O_NO_BACKING, drv);
 789         if (ret != 0) {
 790             error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file);
 791             goto delete_and_fail;
 792         }
 793     }
 794
 795
 796     /* Now we are going to do the actual pivot.  Everything up to this point
 797      * is reversible, but we are committed at this point */
 798     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 799         /* This removes our old bs from the bdrv_states, and adds the new bs */
 800         bdrv_append(states->new_bs, states->old_bs);
 801     }
 802
 803     /* success */
 804     goto exit;
 805
 806 delete_and_fail:
 807     /*
 808     * failure, and it is all-or-none; abandon each new bs, and keep using
 809     * the original bs for all images
 810     */
 811     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
 812         if (states->new_bs) {
 813              bdrv_delete(states->new_bs);
 814         }
 815     }
 816 exit:
 817     QSIMPLEQ_FOREACH_SAFE(states, &snap_bdrv_states, entry, next) {
 818         g_free(states);
 819     }
 820     return;
 821 }
 822
 823
 824 static void eject_device(BlockDriverState *bs, int force, Error **errp)
 825 {
 826     if (bdrv_in_use(bs)) {
 827         error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 828         return;
 829     }
 830     if (!bdrv_dev_has_removable_media(bs)) {
 831         error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs));
 832         return;
 833     }
 834
 835     if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) {
 836         bdrv_dev_eject_request(bs, force);
 837         if (!force) {
 838             error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
 839             return;
 840         }
 841     }
 842
 843     bdrv_close(bs);
 844 }
 845
 846 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
 847 {
 848     BlockDriverState *bs;
 849
 850     bs = bdrv_find(device);
 851     if (!bs) {
 852         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 853         return;
 854     }
 855
 856     eject_device(bs, force, errp);
 857 }
 858
 859 void qmp_block_passwd(const char *device, const char *password, Error **errp)
 860 {
 861     BlockDriverState *bs;
 862     int err;
 863
 864     bs = bdrv_find(device);
 865     if (!bs) {
 866         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 867         return;
 868     }
 869
 870     err = bdrv_set_key(bs, password);
 871     if (err == -EINVAL) {
 872         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
 873         return;
 874     } else if (err < 0) {
 875         error_set(errp, QERR_INVALID_PASSWORD);
 876         return;
 877     }
 878 }
 879
 880 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
 881                                     int bdrv_flags, BlockDriver *drv,
 882                                     const char *password, Error **errp)
 883 {
 884     if (bdrv_open(bs, filename, bdrv_flags, drv) < 0) {
 885         error_set(errp, QERR_OPEN_FILE_FAILED, filename);
 886         return;
 887     }
 888
 889     if (bdrv_key_required(bs)) {
 890         if (password) {
 891             if (bdrv_set_key(bs, password) < 0) {
 892                 error_set(errp, QERR_INVALID_PASSWORD);
 893             }
 894         } else {
 895             error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs),
 896                       bdrv_get_encrypted_filename(bs));
 897         }
 898     } else if (password) {
 899         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
 900     }
 901 }
 902
 903 void qmp_change_blockdev(const char *device, const char *filename,
 904                          bool has_format, const char *format, Error **errp)
 905 {
 906     BlockDriverState *bs;
 907     BlockDriver *drv = NULL;
 908     int bdrv_flags;
 909     Error *err = NULL;
 910
 911     bs = bdrv_find(device);
 912     if (!bs) {
 913         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 914         return;
 915     }
 916
 917     if (format) {
 918         drv = bdrv_find_whitelisted_format(format);
 919         if (!drv) {
 920             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
 921             return;
 922         }
 923     }
 924
 925     eject_device(bs, 0, &err);
 926     if (error_is_set(&err)) {
 927         error_propagate(errp, err);
 928         return;
 929     }
 930
 931     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
 932     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
 933
 934     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
 935 }
 936
 937 /* throttling disk I/O limits */
 938 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
 939                                int64_t bps_wr, int64_t iops, int64_t iops_rd,
 940                                int64_t iops_wr, Error **errp)
 941 {
 942     BlockIOLimit io_limits;
 943     BlockDriverState *bs;
 944
 945     bs = bdrv_find(device);
 946     if (!bs) {
 947         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
 948         return;
 949     }
 950
 951     io_limits.bps[BLOCK_IO_LIMIT_TOTAL] = bps;
 952     io_limits.bps[BLOCK_IO_LIMIT_READ]  = bps_rd;
 953     io_limits.bps[BLOCK_IO_LIMIT_WRITE] = bps_wr;
 954     io_limits.iops[BLOCK_IO_LIMIT_TOTAL]= iops;
 955     io_limits.iops[BLOCK_IO_LIMIT_READ] = iops_rd;
 956     io_limits.iops[BLOCK_IO_LIMIT_WRITE]= iops_wr;
 957
 958     if (!do_check_io_limits(&io_limits)) {
 959         error_set(errp, QERR_INVALID_PARAMETER_COMBINATION);
 960         return;
 961     }
 962
 963     bs->io_limits = io_limits;
 964     bs->slice_time = BLOCK_IO_SLICE_TIME;
 965
 966     if (!bs->io_limits_enabled && bdrv_io_limits_enabled(bs)) {
 967         bdrv_io_limits_enable(bs);
 968     } else if (bs->io_limits_enabled && !bdrv_io_limits_enabled(bs)) {
 969         bdrv_io_limits_disable(bs);
 970     } else {
 971         if (bs->block_timer) {
 972             qemu_mod_timer(bs->block_timer, qemu_get_clock_ns(vm_clock));
 973         }
 974     }
 975 }
 976
 977 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
 978 {
 979     const char *id = qdict_get_str(qdict, "id");
 980     BlockDriverState *bs;
 981
 982     bs = bdrv_find(id);
 983     if (!bs) {
 984         qerror_report(QERR_DEVICE_NOT_FOUND, id);
 985         return -1;
 986     }
 987     if (bdrv_in_use(bs)) {
 988         qerror_report(QERR_DEVICE_IN_USE, id);
 989         return -1;
 990     }
 991
 992     /* quiesce block driver; prevent further io */
 993     bdrv_drain_all();
 994     bdrv_flush(bs);
 995     bdrv_close(bs);
 996
 997     /* if we have a device attached to this BlockDriverState
 998      * then we need to make the drive anonymous until the device
 999      * can be removed.  If this is a drive with no device backing
1000      * then we can just get rid of the block driver state right here.
1001      */
1002     if (bdrv_get_attached_dev(bs)) {
1003         bdrv_make_anon(bs);
1004     } else {
1005         drive_uninit(drive_get_by_blockdev(bs));
1006     }
1007
1008     return 0;
1009 }
1010
1011 void qmp_block_resize(const char *device, int64_t size, Error **errp)
1012 {
1013     BlockDriverState *bs;
1014
1015     bs = bdrv_find(device);
1016     if (!bs) {
1017         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1018         return;
1019     }
1020
1021     if (size < 0) {
1022         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
1023         return;
1024     }
1025
1026     switch (bdrv_truncate(bs, size)) {
1027     case 0:
1028         break;
1029     case -ENOMEDIUM:
1030         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1031         break;
1032     case -ENOTSUP:
1033         error_set(errp, QERR_UNSUPPORTED);
1034         break;
1035     case -EACCES:
1036         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1037         break;
1038     case -EBUSY:
1039         error_set(errp, QERR_DEVICE_IN_USE, device);
1040         break;
1041     default:
1042         error_set(errp, QERR_UNDEFINED_ERROR);
1043         break;
1044     }
1045 }
1046
1047 static QObject *qobject_from_block_job(BlockJob *job)
1048 {
1049     return qobject_from_jsonf("{ 'type': %s,"
1050                               "'device': %s,"
1051                               "'len': %" PRId64 ","
1052                               "'offset': %" PRId64 ","
1053                               "'speed': %" PRId64 " }",
1054                               job->job_type->job_type,
1055                               bdrv_get_device_name(job->bs),
1056                               job->len,
1057                               job->offset,
1058                               job->speed);
1059 }
1060
1061 static void block_stream_cb(void *opaque, int ret)
1062 {
1063     BlockDriverState *bs = opaque;
1064     QObject *obj;
1065
1066     trace_block_stream_cb(bs, bs->job, ret);
1067
1068     assert(bs->job);
1069     obj = qobject_from_block_job(bs->job);
1070     if (ret < 0) {
1071         QDict *dict = qobject_to_qdict(obj);
1072         qdict_put(dict, "error", qstring_from_str(strerror(-ret)));
1073     }
1074
1075     if (block_job_is_cancelled(bs->job)) {
1076         monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj);
1077     } else {
1078         monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj);
1079     }
1080     qobject_decref(obj);
1081
1082     drive_put_ref_bh_schedule(drive_get_by_blockdev(bs));
1083 }
1084
1085 void qmp_block_stream(const char *device, bool has_base,
1086                       const char *base, Error **errp)
1087 {
1088     BlockDriverState *bs;
1089     BlockDriverState *base_bs = NULL;
1090     int ret;
1091
1092     bs = bdrv_find(device);
1093     if (!bs) {
1094         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1095         return;
1096     }
1097
1098     if (base) {
1099         base_bs = bdrv_find_backing_image(bs, base);
1100         if (base_bs == NULL) {
1101             error_set(errp, QERR_BASE_NOT_FOUND, base);
1102             return;
1103         }
1104     }
1105
1106     ret = stream_start(bs, base_bs, base, block_stream_cb, bs);
1107     if (ret < 0) {
1108         switch (ret) {
1109         case -EBUSY:
1110             error_set(errp, QERR_DEVICE_IN_USE, device);
1111             return;
1112         default:
1113             error_set(errp, QERR_NOT_SUPPORTED);
1114             return;
1115         }
1116     }
1117
1118     /* Grab a reference so hotplug does not delete the BlockDriverState from
1119      * underneath us.
1120      */
1121     drive_get_ref(drive_get_by_blockdev(bs));
1122
1123     trace_qmp_block_stream(bs, bs->job);
1124 }
1125
1126 static BlockJob *find_block_job(const char *device)
1127 {
1128     BlockDriverState *bs;
1129
1130     bs = bdrv_find(device);
1131     if (!bs || !bs->job) {
1132         return NULL;
1133     }
1134     return bs->job;
1135 }
1136
1137 void qmp_block_job_set_speed(const char *device, int64_t value, Error **errp)
1138 {
1139     BlockJob *job = find_block_job(device);
1140
1141     if (!job) {
1142         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1143         return;
1144     }
1145
1146     if (block_job_set_speed(job, value) < 0) {
1147         error_set(errp, QERR_NOT_SUPPORTED);
1148     }
1149 }
1150
1151 void qmp_block_job_cancel(const char *device, Error **errp)
1152 {
1153     BlockJob *job = find_block_job(device);
1154
1155     if (!job) {
1156         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1157         return;
1158     }
1159
1160     trace_qmp_block_job_cancel(job);
1161     block_job_cancel(job);
1162 }
1163
1164 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
1165 {
1166     BlockJobInfoList **prev = opaque;
1167     BlockJob *job = bs->job;
1168
1169     if (job) {
1170         BlockJobInfoList *elem;
1171         BlockJobInfo *info = g_new(BlockJobInfo, 1);
1172         *info = (BlockJobInfo){
1173             .type   = g_strdup(job->job_type->job_type),
1174             .device = g_strdup(bdrv_get_device_name(bs)),
1175             .len    = job->len,
1176             .offset = job->offset,
1177             .speed  = job->speed,
1178         };
1179
1180         elem = g_new0(BlockJobInfoList, 1);
1181         elem->value = info;
1182
1183         (*prev)->next = elem;
1184         *prev = elem;
1185     }
1186 }
1187
1188 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
1189 {
1190     /* Dummy is a fake list element for holding the head pointer */
1191     BlockJobInfoList dummy = {};
1192     BlockJobInfoList *prev = &dummy;
1193     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
1194     return dummy.next;
1195 }