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, bool is_read) 242 { 243 if (!strcmp(buf, "ignore")) { 244 return BLOCKDEV_ON_ERROR_IGNORE; 245 } else if (!is_read && !strcmp(buf, "enospc")) { 246 return BLOCKDEV_ON_ERROR_ENOSPC; 247 } else if (!strcmp(buf, "stop")) { 248 return BLOCKDEV_ON_ERROR_STOP; 249 } else if (!strcmp(buf, "report")) { 250 return BLOCKDEV_ON_ERROR_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 if (qemu_opt_get(opts, "boot") != NULL) { 437 fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be " 438 "ignored. Future versions will reject this parameter. Please " 439 "update your scripts.\n"); 440 } 441 442 on_write_error = BLOCKDEV_ON_ERROR_ENOSPC; 443 if ((buf = qemu_opt_get(opts, "werror")) != NULL) { 444 if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) { 445 error_report("werror is not supported by this bus type"); 446 return NULL; 447 } 448 449 on_write_error = parse_block_error_action(buf, 0); 450 if (on_write_error < 0) { 451 return NULL; 452 } 453 } 454 455 on_read_error = BLOCKDEV_ON_ERROR_REPORT; 456 if ((buf = qemu_opt_get(opts, "rerror")) != NULL) { 457 if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) { 458 error_report("rerror is not supported by this bus type"); 459 return NULL; 460 } 461 462 on_read_error = parse_block_error_action(buf, 1); 463 if (on_read_error < 0) { 464 return NULL; 465 } 466 } 467 468 if ((devaddr = qemu_opt_get(opts, "addr")) != NULL) { 469 if (type != IF_VIRTIO) { 470 error_report("addr is not supported by this bus type"); 471 return NULL; 472 } 473 } 474 475 /* compute bus and unit according index */ 476 477 if (index != -1) { 478 if (bus_id != 0 || unit_id != -1) { 479 error_report("index cannot be used with bus and unit"); 480 return NULL; 481 } 482 bus_id = drive_index_to_bus_id(type, index); 483 unit_id = drive_index_to_unit_id(type, index); 484 } 485 486 /* if user doesn't specify a unit_id, 487 * try to find the first free 488 */ 489 490 if (unit_id == -1) { 491 unit_id = 0; 492 while (drive_get(type, bus_id, unit_id) != NULL) { 493 unit_id++; 494 if (max_devs && unit_id >= max_devs) { 495 unit_id -= max_devs; 496 bus_id++; 497 } 498 } 499 } 500 501 /* check unit id */ 502 503 if (max_devs && unit_id >= max_devs) { 504 error_report("unit %d too big (max is %d)", 505 unit_id, max_devs - 1); 506 return NULL; 507 } 508 509 /* 510 * catch multiple definitions 511 */ 512 513 if (drive_get(type, bus_id, unit_id) != NULL) { 514 error_report("drive with bus=%d, unit=%d (index=%d) exists", 515 bus_id, unit_id, index); 516 return NULL; 517 } 518 519 /* init */ 520 521 dinfo = g_malloc0(sizeof(*dinfo)); 522 if ((buf = qemu_opts_id(opts)) != NULL) { 523 dinfo->id = g_strdup(buf); 524 } else { 525 /* no id supplied -> create one */ 526 dinfo->id = g_malloc0(32); 527 if (type == IF_IDE || type == IF_SCSI) 528 mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd"; 529 if (max_devs) 530 snprintf(dinfo->id, 32, "%s%i%s%i", 531 if_name[type], bus_id, mediastr, unit_id); 532 else 533 snprintf(dinfo->id, 32, "%s%s%i", 534 if_name[type], mediastr, unit_id); 535 } 536 dinfo->bdrv = bdrv_new(dinfo->id); 537 dinfo->bdrv->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0; 538 dinfo->bdrv->read_only = ro; 539 dinfo->devaddr = devaddr; 540 dinfo->type = type; 541 dinfo->bus = bus_id; 542 dinfo->unit = unit_id; 543 dinfo->cyls = cyls; 544 dinfo->heads = heads; 545 dinfo->secs = secs; 546 dinfo->trans = translation; 547 dinfo->opts = opts; 548 dinfo->refcount = 1; 549 dinfo->serial = serial; 550 QTAILQ_INSERT_TAIL(&drives, dinfo, next); 551 552 bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error); 553 554 /* disk I/O throttling */ 555 bdrv_set_io_limits(dinfo->bdrv, &io_limits); 556 557 switch(type) { 558 case IF_IDE: 559 case IF_SCSI: 560 case IF_XEN: 561 case IF_NONE: 562 dinfo->media_cd = media == MEDIA_CDROM; 563 break; 564 case IF_SD: 565 case IF_FLOPPY: 566 case IF_PFLASH: 567 case IF_MTD: 568 break; 569 case IF_VIRTIO: 570 /* add virtio block device */ 571 opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0, NULL); 572 if (arch_type == QEMU_ARCH_S390X) { 573 qemu_opt_set(opts, "driver", "virtio-blk-s390"); 574 } else { 575 qemu_opt_set(opts, "driver", "virtio-blk-pci"); 576 } 577 qemu_opt_set(opts, "drive", dinfo->id); 578 if (devaddr) 579 qemu_opt_set(opts, "addr", devaddr); 580 break; 581 default: 582 abort(); 583 } 584 if (!file || !*file) { 585 return dinfo; 586 } 587 if (snapshot) { 588 /* always use cache=unsafe with snapshot */ 589 bdrv_flags &= ~BDRV_O_CACHE_MASK; 590 bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH); 591 } 592 593 if (copy_on_read) { 594 bdrv_flags |= BDRV_O_COPY_ON_READ; 595 } 596 597 if (runstate_check(RUN_STATE_INMIGRATE)) { 598 bdrv_flags |= BDRV_O_INCOMING; 599 } 600 601 if (media == MEDIA_CDROM) { 602 /* CDROM is fine for any interface, don't check. */ 603 ro = 1; 604 } else if (ro == 1) { 605 if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY && 606 type != IF_NONE && type != IF_PFLASH) { 607 error_report("readonly not supported by this bus type"); 608 goto err; 609 } 610 } 611 612 bdrv_flags |= ro ? 0 : BDRV_O_RDWR; 613 614 if (ro && copy_on_read) { 615 error_report("warning: disabling copy_on_read on readonly drive"); 616 } 617 618 ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv); 619 if (ret < 0) { 620 error_report("could not open disk image %s: %s", 621 file, strerror(-ret)); 622 goto err; 623 } 624 625 if (bdrv_key_required(dinfo->bdrv)) 626 autostart = 0; 627 return dinfo; 628 629 err: 630 bdrv_delete(dinfo->bdrv); 631 g_free(dinfo->id); 632 QTAILQ_REMOVE(&drives, dinfo, next); 633 g_free(dinfo); 634 return NULL; 635 } 636 637 void do_commit(Monitor *mon, const QDict *qdict) 638 { 639 const char *device = qdict_get_str(qdict, "device"); 640 BlockDriverState *bs; 641 int ret; 642 643 if (!strcmp(device, "all")) { 644 ret = bdrv_commit_all(); 645 if (ret == -EBUSY) { 646 qerror_report(QERR_DEVICE_IN_USE, device); 647 return; 648 } 649 } else { 650 bs = bdrv_find(device); 651 if (!bs) { 652 qerror_report(QERR_DEVICE_NOT_FOUND, device); 653 return; 654 } 655 ret = bdrv_commit(bs); 656 if (ret == -EBUSY) { 657 qerror_report(QERR_DEVICE_IN_USE, device); 658 return; 659 } 660 } 661 } 662 663 static void blockdev_do_action(int kind, void *data, Error **errp) 664 { 665 BlockdevAction action; 666 BlockdevActionList list; 667 668 action.kind = kind; 669 action.data = data; 670 list.value = &action; 671 list.next = NULL; 672 qmp_transaction(&list, errp); 673 } 674 675 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file, 676 bool has_format, const char *format, 677 bool has_mode, enum NewImageMode mode, 678 Error **errp) 679 { 680 BlockdevSnapshot snapshot = { 681 .device = (char *) device, 682 .snapshot_file = (char *) snapshot_file, 683 .has_format = has_format, 684 .format = (char *) format, 685 .has_mode = has_mode, 686 .mode = mode, 687 }; 688 blockdev_do_action(BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC, &snapshot, 689 errp); 690 } 691 692 693 /* New and old BlockDriverState structs for group snapshots */ 694 typedef struct BlkTransactionStates { 695 BlockDriverState *old_bs; 696 BlockDriverState *new_bs; 697 QSIMPLEQ_ENTRY(BlkTransactionStates) entry; 698 } BlkTransactionStates; 699 700 /* 701 * 'Atomic' group snapshots. The snapshots are taken as a set, and if any fail 702 * then we do not pivot any of the devices in the group, and abandon the 703 * snapshots 704 */ 705 void qmp_transaction(BlockdevActionList *dev_list, Error **errp) 706 { 707 int ret = 0; 708 BlockdevActionList *dev_entry = dev_list; 709 BlkTransactionStates *states, *next; 710 711 QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionStates) snap_bdrv_states; 712 QSIMPLEQ_INIT(&snap_bdrv_states); 713 714 /* drain all i/o before any snapshots */ 715 bdrv_drain_all(); 716 717 /* We don't do anything in this loop that commits us to the snapshot */ 718 while (NULL != dev_entry) { 719 BlockdevAction *dev_info = NULL; 720 BlockDriver *proto_drv; 721 BlockDriver *drv; 722 int flags; 723 enum NewImageMode mode; 724 const char *new_image_file; 725 const char *device; 726 const char *format = "qcow2"; 727 728 dev_info = dev_entry->value; 729 dev_entry = dev_entry->next; 730 731 states = g_malloc0(sizeof(BlkTransactionStates)); 732 QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, states, entry); 733 734 switch (dev_info->kind) { 735 case BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC: 736 device = dev_info->blockdev_snapshot_sync->device; 737 if (!dev_info->blockdev_snapshot_sync->has_mode) { 738 dev_info->blockdev_snapshot_sync->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS; 739 } 740 new_image_file = dev_info->blockdev_snapshot_sync->snapshot_file; 741 if (dev_info->blockdev_snapshot_sync->has_format) { 742 format = dev_info->blockdev_snapshot_sync->format; 743 } 744 mode = dev_info->blockdev_snapshot_sync->mode; 745 break; 746 default: 747 abort(); 748 } 749 750 drv = bdrv_find_format(format); 751 if (!drv) { 752 error_set(errp, QERR_INVALID_BLOCK_FORMAT, format); 753 goto delete_and_fail; 754 } 755 756 states->old_bs = bdrv_find(device); 757 if (!states->old_bs) { 758 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 759 goto delete_and_fail; 760 } 761 762 if (!bdrv_is_inserted(states->old_bs)) { 763 error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device); 764 goto delete_and_fail; 765 } 766 767 if (bdrv_in_use(states->old_bs)) { 768 error_set(errp, QERR_DEVICE_IN_USE, device); 769 goto delete_and_fail; 770 } 771 772 if (!bdrv_is_read_only(states->old_bs)) { 773 if (bdrv_flush(states->old_bs)) { 774 error_set(errp, QERR_IO_ERROR); 775 goto delete_and_fail; 776 } 777 } 778 779 flags = states->old_bs->open_flags; 780 781 proto_drv = bdrv_find_protocol(new_image_file); 782 if (!proto_drv) { 783 error_set(errp, QERR_INVALID_BLOCK_FORMAT, format); 784 goto delete_and_fail; 785 } 786 787 /* create new image w/backing file */ 788 if (mode != NEW_IMAGE_MODE_EXISTING) { 789 ret = bdrv_img_create(new_image_file, format, 790 states->old_bs->filename, 791 states->old_bs->drv->format_name, 792 NULL, -1, flags); 793 if (ret) { 794 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file); 795 goto delete_and_fail; 796 } 797 } 798 799 /* We will manually add the backing_hd field to the bs later */ 800 states->new_bs = bdrv_new(""); 801 ret = bdrv_open(states->new_bs, new_image_file, 802 flags | BDRV_O_NO_BACKING, drv); 803 if (ret != 0) { 804 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file); 805 goto delete_and_fail; 806 } 807 } 808 809 810 /* Now we are going to do the actual pivot. Everything up to this point 811 * is reversible, but we are committed at this point */ 812 QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) { 813 /* This removes our old bs from the bdrv_states, and adds the new bs */ 814 bdrv_append(states->new_bs, states->old_bs); 815 /* We don't need (or want) to use the transactional 816 * bdrv_reopen_multiple() across all the entries at once, because we 817 * don't want to abort all of them if one of them fails the reopen */ 818 bdrv_reopen(states->new_bs, states->new_bs->open_flags & ~BDRV_O_RDWR, 819 NULL); 820 } 821 822 /* success */ 823 goto exit; 824 825 delete_and_fail: 826 /* 827 * failure, and it is all-or-none; abandon each new bs, and keep using 828 * the original bs for all images 829 */ 830 QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) { 831 if (states->new_bs) { 832 bdrv_delete(states->new_bs); 833 } 834 } 835 exit: 836 QSIMPLEQ_FOREACH_SAFE(states, &snap_bdrv_states, entry, next) { 837 g_free(states); 838 } 839 } 840 841 842 static void eject_device(BlockDriverState *bs, int force, Error **errp) 843 { 844 if (bdrv_in_use(bs)) { 845 error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs)); 846 return; 847 } 848 if (!bdrv_dev_has_removable_media(bs)) { 849 error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs)); 850 return; 851 } 852 853 if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) { 854 bdrv_dev_eject_request(bs, force); 855 if (!force) { 856 error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs)); 857 return; 858 } 859 } 860 861 bdrv_close(bs); 862 } 863 864 void qmp_eject(const char *device, bool has_force, bool force, Error **errp) 865 { 866 BlockDriverState *bs; 867 868 bs = bdrv_find(device); 869 if (!bs) { 870 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 871 return; 872 } 873 874 eject_device(bs, force, errp); 875 } 876 877 void qmp_block_passwd(const char *device, const char *password, Error **errp) 878 { 879 BlockDriverState *bs; 880 int err; 881 882 bs = bdrv_find(device); 883 if (!bs) { 884 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 885 return; 886 } 887 888 err = bdrv_set_key(bs, password); 889 if (err == -EINVAL) { 890 error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs)); 891 return; 892 } else if (err < 0) { 893 error_set(errp, QERR_INVALID_PASSWORD); 894 return; 895 } 896 } 897 898 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename, 899 int bdrv_flags, BlockDriver *drv, 900 const char *password, Error **errp) 901 { 902 if (bdrv_open(bs, filename, bdrv_flags, drv) < 0) { 903 error_set(errp, QERR_OPEN_FILE_FAILED, filename); 904 return; 905 } 906 907 if (bdrv_key_required(bs)) { 908 if (password) { 909 if (bdrv_set_key(bs, password) < 0) { 910 error_set(errp, QERR_INVALID_PASSWORD); 911 } 912 } else { 913 error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs), 914 bdrv_get_encrypted_filename(bs)); 915 } 916 } else if (password) { 917 error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs)); 918 } 919 } 920 921 void qmp_change_blockdev(const char *device, const char *filename, 922 bool has_format, const char *format, Error **errp) 923 { 924 BlockDriverState *bs; 925 BlockDriver *drv = NULL; 926 int bdrv_flags; 927 Error *err = NULL; 928 929 bs = bdrv_find(device); 930 if (!bs) { 931 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 932 return; 933 } 934 935 if (format) { 936 drv = bdrv_find_whitelisted_format(format); 937 if (!drv) { 938 error_set(errp, QERR_INVALID_BLOCK_FORMAT, format); 939 return; 940 } 941 } 942 943 eject_device(bs, 0, &err); 944 if (error_is_set(&err)) { 945 error_propagate(errp, err); 946 return; 947 } 948 949 bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR; 950 bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0; 951 952 qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp); 953 } 954 955 /* throttling disk I/O limits */ 956 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd, 957 int64_t bps_wr, int64_t iops, int64_t iops_rd, 958 int64_t iops_wr, Error **errp) 959 { 960 BlockIOLimit io_limits; 961 BlockDriverState *bs; 962 963 bs = bdrv_find(device); 964 if (!bs) { 965 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 966 return; 967 } 968 969 io_limits.bps[BLOCK_IO_LIMIT_TOTAL] = bps; 970 io_limits.bps[BLOCK_IO_LIMIT_READ] = bps_rd; 971 io_limits.bps[BLOCK_IO_LIMIT_WRITE] = bps_wr; 972 io_limits.iops[BLOCK_IO_LIMIT_TOTAL]= iops; 973 io_limits.iops[BLOCK_IO_LIMIT_READ] = iops_rd; 974 io_limits.iops[BLOCK_IO_LIMIT_WRITE]= iops_wr; 975 976 if (!do_check_io_limits(&io_limits)) { 977 error_set(errp, QERR_INVALID_PARAMETER_COMBINATION); 978 return; 979 } 980 981 bs->io_limits = io_limits; 982 bs->slice_time = BLOCK_IO_SLICE_TIME; 983 984 if (!bs->io_limits_enabled && bdrv_io_limits_enabled(bs)) { 985 bdrv_io_limits_enable(bs); 986 } else if (bs->io_limits_enabled && !bdrv_io_limits_enabled(bs)) { 987 bdrv_io_limits_disable(bs); 988 } else { 989 if (bs->block_timer) { 990 qemu_mod_timer(bs->block_timer, qemu_get_clock_ns(vm_clock)); 991 } 992 } 993 } 994 995 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data) 996 { 997 const char *id = qdict_get_str(qdict, "id"); 998 BlockDriverState *bs; 999 1000 bs = bdrv_find(id); 1001 if (!bs) { 1002 qerror_report(QERR_DEVICE_NOT_FOUND, id); 1003 return -1; 1004 } 1005 if (bdrv_in_use(bs)) { 1006 qerror_report(QERR_DEVICE_IN_USE, id); 1007 return -1; 1008 } 1009 1010 /* quiesce block driver; prevent further io */ 1011 bdrv_drain_all(); 1012 bdrv_flush(bs); 1013 bdrv_close(bs); 1014 1015 /* if we have a device attached to this BlockDriverState 1016 * then we need to make the drive anonymous until the device 1017 * can be removed. If this is a drive with no device backing 1018 * then we can just get rid of the block driver state right here. 1019 */ 1020 if (bdrv_get_attached_dev(bs)) { 1021 bdrv_make_anon(bs); 1022 } else { 1023 drive_uninit(drive_get_by_blockdev(bs)); 1024 } 1025 1026 return 0; 1027 } 1028 1029 void qmp_block_resize(const char *device, int64_t size, Error **errp) 1030 { 1031 BlockDriverState *bs; 1032 1033 bs = bdrv_find(device); 1034 if (!bs) { 1035 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 1036 return; 1037 } 1038 1039 if (size < 0) { 1040 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size"); 1041 return; 1042 } 1043 1044 switch (bdrv_truncate(bs, size)) { 1045 case 0: 1046 break; 1047 case -ENOMEDIUM: 1048 error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device); 1049 break; 1050 case -ENOTSUP: 1051 error_set(errp, QERR_UNSUPPORTED); 1052 break; 1053 case -EACCES: 1054 error_set(errp, QERR_DEVICE_IS_READ_ONLY, device); 1055 break; 1056 case -EBUSY: 1057 error_set(errp, QERR_DEVICE_IN_USE, device); 1058 break; 1059 default: 1060 error_set(errp, QERR_UNDEFINED_ERROR); 1061 break; 1062 } 1063 } 1064 1065 static void block_job_cb(void *opaque, int ret) 1066 { 1067 BlockDriverState *bs = opaque; 1068 QObject *obj; 1069 1070 trace_block_job_cb(bs, bs->job, ret); 1071 1072 assert(bs->job); 1073 obj = qobject_from_block_job(bs->job); 1074 if (ret < 0) { 1075 QDict *dict = qobject_to_qdict(obj); 1076 qdict_put(dict, "error", qstring_from_str(strerror(-ret))); 1077 } 1078 1079 if (block_job_is_cancelled(bs->job)) { 1080 monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj); 1081 } else { 1082 monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj); 1083 } 1084 qobject_decref(obj); 1085 1086 drive_put_ref_bh_schedule(drive_get_by_blockdev(bs)); 1087 } 1088 1089 void qmp_block_stream(const char *device, bool has_base, 1090 const char *base, bool has_speed, int64_t speed, 1091 bool has_on_error, BlockdevOnError on_error, 1092 Error **errp) 1093 { 1094 BlockDriverState *bs; 1095 BlockDriverState *base_bs = NULL; 1096 Error *local_err = NULL; 1097 1098 if (!has_on_error) { 1099 on_error = BLOCKDEV_ON_ERROR_REPORT; 1100 } 1101 1102 bs = bdrv_find(device); 1103 if (!bs) { 1104 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 1105 return; 1106 } 1107 1108 if (base) { 1109 base_bs = bdrv_find_backing_image(bs, base); 1110 if (base_bs == NULL) { 1111 error_set(errp, QERR_BASE_NOT_FOUND, base); 1112 return; 1113 } 1114 } 1115 1116 stream_start(bs, base_bs, base, has_speed ? speed : 0, 1117 on_error, block_job_cb, bs, &local_err); 1118 if (error_is_set(&local_err)) { 1119 error_propagate(errp, local_err); 1120 return; 1121 } 1122 1123 /* Grab a reference so hotplug does not delete the BlockDriverState from 1124 * underneath us. 1125 */ 1126 drive_get_ref(drive_get_by_blockdev(bs)); 1127 1128 trace_qmp_block_stream(bs, bs->job); 1129 } 1130 1131 void qmp_block_commit(const char *device, 1132 bool has_base, const char *base, const char *top, 1133 bool has_speed, int64_t speed, 1134 Error **errp) 1135 { 1136 BlockDriverState *bs; 1137 BlockDriverState *base_bs, *top_bs; 1138 Error *local_err = NULL; 1139 /* This will be part of the QMP command, if/when the 1140 * BlockdevOnError change for blkmirror makes it in 1141 */ 1142 BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT; 1143 1144 /* drain all i/o before commits */ 1145 bdrv_drain_all(); 1146 1147 bs = bdrv_find(device); 1148 if (!bs) { 1149 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 1150 return; 1151 } 1152 1153 /* default top_bs is the active layer */ 1154 top_bs = bs; 1155 1156 if (top) { 1157 if (strcmp(bs->filename, top) != 0) { 1158 top_bs = bdrv_find_backing_image(bs, top); 1159 } 1160 } 1161 1162 if (top_bs == NULL) { 1163 error_setg(errp, "Top image file %s not found", top ? top : "NULL"); 1164 return; 1165 } 1166 1167 if (has_base && base) { 1168 base_bs = bdrv_find_backing_image(top_bs, base); 1169 } else { 1170 base_bs = bdrv_find_base(top_bs); 1171 } 1172 1173 if (base_bs == NULL) { 1174 error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL"); 1175 return; 1176 } 1177 1178 commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs, 1179 &local_err); 1180 if (local_err != NULL) { 1181 error_propagate(errp, local_err); 1182 return; 1183 } 1184 /* Grab a reference so hotplug does not delete the BlockDriverState from 1185 * underneath us. 1186 */ 1187 drive_get_ref(drive_get_by_blockdev(bs)); 1188 } 1189 1190 void qmp_drive_mirror(const char *device, const char *target, 1191 bool has_format, const char *format, 1192 enum MirrorSyncMode sync, 1193 bool has_mode, enum NewImageMode mode, 1194 bool has_speed, int64_t speed, 1195 bool has_on_source_error, BlockdevOnError on_source_error, 1196 bool has_on_target_error, BlockdevOnError on_target_error, 1197 Error **errp) 1198 { 1199 BlockDriverInfo bdi; 1200 BlockDriverState *bs; 1201 BlockDriverState *source, *target_bs; 1202 BlockDriver *proto_drv; 1203 BlockDriver *drv = NULL; 1204 Error *local_err = NULL; 1205 int flags; 1206 uint64_t size; 1207 int ret; 1208 1209 if (!has_speed) { 1210 speed = 0; 1211 } 1212 if (!has_on_source_error) { 1213 on_source_error = BLOCKDEV_ON_ERROR_REPORT; 1214 } 1215 if (!has_on_target_error) { 1216 on_target_error = BLOCKDEV_ON_ERROR_REPORT; 1217 } 1218 if (!has_mode) { 1219 mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS; 1220 } 1221 1222 bs = bdrv_find(device); 1223 if (!bs) { 1224 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 1225 return; 1226 } 1227 1228 if (!bdrv_is_inserted(bs)) { 1229 error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device); 1230 return; 1231 } 1232 1233 if (!has_format) { 1234 format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name; 1235 } 1236 if (format) { 1237 drv = bdrv_find_format(format); 1238 if (!drv) { 1239 error_set(errp, QERR_INVALID_BLOCK_FORMAT, format); 1240 return; 1241 } 1242 } 1243 1244 if (bdrv_in_use(bs)) { 1245 error_set(errp, QERR_DEVICE_IN_USE, device); 1246 return; 1247 } 1248 1249 flags = bs->open_flags | BDRV_O_RDWR; 1250 source = bs->backing_hd; 1251 if (!source && sync == MIRROR_SYNC_MODE_TOP) { 1252 sync = MIRROR_SYNC_MODE_FULL; 1253 } 1254 1255 proto_drv = bdrv_find_protocol(target); 1256 if (!proto_drv) { 1257 error_set(errp, QERR_INVALID_BLOCK_FORMAT, format); 1258 return; 1259 } 1260 1261 if (sync == MIRROR_SYNC_MODE_FULL && mode != NEW_IMAGE_MODE_EXISTING) { 1262 /* create new image w/o backing file */ 1263 assert(format && drv); 1264 bdrv_get_geometry(bs, &size); 1265 size *= 512; 1266 ret = bdrv_img_create(target, format, 1267 NULL, NULL, NULL, size, flags); 1268 } else { 1269 switch (mode) { 1270 case NEW_IMAGE_MODE_EXISTING: 1271 ret = 0; 1272 break; 1273 case NEW_IMAGE_MODE_ABSOLUTE_PATHS: 1274 /* create new image with backing file */ 1275 ret = bdrv_img_create(target, format, 1276 source->filename, 1277 source->drv->format_name, 1278 NULL, -1, flags); 1279 break; 1280 default: 1281 abort(); 1282 } 1283 } 1284 1285 if (ret) { 1286 error_set(errp, QERR_OPEN_FILE_FAILED, target); 1287 return; 1288 } 1289 1290 target_bs = bdrv_new(""); 1291 ret = bdrv_open(target_bs, target, flags | BDRV_O_NO_BACKING, drv); 1292 1293 if (ret < 0) { 1294 bdrv_delete(target_bs); 1295 error_set(errp, QERR_OPEN_FILE_FAILED, target); 1296 return; 1297 } 1298 1299 /* We need a backing file if we will copy parts of a cluster. */ 1300 if (bdrv_get_info(target_bs, &bdi) >= 0 && bdi.cluster_size != 0 && 1301 bdi.cluster_size >= BDRV_SECTORS_PER_DIRTY_CHUNK * 512) { 1302 ret = bdrv_open_backing_file(target_bs); 1303 if (ret < 0) { 1304 bdrv_delete(target_bs); 1305 error_set(errp, QERR_OPEN_FILE_FAILED, target); 1306 return; 1307 } 1308 } 1309 1310 mirror_start(bs, target_bs, speed, sync, on_source_error, on_target_error, 1311 block_job_cb, bs, &local_err); 1312 if (local_err != NULL) { 1313 bdrv_delete(target_bs); 1314 error_propagate(errp, local_err); 1315 return; 1316 } 1317 1318 /* Grab a reference so hotplug does not delete the BlockDriverState from 1319 * underneath us. 1320 */ 1321 drive_get_ref(drive_get_by_blockdev(bs)); 1322 } 1323 1324 static BlockJob *find_block_job(const char *device) 1325 { 1326 BlockDriverState *bs; 1327 1328 bs = bdrv_find(device); 1329 if (!bs || !bs->job) { 1330 return NULL; 1331 } 1332 return bs->job; 1333 } 1334 1335 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp) 1336 { 1337 BlockJob *job = find_block_job(device); 1338 1339 if (!job) { 1340 error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device); 1341 return; 1342 } 1343 1344 block_job_set_speed(job, speed, errp); 1345 } 1346 1347 void qmp_block_job_cancel(const char *device, 1348 bool has_force, bool force, Error **errp) 1349 { 1350 BlockJob *job = find_block_job(device); 1351 1352 if (!has_force) { 1353 force = false; 1354 } 1355 1356 if (!job) { 1357 error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device); 1358 return; 1359 } 1360 if (job->paused && !force) { 1361 error_set(errp, QERR_BLOCK_JOB_PAUSED, device); 1362 return; 1363 } 1364 1365 trace_qmp_block_job_cancel(job); 1366 block_job_cancel(job); 1367 } 1368 1369 void qmp_block_job_pause(const char *device, Error **errp) 1370 { 1371 BlockJob *job = find_block_job(device); 1372 1373 if (!job) { 1374 error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device); 1375 return; 1376 } 1377 1378 trace_qmp_block_job_pause(job); 1379 block_job_pause(job); 1380 } 1381 1382 void qmp_block_job_resume(const char *device, Error **errp) 1383 { 1384 BlockJob *job = find_block_job(device); 1385 1386 if (!job) { 1387 error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device); 1388 return; 1389 } 1390 1391 trace_qmp_block_job_resume(job); 1392 block_job_resume(job); 1393 } 1394 1395 void qmp_block_job_complete(const char *device, Error **errp) 1396 { 1397 BlockJob *job = find_block_job(device); 1398 1399 if (!job) { 1400 error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device); 1401 return; 1402 } 1403 1404 trace_qmp_block_job_complete(job); 1405 block_job_complete(job, errp); 1406 } 1407 1408 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs) 1409 { 1410 BlockJobInfoList **prev = opaque; 1411 BlockJob *job = bs->job; 1412 1413 if (job) { 1414 BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1); 1415 elem->value = block_job_query(bs->job); 1416 (*prev)->next = elem; 1417 *prev = elem; 1418 } 1419 } 1420 1421 BlockJobInfoList *qmp_query_block_jobs(Error **errp) 1422 { 1423 /* Dummy is a fake list element for holding the head pointer */ 1424 BlockJobInfoList dummy = {}; 1425 BlockJobInfoList *prev = &dummy; 1426 bdrv_iterate(do_qmp_query_block_jobs_one, &prev); 1427 return dummy.next; 1428 } 1429