1 /* 2 * QEMU host block devices 3 * 4 * Copyright (c) 2003-2008 Fabrice Bellard 5 * 6 * This work is licensed under the terms of the GNU GPL, version 2 or 7 * later. See the COPYING file in the top-level directory. 8 */ 9 10 #include "block.h" 11 #include "blockdev.h" 12 #include "monitor.h" 13 #include "qerror.h" 14 #include "qemu-option.h" 15 #include "qemu-config.h" 16 #include "qemu-objects.h" 17 #include "sysemu.h" 18 #include "block_int.h" 19 #include "qmp-commands.h" 20 #include "trace.h" 21 #include "arch_init.h" 22 23 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives); 24 25 static const char *const if_name[IF_COUNT] = { 26 [IF_NONE] = "none", 27 [IF_IDE] = "ide", 28 [IF_SCSI] = "scsi", 29 [IF_FLOPPY] = "floppy", 30 [IF_PFLASH] = "pflash", 31 [IF_MTD] = "mtd", 32 [IF_SD] = "sd", 33 [IF_VIRTIO] = "virtio", 34 [IF_XEN] = "xen", 35 }; 36 37 static const int if_max_devs[IF_COUNT] = { 38 /* 39 * Do not change these numbers! They govern how drive option 40 * index maps to unit and bus. That mapping is ABI. 41 * 42 * All controllers used to imlement if=T drives need to support 43 * if_max_devs[T] units, for any T with if_max_devs[T] != 0. 44 * Otherwise, some index values map to "impossible" bus, unit 45 * values. 46 * 47 * For instance, if you change [IF_SCSI] to 255, -drive 48 * if=scsi,index=12 no longer means bus=1,unit=5, but 49 * bus=0,unit=12. With an lsi53c895a controller (7 units max), 50 * the drive can't be set up. Regression. 51 */ 52 [IF_IDE] = 2, 53 [IF_SCSI] = 7, 54 }; 55 56 /* 57 * We automatically delete the drive when a device using it gets 58 * unplugged. Questionable feature, but we can't just drop it. 59 * Device models call blockdev_mark_auto_del() to schedule the 60 * automatic deletion, and generic qdev code calls blockdev_auto_del() 61 * when deletion is actually safe. 62 */ 63 void blockdev_mark_auto_del(BlockDriverState *bs) 64 { 65 DriveInfo *dinfo = drive_get_by_blockdev(bs); 66 67 if (bs->job) { 68 block_job_cancel(bs->job); 69 } 70 if (dinfo) { 71 dinfo->auto_del = 1; 72 } 73 } 74 75 void blockdev_auto_del(BlockDriverState *bs) 76 { 77 DriveInfo *dinfo = drive_get_by_blockdev(bs); 78 79 if (dinfo && dinfo->auto_del) { 80 drive_put_ref(dinfo); 81 } 82 } 83 84 static int drive_index_to_bus_id(BlockInterfaceType type, int index) 85 { 86 int max_devs = if_max_devs[type]; 87 return max_devs ? index / max_devs : 0; 88 } 89 90 static int drive_index_to_unit_id(BlockInterfaceType type, int index) 91 { 92 int max_devs = if_max_devs[type]; 93 return max_devs ? index % max_devs : index; 94 } 95 96 QemuOpts *drive_def(const char *optstr) 97 { 98 return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0); 99 } 100 101 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file, 102 const char *optstr) 103 { 104 QemuOpts *opts; 105 char buf[32]; 106 107 opts = drive_def(optstr); 108 if (!opts) { 109 return NULL; 110 } 111 if (type != IF_DEFAULT) { 112 qemu_opt_set(opts, "if", if_name[type]); 113 } 114 if (index >= 0) { 115 snprintf(buf, sizeof(buf), "%d", index); 116 qemu_opt_set(opts, "index", buf); 117 } 118 if (file) 119 qemu_opt_set(opts, "file", file); 120 return opts; 121 } 122 123 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit) 124 { 125 DriveInfo *dinfo; 126 127 /* seek interface, bus and unit */ 128 129 QTAILQ_FOREACH(dinfo, &drives, next) { 130 if (dinfo->type == type && 131 dinfo->bus == bus && 132 dinfo->unit == unit) 133 return dinfo; 134 } 135 136 return NULL; 137 } 138 139 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index) 140 { 141 return drive_get(type, 142 drive_index_to_bus_id(type, index), 143 drive_index_to_unit_id(type, index)); 144 } 145 146 int drive_get_max_bus(BlockInterfaceType type) 147 { 148 int max_bus; 149 DriveInfo *dinfo; 150 151 max_bus = -1; 152 QTAILQ_FOREACH(dinfo, &drives, next) { 153 if(dinfo->type == type && 154 dinfo->bus > max_bus) 155 max_bus = dinfo->bus; 156 } 157 return max_bus; 158 } 159 160 /* Get a block device. This should only be used for single-drive devices 161 (e.g. SD/Floppy/MTD). Multi-disk devices (scsi/ide) should use the 162 appropriate bus. */ 163 DriveInfo *drive_get_next(BlockInterfaceType type) 164 { 165 static int next_block_unit[IF_COUNT]; 166 167 return drive_get(type, 0, next_block_unit[type]++); 168 } 169 170 DriveInfo *drive_get_by_blockdev(BlockDriverState *bs) 171 { 172 DriveInfo *dinfo; 173 174 QTAILQ_FOREACH(dinfo, &drives, next) { 175 if (dinfo->bdrv == bs) { 176 return dinfo; 177 } 178 } 179 return NULL; 180 } 181 182 static void bdrv_format_print(void *opaque, const char *name) 183 { 184 error_printf(" %s", name); 185 } 186 187 static void drive_uninit(DriveInfo *dinfo) 188 { 189 qemu_opts_del(dinfo->opts); 190 bdrv_delete(dinfo->bdrv); 191 g_free(dinfo->id); 192 QTAILQ_REMOVE(&drives, dinfo, next); 193 g_free(dinfo); 194 } 195 196 void drive_put_ref(DriveInfo *dinfo) 197 { 198 assert(dinfo->refcount); 199 if (--dinfo->refcount == 0) { 200 drive_uninit(dinfo); 201 } 202 } 203 204 void drive_get_ref(DriveInfo *dinfo) 205 { 206 dinfo->refcount++; 207 } 208 209 typedef struct { 210 QEMUBH *bh; 211 DriveInfo *dinfo; 212 } DrivePutRefBH; 213 214 static void drive_put_ref_bh(void *opaque) 215 { 216 DrivePutRefBH *s = opaque; 217 218 drive_put_ref(s->dinfo); 219 qemu_bh_delete(s->bh); 220 g_free(s); 221 } 222 223 /* 224 * Release a drive reference in a BH 225 * 226 * It is not possible to use drive_put_ref() from a callback function when the 227 * callers still need the drive. In such cases we schedule a BH to release the 228 * reference. 229 */ 230 static void drive_put_ref_bh_schedule(DriveInfo *dinfo) 231 { 232 DrivePutRefBH *s; 233 234 s = g_new(DrivePutRefBH, 1); 235 s->bh = qemu_bh_new(drive_put_ref_bh, s); 236 s->dinfo = dinfo; 237 qemu_bh_schedule(s->bh); 238 } 239 240 static int parse_block_error_action(const char *buf, int is_read) 241 { 242 if (!strcmp(buf, "ignore")) { 243 return BLOCK_ERR_IGNORE; 244 } else if (!is_read && !strcmp(buf, "enospc")) { 245 return BLOCK_ERR_STOP_ENOSPC; 246 } else if (!strcmp(buf, "stop")) { 247 return BLOCK_ERR_STOP_ANY; 248 } else if (!strcmp(buf, "report")) { 249 return BLOCK_ERR_REPORT; 250 } else { 251 error_report("'%s' invalid %s error action", 252 buf, is_read ? "read" : "write"); 253 return -1; 254 } 255 } 256 257 static bool do_check_io_limits(BlockIOLimit *io_limits) 258 { 259 bool bps_flag; 260 bool iops_flag; 261 262 assert(io_limits); 263 264 bps_flag = (io_limits->bps[BLOCK_IO_LIMIT_TOTAL] != 0) 265 && ((io_limits->bps[BLOCK_IO_LIMIT_READ] != 0) 266 || (io_limits->bps[BLOCK_IO_LIMIT_WRITE] != 0)); 267 iops_flag = (io_limits->iops[BLOCK_IO_LIMIT_TOTAL] != 0) 268 && ((io_limits->iops[BLOCK_IO_LIMIT_READ] != 0) 269 || (io_limits->iops[BLOCK_IO_LIMIT_WRITE] != 0)); 270 if (bps_flag || iops_flag) { 271 return false; 272 } 273 274 return true; 275 } 276 277 DriveInfo *drive_init(QemuOpts *opts, int default_to_scsi) 278 { 279 const char *buf; 280 const char *file = NULL; 281 const char *serial; 282 const char *mediastr = ""; 283 BlockInterfaceType type; 284 enum { MEDIA_DISK, MEDIA_CDROM } media; 285 int bus_id, unit_id; 286 int cyls, heads, secs, translation; 287 BlockDriver *drv = NULL; 288 int max_devs; 289 int index; 290 int ro = 0; 291 int bdrv_flags = 0; 292 int on_read_error, on_write_error; 293 const char *devaddr; 294 DriveInfo *dinfo; 295 BlockIOLimit io_limits; 296 int snapshot = 0; 297 bool copy_on_read; 298 int ret; 299 300 translation = BIOS_ATA_TRANSLATION_AUTO; 301 media = MEDIA_DISK; 302 303 /* extract parameters */ 304 bus_id = qemu_opt_get_number(opts, "bus", 0); 305 unit_id = qemu_opt_get_number(opts, "unit", -1); 306 index = qemu_opt_get_number(opts, "index", -1); 307 308 cyls = qemu_opt_get_number(opts, "cyls", 0); 309 heads = qemu_opt_get_number(opts, "heads", 0); 310 secs = qemu_opt_get_number(opts, "secs", 0); 311 312 snapshot = qemu_opt_get_bool(opts, "snapshot", 0); 313 ro = qemu_opt_get_bool(opts, "readonly", 0); 314 copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false); 315 316 file = qemu_opt_get(opts, "file"); 317 serial = qemu_opt_get(opts, "serial"); 318 319 if ((buf = qemu_opt_get(opts, "if")) != NULL) { 320 for (type = 0; type < IF_COUNT && strcmp(buf, if_name[type]); type++) 321 ; 322 if (type == IF_COUNT) { 323 error_report("unsupported bus type '%s'", buf); 324 return NULL; 325 } 326 } else { 327 type = default_to_scsi ? IF_SCSI : IF_IDE; 328 } 329 330 max_devs = if_max_devs[type]; 331 332 if (cyls || heads || secs) { 333 if (cyls < 1 || (type == IF_IDE && cyls > 16383)) { 334 error_report("invalid physical cyls number"); 335 return NULL; 336 } 337 if (heads < 1 || (type == IF_IDE && heads > 16)) { 338 error_report("invalid physical heads number"); 339 return NULL; 340 } 341 if (secs < 1 || (type == IF_IDE && secs > 63)) { 342 error_report("invalid physical secs number"); 343 return NULL; 344 } 345 } 346 347 if ((buf = qemu_opt_get(opts, "trans")) != NULL) { 348 if (!cyls) { 349 error_report("'%s' trans must be used with cyls, heads and secs", 350 buf); 351 return NULL; 352 } 353 if (!strcmp(buf, "none")) 354 translation = BIOS_ATA_TRANSLATION_NONE; 355 else if (!strcmp(buf, "lba")) 356 translation = BIOS_ATA_TRANSLATION_LBA; 357 else if (!strcmp(buf, "auto")) 358 translation = BIOS_ATA_TRANSLATION_AUTO; 359 else { 360 error_report("'%s' invalid translation type", buf); 361 return NULL; 362 } 363 } 364 365 if ((buf = qemu_opt_get(opts, "media")) != NULL) { 366 if (!strcmp(buf, "disk")) { 367 media = MEDIA_DISK; 368 } else if (!strcmp(buf, "cdrom")) { 369 if (cyls || secs || heads) { 370 error_report("CHS can't be set with media=%s", buf); 371 return NULL; 372 } 373 media = MEDIA_CDROM; 374 } else { 375 error_report("'%s' invalid media", buf); 376 return NULL; 377 } 378 } 379 380 if ((buf = qemu_opt_get(opts, "cache")) != NULL) { 381 if (bdrv_parse_cache_flags(buf, &bdrv_flags) != 0) { 382 error_report("invalid cache option"); 383 return NULL; 384 } 385 } 386 387 #ifdef CONFIG_LINUX_AIO 388 if ((buf = qemu_opt_get(opts, "aio")) != NULL) { 389 if (!strcmp(buf, "native")) { 390 bdrv_flags |= BDRV_O_NATIVE_AIO; 391 } else if (!strcmp(buf, "threads")) { 392 /* this is the default */ 393 } else { 394 error_report("invalid aio option"); 395 return NULL; 396 } 397 } 398 #endif 399 400 if ((buf = qemu_opt_get(opts, "format")) != NULL) { 401 if (strcmp(buf, "?") == 0) { 402 error_printf("Supported formats:"); 403 bdrv_iterate_format(bdrv_format_print, NULL); 404 error_printf("\n"); 405 return NULL; 406 } 407 drv = bdrv_find_whitelisted_format(buf); 408 if (!drv) { 409 error_report("'%s' invalid format", buf); 410 return NULL; 411 } 412 } 413 414 /* disk I/O throttling */ 415 io_limits.bps[BLOCK_IO_LIMIT_TOTAL] = 416 qemu_opt_get_number(opts, "bps", 0); 417 io_limits.bps[BLOCK_IO_LIMIT_READ] = 418 qemu_opt_get_number(opts, "bps_rd", 0); 419 io_limits.bps[BLOCK_IO_LIMIT_WRITE] = 420 qemu_opt_get_number(opts, "bps_wr", 0); 421 io_limits.iops[BLOCK_IO_LIMIT_TOTAL] = 422 qemu_opt_get_number(opts, "iops", 0); 423 io_limits.iops[BLOCK_IO_LIMIT_READ] = 424 qemu_opt_get_number(opts, "iops_rd", 0); 425 io_limits.iops[BLOCK_IO_LIMIT_WRITE] = 426 qemu_opt_get_number(opts, "iops_wr", 0); 427 428 if (!do_check_io_limits(&io_limits)) { 429 error_report("bps(iops) and bps_rd/bps_wr(iops_rd/iops_wr) " 430 "cannot be used at the same time"); 431 return NULL; 432 } 433 434 on_write_error = BLOCK_ERR_STOP_ENOSPC; 435 if ((buf = qemu_opt_get(opts, "werror")) != NULL) { 436 if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) { 437 error_report("werror is not supported by this bus type"); 438 return NULL; 439 } 440 441 on_write_error = parse_block_error_action(buf, 0); 442 if (on_write_error < 0) { 443 return NULL; 444 } 445 } 446 447 on_read_error = BLOCK_ERR_REPORT; 448 if ((buf = qemu_opt_get(opts, "rerror")) != NULL) { 449 if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) { 450 error_report("rerror is not supported by this bus type"); 451 return NULL; 452 } 453 454 on_read_error = parse_block_error_action(buf, 1); 455 if (on_read_error < 0) { 456 return NULL; 457 } 458 } 459 460 if ((devaddr = qemu_opt_get(opts, "addr")) != NULL) { 461 if (type != IF_VIRTIO) { 462 error_report("addr is not supported by this bus type"); 463 return NULL; 464 } 465 } 466 467 /* compute bus and unit according index */ 468 469 if (index != -1) { 470 if (bus_id != 0 || unit_id != -1) { 471 error_report("index cannot be used with bus and unit"); 472 return NULL; 473 } 474 bus_id = drive_index_to_bus_id(type, index); 475 unit_id = drive_index_to_unit_id(type, index); 476 } 477 478 /* if user doesn't specify a unit_id, 479 * try to find the first free 480 */ 481 482 if (unit_id == -1) { 483 unit_id = 0; 484 while (drive_get(type, bus_id, unit_id) != NULL) { 485 unit_id++; 486 if (max_devs && unit_id >= max_devs) { 487 unit_id -= max_devs; 488 bus_id++; 489 } 490 } 491 } 492 493 /* check unit id */ 494 495 if (max_devs && unit_id >= max_devs) { 496 error_report("unit %d too big (max is %d)", 497 unit_id, max_devs - 1); 498 return NULL; 499 } 500 501 /* 502 * catch multiple definitions 503 */ 504 505 if (drive_get(type, bus_id, unit_id) != NULL) { 506 error_report("drive with bus=%d, unit=%d (index=%d) exists", 507 bus_id, unit_id, index); 508 return NULL; 509 } 510 511 /* init */ 512 513 dinfo = g_malloc0(sizeof(*dinfo)); 514 if ((buf = qemu_opts_id(opts)) != NULL) { 515 dinfo->id = g_strdup(buf); 516 } else { 517 /* no id supplied -> create one */ 518 dinfo->id = g_malloc0(32); 519 if (type == IF_IDE || type == IF_SCSI) 520 mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd"; 521 if (max_devs) 522 snprintf(dinfo->id, 32, "%s%i%s%i", 523 if_name[type], bus_id, mediastr, unit_id); 524 else 525 snprintf(dinfo->id, 32, "%s%s%i", 526 if_name[type], mediastr, unit_id); 527 } 528 dinfo->bdrv = bdrv_new(dinfo->id); 529 dinfo->devaddr = devaddr; 530 dinfo->type = type; 531 dinfo->bus = bus_id; 532 dinfo->unit = unit_id; 533 dinfo->opts = opts; 534 dinfo->refcount = 1; 535 if (serial) { 536 pstrcpy(dinfo->serial, sizeof(dinfo->serial), serial); 537 } 538 QTAILQ_INSERT_TAIL(&drives, dinfo, next); 539 540 bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error); 541 542 /* disk I/O throttling */ 543 bdrv_set_io_limits(dinfo->bdrv, &io_limits); 544 545 switch(type) { 546 case IF_IDE: 547 case IF_SCSI: 548 case IF_XEN: 549 case IF_NONE: 550 switch(media) { 551 case MEDIA_DISK: 552 if (cyls != 0) { 553 bdrv_set_geometry_hint(dinfo->bdrv, cyls, heads, secs); 554 bdrv_set_translation_hint(dinfo->bdrv, translation); 555 } 556 break; 557 case MEDIA_CDROM: 558 dinfo->media_cd = 1; 559 break; 560 } 561 break; 562 case IF_SD: 563 case IF_FLOPPY: 564 case IF_PFLASH: 565 case IF_MTD: 566 break; 567 case IF_VIRTIO: 568 /* add virtio block device */ 569 opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0, NULL); 570 if (arch_type == QEMU_ARCH_S390X) { 571 qemu_opt_set(opts, "driver", "virtio-blk-s390"); 572 } else { 573 qemu_opt_set(opts, "driver", "virtio-blk-pci"); 574 } 575 qemu_opt_set(opts, "drive", dinfo->id); 576 if (devaddr) 577 qemu_opt_set(opts, "addr", devaddr); 578 break; 579 default: 580 abort(); 581 } 582 if (!file || !*file) { 583 return dinfo; 584 } 585 if (snapshot) { 586 /* always use cache=unsafe with snapshot */ 587 bdrv_flags &= ~BDRV_O_CACHE_MASK; 588 bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH); 589 } 590 591 if (copy_on_read) { 592 bdrv_flags |= BDRV_O_COPY_ON_READ; 593 } 594 595 if (runstate_check(RUN_STATE_INMIGRATE)) { 596 bdrv_flags |= BDRV_O_INCOMING; 597 } 598 599 if (media == MEDIA_CDROM) { 600 /* CDROM is fine for any interface, don't check. */ 601 ro = 1; 602 } else if (ro == 1) { 603 if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY && 604 type != IF_NONE && type != IF_PFLASH) { 605 error_report("readonly not supported by this bus type"); 606 goto err; 607 } 608 } 609 610 bdrv_flags |= ro ? 0 : BDRV_O_RDWR; 611 612 ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv); 613 if (ret < 0) { 614 error_report("could not open disk image %s: %s", 615 file, strerror(-ret)); 616 goto err; 617 } 618 619 if (bdrv_key_required(dinfo->bdrv)) 620 autostart = 0; 621 return dinfo; 622 623 err: 624 bdrv_delete(dinfo->bdrv); 625 g_free(dinfo->id); 626 QTAILQ_REMOVE(&drives, dinfo, next); 627 g_free(dinfo); 628 return NULL; 629 } 630 631 void do_commit(Monitor *mon, const QDict *qdict) 632 { 633 const char *device = qdict_get_str(qdict, "device"); 634 BlockDriverState *bs; 635 int ret; 636 637 if (!strcmp(device, "all")) { 638 ret = bdrv_commit_all(); 639 if (ret == -EBUSY) { 640 qerror_report(QERR_DEVICE_IN_USE, device); 641 return; 642 } 643 } else { 644 bs = bdrv_find(device); 645 if (!bs) { 646 qerror_report(QERR_DEVICE_NOT_FOUND, device); 647 return; 648 } 649 ret = bdrv_commit(bs); 650 if (ret == -EBUSY) { 651 qerror_report(QERR_DEVICE_IN_USE, device); 652 return; 653 } 654 } 655 } 656 657 static void blockdev_do_action(int kind, void *data, Error **errp) 658 { 659 BlockdevAction action; 660 BlockdevActionList list; 661 662 action.kind = kind; 663 action.data = data; 664 list.value = &action; 665 list.next = NULL; 666 qmp_transaction(&list, errp); 667 } 668 669 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file, 670 bool has_format, const char *format, 671 bool has_mode, enum NewImageMode mode, 672 Error **errp) 673 { 674 BlockdevSnapshot snapshot = { 675 .device = (char *) device, 676 .snapshot_file = (char *) snapshot_file, 677 .has_format = has_format, 678 .format = (char *) format, 679 .has_mode = has_mode, 680 .mode = mode, 681 }; 682 blockdev_do_action(BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC, &snapshot, 683 errp); 684 } 685 686 687 /* New and old BlockDriverState structs for group snapshots */ 688 typedef struct BlkTransactionStates { 689 BlockDriverState *old_bs; 690 BlockDriverState *new_bs; 691 QSIMPLEQ_ENTRY(BlkTransactionStates) entry; 692 } BlkTransactionStates; 693 694 /* 695 * 'Atomic' group snapshots. The snapshots are taken as a set, and if any fail 696 * then we do not pivot any of the devices in the group, and abandon the 697 * snapshots 698 */ 699 void qmp_transaction(BlockdevActionList *dev_list, Error **errp) 700 { 701 int ret = 0; 702 BlockdevActionList *dev_entry = dev_list; 703 BlkTransactionStates *states, *next; 704 705 QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionStates) snap_bdrv_states; 706 QSIMPLEQ_INIT(&snap_bdrv_states); 707 708 /* drain all i/o before any snapshots */ 709 bdrv_drain_all(); 710 711 /* We don't do anything in this loop that commits us to the snapshot */ 712 while (NULL != dev_entry) { 713 BlockdevAction *dev_info = NULL; 714 BlockDriver *proto_drv; 715 BlockDriver *drv; 716 int flags; 717 enum NewImageMode mode; 718 const char *new_image_file; 719 const char *device; 720 const char *format = "qcow2"; 721 722 dev_info = dev_entry->value; 723 dev_entry = dev_entry->next; 724 725 states = g_malloc0(sizeof(BlkTransactionStates)); 726 QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, states, entry); 727 728 switch (dev_info->kind) { 729 case BLOCKDEV_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC: 730 device = dev_info->blockdev_snapshot_sync->device; 731 if (!dev_info->blockdev_snapshot_sync->has_mode) { 732 dev_info->blockdev_snapshot_sync->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS; 733 } 734 new_image_file = dev_info->blockdev_snapshot_sync->snapshot_file; 735 if (dev_info->blockdev_snapshot_sync->has_format) { 736 format = dev_info->blockdev_snapshot_sync->format; 737 } 738 mode = dev_info->blockdev_snapshot_sync->mode; 739 break; 740 default: 741 abort(); 742 } 743 744 drv = bdrv_find_format(format); 745 if (!drv) { 746 error_set(errp, QERR_INVALID_BLOCK_FORMAT, format); 747 goto delete_and_fail; 748 } 749 750 states->old_bs = bdrv_find(device); 751 if (!states->old_bs) { 752 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 753 goto delete_and_fail; 754 } 755 756 if (!bdrv_is_inserted(states->old_bs)) { 757 error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device); 758 goto delete_and_fail; 759 } 760 761 if (bdrv_in_use(states->old_bs)) { 762 error_set(errp, QERR_DEVICE_IN_USE, device); 763 goto delete_and_fail; 764 } 765 766 if (!bdrv_is_read_only(states->old_bs)) { 767 if (bdrv_flush(states->old_bs)) { 768 error_set(errp, QERR_IO_ERROR); 769 goto delete_and_fail; 770 } 771 } 772 773 flags = states->old_bs->open_flags; 774 775 proto_drv = bdrv_find_protocol(new_image_file); 776 if (!proto_drv) { 777 error_set(errp, QERR_INVALID_BLOCK_FORMAT, format); 778 goto delete_and_fail; 779 } 780 781 /* create new image w/backing file */ 782 if (mode != NEW_IMAGE_MODE_EXISTING) { 783 ret = bdrv_img_create(new_image_file, format, 784 states->old_bs->filename, 785 states->old_bs->drv->format_name, 786 NULL, -1, flags); 787 if (ret) { 788 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file); 789 goto delete_and_fail; 790 } 791 } 792 793 /* We will manually add the backing_hd field to the bs later */ 794 states->new_bs = bdrv_new(""); 795 ret = bdrv_open(states->new_bs, new_image_file, 796 flags | BDRV_O_NO_BACKING, drv); 797 if (ret != 0) { 798 error_set(errp, QERR_OPEN_FILE_FAILED, new_image_file); 799 goto delete_and_fail; 800 } 801 } 802 803 804 /* Now we are going to do the actual pivot. Everything up to this point 805 * is reversible, but we are committed at this point */ 806 QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) { 807 /* This removes our old bs from the bdrv_states, and adds the new bs */ 808 bdrv_append(states->new_bs, states->old_bs); 809 } 810 811 /* success */ 812 goto exit; 813 814 delete_and_fail: 815 /* 816 * failure, and it is all-or-none; abandon each new bs, and keep using 817 * the original bs for all images 818 */ 819 QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) { 820 if (states->new_bs) { 821 bdrv_delete(states->new_bs); 822 } 823 } 824 exit: 825 QSIMPLEQ_FOREACH_SAFE(states, &snap_bdrv_states, entry, next) { 826 g_free(states); 827 } 828 return; 829 } 830 831 832 static void eject_device(BlockDriverState *bs, int force, Error **errp) 833 { 834 if (bdrv_in_use(bs)) { 835 error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs)); 836 return; 837 } 838 if (!bdrv_dev_has_removable_media(bs)) { 839 error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs)); 840 return; 841 } 842 843 if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) { 844 bdrv_dev_eject_request(bs, force); 845 if (!force) { 846 error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs)); 847 return; 848 } 849 } 850 851 bdrv_close(bs); 852 } 853 854 void qmp_eject(const char *device, bool has_force, bool force, Error **errp) 855 { 856 BlockDriverState *bs; 857 858 bs = bdrv_find(device); 859 if (!bs) { 860 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 861 return; 862 } 863 864 eject_device(bs, force, errp); 865 } 866 867 void qmp_block_passwd(const char *device, const char *password, Error **errp) 868 { 869 BlockDriverState *bs; 870 int err; 871 872 bs = bdrv_find(device); 873 if (!bs) { 874 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 875 return; 876 } 877 878 err = bdrv_set_key(bs, password); 879 if (err == -EINVAL) { 880 error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs)); 881 return; 882 } else if (err < 0) { 883 error_set(errp, QERR_INVALID_PASSWORD); 884 return; 885 } 886 } 887 888 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename, 889 int bdrv_flags, BlockDriver *drv, 890 const char *password, Error **errp) 891 { 892 if (bdrv_open(bs, filename, bdrv_flags, drv) < 0) { 893 error_set(errp, QERR_OPEN_FILE_FAILED, filename); 894 return; 895 } 896 897 if (bdrv_key_required(bs)) { 898 if (password) { 899 if (bdrv_set_key(bs, password) < 0) { 900 error_set(errp, QERR_INVALID_PASSWORD); 901 } 902 } else { 903 error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs), 904 bdrv_get_encrypted_filename(bs)); 905 } 906 } else if (password) { 907 error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs)); 908 } 909 } 910 911 void qmp_change_blockdev(const char *device, const char *filename, 912 bool has_format, const char *format, Error **errp) 913 { 914 BlockDriverState *bs; 915 BlockDriver *drv = NULL; 916 int bdrv_flags; 917 Error *err = NULL; 918 919 bs = bdrv_find(device); 920 if (!bs) { 921 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 922 return; 923 } 924 925 if (format) { 926 drv = bdrv_find_whitelisted_format(format); 927 if (!drv) { 928 error_set(errp, QERR_INVALID_BLOCK_FORMAT, format); 929 return; 930 } 931 } 932 933 eject_device(bs, 0, &err); 934 if (error_is_set(&err)) { 935 error_propagate(errp, err); 936 return; 937 } 938 939 bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR; 940 bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0; 941 942 qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp); 943 } 944 945 /* throttling disk I/O limits */ 946 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd, 947 int64_t bps_wr, int64_t iops, int64_t iops_rd, 948 int64_t iops_wr, Error **errp) 949 { 950 BlockIOLimit io_limits; 951 BlockDriverState *bs; 952 953 bs = bdrv_find(device); 954 if (!bs) { 955 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 956 return; 957 } 958 959 io_limits.bps[BLOCK_IO_LIMIT_TOTAL] = bps; 960 io_limits.bps[BLOCK_IO_LIMIT_READ] = bps_rd; 961 io_limits.bps[BLOCK_IO_LIMIT_WRITE] = bps_wr; 962 io_limits.iops[BLOCK_IO_LIMIT_TOTAL]= iops; 963 io_limits.iops[BLOCK_IO_LIMIT_READ] = iops_rd; 964 io_limits.iops[BLOCK_IO_LIMIT_WRITE]= iops_wr; 965 966 if (!do_check_io_limits(&io_limits)) { 967 error_set(errp, QERR_INVALID_PARAMETER_COMBINATION); 968 return; 969 } 970 971 bs->io_limits = io_limits; 972 bs->slice_time = BLOCK_IO_SLICE_TIME; 973 974 if (!bs->io_limits_enabled && bdrv_io_limits_enabled(bs)) { 975 bdrv_io_limits_enable(bs); 976 } else if (bs->io_limits_enabled && !bdrv_io_limits_enabled(bs)) { 977 bdrv_io_limits_disable(bs); 978 } else { 979 if (bs->block_timer) { 980 qemu_mod_timer(bs->block_timer, qemu_get_clock_ns(vm_clock)); 981 } 982 } 983 } 984 985 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data) 986 { 987 const char *id = qdict_get_str(qdict, "id"); 988 BlockDriverState *bs; 989 990 bs = bdrv_find(id); 991 if (!bs) { 992 qerror_report(QERR_DEVICE_NOT_FOUND, id); 993 return -1; 994 } 995 if (bdrv_in_use(bs)) { 996 qerror_report(QERR_DEVICE_IN_USE, id); 997 return -1; 998 } 999 1000 /* quiesce block driver; prevent further io */ 1001 bdrv_drain_all(); 1002 bdrv_flush(bs); 1003 bdrv_close(bs); 1004 1005 /* if we have a device attached to this BlockDriverState 1006 * then we need to make the drive anonymous until the device 1007 * can be removed. If this is a drive with no device backing 1008 * then we can just get rid of the block driver state right here. 1009 */ 1010 if (bdrv_get_attached_dev(bs)) { 1011 bdrv_make_anon(bs); 1012 } else { 1013 drive_uninit(drive_get_by_blockdev(bs)); 1014 } 1015 1016 return 0; 1017 } 1018 1019 void qmp_block_resize(const char *device, int64_t size, Error **errp) 1020 { 1021 BlockDriverState *bs; 1022 1023 bs = bdrv_find(device); 1024 if (!bs) { 1025 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 1026 return; 1027 } 1028 1029 if (size < 0) { 1030 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size"); 1031 return; 1032 } 1033 1034 switch (bdrv_truncate(bs, size)) { 1035 case 0: 1036 break; 1037 case -ENOMEDIUM: 1038 error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device); 1039 break; 1040 case -ENOTSUP: 1041 error_set(errp, QERR_UNSUPPORTED); 1042 break; 1043 case -EACCES: 1044 error_set(errp, QERR_DEVICE_IS_READ_ONLY, device); 1045 break; 1046 case -EBUSY: 1047 error_set(errp, QERR_DEVICE_IN_USE, device); 1048 break; 1049 default: 1050 error_set(errp, QERR_UNDEFINED_ERROR); 1051 break; 1052 } 1053 } 1054 1055 static QObject *qobject_from_block_job(BlockJob *job) 1056 { 1057 return qobject_from_jsonf("{ 'type': %s," 1058 "'device': %s," 1059 "'len': %" PRId64 "," 1060 "'offset': %" PRId64 "," 1061 "'speed': %" PRId64 " }", 1062 job->job_type->job_type, 1063 bdrv_get_device_name(job->bs), 1064 job->len, 1065 job->offset, 1066 job->speed); 1067 } 1068 1069 static void block_stream_cb(void *opaque, int ret) 1070 { 1071 BlockDriverState *bs = opaque; 1072 QObject *obj; 1073 1074 trace_block_stream_cb(bs, bs->job, ret); 1075 1076 assert(bs->job); 1077 obj = qobject_from_block_job(bs->job); 1078 if (ret < 0) { 1079 QDict *dict = qobject_to_qdict(obj); 1080 qdict_put(dict, "error", qstring_from_str(strerror(-ret))); 1081 } 1082 1083 if (block_job_is_cancelled(bs->job)) { 1084 monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj); 1085 } else { 1086 monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj); 1087 } 1088 qobject_decref(obj); 1089 1090 drive_put_ref_bh_schedule(drive_get_by_blockdev(bs)); 1091 } 1092 1093 void qmp_block_stream(const char *device, bool has_base, 1094 const char *base, bool has_speed, 1095 int64_t speed, Error **errp) 1096 { 1097 BlockDriverState *bs; 1098 BlockDriverState *base_bs = NULL; 1099 Error *local_err = NULL; 1100 1101 bs = bdrv_find(device); 1102 if (!bs) { 1103 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 1104 return; 1105 } 1106 1107 if (base) { 1108 base_bs = bdrv_find_backing_image(bs, base); 1109 if (base_bs == NULL) { 1110 error_set(errp, QERR_BASE_NOT_FOUND, base); 1111 return; 1112 } 1113 } 1114 1115 stream_start(bs, base_bs, base, has_speed ? speed : 0, 1116 block_stream_cb, bs, &local_err); 1117 if (error_is_set(&local_err)) { 1118 error_propagate(errp, local_err); 1119 return; 1120 } 1121 1122 /* Grab a reference so hotplug does not delete the BlockDriverState from 1123 * underneath us. 1124 */ 1125 drive_get_ref(drive_get_by_blockdev(bs)); 1126 1127 trace_qmp_block_stream(bs, bs->job); 1128 } 1129 1130 static BlockJob *find_block_job(const char *device) 1131 { 1132 BlockDriverState *bs; 1133 1134 bs = bdrv_find(device); 1135 if (!bs || !bs->job) { 1136 return NULL; 1137 } 1138 return bs->job; 1139 } 1140 1141 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp) 1142 { 1143 BlockJob *job = find_block_job(device); 1144 1145 if (!job) { 1146 error_set(errp, QERR_DEVICE_NOT_ACTIVE, device); 1147 return; 1148 } 1149 1150 block_job_set_speed(job, speed, errp); 1151 } 1152 1153 void qmp_block_job_cancel(const char *device, Error **errp) 1154 { 1155 BlockJob *job = find_block_job(device); 1156 1157 if (!job) { 1158 error_set(errp, QERR_DEVICE_NOT_ACTIVE, device); 1159 return; 1160 } 1161 1162 trace_qmp_block_job_cancel(job); 1163 block_job_cancel(job); 1164 } 1165 1166 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs) 1167 { 1168 BlockJobInfoList **prev = opaque; 1169 BlockJob *job = bs->job; 1170 1171 if (job) { 1172 BlockJobInfoList *elem; 1173 BlockJobInfo *info = g_new(BlockJobInfo, 1); 1174 *info = (BlockJobInfo){ 1175 .type = g_strdup(job->job_type->job_type), 1176 .device = g_strdup(bdrv_get_device_name(bs)), 1177 .len = job->len, 1178 .offset = job->offset, 1179 .speed = job->speed, 1180 }; 1181 1182 elem = g_new0(BlockJobInfoList, 1); 1183 elem->value = info; 1184 1185 (*prev)->next = elem; 1186 *prev = elem; 1187 } 1188 } 1189 1190 BlockJobInfoList *qmp_query_block_jobs(Error **errp) 1191 { 1192 /* Dummy is a fake list element for holding the head pointer */ 1193 BlockJobInfoList dummy = {}; 1194 BlockJobInfoList *prev = &dummy; 1195 bdrv_iterate(do_qmp_query_block_jobs_one, &prev); 1196 return dummy.next; 1197 } 1198