1 /* 2 * QEMU System Emulator block driver 3 * 4 * Copyright (c) 2003 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 #include "qemu/osdep.h" 25 #include "trace.h" 26 #include "block/block_int.h" 27 #include "block/blockjob.h" 28 #include "block/nbd.h" 29 #include "qemu/error-report.h" 30 #include "module_block.h" 31 #include "qemu/module.h" 32 #include "qapi/qmp/qerror.h" 33 #include "qapi/qmp/qbool.h" 34 #include "qapi/qmp/qjson.h" 35 #include "sysemu/block-backend.h" 36 #include "sysemu/sysemu.h" 37 #include "qemu/notify.h" 38 #include "qemu/coroutine.h" 39 #include "block/qapi.h" 40 #include "qmp-commands.h" 41 #include "qemu/timer.h" 42 #include "qapi-event.h" 43 #include "qemu/cutils.h" 44 #include "qemu/id.h" 45 #include "qapi/util.h" 46 47 #ifdef CONFIG_BSD 48 #include <sys/ioctl.h> 49 #include <sys/queue.h> 50 #ifndef __DragonFly__ 51 #include <sys/disk.h> 52 #endif 53 #endif 54 55 #ifdef _WIN32 56 #include <windows.h> 57 #endif 58 59 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ 60 61 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states = 62 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states); 63 64 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states = 65 QTAILQ_HEAD_INITIALIZER(all_bdrv_states); 66 67 static QLIST_HEAD(, BlockDriver) bdrv_drivers = 68 QLIST_HEAD_INITIALIZER(bdrv_drivers); 69 70 static BlockDriverState *bdrv_open_inherit(const char *filename, 71 const char *reference, 72 QDict *options, int flags, 73 BlockDriverState *parent, 74 const BdrvChildRole *child_role, 75 Error **errp); 76 77 /* If non-zero, use only whitelisted block drivers */ 78 static int use_bdrv_whitelist; 79 80 #ifdef _WIN32 81 static int is_windows_drive_prefix(const char *filename) 82 { 83 return (((filename[0] >= 'a' && filename[0] <= 'z') || 84 (filename[0] >= 'A' && filename[0] <= 'Z')) && 85 filename[1] == ':'); 86 } 87 88 int is_windows_drive(const char *filename) 89 { 90 if (is_windows_drive_prefix(filename) && 91 filename[2] == '\0') 92 return 1; 93 if (strstart(filename, "\\\\.\\", NULL) || 94 strstart(filename, "//./", NULL)) 95 return 1; 96 return 0; 97 } 98 #endif 99 100 size_t bdrv_opt_mem_align(BlockDriverState *bs) 101 { 102 if (!bs || !bs->drv) { 103 /* page size or 4k (hdd sector size) should be on the safe side */ 104 return MAX(4096, getpagesize()); 105 } 106 107 return bs->bl.opt_mem_alignment; 108 } 109 110 size_t bdrv_min_mem_align(BlockDriverState *bs) 111 { 112 if (!bs || !bs->drv) { 113 /* page size or 4k (hdd sector size) should be on the safe side */ 114 return MAX(4096, getpagesize()); 115 } 116 117 return bs->bl.min_mem_alignment; 118 } 119 120 /* check if the path starts with "<protocol>:" */ 121 int path_has_protocol(const char *path) 122 { 123 const char *p; 124 125 #ifdef _WIN32 126 if (is_windows_drive(path) || 127 is_windows_drive_prefix(path)) { 128 return 0; 129 } 130 p = path + strcspn(path, ":/\\"); 131 #else 132 p = path + strcspn(path, ":/"); 133 #endif 134 135 return *p == ':'; 136 } 137 138 int path_is_absolute(const char *path) 139 { 140 #ifdef _WIN32 141 /* specific case for names like: "\\.\d:" */ 142 if (is_windows_drive(path) || is_windows_drive_prefix(path)) { 143 return 1; 144 } 145 return (*path == '/' || *path == '\\'); 146 #else 147 return (*path == '/'); 148 #endif 149 } 150 151 /* if filename is absolute, just copy it to dest. Otherwise, build a 152 path to it by considering it is relative to base_path. URL are 153 supported. */ 154 void path_combine(char *dest, int dest_size, 155 const char *base_path, 156 const char *filename) 157 { 158 const char *p, *p1; 159 int len; 160 161 if (dest_size <= 0) 162 return; 163 if (path_is_absolute(filename)) { 164 pstrcpy(dest, dest_size, filename); 165 } else { 166 p = strchr(base_path, ':'); 167 if (p) 168 p++; 169 else 170 p = base_path; 171 p1 = strrchr(base_path, '/'); 172 #ifdef _WIN32 173 { 174 const char *p2; 175 p2 = strrchr(base_path, '\\'); 176 if (!p1 || p2 > p1) 177 p1 = p2; 178 } 179 #endif 180 if (p1) 181 p1++; 182 else 183 p1 = base_path; 184 if (p1 > p) 185 p = p1; 186 len = p - base_path; 187 if (len > dest_size - 1) 188 len = dest_size - 1; 189 memcpy(dest, base_path, len); 190 dest[len] = '\0'; 191 pstrcat(dest, dest_size, filename); 192 } 193 } 194 195 void bdrv_get_full_backing_filename_from_filename(const char *backed, 196 const char *backing, 197 char *dest, size_t sz, 198 Error **errp) 199 { 200 if (backing[0] == '\0' || path_has_protocol(backing) || 201 path_is_absolute(backing)) 202 { 203 pstrcpy(dest, sz, backing); 204 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) { 205 error_setg(errp, "Cannot use relative backing file names for '%s'", 206 backed); 207 } else { 208 path_combine(dest, sz, backed, backing); 209 } 210 } 211 212 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz, 213 Error **errp) 214 { 215 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename; 216 217 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file, 218 dest, sz, errp); 219 } 220 221 void bdrv_register(BlockDriver *bdrv) 222 { 223 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); 224 } 225 226 BlockDriverState *bdrv_new(void) 227 { 228 BlockDriverState *bs; 229 int i; 230 231 bs = g_new0(BlockDriverState, 1); 232 QLIST_INIT(&bs->dirty_bitmaps); 233 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { 234 QLIST_INIT(&bs->op_blockers[i]); 235 } 236 notifier_with_return_list_init(&bs->before_write_notifiers); 237 bs->refcnt = 1; 238 bs->aio_context = qemu_get_aio_context(); 239 240 qemu_co_queue_init(&bs->flush_queue); 241 242 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list); 243 244 return bs; 245 } 246 247 static BlockDriver *bdrv_do_find_format(const char *format_name) 248 { 249 BlockDriver *drv1; 250 251 QLIST_FOREACH(drv1, &bdrv_drivers, list) { 252 if (!strcmp(drv1->format_name, format_name)) { 253 return drv1; 254 } 255 } 256 257 return NULL; 258 } 259 260 BlockDriver *bdrv_find_format(const char *format_name) 261 { 262 BlockDriver *drv1; 263 int i; 264 265 drv1 = bdrv_do_find_format(format_name); 266 if (drv1) { 267 return drv1; 268 } 269 270 /* The driver isn't registered, maybe we need to load a module */ 271 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { 272 if (!strcmp(block_driver_modules[i].format_name, format_name)) { 273 block_module_load_one(block_driver_modules[i].library_name); 274 break; 275 } 276 } 277 278 return bdrv_do_find_format(format_name); 279 } 280 281 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) 282 { 283 static const char *whitelist_rw[] = { 284 CONFIG_BDRV_RW_WHITELIST 285 }; 286 static const char *whitelist_ro[] = { 287 CONFIG_BDRV_RO_WHITELIST 288 }; 289 const char **p; 290 291 if (!whitelist_rw[0] && !whitelist_ro[0]) { 292 return 1; /* no whitelist, anything goes */ 293 } 294 295 for (p = whitelist_rw; *p; p++) { 296 if (!strcmp(drv->format_name, *p)) { 297 return 1; 298 } 299 } 300 if (read_only) { 301 for (p = whitelist_ro; *p; p++) { 302 if (!strcmp(drv->format_name, *p)) { 303 return 1; 304 } 305 } 306 } 307 return 0; 308 } 309 310 bool bdrv_uses_whitelist(void) 311 { 312 return use_bdrv_whitelist; 313 } 314 315 typedef struct CreateCo { 316 BlockDriver *drv; 317 char *filename; 318 QemuOpts *opts; 319 int ret; 320 Error *err; 321 } CreateCo; 322 323 static void coroutine_fn bdrv_create_co_entry(void *opaque) 324 { 325 Error *local_err = NULL; 326 int ret; 327 328 CreateCo *cco = opaque; 329 assert(cco->drv); 330 331 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err); 332 error_propagate(&cco->err, local_err); 333 cco->ret = ret; 334 } 335 336 int bdrv_create(BlockDriver *drv, const char* filename, 337 QemuOpts *opts, Error **errp) 338 { 339 int ret; 340 341 Coroutine *co; 342 CreateCo cco = { 343 .drv = drv, 344 .filename = g_strdup(filename), 345 .opts = opts, 346 .ret = NOT_DONE, 347 .err = NULL, 348 }; 349 350 if (!drv->bdrv_create) { 351 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name); 352 ret = -ENOTSUP; 353 goto out; 354 } 355 356 if (qemu_in_coroutine()) { 357 /* Fast-path if already in coroutine context */ 358 bdrv_create_co_entry(&cco); 359 } else { 360 co = qemu_coroutine_create(bdrv_create_co_entry, &cco); 361 qemu_coroutine_enter(co); 362 while (cco.ret == NOT_DONE) { 363 aio_poll(qemu_get_aio_context(), true); 364 } 365 } 366 367 ret = cco.ret; 368 if (ret < 0) { 369 if (cco.err) { 370 error_propagate(errp, cco.err); 371 } else { 372 error_setg_errno(errp, -ret, "Could not create image"); 373 } 374 } 375 376 out: 377 g_free(cco.filename); 378 return ret; 379 } 380 381 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp) 382 { 383 BlockDriver *drv; 384 Error *local_err = NULL; 385 int ret; 386 387 drv = bdrv_find_protocol(filename, true, errp); 388 if (drv == NULL) { 389 return -ENOENT; 390 } 391 392 ret = bdrv_create(drv, filename, opts, &local_err); 393 error_propagate(errp, local_err); 394 return ret; 395 } 396 397 /** 398 * Try to get @bs's logical and physical block size. 399 * On success, store them in @bsz struct and return 0. 400 * On failure return -errno. 401 * @bs must not be empty. 402 */ 403 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 404 { 405 BlockDriver *drv = bs->drv; 406 407 if (drv && drv->bdrv_probe_blocksizes) { 408 return drv->bdrv_probe_blocksizes(bs, bsz); 409 } 410 411 return -ENOTSUP; 412 } 413 414 /** 415 * Try to get @bs's geometry (cyls, heads, sectors). 416 * On success, store them in @geo struct and return 0. 417 * On failure return -errno. 418 * @bs must not be empty. 419 */ 420 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 421 { 422 BlockDriver *drv = bs->drv; 423 424 if (drv && drv->bdrv_probe_geometry) { 425 return drv->bdrv_probe_geometry(bs, geo); 426 } 427 428 return -ENOTSUP; 429 } 430 431 /* 432 * Create a uniquely-named empty temporary file. 433 * Return 0 upon success, otherwise a negative errno value. 434 */ 435 int get_tmp_filename(char *filename, int size) 436 { 437 #ifdef _WIN32 438 char temp_dir[MAX_PATH]; 439 /* GetTempFileName requires that its output buffer (4th param) 440 have length MAX_PATH or greater. */ 441 assert(size >= MAX_PATH); 442 return (GetTempPath(MAX_PATH, temp_dir) 443 && GetTempFileName(temp_dir, "qem", 0, filename) 444 ? 0 : -GetLastError()); 445 #else 446 int fd; 447 const char *tmpdir; 448 tmpdir = getenv("TMPDIR"); 449 if (!tmpdir) { 450 tmpdir = "/var/tmp"; 451 } 452 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) { 453 return -EOVERFLOW; 454 } 455 fd = mkstemp(filename); 456 if (fd < 0) { 457 return -errno; 458 } 459 if (close(fd) != 0) { 460 unlink(filename); 461 return -errno; 462 } 463 return 0; 464 #endif 465 } 466 467 /* 468 * Detect host devices. By convention, /dev/cdrom[N] is always 469 * recognized as a host CDROM. 470 */ 471 static BlockDriver *find_hdev_driver(const char *filename) 472 { 473 int score_max = 0, score; 474 BlockDriver *drv = NULL, *d; 475 476 QLIST_FOREACH(d, &bdrv_drivers, list) { 477 if (d->bdrv_probe_device) { 478 score = d->bdrv_probe_device(filename); 479 if (score > score_max) { 480 score_max = score; 481 drv = d; 482 } 483 } 484 } 485 486 return drv; 487 } 488 489 static BlockDriver *bdrv_do_find_protocol(const char *protocol) 490 { 491 BlockDriver *drv1; 492 493 QLIST_FOREACH(drv1, &bdrv_drivers, list) { 494 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) { 495 return drv1; 496 } 497 } 498 499 return NULL; 500 } 501 502 BlockDriver *bdrv_find_protocol(const char *filename, 503 bool allow_protocol_prefix, 504 Error **errp) 505 { 506 BlockDriver *drv1; 507 char protocol[128]; 508 int len; 509 const char *p; 510 int i; 511 512 /* TODO Drivers without bdrv_file_open must be specified explicitly */ 513 514 /* 515 * XXX(hch): we really should not let host device detection 516 * override an explicit protocol specification, but moving this 517 * later breaks access to device names with colons in them. 518 * Thanks to the brain-dead persistent naming schemes on udev- 519 * based Linux systems those actually are quite common. 520 */ 521 drv1 = find_hdev_driver(filename); 522 if (drv1) { 523 return drv1; 524 } 525 526 if (!path_has_protocol(filename) || !allow_protocol_prefix) { 527 return &bdrv_file; 528 } 529 530 p = strchr(filename, ':'); 531 assert(p != NULL); 532 len = p - filename; 533 if (len > sizeof(protocol) - 1) 534 len = sizeof(protocol) - 1; 535 memcpy(protocol, filename, len); 536 protocol[len] = '\0'; 537 538 drv1 = bdrv_do_find_protocol(protocol); 539 if (drv1) { 540 return drv1; 541 } 542 543 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { 544 if (block_driver_modules[i].protocol_name && 545 !strcmp(block_driver_modules[i].protocol_name, protocol)) { 546 block_module_load_one(block_driver_modules[i].library_name); 547 break; 548 } 549 } 550 551 drv1 = bdrv_do_find_protocol(protocol); 552 if (!drv1) { 553 error_setg(errp, "Unknown protocol '%s'", protocol); 554 } 555 return drv1; 556 } 557 558 /* 559 * Guess image format by probing its contents. 560 * This is not a good idea when your image is raw (CVE-2008-2004), but 561 * we do it anyway for backward compatibility. 562 * 563 * @buf contains the image's first @buf_size bytes. 564 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE, 565 * but can be smaller if the image file is smaller) 566 * @filename is its filename. 567 * 568 * For all block drivers, call the bdrv_probe() method to get its 569 * probing score. 570 * Return the first block driver with the highest probing score. 571 */ 572 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, 573 const char *filename) 574 { 575 int score_max = 0, score; 576 BlockDriver *drv = NULL, *d; 577 578 QLIST_FOREACH(d, &bdrv_drivers, list) { 579 if (d->bdrv_probe) { 580 score = d->bdrv_probe(buf, buf_size, filename); 581 if (score > score_max) { 582 score_max = score; 583 drv = d; 584 } 585 } 586 } 587 588 return drv; 589 } 590 591 static int find_image_format(BdrvChild *file, const char *filename, 592 BlockDriver **pdrv, Error **errp) 593 { 594 BlockDriverState *bs = file->bs; 595 BlockDriver *drv; 596 uint8_t buf[BLOCK_PROBE_BUF_SIZE]; 597 int ret = 0; 598 599 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ 600 if (bdrv_is_sg(bs) || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) { 601 *pdrv = &bdrv_raw; 602 return ret; 603 } 604 605 ret = bdrv_pread(file, 0, buf, sizeof(buf)); 606 if (ret < 0) { 607 error_setg_errno(errp, -ret, "Could not read image for determining its " 608 "format"); 609 *pdrv = NULL; 610 return ret; 611 } 612 613 drv = bdrv_probe_all(buf, ret, filename); 614 if (!drv) { 615 error_setg(errp, "Could not determine image format: No compatible " 616 "driver found"); 617 ret = -ENOENT; 618 } 619 *pdrv = drv; 620 return ret; 621 } 622 623 /** 624 * Set the current 'total_sectors' value 625 * Return 0 on success, -errno on error. 626 */ 627 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint) 628 { 629 BlockDriver *drv = bs->drv; 630 631 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ 632 if (bdrv_is_sg(bs)) 633 return 0; 634 635 /* query actual device if possible, otherwise just trust the hint */ 636 if (drv->bdrv_getlength) { 637 int64_t length = drv->bdrv_getlength(bs); 638 if (length < 0) { 639 return length; 640 } 641 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); 642 } 643 644 bs->total_sectors = hint; 645 return 0; 646 } 647 648 /** 649 * Combines a QDict of new block driver @options with any missing options taken 650 * from @old_options, so that leaving out an option defaults to its old value. 651 */ 652 static void bdrv_join_options(BlockDriverState *bs, QDict *options, 653 QDict *old_options) 654 { 655 if (bs->drv && bs->drv->bdrv_join_options) { 656 bs->drv->bdrv_join_options(options, old_options); 657 } else { 658 qdict_join(options, old_options, false); 659 } 660 } 661 662 /** 663 * Set open flags for a given discard mode 664 * 665 * Return 0 on success, -1 if the discard mode was invalid. 666 */ 667 int bdrv_parse_discard_flags(const char *mode, int *flags) 668 { 669 *flags &= ~BDRV_O_UNMAP; 670 671 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { 672 /* do nothing */ 673 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { 674 *flags |= BDRV_O_UNMAP; 675 } else { 676 return -1; 677 } 678 679 return 0; 680 } 681 682 /** 683 * Set open flags for a given cache mode 684 * 685 * Return 0 on success, -1 if the cache mode was invalid. 686 */ 687 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) 688 { 689 *flags &= ~BDRV_O_CACHE_MASK; 690 691 if (!strcmp(mode, "off") || !strcmp(mode, "none")) { 692 *writethrough = false; 693 *flags |= BDRV_O_NOCACHE; 694 } else if (!strcmp(mode, "directsync")) { 695 *writethrough = true; 696 *flags |= BDRV_O_NOCACHE; 697 } else if (!strcmp(mode, "writeback")) { 698 *writethrough = false; 699 } else if (!strcmp(mode, "unsafe")) { 700 *writethrough = false; 701 *flags |= BDRV_O_NO_FLUSH; 702 } else if (!strcmp(mode, "writethrough")) { 703 *writethrough = true; 704 } else { 705 return -1; 706 } 707 708 return 0; 709 } 710 711 static void bdrv_child_cb_drained_begin(BdrvChild *child) 712 { 713 BlockDriverState *bs = child->opaque; 714 bdrv_drained_begin(bs); 715 } 716 717 static void bdrv_child_cb_drained_end(BdrvChild *child) 718 { 719 BlockDriverState *bs = child->opaque; 720 bdrv_drained_end(bs); 721 } 722 723 /* 724 * Returns the options and flags that a temporary snapshot should get, based on 725 * the originally requested flags (the originally requested image will have 726 * flags like a backing file) 727 */ 728 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, 729 int parent_flags, QDict *parent_options) 730 { 731 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; 732 733 /* For temporary files, unconditional cache=unsafe is fine */ 734 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); 735 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); 736 737 /* Copy the read-only option from the parent */ 738 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); 739 740 /* aio=native doesn't work for cache.direct=off, so disable it for the 741 * temporary snapshot */ 742 *child_flags &= ~BDRV_O_NATIVE_AIO; 743 } 744 745 /* 746 * Returns the options and flags that bs->file should get if a protocol driver 747 * is expected, based on the given options and flags for the parent BDS 748 */ 749 static void bdrv_inherited_options(int *child_flags, QDict *child_options, 750 int parent_flags, QDict *parent_options) 751 { 752 int flags = parent_flags; 753 754 /* Enable protocol handling, disable format probing for bs->file */ 755 flags |= BDRV_O_PROTOCOL; 756 757 /* If the cache mode isn't explicitly set, inherit direct and no-flush from 758 * the parent. */ 759 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); 760 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); 761 762 /* Inherit the read-only option from the parent if it's not set */ 763 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); 764 765 /* Our block drivers take care to send flushes and respect unmap policy, 766 * so we can default to enable both on lower layers regardless of the 767 * corresponding parent options. */ 768 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); 769 770 /* Clear flags that only apply to the top layer */ 771 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ | 772 BDRV_O_NO_IO); 773 774 *child_flags = flags; 775 } 776 777 const BdrvChildRole child_file = { 778 .inherit_options = bdrv_inherited_options, 779 .drained_begin = bdrv_child_cb_drained_begin, 780 .drained_end = bdrv_child_cb_drained_end, 781 }; 782 783 /* 784 * Returns the options and flags that bs->file should get if the use of formats 785 * (and not only protocols) is permitted for it, based on the given options and 786 * flags for the parent BDS 787 */ 788 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options, 789 int parent_flags, QDict *parent_options) 790 { 791 child_file.inherit_options(child_flags, child_options, 792 parent_flags, parent_options); 793 794 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO); 795 } 796 797 const BdrvChildRole child_format = { 798 .inherit_options = bdrv_inherited_fmt_options, 799 .drained_begin = bdrv_child_cb_drained_begin, 800 .drained_end = bdrv_child_cb_drained_end, 801 }; 802 803 /* 804 * Returns the options and flags that bs->backing should get, based on the 805 * given options and flags for the parent BDS 806 */ 807 static void bdrv_backing_options(int *child_flags, QDict *child_options, 808 int parent_flags, QDict *parent_options) 809 { 810 int flags = parent_flags; 811 812 /* The cache mode is inherited unmodified for backing files; except WCE, 813 * which is only applied on the top level (BlockBackend) */ 814 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); 815 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); 816 817 /* backing files always opened read-only */ 818 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); 819 flags &= ~BDRV_O_COPY_ON_READ; 820 821 /* snapshot=on is handled on the top layer */ 822 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY); 823 824 *child_flags = flags; 825 } 826 827 static const BdrvChildRole child_backing = { 828 .inherit_options = bdrv_backing_options, 829 .drained_begin = bdrv_child_cb_drained_begin, 830 .drained_end = bdrv_child_cb_drained_end, 831 }; 832 833 static int bdrv_open_flags(BlockDriverState *bs, int flags) 834 { 835 int open_flags = flags; 836 837 /* 838 * Clear flags that are internal to the block layer before opening the 839 * image. 840 */ 841 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); 842 843 /* 844 * Snapshots should be writable. 845 */ 846 if (flags & BDRV_O_TEMPORARY) { 847 open_flags |= BDRV_O_RDWR; 848 } 849 850 return open_flags; 851 } 852 853 static void update_flags_from_options(int *flags, QemuOpts *opts) 854 { 855 *flags &= ~BDRV_O_CACHE_MASK; 856 857 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH)); 858 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { 859 *flags |= BDRV_O_NO_FLUSH; 860 } 861 862 assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT)); 863 if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) { 864 *flags |= BDRV_O_NOCACHE; 865 } 866 867 *flags &= ~BDRV_O_RDWR; 868 869 assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY)); 870 if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) { 871 *flags |= BDRV_O_RDWR; 872 } 873 874 } 875 876 static void update_options_from_flags(QDict *options, int flags) 877 { 878 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { 879 qdict_put(options, BDRV_OPT_CACHE_DIRECT, 880 qbool_from_bool(flags & BDRV_O_NOCACHE)); 881 } 882 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { 883 qdict_put(options, BDRV_OPT_CACHE_NO_FLUSH, 884 qbool_from_bool(flags & BDRV_O_NO_FLUSH)); 885 } 886 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { 887 qdict_put(options, BDRV_OPT_READ_ONLY, 888 qbool_from_bool(!(flags & BDRV_O_RDWR))); 889 } 890 } 891 892 static void bdrv_assign_node_name(BlockDriverState *bs, 893 const char *node_name, 894 Error **errp) 895 { 896 char *gen_node_name = NULL; 897 898 if (!node_name) { 899 node_name = gen_node_name = id_generate(ID_BLOCK); 900 } else if (!id_wellformed(node_name)) { 901 /* 902 * Check for empty string or invalid characters, but not if it is 903 * generated (generated names use characters not available to the user) 904 */ 905 error_setg(errp, "Invalid node name"); 906 return; 907 } 908 909 /* takes care of avoiding namespaces collisions */ 910 if (blk_by_name(node_name)) { 911 error_setg(errp, "node-name=%s is conflicting with a device id", 912 node_name); 913 goto out; 914 } 915 916 /* takes care of avoiding duplicates node names */ 917 if (bdrv_find_node(node_name)) { 918 error_setg(errp, "Duplicate node name"); 919 goto out; 920 } 921 922 /* copy node name into the bs and insert it into the graph list */ 923 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); 924 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); 925 out: 926 g_free(gen_node_name); 927 } 928 929 QemuOptsList bdrv_runtime_opts = { 930 .name = "bdrv_common", 931 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), 932 .desc = { 933 { 934 .name = "node-name", 935 .type = QEMU_OPT_STRING, 936 .help = "Node name of the block device node", 937 }, 938 { 939 .name = "driver", 940 .type = QEMU_OPT_STRING, 941 .help = "Block driver to use for the node", 942 }, 943 { 944 .name = BDRV_OPT_CACHE_DIRECT, 945 .type = QEMU_OPT_BOOL, 946 .help = "Bypass software writeback cache on the host", 947 }, 948 { 949 .name = BDRV_OPT_CACHE_NO_FLUSH, 950 .type = QEMU_OPT_BOOL, 951 .help = "Ignore flush requests", 952 }, 953 { 954 .name = BDRV_OPT_READ_ONLY, 955 .type = QEMU_OPT_BOOL, 956 .help = "Node is opened in read-only mode", 957 }, 958 { 959 .name = "detect-zeroes", 960 .type = QEMU_OPT_STRING, 961 .help = "try to optimize zero writes (off, on, unmap)", 962 }, 963 { 964 .name = "discard", 965 .type = QEMU_OPT_STRING, 966 .help = "discard operation (ignore/off, unmap/on)", 967 }, 968 { /* end of list */ } 969 }, 970 }; 971 972 /* 973 * Common part for opening disk images and files 974 * 975 * Removes all processed options from *options. 976 */ 977 static int bdrv_open_common(BlockDriverState *bs, BdrvChild *file, 978 QDict *options, Error **errp) 979 { 980 int ret, open_flags; 981 const char *filename; 982 const char *driver_name = NULL; 983 const char *node_name = NULL; 984 const char *discard; 985 const char *detect_zeroes; 986 QemuOpts *opts; 987 BlockDriver *drv; 988 Error *local_err = NULL; 989 990 assert(bs->file == NULL); 991 assert(options != NULL && bs->options != options); 992 993 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); 994 qemu_opts_absorb_qdict(opts, options, &local_err); 995 if (local_err) { 996 error_propagate(errp, local_err); 997 ret = -EINVAL; 998 goto fail_opts; 999 } 1000 1001 update_flags_from_options(&bs->open_flags, opts); 1002 1003 driver_name = qemu_opt_get(opts, "driver"); 1004 drv = bdrv_find_format(driver_name); 1005 assert(drv != NULL); 1006 1007 if (file != NULL) { 1008 filename = file->bs->filename; 1009 } else { 1010 filename = qdict_get_try_str(options, "filename"); 1011 } 1012 1013 if (drv->bdrv_needs_filename && !filename) { 1014 error_setg(errp, "The '%s' block driver requires a file name", 1015 drv->format_name); 1016 ret = -EINVAL; 1017 goto fail_opts; 1018 } 1019 1020 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, 1021 drv->format_name); 1022 1023 node_name = qemu_opt_get(opts, "node-name"); 1024 bdrv_assign_node_name(bs, node_name, &local_err); 1025 if (local_err) { 1026 error_propagate(errp, local_err); 1027 ret = -EINVAL; 1028 goto fail_opts; 1029 } 1030 1031 bs->read_only = !(bs->open_flags & BDRV_O_RDWR); 1032 1033 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) { 1034 error_setg(errp, 1035 !bs->read_only && bdrv_is_whitelisted(drv, true) 1036 ? "Driver '%s' can only be used for read-only devices" 1037 : "Driver '%s' is not whitelisted", 1038 drv->format_name); 1039 ret = -ENOTSUP; 1040 goto fail_opts; 1041 } 1042 1043 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */ 1044 if (bs->open_flags & BDRV_O_COPY_ON_READ) { 1045 if (!bs->read_only) { 1046 bdrv_enable_copy_on_read(bs); 1047 } else { 1048 error_setg(errp, "Can't use copy-on-read on read-only device"); 1049 ret = -EINVAL; 1050 goto fail_opts; 1051 } 1052 } 1053 1054 discard = qemu_opt_get(opts, "discard"); 1055 if (discard != NULL) { 1056 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { 1057 error_setg(errp, "Invalid discard option"); 1058 ret = -EINVAL; 1059 goto fail_opts; 1060 } 1061 } 1062 1063 detect_zeroes = qemu_opt_get(opts, "detect-zeroes"); 1064 if (detect_zeroes) { 1065 BlockdevDetectZeroesOptions value = 1066 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup, 1067 detect_zeroes, 1068 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX, 1069 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, 1070 &local_err); 1071 if (local_err) { 1072 error_propagate(errp, local_err); 1073 ret = -EINVAL; 1074 goto fail_opts; 1075 } 1076 1077 if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && 1078 !(bs->open_flags & BDRV_O_UNMAP)) 1079 { 1080 error_setg(errp, "setting detect-zeroes to unmap is not allowed " 1081 "without setting discard operation to unmap"); 1082 ret = -EINVAL; 1083 goto fail_opts; 1084 } 1085 1086 bs->detect_zeroes = value; 1087 } 1088 1089 if (filename != NULL) { 1090 pstrcpy(bs->filename, sizeof(bs->filename), filename); 1091 } else { 1092 bs->filename[0] = '\0'; 1093 } 1094 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); 1095 1096 bs->drv = drv; 1097 bs->opaque = g_malloc0(drv->instance_size); 1098 1099 /* Open the image, either directly or using a protocol */ 1100 open_flags = bdrv_open_flags(bs, bs->open_flags); 1101 if (drv->bdrv_file_open) { 1102 assert(file == NULL); 1103 assert(!drv->bdrv_needs_filename || filename != NULL); 1104 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); 1105 } else { 1106 if (file == NULL) { 1107 error_setg(errp, "Can't use '%s' as a block driver for the " 1108 "protocol level", drv->format_name); 1109 ret = -EINVAL; 1110 goto free_and_fail; 1111 } 1112 bs->file = file; 1113 ret = drv->bdrv_open(bs, options, open_flags, &local_err); 1114 } 1115 1116 if (ret < 0) { 1117 if (local_err) { 1118 error_propagate(errp, local_err); 1119 } else if (bs->filename[0]) { 1120 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); 1121 } else { 1122 error_setg_errno(errp, -ret, "Could not open image"); 1123 } 1124 goto free_and_fail; 1125 } 1126 1127 ret = refresh_total_sectors(bs, bs->total_sectors); 1128 if (ret < 0) { 1129 error_setg_errno(errp, -ret, "Could not refresh total sector count"); 1130 goto free_and_fail; 1131 } 1132 1133 bdrv_refresh_limits(bs, &local_err); 1134 if (local_err) { 1135 error_propagate(errp, local_err); 1136 ret = -EINVAL; 1137 goto free_and_fail; 1138 } 1139 1140 assert(bdrv_opt_mem_align(bs) != 0); 1141 assert(bdrv_min_mem_align(bs) != 0); 1142 assert(is_power_of_2(bs->bl.request_alignment)); 1143 1144 qemu_opts_del(opts); 1145 return 0; 1146 1147 free_and_fail: 1148 bs->file = NULL; 1149 g_free(bs->opaque); 1150 bs->opaque = NULL; 1151 bs->drv = NULL; 1152 fail_opts: 1153 qemu_opts_del(opts); 1154 return ret; 1155 } 1156 1157 static QDict *parse_json_filename(const char *filename, Error **errp) 1158 { 1159 QObject *options_obj; 1160 QDict *options; 1161 int ret; 1162 1163 ret = strstart(filename, "json:", &filename); 1164 assert(ret); 1165 1166 options_obj = qobject_from_json(filename); 1167 if (!options_obj) { 1168 error_setg(errp, "Could not parse the JSON options"); 1169 return NULL; 1170 } 1171 1172 if (qobject_type(options_obj) != QTYPE_QDICT) { 1173 qobject_decref(options_obj); 1174 error_setg(errp, "Invalid JSON object given"); 1175 return NULL; 1176 } 1177 1178 options = qobject_to_qdict(options_obj); 1179 qdict_flatten(options); 1180 1181 return options; 1182 } 1183 1184 static void parse_json_protocol(QDict *options, const char **pfilename, 1185 Error **errp) 1186 { 1187 QDict *json_options; 1188 Error *local_err = NULL; 1189 1190 /* Parse json: pseudo-protocol */ 1191 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { 1192 return; 1193 } 1194 1195 json_options = parse_json_filename(*pfilename, &local_err); 1196 if (local_err) { 1197 error_propagate(errp, local_err); 1198 return; 1199 } 1200 1201 /* Options given in the filename have lower priority than options 1202 * specified directly */ 1203 qdict_join(options, json_options, false); 1204 QDECREF(json_options); 1205 *pfilename = NULL; 1206 } 1207 1208 /* 1209 * Fills in default options for opening images and converts the legacy 1210 * filename/flags pair to option QDict entries. 1211 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a 1212 * block driver has been specified explicitly. 1213 */ 1214 static int bdrv_fill_options(QDict **options, const char *filename, 1215 int *flags, Error **errp) 1216 { 1217 const char *drvname; 1218 bool protocol = *flags & BDRV_O_PROTOCOL; 1219 bool parse_filename = false; 1220 BlockDriver *drv = NULL; 1221 Error *local_err = NULL; 1222 1223 drvname = qdict_get_try_str(*options, "driver"); 1224 if (drvname) { 1225 drv = bdrv_find_format(drvname); 1226 if (!drv) { 1227 error_setg(errp, "Unknown driver '%s'", drvname); 1228 return -ENOENT; 1229 } 1230 /* If the user has explicitly specified the driver, this choice should 1231 * override the BDRV_O_PROTOCOL flag */ 1232 protocol = drv->bdrv_file_open; 1233 } 1234 1235 if (protocol) { 1236 *flags |= BDRV_O_PROTOCOL; 1237 } else { 1238 *flags &= ~BDRV_O_PROTOCOL; 1239 } 1240 1241 /* Translate cache options from flags into options */ 1242 update_options_from_flags(*options, *flags); 1243 1244 /* Fetch the file name from the options QDict if necessary */ 1245 if (protocol && filename) { 1246 if (!qdict_haskey(*options, "filename")) { 1247 qdict_put(*options, "filename", qstring_from_str(filename)); 1248 parse_filename = true; 1249 } else { 1250 error_setg(errp, "Can't specify 'file' and 'filename' options at " 1251 "the same time"); 1252 return -EINVAL; 1253 } 1254 } 1255 1256 /* Find the right block driver */ 1257 filename = qdict_get_try_str(*options, "filename"); 1258 1259 if (!drvname && protocol) { 1260 if (filename) { 1261 drv = bdrv_find_protocol(filename, parse_filename, errp); 1262 if (!drv) { 1263 return -EINVAL; 1264 } 1265 1266 drvname = drv->format_name; 1267 qdict_put(*options, "driver", qstring_from_str(drvname)); 1268 } else { 1269 error_setg(errp, "Must specify either driver or file"); 1270 return -EINVAL; 1271 } 1272 } 1273 1274 assert(drv || !protocol); 1275 1276 /* Driver-specific filename parsing */ 1277 if (drv && drv->bdrv_parse_filename && parse_filename) { 1278 drv->bdrv_parse_filename(filename, *options, &local_err); 1279 if (local_err) { 1280 error_propagate(errp, local_err); 1281 return -EINVAL; 1282 } 1283 1284 if (!drv->bdrv_needs_filename) { 1285 qdict_del(*options, "filename"); 1286 } 1287 } 1288 1289 return 0; 1290 } 1291 1292 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs) 1293 { 1294 BlockDriverState *old_bs = child->bs; 1295 1296 if (old_bs) { 1297 if (old_bs->quiesce_counter && child->role->drained_end) { 1298 child->role->drained_end(child); 1299 } 1300 QLIST_REMOVE(child, next_parent); 1301 } 1302 1303 child->bs = new_bs; 1304 1305 if (new_bs) { 1306 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); 1307 if (new_bs->quiesce_counter && child->role->drained_begin) { 1308 child->role->drained_begin(child); 1309 } 1310 } 1311 } 1312 1313 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, 1314 const char *child_name, 1315 const BdrvChildRole *child_role, 1316 void *opaque) 1317 { 1318 BdrvChild *child = g_new(BdrvChild, 1); 1319 *child = (BdrvChild) { 1320 .bs = NULL, 1321 .name = g_strdup(child_name), 1322 .role = child_role, 1323 .opaque = opaque, 1324 }; 1325 1326 bdrv_replace_child(child, child_bs); 1327 1328 return child; 1329 } 1330 1331 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, 1332 BlockDriverState *child_bs, 1333 const char *child_name, 1334 const BdrvChildRole *child_role) 1335 { 1336 BdrvChild *child = bdrv_root_attach_child(child_bs, child_name, child_role, 1337 parent_bs); 1338 QLIST_INSERT_HEAD(&parent_bs->children, child, next); 1339 return child; 1340 } 1341 1342 static void bdrv_detach_child(BdrvChild *child) 1343 { 1344 if (child->next.le_prev) { 1345 QLIST_REMOVE(child, next); 1346 child->next.le_prev = NULL; 1347 } 1348 1349 bdrv_replace_child(child, NULL); 1350 1351 g_free(child->name); 1352 g_free(child); 1353 } 1354 1355 void bdrv_root_unref_child(BdrvChild *child) 1356 { 1357 BlockDriverState *child_bs; 1358 1359 child_bs = child->bs; 1360 bdrv_detach_child(child); 1361 bdrv_unref(child_bs); 1362 } 1363 1364 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) 1365 { 1366 if (child == NULL) { 1367 return; 1368 } 1369 1370 if (child->bs->inherits_from == parent) { 1371 child->bs->inherits_from = NULL; 1372 } 1373 1374 bdrv_root_unref_child(child); 1375 } 1376 1377 1378 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) 1379 { 1380 BdrvChild *c; 1381 QLIST_FOREACH(c, &bs->parents, next_parent) { 1382 if (c->role->change_media) { 1383 c->role->change_media(c, load); 1384 } 1385 } 1386 } 1387 1388 static void bdrv_parent_cb_resize(BlockDriverState *bs) 1389 { 1390 BdrvChild *c; 1391 QLIST_FOREACH(c, &bs->parents, next_parent) { 1392 if (c->role->resize) { 1393 c->role->resize(c); 1394 } 1395 } 1396 } 1397 1398 /* 1399 * Sets the backing file link of a BDS. A new reference is created; callers 1400 * which don't need their own reference any more must call bdrv_unref(). 1401 */ 1402 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd) 1403 { 1404 if (backing_hd) { 1405 bdrv_ref(backing_hd); 1406 } 1407 1408 if (bs->backing) { 1409 assert(bs->backing_blocker); 1410 bdrv_op_unblock_all(bs->backing->bs, bs->backing_blocker); 1411 bdrv_unref_child(bs, bs->backing); 1412 } else if (backing_hd) { 1413 error_setg(&bs->backing_blocker, 1414 "node is used as backing hd of '%s'", 1415 bdrv_get_device_or_node_name(bs)); 1416 } 1417 1418 if (!backing_hd) { 1419 error_free(bs->backing_blocker); 1420 bs->backing_blocker = NULL; 1421 bs->backing = NULL; 1422 goto out; 1423 } 1424 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing); 1425 bs->open_flags &= ~BDRV_O_NO_BACKING; 1426 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename); 1427 pstrcpy(bs->backing_format, sizeof(bs->backing_format), 1428 backing_hd->drv ? backing_hd->drv->format_name : ""); 1429 1430 bdrv_op_block_all(backing_hd, bs->backing_blocker); 1431 /* Otherwise we won't be able to commit due to check in bdrv_commit */ 1432 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, 1433 bs->backing_blocker); 1434 /* 1435 * We do backup in 3 ways: 1436 * 1. drive backup 1437 * The target bs is new opened, and the source is top BDS 1438 * 2. blockdev backup 1439 * Both the source and the target are top BDSes. 1440 * 3. internal backup(used for block replication) 1441 * Both the source and the target are backing file 1442 * 1443 * In case 1 and 2, neither the source nor the target is the backing file. 1444 * In case 3, we will block the top BDS, so there is only one block job 1445 * for the top BDS and its backing chain. 1446 */ 1447 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, 1448 bs->backing_blocker); 1449 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, 1450 bs->backing_blocker); 1451 out: 1452 bdrv_refresh_limits(bs, NULL); 1453 } 1454 1455 /* 1456 * Opens the backing file for a BlockDriverState if not yet open 1457 * 1458 * bdref_key specifies the key for the image's BlockdevRef in the options QDict. 1459 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict 1460 * itself, all options starting with "${bdref_key}." are considered part of the 1461 * BlockdevRef. 1462 * 1463 * TODO Can this be unified with bdrv_open_image()? 1464 */ 1465 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, 1466 const char *bdref_key, Error **errp) 1467 { 1468 char *backing_filename = g_malloc0(PATH_MAX); 1469 char *bdref_key_dot; 1470 const char *reference = NULL; 1471 int ret = 0; 1472 BlockDriverState *backing_hd; 1473 QDict *options; 1474 QDict *tmp_parent_options = NULL; 1475 Error *local_err = NULL; 1476 1477 if (bs->backing != NULL) { 1478 goto free_exit; 1479 } 1480 1481 /* NULL means an empty set of options */ 1482 if (parent_options == NULL) { 1483 tmp_parent_options = qdict_new(); 1484 parent_options = tmp_parent_options; 1485 } 1486 1487 bs->open_flags &= ~BDRV_O_NO_BACKING; 1488 1489 bdref_key_dot = g_strdup_printf("%s.", bdref_key); 1490 qdict_extract_subqdict(parent_options, &options, bdref_key_dot); 1491 g_free(bdref_key_dot); 1492 1493 reference = qdict_get_try_str(parent_options, bdref_key); 1494 if (reference || qdict_haskey(options, "file.filename")) { 1495 backing_filename[0] = '\0'; 1496 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { 1497 QDECREF(options); 1498 goto free_exit; 1499 } else { 1500 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX, 1501 &local_err); 1502 if (local_err) { 1503 ret = -EINVAL; 1504 error_propagate(errp, local_err); 1505 QDECREF(options); 1506 goto free_exit; 1507 } 1508 } 1509 1510 if (!bs->drv || !bs->drv->supports_backing) { 1511 ret = -EINVAL; 1512 error_setg(errp, "Driver doesn't support backing files"); 1513 QDECREF(options); 1514 goto free_exit; 1515 } 1516 1517 if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { 1518 qdict_put(options, "driver", qstring_from_str(bs->backing_format)); 1519 } 1520 1521 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL, 1522 reference, options, 0, bs, &child_backing, 1523 errp); 1524 if (!backing_hd) { 1525 bs->open_flags |= BDRV_O_NO_BACKING; 1526 error_prepend(errp, "Could not open backing file: "); 1527 ret = -EINVAL; 1528 goto free_exit; 1529 } 1530 1531 /* Hook up the backing file link; drop our reference, bs owns the 1532 * backing_hd reference now */ 1533 bdrv_set_backing_hd(bs, backing_hd); 1534 bdrv_unref(backing_hd); 1535 1536 qdict_del(parent_options, bdref_key); 1537 1538 free_exit: 1539 g_free(backing_filename); 1540 QDECREF(tmp_parent_options); 1541 return ret; 1542 } 1543 1544 /* 1545 * Opens a disk image whose options are given as BlockdevRef in another block 1546 * device's options. 1547 * 1548 * If allow_none is true, no image will be opened if filename is false and no 1549 * BlockdevRef is given. NULL will be returned, but errp remains unset. 1550 * 1551 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. 1552 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict 1553 * itself, all options starting with "${bdref_key}." are considered part of the 1554 * BlockdevRef. 1555 * 1556 * The BlockdevRef will be removed from the options QDict. 1557 */ 1558 BdrvChild *bdrv_open_child(const char *filename, 1559 QDict *options, const char *bdref_key, 1560 BlockDriverState* parent, 1561 const BdrvChildRole *child_role, 1562 bool allow_none, Error **errp) 1563 { 1564 BdrvChild *c = NULL; 1565 BlockDriverState *bs; 1566 QDict *image_options; 1567 char *bdref_key_dot; 1568 const char *reference; 1569 1570 assert(child_role != NULL); 1571 1572 bdref_key_dot = g_strdup_printf("%s.", bdref_key); 1573 qdict_extract_subqdict(options, &image_options, bdref_key_dot); 1574 g_free(bdref_key_dot); 1575 1576 reference = qdict_get_try_str(options, bdref_key); 1577 if (!filename && !reference && !qdict_size(image_options)) { 1578 if (!allow_none) { 1579 error_setg(errp, "A block device must be specified for \"%s\"", 1580 bdref_key); 1581 } 1582 QDECREF(image_options); 1583 goto done; 1584 } 1585 1586 bs = bdrv_open_inherit(filename, reference, image_options, 0, 1587 parent, child_role, errp); 1588 if (!bs) { 1589 goto done; 1590 } 1591 1592 c = bdrv_attach_child(parent, bs, bdref_key, child_role); 1593 1594 done: 1595 qdict_del(options, bdref_key); 1596 return c; 1597 } 1598 1599 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, 1600 int flags, 1601 QDict *snapshot_options, 1602 Error **errp) 1603 { 1604 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */ 1605 char *tmp_filename = g_malloc0(PATH_MAX + 1); 1606 int64_t total_size; 1607 QemuOpts *opts = NULL; 1608 BlockDriverState *bs_snapshot; 1609 int ret; 1610 1611 /* if snapshot, we create a temporary backing file and open it 1612 instead of opening 'filename' directly */ 1613 1614 /* Get the required size from the image */ 1615 total_size = bdrv_getlength(bs); 1616 if (total_size < 0) { 1617 error_setg_errno(errp, -total_size, "Could not get image size"); 1618 goto out; 1619 } 1620 1621 /* Create the temporary image */ 1622 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1); 1623 if (ret < 0) { 1624 error_setg_errno(errp, -ret, "Could not get temporary filename"); 1625 goto out; 1626 } 1627 1628 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, 1629 &error_abort); 1630 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); 1631 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); 1632 qemu_opts_del(opts); 1633 if (ret < 0) { 1634 error_prepend(errp, "Could not create temporary overlay '%s': ", 1635 tmp_filename); 1636 goto out; 1637 } 1638 1639 /* Prepare options QDict for the temporary file */ 1640 qdict_put(snapshot_options, "file.driver", 1641 qstring_from_str("file")); 1642 qdict_put(snapshot_options, "file.filename", 1643 qstring_from_str(tmp_filename)); 1644 qdict_put(snapshot_options, "driver", 1645 qstring_from_str("qcow2")); 1646 1647 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); 1648 snapshot_options = NULL; 1649 if (!bs_snapshot) { 1650 ret = -EINVAL; 1651 goto out; 1652 } 1653 1654 /* bdrv_append() consumes a strong reference to bs_snapshot (i.e. it will 1655 * call bdrv_unref() on it), so in order to be able to return one, we have 1656 * to increase bs_snapshot's refcount here */ 1657 bdrv_ref(bs_snapshot); 1658 bdrv_append(bs_snapshot, bs); 1659 1660 g_free(tmp_filename); 1661 return bs_snapshot; 1662 1663 out: 1664 QDECREF(snapshot_options); 1665 g_free(tmp_filename); 1666 return NULL; 1667 } 1668 1669 /* 1670 * Opens a disk image (raw, qcow2, vmdk, ...) 1671 * 1672 * options is a QDict of options to pass to the block drivers, or NULL for an 1673 * empty set of options. The reference to the QDict belongs to the block layer 1674 * after the call (even on failure), so if the caller intends to reuse the 1675 * dictionary, it needs to use QINCREF() before calling bdrv_open. 1676 * 1677 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. 1678 * If it is not NULL, the referenced BDS will be reused. 1679 * 1680 * The reference parameter may be used to specify an existing block device which 1681 * should be opened. If specified, neither options nor a filename may be given, 1682 * nor can an existing BDS be reused (that is, *pbs has to be NULL). 1683 */ 1684 static BlockDriverState *bdrv_open_inherit(const char *filename, 1685 const char *reference, 1686 QDict *options, int flags, 1687 BlockDriverState *parent, 1688 const BdrvChildRole *child_role, 1689 Error **errp) 1690 { 1691 int ret; 1692 BdrvChild *file = NULL; 1693 BlockDriverState *bs; 1694 BlockDriver *drv = NULL; 1695 const char *drvname; 1696 const char *backing; 1697 Error *local_err = NULL; 1698 QDict *snapshot_options = NULL; 1699 int snapshot_flags = 0; 1700 1701 assert(!child_role || !flags); 1702 assert(!child_role == !parent); 1703 1704 if (reference) { 1705 bool options_non_empty = options ? qdict_size(options) : false; 1706 QDECREF(options); 1707 1708 if (filename || options_non_empty) { 1709 error_setg(errp, "Cannot reference an existing block device with " 1710 "additional options or a new filename"); 1711 return NULL; 1712 } 1713 1714 bs = bdrv_lookup_bs(reference, reference, errp); 1715 if (!bs) { 1716 return NULL; 1717 } 1718 1719 bdrv_ref(bs); 1720 return bs; 1721 } 1722 1723 bs = bdrv_new(); 1724 1725 /* NULL means an empty set of options */ 1726 if (options == NULL) { 1727 options = qdict_new(); 1728 } 1729 1730 /* json: syntax counts as explicit options, as if in the QDict */ 1731 parse_json_protocol(options, &filename, &local_err); 1732 if (local_err) { 1733 goto fail; 1734 } 1735 1736 bs->explicit_options = qdict_clone_shallow(options); 1737 1738 if (child_role) { 1739 bs->inherits_from = parent; 1740 child_role->inherit_options(&flags, options, 1741 parent->open_flags, parent->options); 1742 } 1743 1744 ret = bdrv_fill_options(&options, filename, &flags, &local_err); 1745 if (local_err) { 1746 goto fail; 1747 } 1748 1749 /* Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. 1750 * FIXME: we're parsing the QDict to avoid having to create a 1751 * QemuOpts just for this, but neither option is optimal. */ 1752 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && 1753 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { 1754 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); 1755 } else { 1756 flags &= ~BDRV_O_RDWR; 1757 } 1758 1759 if (flags & BDRV_O_SNAPSHOT) { 1760 snapshot_options = qdict_new(); 1761 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, 1762 flags, options); 1763 /* Let bdrv_backing_options() override "read-only" */ 1764 qdict_del(options, BDRV_OPT_READ_ONLY); 1765 bdrv_backing_options(&flags, options, flags, options); 1766 } 1767 1768 bs->open_flags = flags; 1769 bs->options = options; 1770 options = qdict_clone_shallow(options); 1771 1772 /* Find the right image format driver */ 1773 drvname = qdict_get_try_str(options, "driver"); 1774 if (drvname) { 1775 drv = bdrv_find_format(drvname); 1776 if (!drv) { 1777 error_setg(errp, "Unknown driver: '%s'", drvname); 1778 goto fail; 1779 } 1780 } 1781 1782 assert(drvname || !(flags & BDRV_O_PROTOCOL)); 1783 1784 backing = qdict_get_try_str(options, "backing"); 1785 if (backing && *backing == '\0') { 1786 flags |= BDRV_O_NO_BACKING; 1787 qdict_del(options, "backing"); 1788 } 1789 1790 /* Open image file without format layer */ 1791 if ((flags & BDRV_O_PROTOCOL) == 0) { 1792 file = bdrv_open_child(filename, options, "file", bs, 1793 &child_file, true, &local_err); 1794 if (local_err) { 1795 goto fail; 1796 } 1797 } 1798 1799 /* Image format probing */ 1800 bs->probed = !drv; 1801 if (!drv && file) { 1802 ret = find_image_format(file, filename, &drv, &local_err); 1803 if (ret < 0) { 1804 goto fail; 1805 } 1806 /* 1807 * This option update would logically belong in bdrv_fill_options(), 1808 * but we first need to open bs->file for the probing to work, while 1809 * opening bs->file already requires the (mostly) final set of options 1810 * so that cache mode etc. can be inherited. 1811 * 1812 * Adding the driver later is somewhat ugly, but it's not an option 1813 * that would ever be inherited, so it's correct. We just need to make 1814 * sure to update both bs->options (which has the full effective 1815 * options for bs) and options (which has file.* already removed). 1816 */ 1817 qdict_put(bs->options, "driver", qstring_from_str(drv->format_name)); 1818 qdict_put(options, "driver", qstring_from_str(drv->format_name)); 1819 } else if (!drv) { 1820 error_setg(errp, "Must specify either driver or file"); 1821 goto fail; 1822 } 1823 1824 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ 1825 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); 1826 /* file must be NULL if a protocol BDS is about to be created 1827 * (the inverse results in an error message from bdrv_open_common()) */ 1828 assert(!(flags & BDRV_O_PROTOCOL) || !file); 1829 1830 /* Open the image */ 1831 ret = bdrv_open_common(bs, file, options, &local_err); 1832 if (ret < 0) { 1833 goto fail; 1834 } 1835 1836 if (file && (bs->file != file)) { 1837 bdrv_unref_child(bs, file); 1838 file = NULL; 1839 } 1840 1841 /* If there is a backing file, use it */ 1842 if ((flags & BDRV_O_NO_BACKING) == 0) { 1843 ret = bdrv_open_backing_file(bs, options, "backing", &local_err); 1844 if (ret < 0) { 1845 goto close_and_fail; 1846 } 1847 } 1848 1849 bdrv_refresh_filename(bs); 1850 1851 /* Check if any unknown options were used */ 1852 if (options && (qdict_size(options) != 0)) { 1853 const QDictEntry *entry = qdict_first(options); 1854 if (flags & BDRV_O_PROTOCOL) { 1855 error_setg(errp, "Block protocol '%s' doesn't support the option " 1856 "'%s'", drv->format_name, entry->key); 1857 } else { 1858 error_setg(errp, 1859 "Block format '%s' does not support the option '%s'", 1860 drv->format_name, entry->key); 1861 } 1862 1863 goto close_and_fail; 1864 } 1865 1866 if (!bdrv_key_required(bs)) { 1867 bdrv_parent_cb_change_media(bs, true); 1868 } else if (!runstate_check(RUN_STATE_PRELAUNCH) 1869 && !runstate_check(RUN_STATE_INMIGRATE) 1870 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */ 1871 error_setg(errp, 1872 "Guest must be stopped for opening of encrypted image"); 1873 goto close_and_fail; 1874 } 1875 1876 QDECREF(options); 1877 1878 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the 1879 * temporary snapshot afterwards. */ 1880 if (snapshot_flags) { 1881 BlockDriverState *snapshot_bs; 1882 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, 1883 snapshot_options, &local_err); 1884 snapshot_options = NULL; 1885 if (local_err) { 1886 goto close_and_fail; 1887 } 1888 /* We are not going to return bs but the overlay on top of it 1889 * (snapshot_bs); thus, we have to drop the strong reference to bs 1890 * (which we obtained by calling bdrv_new()). bs will not be deleted, 1891 * though, because the overlay still has a reference to it. */ 1892 bdrv_unref(bs); 1893 bs = snapshot_bs; 1894 } 1895 1896 return bs; 1897 1898 fail: 1899 if (file != NULL) { 1900 bdrv_unref_child(bs, file); 1901 } 1902 QDECREF(snapshot_options); 1903 QDECREF(bs->explicit_options); 1904 QDECREF(bs->options); 1905 QDECREF(options); 1906 bs->options = NULL; 1907 bdrv_unref(bs); 1908 error_propagate(errp, local_err); 1909 return NULL; 1910 1911 close_and_fail: 1912 bdrv_unref(bs); 1913 QDECREF(snapshot_options); 1914 QDECREF(options); 1915 error_propagate(errp, local_err); 1916 return NULL; 1917 } 1918 1919 BlockDriverState *bdrv_open(const char *filename, const char *reference, 1920 QDict *options, int flags, Error **errp) 1921 { 1922 return bdrv_open_inherit(filename, reference, options, flags, NULL, 1923 NULL, errp); 1924 } 1925 1926 typedef struct BlockReopenQueueEntry { 1927 bool prepared; 1928 BDRVReopenState state; 1929 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry; 1930 } BlockReopenQueueEntry; 1931 1932 /* 1933 * Adds a BlockDriverState to a simple queue for an atomic, transactional 1934 * reopen of multiple devices. 1935 * 1936 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT 1937 * already performed, or alternatively may be NULL a new BlockReopenQueue will 1938 * be created and initialized. This newly created BlockReopenQueue should be 1939 * passed back in for subsequent calls that are intended to be of the same 1940 * atomic 'set'. 1941 * 1942 * bs is the BlockDriverState to add to the reopen queue. 1943 * 1944 * options contains the changed options for the associated bs 1945 * (the BlockReopenQueue takes ownership) 1946 * 1947 * flags contains the open flags for the associated bs 1948 * 1949 * returns a pointer to bs_queue, which is either the newly allocated 1950 * bs_queue, or the existing bs_queue being used. 1951 * 1952 */ 1953 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, 1954 BlockDriverState *bs, 1955 QDict *options, 1956 int flags, 1957 const BdrvChildRole *role, 1958 QDict *parent_options, 1959 int parent_flags) 1960 { 1961 assert(bs != NULL); 1962 1963 BlockReopenQueueEntry *bs_entry; 1964 BdrvChild *child; 1965 QDict *old_options, *explicit_options; 1966 1967 if (bs_queue == NULL) { 1968 bs_queue = g_new0(BlockReopenQueue, 1); 1969 QSIMPLEQ_INIT(bs_queue); 1970 } 1971 1972 if (!options) { 1973 options = qdict_new(); 1974 } 1975 1976 /* Check if this BlockDriverState is already in the queue */ 1977 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { 1978 if (bs == bs_entry->state.bs) { 1979 break; 1980 } 1981 } 1982 1983 /* 1984 * Precedence of options: 1985 * 1. Explicitly passed in options (highest) 1986 * 2. Set in flags (only for top level) 1987 * 3. Retained from explicitly set options of bs 1988 * 4. Inherited from parent node 1989 * 5. Retained from effective options of bs 1990 */ 1991 1992 if (!parent_options) { 1993 /* 1994 * Any setting represented by flags is always updated. If the 1995 * corresponding QDict option is set, it takes precedence. Otherwise 1996 * the flag is translated into a QDict option. The old setting of bs is 1997 * not considered. 1998 */ 1999 update_options_from_flags(options, flags); 2000 } 2001 2002 /* Old explicitly set values (don't overwrite by inherited value) */ 2003 if (bs_entry) { 2004 old_options = qdict_clone_shallow(bs_entry->state.explicit_options); 2005 } else { 2006 old_options = qdict_clone_shallow(bs->explicit_options); 2007 } 2008 bdrv_join_options(bs, options, old_options); 2009 QDECREF(old_options); 2010 2011 explicit_options = qdict_clone_shallow(options); 2012 2013 /* Inherit from parent node */ 2014 if (parent_options) { 2015 assert(!flags); 2016 role->inherit_options(&flags, options, parent_flags, parent_options); 2017 } 2018 2019 /* Old values are used for options that aren't set yet */ 2020 old_options = qdict_clone_shallow(bs->options); 2021 bdrv_join_options(bs, options, old_options); 2022 QDECREF(old_options); 2023 2024 /* bdrv_open() masks this flag out */ 2025 flags &= ~BDRV_O_PROTOCOL; 2026 2027 QLIST_FOREACH(child, &bs->children, next) { 2028 QDict *new_child_options; 2029 char *child_key_dot; 2030 2031 /* reopen can only change the options of block devices that were 2032 * implicitly created and inherited options. For other (referenced) 2033 * block devices, a syntax like "backing.foo" results in an error. */ 2034 if (child->bs->inherits_from != bs) { 2035 continue; 2036 } 2037 2038 child_key_dot = g_strdup_printf("%s.", child->name); 2039 qdict_extract_subqdict(options, &new_child_options, child_key_dot); 2040 g_free(child_key_dot); 2041 2042 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0, 2043 child->role, options, flags); 2044 } 2045 2046 if (!bs_entry) { 2047 bs_entry = g_new0(BlockReopenQueueEntry, 1); 2048 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry); 2049 } else { 2050 QDECREF(bs_entry->state.options); 2051 QDECREF(bs_entry->state.explicit_options); 2052 } 2053 2054 bs_entry->state.bs = bs; 2055 bs_entry->state.options = options; 2056 bs_entry->state.explicit_options = explicit_options; 2057 bs_entry->state.flags = flags; 2058 2059 return bs_queue; 2060 } 2061 2062 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, 2063 BlockDriverState *bs, 2064 QDict *options, int flags) 2065 { 2066 return bdrv_reopen_queue_child(bs_queue, bs, options, flags, 2067 NULL, NULL, 0); 2068 } 2069 2070 /* 2071 * Reopen multiple BlockDriverStates atomically & transactionally. 2072 * 2073 * The queue passed in (bs_queue) must have been built up previous 2074 * via bdrv_reopen_queue(). 2075 * 2076 * Reopens all BDS specified in the queue, with the appropriate 2077 * flags. All devices are prepared for reopen, and failure of any 2078 * device will cause all device changes to be abandonded, and intermediate 2079 * data cleaned up. 2080 * 2081 * If all devices prepare successfully, then the changes are committed 2082 * to all devices. 2083 * 2084 */ 2085 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) 2086 { 2087 int ret = -1; 2088 BlockReopenQueueEntry *bs_entry, *next; 2089 Error *local_err = NULL; 2090 2091 assert(bs_queue != NULL); 2092 2093 bdrv_drain_all(); 2094 2095 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { 2096 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) { 2097 error_propagate(errp, local_err); 2098 goto cleanup; 2099 } 2100 bs_entry->prepared = true; 2101 } 2102 2103 /* If we reach this point, we have success and just need to apply the 2104 * changes 2105 */ 2106 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { 2107 bdrv_reopen_commit(&bs_entry->state); 2108 } 2109 2110 ret = 0; 2111 2112 cleanup: 2113 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { 2114 if (ret && bs_entry->prepared) { 2115 bdrv_reopen_abort(&bs_entry->state); 2116 } else if (ret) { 2117 QDECREF(bs_entry->state.explicit_options); 2118 } 2119 QDECREF(bs_entry->state.options); 2120 g_free(bs_entry); 2121 } 2122 g_free(bs_queue); 2123 return ret; 2124 } 2125 2126 2127 /* Reopen a single BlockDriverState with the specified flags. */ 2128 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp) 2129 { 2130 int ret = -1; 2131 Error *local_err = NULL; 2132 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags); 2133 2134 ret = bdrv_reopen_multiple(queue, &local_err); 2135 if (local_err != NULL) { 2136 error_propagate(errp, local_err); 2137 } 2138 return ret; 2139 } 2140 2141 2142 /* 2143 * Prepares a BlockDriverState for reopen. All changes are staged in the 2144 * 'opaque' field of the BDRVReopenState, which is used and allocated by 2145 * the block driver layer .bdrv_reopen_prepare() 2146 * 2147 * bs is the BlockDriverState to reopen 2148 * flags are the new open flags 2149 * queue is the reopen queue 2150 * 2151 * Returns 0 on success, non-zero on error. On error errp will be set 2152 * as well. 2153 * 2154 * On failure, bdrv_reopen_abort() will be called to clean up any data. 2155 * It is the responsibility of the caller to then call the abort() or 2156 * commit() for any other BDS that have been left in a prepare() state 2157 * 2158 */ 2159 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue, 2160 Error **errp) 2161 { 2162 int ret = -1; 2163 Error *local_err = NULL; 2164 BlockDriver *drv; 2165 QemuOpts *opts; 2166 const char *value; 2167 2168 assert(reopen_state != NULL); 2169 assert(reopen_state->bs->drv != NULL); 2170 drv = reopen_state->bs->drv; 2171 2172 /* Process generic block layer options */ 2173 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); 2174 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err); 2175 if (local_err) { 2176 error_propagate(errp, local_err); 2177 ret = -EINVAL; 2178 goto error; 2179 } 2180 2181 update_flags_from_options(&reopen_state->flags, opts); 2182 2183 /* node-name and driver must be unchanged. Put them back into the QDict, so 2184 * that they are checked at the end of this function. */ 2185 value = qemu_opt_get(opts, "node-name"); 2186 if (value) { 2187 qdict_put(reopen_state->options, "node-name", qstring_from_str(value)); 2188 } 2189 2190 value = qemu_opt_get(opts, "driver"); 2191 if (value) { 2192 qdict_put(reopen_state->options, "driver", qstring_from_str(value)); 2193 } 2194 2195 /* if we are to stay read-only, do not allow permission change 2196 * to r/w */ 2197 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) && 2198 reopen_state->flags & BDRV_O_RDWR) { 2199 error_setg(errp, "Node '%s' is read only", 2200 bdrv_get_device_or_node_name(reopen_state->bs)); 2201 goto error; 2202 } 2203 2204 2205 ret = bdrv_flush(reopen_state->bs); 2206 if (ret) { 2207 error_setg_errno(errp, -ret, "Error flushing drive"); 2208 goto error; 2209 } 2210 2211 if (drv->bdrv_reopen_prepare) { 2212 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); 2213 if (ret) { 2214 if (local_err != NULL) { 2215 error_propagate(errp, local_err); 2216 } else { 2217 error_setg(errp, "failed while preparing to reopen image '%s'", 2218 reopen_state->bs->filename); 2219 } 2220 goto error; 2221 } 2222 } else { 2223 /* It is currently mandatory to have a bdrv_reopen_prepare() 2224 * handler for each supported drv. */ 2225 error_setg(errp, "Block format '%s' used by node '%s' " 2226 "does not support reopening files", drv->format_name, 2227 bdrv_get_device_or_node_name(reopen_state->bs)); 2228 ret = -1; 2229 goto error; 2230 } 2231 2232 /* Options that are not handled are only okay if they are unchanged 2233 * compared to the old state. It is expected that some options are only 2234 * used for the initial open, but not reopen (e.g. filename) */ 2235 if (qdict_size(reopen_state->options)) { 2236 const QDictEntry *entry = qdict_first(reopen_state->options); 2237 2238 do { 2239 QString *new_obj = qobject_to_qstring(entry->value); 2240 const char *new = qstring_get_str(new_obj); 2241 const char *old = qdict_get_try_str(reopen_state->bs->options, 2242 entry->key); 2243 2244 if (!old || strcmp(new, old)) { 2245 error_setg(errp, "Cannot change the option '%s'", entry->key); 2246 ret = -EINVAL; 2247 goto error; 2248 } 2249 } while ((entry = qdict_next(reopen_state->options, entry))); 2250 } 2251 2252 ret = 0; 2253 2254 error: 2255 qemu_opts_del(opts); 2256 return ret; 2257 } 2258 2259 /* 2260 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and 2261 * makes them final by swapping the staging BlockDriverState contents into 2262 * the active BlockDriverState contents. 2263 */ 2264 void bdrv_reopen_commit(BDRVReopenState *reopen_state) 2265 { 2266 BlockDriver *drv; 2267 2268 assert(reopen_state != NULL); 2269 drv = reopen_state->bs->drv; 2270 assert(drv != NULL); 2271 2272 /* If there are any driver level actions to take */ 2273 if (drv->bdrv_reopen_commit) { 2274 drv->bdrv_reopen_commit(reopen_state); 2275 } 2276 2277 /* set BDS specific flags now */ 2278 QDECREF(reopen_state->bs->explicit_options); 2279 2280 reopen_state->bs->explicit_options = reopen_state->explicit_options; 2281 reopen_state->bs->open_flags = reopen_state->flags; 2282 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR); 2283 2284 bdrv_refresh_limits(reopen_state->bs, NULL); 2285 } 2286 2287 /* 2288 * Abort the reopen, and delete and free the staged changes in 2289 * reopen_state 2290 */ 2291 void bdrv_reopen_abort(BDRVReopenState *reopen_state) 2292 { 2293 BlockDriver *drv; 2294 2295 assert(reopen_state != NULL); 2296 drv = reopen_state->bs->drv; 2297 assert(drv != NULL); 2298 2299 if (drv->bdrv_reopen_abort) { 2300 drv->bdrv_reopen_abort(reopen_state); 2301 } 2302 2303 QDECREF(reopen_state->explicit_options); 2304 } 2305 2306 2307 static void bdrv_close(BlockDriverState *bs) 2308 { 2309 BdrvAioNotifier *ban, *ban_next; 2310 2311 assert(!bs->job); 2312 assert(!bs->refcnt); 2313 2314 bdrv_drained_begin(bs); /* complete I/O */ 2315 bdrv_flush(bs); 2316 bdrv_drain(bs); /* in case flush left pending I/O */ 2317 2318 bdrv_release_named_dirty_bitmaps(bs); 2319 assert(QLIST_EMPTY(&bs->dirty_bitmaps)); 2320 2321 if (bs->drv) { 2322 BdrvChild *child, *next; 2323 2324 bs->drv->bdrv_close(bs); 2325 bs->drv = NULL; 2326 2327 bdrv_set_backing_hd(bs, NULL); 2328 2329 if (bs->file != NULL) { 2330 bdrv_unref_child(bs, bs->file); 2331 bs->file = NULL; 2332 } 2333 2334 QLIST_FOREACH_SAFE(child, &bs->children, next, next) { 2335 /* TODO Remove bdrv_unref() from drivers' close function and use 2336 * bdrv_unref_child() here */ 2337 if (child->bs->inherits_from == bs) { 2338 child->bs->inherits_from = NULL; 2339 } 2340 bdrv_detach_child(child); 2341 } 2342 2343 g_free(bs->opaque); 2344 bs->opaque = NULL; 2345 bs->copy_on_read = 0; 2346 bs->backing_file[0] = '\0'; 2347 bs->backing_format[0] = '\0'; 2348 bs->total_sectors = 0; 2349 bs->encrypted = false; 2350 bs->valid_key = false; 2351 bs->sg = false; 2352 QDECREF(bs->options); 2353 QDECREF(bs->explicit_options); 2354 bs->options = NULL; 2355 QDECREF(bs->full_open_options); 2356 bs->full_open_options = NULL; 2357 } 2358 2359 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { 2360 g_free(ban); 2361 } 2362 QLIST_INIT(&bs->aio_notifiers); 2363 bdrv_drained_end(bs); 2364 } 2365 2366 void bdrv_close_all(void) 2367 { 2368 block_job_cancel_sync_all(); 2369 nbd_export_close_all(); 2370 2371 /* Drop references from requests still in flight, such as canceled block 2372 * jobs whose AIO context has not been polled yet */ 2373 bdrv_drain_all(); 2374 2375 blk_remove_all_bs(); 2376 blockdev_close_all_bdrv_states(); 2377 2378 assert(QTAILQ_EMPTY(&all_bdrv_states)); 2379 } 2380 2381 static void change_parent_backing_link(BlockDriverState *from, 2382 BlockDriverState *to) 2383 { 2384 BdrvChild *c, *next, *to_c; 2385 2386 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { 2387 if (c->role == &child_backing) { 2388 /* @from is generally not allowed to be a backing file, except for 2389 * when @to is the overlay. In that case, @from may not be replaced 2390 * by @to as @to's backing node. */ 2391 QLIST_FOREACH(to_c, &to->children, next) { 2392 if (to_c == c) { 2393 break; 2394 } 2395 } 2396 if (to_c) { 2397 continue; 2398 } 2399 } 2400 2401 assert(c->role != &child_backing); 2402 bdrv_ref(to); 2403 bdrv_replace_child(c, to); 2404 bdrv_unref(from); 2405 } 2406 } 2407 2408 /* 2409 * Add new bs contents at the top of an image chain while the chain is 2410 * live, while keeping required fields on the top layer. 2411 * 2412 * This will modify the BlockDriverState fields, and swap contents 2413 * between bs_new and bs_top. Both bs_new and bs_top are modified. 2414 * 2415 * bs_new must not be attached to a BlockBackend. 2416 * 2417 * This function does not create any image files. 2418 * 2419 * bdrv_append() takes ownership of a bs_new reference and unrefs it because 2420 * that's what the callers commonly need. bs_new will be referenced by the old 2421 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a 2422 * reference of its own, it must call bdrv_ref(). 2423 */ 2424 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top) 2425 { 2426 assert(!bdrv_requests_pending(bs_top)); 2427 assert(!bdrv_requests_pending(bs_new)); 2428 2429 bdrv_ref(bs_top); 2430 2431 change_parent_backing_link(bs_top, bs_new); 2432 bdrv_set_backing_hd(bs_new, bs_top); 2433 bdrv_unref(bs_top); 2434 2435 /* bs_new is now referenced by its new parents, we don't need the 2436 * additional reference any more. */ 2437 bdrv_unref(bs_new); 2438 } 2439 2440 void bdrv_replace_in_backing_chain(BlockDriverState *old, BlockDriverState *new) 2441 { 2442 assert(!bdrv_requests_pending(old)); 2443 assert(!bdrv_requests_pending(new)); 2444 2445 bdrv_ref(old); 2446 2447 change_parent_backing_link(old, new); 2448 2449 bdrv_unref(old); 2450 } 2451 2452 static void bdrv_delete(BlockDriverState *bs) 2453 { 2454 assert(!bs->job); 2455 assert(bdrv_op_blocker_is_empty(bs)); 2456 assert(!bs->refcnt); 2457 2458 bdrv_close(bs); 2459 2460 /* remove from list, if necessary */ 2461 if (bs->node_name[0] != '\0') { 2462 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); 2463 } 2464 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); 2465 2466 g_free(bs); 2467 } 2468 2469 /* 2470 * Run consistency checks on an image 2471 * 2472 * Returns 0 if the check could be completed (it doesn't mean that the image is 2473 * free of errors) or -errno when an internal error occurred. The results of the 2474 * check are stored in res. 2475 */ 2476 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix) 2477 { 2478 if (bs->drv == NULL) { 2479 return -ENOMEDIUM; 2480 } 2481 if (bs->drv->bdrv_check == NULL) { 2482 return -ENOTSUP; 2483 } 2484 2485 memset(res, 0, sizeof(*res)); 2486 return bs->drv->bdrv_check(bs, res, fix); 2487 } 2488 2489 /* 2490 * Return values: 2491 * 0 - success 2492 * -EINVAL - backing format specified, but no file 2493 * -ENOSPC - can't update the backing file because no space is left in the 2494 * image file header 2495 * -ENOTSUP - format driver doesn't support changing the backing file 2496 */ 2497 int bdrv_change_backing_file(BlockDriverState *bs, 2498 const char *backing_file, const char *backing_fmt) 2499 { 2500 BlockDriver *drv = bs->drv; 2501 int ret; 2502 2503 /* Backing file format doesn't make sense without a backing file */ 2504 if (backing_fmt && !backing_file) { 2505 return -EINVAL; 2506 } 2507 2508 if (drv->bdrv_change_backing_file != NULL) { 2509 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); 2510 } else { 2511 ret = -ENOTSUP; 2512 } 2513 2514 if (ret == 0) { 2515 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 2516 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2517 } 2518 return ret; 2519 } 2520 2521 /* 2522 * Finds the image layer in the chain that has 'bs' as its backing file. 2523 * 2524 * active is the current topmost image. 2525 * 2526 * Returns NULL if bs is not found in active's image chain, 2527 * or if active == bs. 2528 * 2529 * Returns the bottommost base image if bs == NULL. 2530 */ 2531 BlockDriverState *bdrv_find_overlay(BlockDriverState *active, 2532 BlockDriverState *bs) 2533 { 2534 while (active && bs != backing_bs(active)) { 2535 active = backing_bs(active); 2536 } 2537 2538 return active; 2539 } 2540 2541 /* Given a BDS, searches for the base layer. */ 2542 BlockDriverState *bdrv_find_base(BlockDriverState *bs) 2543 { 2544 return bdrv_find_overlay(bs, NULL); 2545 } 2546 2547 /* 2548 * Drops images above 'base' up to and including 'top', and sets the image 2549 * above 'top' to have base as its backing file. 2550 * 2551 * Requires that the overlay to 'top' is opened r/w, so that the backing file 2552 * information in 'bs' can be properly updated. 2553 * 2554 * E.g., this will convert the following chain: 2555 * bottom <- base <- intermediate <- top <- active 2556 * 2557 * to 2558 * 2559 * bottom <- base <- active 2560 * 2561 * It is allowed for bottom==base, in which case it converts: 2562 * 2563 * base <- intermediate <- top <- active 2564 * 2565 * to 2566 * 2567 * base <- active 2568 * 2569 * If backing_file_str is non-NULL, it will be used when modifying top's 2570 * overlay image metadata. 2571 * 2572 * Error conditions: 2573 * if active == top, that is considered an error 2574 * 2575 */ 2576 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top, 2577 BlockDriverState *base, const char *backing_file_str) 2578 { 2579 BlockDriverState *new_top_bs = NULL; 2580 int ret = -EIO; 2581 2582 if (!top->drv || !base->drv) { 2583 goto exit; 2584 } 2585 2586 new_top_bs = bdrv_find_overlay(active, top); 2587 2588 if (new_top_bs == NULL) { 2589 /* we could not find the image above 'top', this is an error */ 2590 goto exit; 2591 } 2592 2593 /* special case of new_top_bs->backing->bs already pointing to base - nothing 2594 * to do, no intermediate images */ 2595 if (backing_bs(new_top_bs) == base) { 2596 ret = 0; 2597 goto exit; 2598 } 2599 2600 /* Make sure that base is in the backing chain of top */ 2601 if (!bdrv_chain_contains(top, base)) { 2602 goto exit; 2603 } 2604 2605 /* success - we can delete the intermediate states, and link top->base */ 2606 backing_file_str = backing_file_str ? backing_file_str : base->filename; 2607 ret = bdrv_change_backing_file(new_top_bs, backing_file_str, 2608 base->drv ? base->drv->format_name : ""); 2609 if (ret) { 2610 goto exit; 2611 } 2612 bdrv_set_backing_hd(new_top_bs, base); 2613 2614 ret = 0; 2615 exit: 2616 return ret; 2617 } 2618 2619 /** 2620 * Truncate file to 'offset' bytes (needed only for file protocols) 2621 */ 2622 int bdrv_truncate(BlockDriverState *bs, int64_t offset) 2623 { 2624 BlockDriver *drv = bs->drv; 2625 int ret; 2626 if (!drv) 2627 return -ENOMEDIUM; 2628 if (!drv->bdrv_truncate) 2629 return -ENOTSUP; 2630 if (bs->read_only) 2631 return -EACCES; 2632 2633 ret = drv->bdrv_truncate(bs, offset); 2634 if (ret == 0) { 2635 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); 2636 bdrv_dirty_bitmap_truncate(bs); 2637 bdrv_parent_cb_resize(bs); 2638 ++bs->write_gen; 2639 } 2640 return ret; 2641 } 2642 2643 /** 2644 * Length of a allocated file in bytes. Sparse files are counted by actual 2645 * allocated space. Return < 0 if error or unknown. 2646 */ 2647 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) 2648 { 2649 BlockDriver *drv = bs->drv; 2650 if (!drv) { 2651 return -ENOMEDIUM; 2652 } 2653 if (drv->bdrv_get_allocated_file_size) { 2654 return drv->bdrv_get_allocated_file_size(bs); 2655 } 2656 if (bs->file) { 2657 return bdrv_get_allocated_file_size(bs->file->bs); 2658 } 2659 return -ENOTSUP; 2660 } 2661 2662 /** 2663 * Return number of sectors on success, -errno on error. 2664 */ 2665 int64_t bdrv_nb_sectors(BlockDriverState *bs) 2666 { 2667 BlockDriver *drv = bs->drv; 2668 2669 if (!drv) 2670 return -ENOMEDIUM; 2671 2672 if (drv->has_variable_length) { 2673 int ret = refresh_total_sectors(bs, bs->total_sectors); 2674 if (ret < 0) { 2675 return ret; 2676 } 2677 } 2678 return bs->total_sectors; 2679 } 2680 2681 /** 2682 * Return length in bytes on success, -errno on error. 2683 * The length is always a multiple of BDRV_SECTOR_SIZE. 2684 */ 2685 int64_t bdrv_getlength(BlockDriverState *bs) 2686 { 2687 int64_t ret = bdrv_nb_sectors(bs); 2688 2689 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret; 2690 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE; 2691 } 2692 2693 /* return 0 as number of sectors if no device present or error */ 2694 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) 2695 { 2696 int64_t nb_sectors = bdrv_nb_sectors(bs); 2697 2698 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors; 2699 } 2700 2701 bool bdrv_is_read_only(BlockDriverState *bs) 2702 { 2703 return bs->read_only; 2704 } 2705 2706 bool bdrv_is_sg(BlockDriverState *bs) 2707 { 2708 return bs->sg; 2709 } 2710 2711 bool bdrv_is_encrypted(BlockDriverState *bs) 2712 { 2713 if (bs->backing && bs->backing->bs->encrypted) { 2714 return true; 2715 } 2716 return bs->encrypted; 2717 } 2718 2719 bool bdrv_key_required(BlockDriverState *bs) 2720 { 2721 BdrvChild *backing = bs->backing; 2722 2723 if (backing && backing->bs->encrypted && !backing->bs->valid_key) { 2724 return true; 2725 } 2726 return (bs->encrypted && !bs->valid_key); 2727 } 2728 2729 int bdrv_set_key(BlockDriverState *bs, const char *key) 2730 { 2731 int ret; 2732 if (bs->backing && bs->backing->bs->encrypted) { 2733 ret = bdrv_set_key(bs->backing->bs, key); 2734 if (ret < 0) 2735 return ret; 2736 if (!bs->encrypted) 2737 return 0; 2738 } 2739 if (!bs->encrypted) { 2740 return -EINVAL; 2741 } else if (!bs->drv || !bs->drv->bdrv_set_key) { 2742 return -ENOMEDIUM; 2743 } 2744 ret = bs->drv->bdrv_set_key(bs, key); 2745 if (ret < 0) { 2746 bs->valid_key = false; 2747 } else if (!bs->valid_key) { 2748 /* call the change callback now, we skipped it on open */ 2749 bs->valid_key = true; 2750 bdrv_parent_cb_change_media(bs, true); 2751 } 2752 return ret; 2753 } 2754 2755 /* 2756 * Provide an encryption key for @bs. 2757 * If @key is non-null: 2758 * If @bs is not encrypted, fail. 2759 * Else if the key is invalid, fail. 2760 * Else set @bs's key to @key, replacing the existing key, if any. 2761 * If @key is null: 2762 * If @bs is encrypted and still lacks a key, fail. 2763 * Else do nothing. 2764 * On failure, store an error object through @errp if non-null. 2765 */ 2766 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp) 2767 { 2768 if (key) { 2769 if (!bdrv_is_encrypted(bs)) { 2770 error_setg(errp, "Node '%s' is not encrypted", 2771 bdrv_get_device_or_node_name(bs)); 2772 } else if (bdrv_set_key(bs, key) < 0) { 2773 error_setg(errp, QERR_INVALID_PASSWORD); 2774 } 2775 } else { 2776 if (bdrv_key_required(bs)) { 2777 error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED, 2778 "'%s' (%s) is encrypted", 2779 bdrv_get_device_or_node_name(bs), 2780 bdrv_get_encrypted_filename(bs)); 2781 } 2782 } 2783 } 2784 2785 const char *bdrv_get_format_name(BlockDriverState *bs) 2786 { 2787 return bs->drv ? bs->drv->format_name : NULL; 2788 } 2789 2790 static int qsort_strcmp(const void *a, const void *b) 2791 { 2792 return strcmp(a, b); 2793 } 2794 2795 void bdrv_iterate_format(void (*it)(void *opaque, const char *name), 2796 void *opaque) 2797 { 2798 BlockDriver *drv; 2799 int count = 0; 2800 int i; 2801 const char **formats = NULL; 2802 2803 QLIST_FOREACH(drv, &bdrv_drivers, list) { 2804 if (drv->format_name) { 2805 bool found = false; 2806 int i = count; 2807 while (formats && i && !found) { 2808 found = !strcmp(formats[--i], drv->format_name); 2809 } 2810 2811 if (!found) { 2812 formats = g_renew(const char *, formats, count + 1); 2813 formats[count++] = drv->format_name; 2814 } 2815 } 2816 } 2817 2818 qsort(formats, count, sizeof(formats[0]), qsort_strcmp); 2819 2820 for (i = 0; i < count; i++) { 2821 it(opaque, formats[i]); 2822 } 2823 2824 g_free(formats); 2825 } 2826 2827 /* This function is to find a node in the bs graph */ 2828 BlockDriverState *bdrv_find_node(const char *node_name) 2829 { 2830 BlockDriverState *bs; 2831 2832 assert(node_name); 2833 2834 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { 2835 if (!strcmp(node_name, bs->node_name)) { 2836 return bs; 2837 } 2838 } 2839 return NULL; 2840 } 2841 2842 /* Put this QMP function here so it can access the static graph_bdrv_states. */ 2843 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp) 2844 { 2845 BlockDeviceInfoList *list, *entry; 2846 BlockDriverState *bs; 2847 2848 list = NULL; 2849 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { 2850 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp); 2851 if (!info) { 2852 qapi_free_BlockDeviceInfoList(list); 2853 return NULL; 2854 } 2855 entry = g_malloc0(sizeof(*entry)); 2856 entry->value = info; 2857 entry->next = list; 2858 list = entry; 2859 } 2860 2861 return list; 2862 } 2863 2864 BlockDriverState *bdrv_lookup_bs(const char *device, 2865 const char *node_name, 2866 Error **errp) 2867 { 2868 BlockBackend *blk; 2869 BlockDriverState *bs; 2870 2871 if (device) { 2872 blk = blk_by_name(device); 2873 2874 if (blk) { 2875 bs = blk_bs(blk); 2876 if (!bs) { 2877 error_setg(errp, "Device '%s' has no medium", device); 2878 } 2879 2880 return bs; 2881 } 2882 } 2883 2884 if (node_name) { 2885 bs = bdrv_find_node(node_name); 2886 2887 if (bs) { 2888 return bs; 2889 } 2890 } 2891 2892 error_setg(errp, "Cannot find device=%s nor node_name=%s", 2893 device ? device : "", 2894 node_name ? node_name : ""); 2895 return NULL; 2896 } 2897 2898 /* If 'base' is in the same chain as 'top', return true. Otherwise, 2899 * return false. If either argument is NULL, return false. */ 2900 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) 2901 { 2902 while (top && top != base) { 2903 top = backing_bs(top); 2904 } 2905 2906 return top != NULL; 2907 } 2908 2909 BlockDriverState *bdrv_next_node(BlockDriverState *bs) 2910 { 2911 if (!bs) { 2912 return QTAILQ_FIRST(&graph_bdrv_states); 2913 } 2914 return QTAILQ_NEXT(bs, node_list); 2915 } 2916 2917 const char *bdrv_get_node_name(const BlockDriverState *bs) 2918 { 2919 return bs->node_name; 2920 } 2921 2922 const char *bdrv_get_parent_name(const BlockDriverState *bs) 2923 { 2924 BdrvChild *c; 2925 const char *name; 2926 2927 /* If multiple parents have a name, just pick the first one. */ 2928 QLIST_FOREACH(c, &bs->parents, next_parent) { 2929 if (c->role->get_name) { 2930 name = c->role->get_name(c); 2931 if (name && *name) { 2932 return name; 2933 } 2934 } 2935 } 2936 2937 return NULL; 2938 } 2939 2940 /* TODO check what callers really want: bs->node_name or blk_name() */ 2941 const char *bdrv_get_device_name(const BlockDriverState *bs) 2942 { 2943 return bdrv_get_parent_name(bs) ?: ""; 2944 } 2945 2946 /* This can be used to identify nodes that might not have a device 2947 * name associated. Since node and device names live in the same 2948 * namespace, the result is unambiguous. The exception is if both are 2949 * absent, then this returns an empty (non-null) string. */ 2950 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) 2951 { 2952 return bdrv_get_parent_name(bs) ?: bs->node_name; 2953 } 2954 2955 int bdrv_get_flags(BlockDriverState *bs) 2956 { 2957 return bs->open_flags; 2958 } 2959 2960 int bdrv_has_zero_init_1(BlockDriverState *bs) 2961 { 2962 return 1; 2963 } 2964 2965 int bdrv_has_zero_init(BlockDriverState *bs) 2966 { 2967 assert(bs->drv); 2968 2969 /* If BS is a copy on write image, it is initialized to 2970 the contents of the base image, which may not be zeroes. */ 2971 if (bs->backing) { 2972 return 0; 2973 } 2974 if (bs->drv->bdrv_has_zero_init) { 2975 return bs->drv->bdrv_has_zero_init(bs); 2976 } 2977 2978 /* safe default */ 2979 return 0; 2980 } 2981 2982 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs) 2983 { 2984 BlockDriverInfo bdi; 2985 2986 if (bs->backing) { 2987 return false; 2988 } 2989 2990 if (bdrv_get_info(bs, &bdi) == 0) { 2991 return bdi.unallocated_blocks_are_zero; 2992 } 2993 2994 return false; 2995 } 2996 2997 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) 2998 { 2999 BlockDriverInfo bdi; 3000 3001 if (!(bs->open_flags & BDRV_O_UNMAP)) { 3002 return false; 3003 } 3004 3005 if (bdrv_get_info(bs, &bdi) == 0) { 3006 return bdi.can_write_zeroes_with_unmap; 3007 } 3008 3009 return false; 3010 } 3011 3012 const char *bdrv_get_encrypted_filename(BlockDriverState *bs) 3013 { 3014 if (bs->backing && bs->backing->bs->encrypted) 3015 return bs->backing_file; 3016 else if (bs->encrypted) 3017 return bs->filename; 3018 else 3019 return NULL; 3020 } 3021 3022 void bdrv_get_backing_filename(BlockDriverState *bs, 3023 char *filename, int filename_size) 3024 { 3025 pstrcpy(filename, filename_size, bs->backing_file); 3026 } 3027 3028 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 3029 { 3030 BlockDriver *drv = bs->drv; 3031 if (!drv) 3032 return -ENOMEDIUM; 3033 if (!drv->bdrv_get_info) 3034 return -ENOTSUP; 3035 memset(bdi, 0, sizeof(*bdi)); 3036 return drv->bdrv_get_info(bs, bdi); 3037 } 3038 3039 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs) 3040 { 3041 BlockDriver *drv = bs->drv; 3042 if (drv && drv->bdrv_get_specific_info) { 3043 return drv->bdrv_get_specific_info(bs); 3044 } 3045 return NULL; 3046 } 3047 3048 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event) 3049 { 3050 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) { 3051 return; 3052 } 3053 3054 bs->drv->bdrv_debug_event(bs, event); 3055 } 3056 3057 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, 3058 const char *tag) 3059 { 3060 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { 3061 bs = bs->file ? bs->file->bs : NULL; 3062 } 3063 3064 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { 3065 return bs->drv->bdrv_debug_breakpoint(bs, event, tag); 3066 } 3067 3068 return -ENOTSUP; 3069 } 3070 3071 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) 3072 { 3073 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) { 3074 bs = bs->file ? bs->file->bs : NULL; 3075 } 3076 3077 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) { 3078 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); 3079 } 3080 3081 return -ENOTSUP; 3082 } 3083 3084 int bdrv_debug_resume(BlockDriverState *bs, const char *tag) 3085 { 3086 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { 3087 bs = bs->file ? bs->file->bs : NULL; 3088 } 3089 3090 if (bs && bs->drv && bs->drv->bdrv_debug_resume) { 3091 return bs->drv->bdrv_debug_resume(bs, tag); 3092 } 3093 3094 return -ENOTSUP; 3095 } 3096 3097 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) 3098 { 3099 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { 3100 bs = bs->file ? bs->file->bs : NULL; 3101 } 3102 3103 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { 3104 return bs->drv->bdrv_debug_is_suspended(bs, tag); 3105 } 3106 3107 return false; 3108 } 3109 3110 /* backing_file can either be relative, or absolute, or a protocol. If it is 3111 * relative, it must be relative to the chain. So, passing in bs->filename 3112 * from a BDS as backing_file should not be done, as that may be relative to 3113 * the CWD rather than the chain. */ 3114 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, 3115 const char *backing_file) 3116 { 3117 char *filename_full = NULL; 3118 char *backing_file_full = NULL; 3119 char *filename_tmp = NULL; 3120 int is_protocol = 0; 3121 BlockDriverState *curr_bs = NULL; 3122 BlockDriverState *retval = NULL; 3123 3124 if (!bs || !bs->drv || !backing_file) { 3125 return NULL; 3126 } 3127 3128 filename_full = g_malloc(PATH_MAX); 3129 backing_file_full = g_malloc(PATH_MAX); 3130 filename_tmp = g_malloc(PATH_MAX); 3131 3132 is_protocol = path_has_protocol(backing_file); 3133 3134 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) { 3135 3136 /* If either of the filename paths is actually a protocol, then 3137 * compare unmodified paths; otherwise make paths relative */ 3138 if (is_protocol || path_has_protocol(curr_bs->backing_file)) { 3139 if (strcmp(backing_file, curr_bs->backing_file) == 0) { 3140 retval = curr_bs->backing->bs; 3141 break; 3142 } 3143 } else { 3144 /* If not an absolute filename path, make it relative to the current 3145 * image's filename path */ 3146 path_combine(filename_tmp, PATH_MAX, curr_bs->filename, 3147 backing_file); 3148 3149 /* We are going to compare absolute pathnames */ 3150 if (!realpath(filename_tmp, filename_full)) { 3151 continue; 3152 } 3153 3154 /* We need to make sure the backing filename we are comparing against 3155 * is relative to the current image filename (or absolute) */ 3156 path_combine(filename_tmp, PATH_MAX, curr_bs->filename, 3157 curr_bs->backing_file); 3158 3159 if (!realpath(filename_tmp, backing_file_full)) { 3160 continue; 3161 } 3162 3163 if (strcmp(backing_file_full, filename_full) == 0) { 3164 retval = curr_bs->backing->bs; 3165 break; 3166 } 3167 } 3168 } 3169 3170 g_free(filename_full); 3171 g_free(backing_file_full); 3172 g_free(filename_tmp); 3173 return retval; 3174 } 3175 3176 int bdrv_get_backing_file_depth(BlockDriverState *bs) 3177 { 3178 if (!bs->drv) { 3179 return 0; 3180 } 3181 3182 if (!bs->backing) { 3183 return 0; 3184 } 3185 3186 return 1 + bdrv_get_backing_file_depth(bs->backing->bs); 3187 } 3188 3189 void bdrv_init(void) 3190 { 3191 module_call_init(MODULE_INIT_BLOCK); 3192 } 3193 3194 void bdrv_init_with_whitelist(void) 3195 { 3196 use_bdrv_whitelist = 1; 3197 bdrv_init(); 3198 } 3199 3200 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp) 3201 { 3202 BdrvChild *child; 3203 Error *local_err = NULL; 3204 int ret; 3205 3206 if (!bs->drv) { 3207 return; 3208 } 3209 3210 if (!(bs->open_flags & BDRV_O_INACTIVE)) { 3211 return; 3212 } 3213 bs->open_flags &= ~BDRV_O_INACTIVE; 3214 3215 if (bs->drv->bdrv_invalidate_cache) { 3216 bs->drv->bdrv_invalidate_cache(bs, &local_err); 3217 if (local_err) { 3218 bs->open_flags |= BDRV_O_INACTIVE; 3219 error_propagate(errp, local_err); 3220 return; 3221 } 3222 } 3223 3224 QLIST_FOREACH(child, &bs->children, next) { 3225 bdrv_invalidate_cache(child->bs, &local_err); 3226 if (local_err) { 3227 bs->open_flags |= BDRV_O_INACTIVE; 3228 error_propagate(errp, local_err); 3229 return; 3230 } 3231 } 3232 3233 ret = refresh_total_sectors(bs, bs->total_sectors); 3234 if (ret < 0) { 3235 bs->open_flags |= BDRV_O_INACTIVE; 3236 error_setg_errno(errp, -ret, "Could not refresh total sector count"); 3237 return; 3238 } 3239 } 3240 3241 void bdrv_invalidate_cache_all(Error **errp) 3242 { 3243 BlockDriverState *bs; 3244 Error *local_err = NULL; 3245 BdrvNextIterator it; 3246 3247 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { 3248 AioContext *aio_context = bdrv_get_aio_context(bs); 3249 3250 aio_context_acquire(aio_context); 3251 bdrv_invalidate_cache(bs, &local_err); 3252 aio_context_release(aio_context); 3253 if (local_err) { 3254 error_propagate(errp, local_err); 3255 return; 3256 } 3257 } 3258 } 3259 3260 static int bdrv_inactivate_recurse(BlockDriverState *bs, 3261 bool setting_flag) 3262 { 3263 BdrvChild *child; 3264 int ret; 3265 3266 if (!setting_flag && bs->drv->bdrv_inactivate) { 3267 ret = bs->drv->bdrv_inactivate(bs); 3268 if (ret < 0) { 3269 return ret; 3270 } 3271 } 3272 3273 QLIST_FOREACH(child, &bs->children, next) { 3274 ret = bdrv_inactivate_recurse(child->bs, setting_flag); 3275 if (ret < 0) { 3276 return ret; 3277 } 3278 } 3279 3280 if (setting_flag) { 3281 bs->open_flags |= BDRV_O_INACTIVE; 3282 } 3283 return 0; 3284 } 3285 3286 int bdrv_inactivate_all(void) 3287 { 3288 BlockDriverState *bs = NULL; 3289 BdrvNextIterator it; 3290 int ret = 0; 3291 int pass; 3292 3293 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { 3294 aio_context_acquire(bdrv_get_aio_context(bs)); 3295 } 3296 3297 /* We do two passes of inactivation. The first pass calls to drivers' 3298 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk; 3299 * the second pass sets the BDRV_O_INACTIVE flag so that no further write 3300 * is allowed. */ 3301 for (pass = 0; pass < 2; pass++) { 3302 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { 3303 ret = bdrv_inactivate_recurse(bs, pass); 3304 if (ret < 0) { 3305 goto out; 3306 } 3307 } 3308 } 3309 3310 out: 3311 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { 3312 aio_context_release(bdrv_get_aio_context(bs)); 3313 } 3314 3315 return ret; 3316 } 3317 3318 /**************************************************************/ 3319 /* removable device support */ 3320 3321 /** 3322 * Return TRUE if the media is present 3323 */ 3324 bool bdrv_is_inserted(BlockDriverState *bs) 3325 { 3326 BlockDriver *drv = bs->drv; 3327 BdrvChild *child; 3328 3329 if (!drv) { 3330 return false; 3331 } 3332 if (drv->bdrv_is_inserted) { 3333 return drv->bdrv_is_inserted(bs); 3334 } 3335 QLIST_FOREACH(child, &bs->children, next) { 3336 if (!bdrv_is_inserted(child->bs)) { 3337 return false; 3338 } 3339 } 3340 return true; 3341 } 3342 3343 /** 3344 * Return whether the media changed since the last call to this 3345 * function, or -ENOTSUP if we don't know. Most drivers don't know. 3346 */ 3347 int bdrv_media_changed(BlockDriverState *bs) 3348 { 3349 BlockDriver *drv = bs->drv; 3350 3351 if (drv && drv->bdrv_media_changed) { 3352 return drv->bdrv_media_changed(bs); 3353 } 3354 return -ENOTSUP; 3355 } 3356 3357 /** 3358 * If eject_flag is TRUE, eject the media. Otherwise, close the tray 3359 */ 3360 void bdrv_eject(BlockDriverState *bs, bool eject_flag) 3361 { 3362 BlockDriver *drv = bs->drv; 3363 3364 if (drv && drv->bdrv_eject) { 3365 drv->bdrv_eject(bs, eject_flag); 3366 } 3367 } 3368 3369 /** 3370 * Lock or unlock the media (if it is locked, the user won't be able 3371 * to eject it manually). 3372 */ 3373 void bdrv_lock_medium(BlockDriverState *bs, bool locked) 3374 { 3375 BlockDriver *drv = bs->drv; 3376 3377 trace_bdrv_lock_medium(bs, locked); 3378 3379 if (drv && drv->bdrv_lock_medium) { 3380 drv->bdrv_lock_medium(bs, locked); 3381 } 3382 } 3383 3384 /* Get a reference to bs */ 3385 void bdrv_ref(BlockDriverState *bs) 3386 { 3387 bs->refcnt++; 3388 } 3389 3390 /* Release a previously grabbed reference to bs. 3391 * If after releasing, reference count is zero, the BlockDriverState is 3392 * deleted. */ 3393 void bdrv_unref(BlockDriverState *bs) 3394 { 3395 if (!bs) { 3396 return; 3397 } 3398 assert(bs->refcnt > 0); 3399 if (--bs->refcnt == 0) { 3400 bdrv_delete(bs); 3401 } 3402 } 3403 3404 struct BdrvOpBlocker { 3405 Error *reason; 3406 QLIST_ENTRY(BdrvOpBlocker) list; 3407 }; 3408 3409 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) 3410 { 3411 BdrvOpBlocker *blocker; 3412 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); 3413 if (!QLIST_EMPTY(&bs->op_blockers[op])) { 3414 blocker = QLIST_FIRST(&bs->op_blockers[op]); 3415 if (errp) { 3416 *errp = error_copy(blocker->reason); 3417 error_prepend(errp, "Node '%s' is busy: ", 3418 bdrv_get_device_or_node_name(bs)); 3419 } 3420 return true; 3421 } 3422 return false; 3423 } 3424 3425 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) 3426 { 3427 BdrvOpBlocker *blocker; 3428 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); 3429 3430 blocker = g_new0(BdrvOpBlocker, 1); 3431 blocker->reason = reason; 3432 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); 3433 } 3434 3435 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) 3436 { 3437 BdrvOpBlocker *blocker, *next; 3438 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); 3439 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { 3440 if (blocker->reason == reason) { 3441 QLIST_REMOVE(blocker, list); 3442 g_free(blocker); 3443 } 3444 } 3445 } 3446 3447 void bdrv_op_block_all(BlockDriverState *bs, Error *reason) 3448 { 3449 int i; 3450 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { 3451 bdrv_op_block(bs, i, reason); 3452 } 3453 } 3454 3455 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) 3456 { 3457 int i; 3458 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { 3459 bdrv_op_unblock(bs, i, reason); 3460 } 3461 } 3462 3463 bool bdrv_op_blocker_is_empty(BlockDriverState *bs) 3464 { 3465 int i; 3466 3467 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { 3468 if (!QLIST_EMPTY(&bs->op_blockers[i])) { 3469 return false; 3470 } 3471 } 3472 return true; 3473 } 3474 3475 void bdrv_img_create(const char *filename, const char *fmt, 3476 const char *base_filename, const char *base_fmt, 3477 char *options, uint64_t img_size, int flags, 3478 Error **errp, bool quiet) 3479 { 3480 QemuOptsList *create_opts = NULL; 3481 QemuOpts *opts = NULL; 3482 const char *backing_fmt, *backing_file; 3483 int64_t size; 3484 BlockDriver *drv, *proto_drv; 3485 Error *local_err = NULL; 3486 int ret = 0; 3487 3488 /* Find driver and parse its options */ 3489 drv = bdrv_find_format(fmt); 3490 if (!drv) { 3491 error_setg(errp, "Unknown file format '%s'", fmt); 3492 return; 3493 } 3494 3495 proto_drv = bdrv_find_protocol(filename, true, errp); 3496 if (!proto_drv) { 3497 return; 3498 } 3499 3500 if (!drv->create_opts) { 3501 error_setg(errp, "Format driver '%s' does not support image creation", 3502 drv->format_name); 3503 return; 3504 } 3505 3506 if (!proto_drv->create_opts) { 3507 error_setg(errp, "Protocol driver '%s' does not support image creation", 3508 proto_drv->format_name); 3509 return; 3510 } 3511 3512 create_opts = qemu_opts_append(create_opts, drv->create_opts); 3513 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); 3514 3515 /* Create parameter list with default values */ 3516 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); 3517 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); 3518 3519 /* Parse -o options */ 3520 if (options) { 3521 qemu_opts_do_parse(opts, options, NULL, &local_err); 3522 if (local_err) { 3523 error_report_err(local_err); 3524 local_err = NULL; 3525 error_setg(errp, "Invalid options for file format '%s'", fmt); 3526 goto out; 3527 } 3528 } 3529 3530 if (base_filename) { 3531 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err); 3532 if (local_err) { 3533 error_setg(errp, "Backing file not supported for file format '%s'", 3534 fmt); 3535 goto out; 3536 } 3537 } 3538 3539 if (base_fmt) { 3540 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err); 3541 if (local_err) { 3542 error_setg(errp, "Backing file format not supported for file " 3543 "format '%s'", fmt); 3544 goto out; 3545 } 3546 } 3547 3548 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 3549 if (backing_file) { 3550 if (!strcmp(filename, backing_file)) { 3551 error_setg(errp, "Error: Trying to create an image with the " 3552 "same filename as the backing file"); 3553 goto out; 3554 } 3555 } 3556 3557 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 3558 3559 // The size for the image must always be specified, with one exception: 3560 // If we are using a backing file, we can obtain the size from there 3561 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 3562 if (size == -1) { 3563 if (backing_file) { 3564 BlockDriverState *bs; 3565 char *full_backing = g_new0(char, PATH_MAX); 3566 int64_t size; 3567 int back_flags; 3568 QDict *backing_options = NULL; 3569 3570 bdrv_get_full_backing_filename_from_filename(filename, backing_file, 3571 full_backing, PATH_MAX, 3572 &local_err); 3573 if (local_err) { 3574 g_free(full_backing); 3575 goto out; 3576 } 3577 3578 /* backing files always opened read-only */ 3579 back_flags = flags; 3580 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); 3581 3582 if (backing_fmt) { 3583 backing_options = qdict_new(); 3584 qdict_put(backing_options, "driver", 3585 qstring_from_str(backing_fmt)); 3586 } 3587 3588 bs = bdrv_open(full_backing, NULL, backing_options, back_flags, 3589 &local_err); 3590 g_free(full_backing); 3591 if (!bs) { 3592 goto out; 3593 } 3594 size = bdrv_getlength(bs); 3595 if (size < 0) { 3596 error_setg_errno(errp, -size, "Could not get size of '%s'", 3597 backing_file); 3598 bdrv_unref(bs); 3599 goto out; 3600 } 3601 3602 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); 3603 3604 bdrv_unref(bs); 3605 } else { 3606 error_setg(errp, "Image creation needs a size parameter"); 3607 goto out; 3608 } 3609 } 3610 3611 if (!quiet) { 3612 printf("Formatting '%s', fmt=%s ", filename, fmt); 3613 qemu_opts_print(opts, " "); 3614 puts(""); 3615 } 3616 3617 ret = bdrv_create(drv, filename, opts, &local_err); 3618 3619 if (ret == -EFBIG) { 3620 /* This is generally a better message than whatever the driver would 3621 * deliver (especially because of the cluster_size_hint), since that 3622 * is most probably not much different from "image too large". */ 3623 const char *cluster_size_hint = ""; 3624 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { 3625 cluster_size_hint = " (try using a larger cluster size)"; 3626 } 3627 error_setg(errp, "The image size is too large for file format '%s'" 3628 "%s", fmt, cluster_size_hint); 3629 error_free(local_err); 3630 local_err = NULL; 3631 } 3632 3633 out: 3634 qemu_opts_del(opts); 3635 qemu_opts_free(create_opts); 3636 error_propagate(errp, local_err); 3637 } 3638 3639 AioContext *bdrv_get_aio_context(BlockDriverState *bs) 3640 { 3641 return bs->aio_context; 3642 } 3643 3644 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) 3645 { 3646 QLIST_REMOVE(ban, list); 3647 g_free(ban); 3648 } 3649 3650 void bdrv_detach_aio_context(BlockDriverState *bs) 3651 { 3652 BdrvAioNotifier *baf, *baf_tmp; 3653 BdrvChild *child; 3654 3655 if (!bs->drv) { 3656 return; 3657 } 3658 3659 assert(!bs->walking_aio_notifiers); 3660 bs->walking_aio_notifiers = true; 3661 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { 3662 if (baf->deleted) { 3663 bdrv_do_remove_aio_context_notifier(baf); 3664 } else { 3665 baf->detach_aio_context(baf->opaque); 3666 } 3667 } 3668 /* Never mind iterating again to check for ->deleted. bdrv_close() will 3669 * remove remaining aio notifiers if we aren't called again. 3670 */ 3671 bs->walking_aio_notifiers = false; 3672 3673 if (bs->drv->bdrv_detach_aio_context) { 3674 bs->drv->bdrv_detach_aio_context(bs); 3675 } 3676 QLIST_FOREACH(child, &bs->children, next) { 3677 bdrv_detach_aio_context(child->bs); 3678 } 3679 3680 bs->aio_context = NULL; 3681 } 3682 3683 void bdrv_attach_aio_context(BlockDriverState *bs, 3684 AioContext *new_context) 3685 { 3686 BdrvAioNotifier *ban, *ban_tmp; 3687 BdrvChild *child; 3688 3689 if (!bs->drv) { 3690 return; 3691 } 3692 3693 bs->aio_context = new_context; 3694 3695 QLIST_FOREACH(child, &bs->children, next) { 3696 bdrv_attach_aio_context(child->bs, new_context); 3697 } 3698 if (bs->drv->bdrv_attach_aio_context) { 3699 bs->drv->bdrv_attach_aio_context(bs, new_context); 3700 } 3701 3702 assert(!bs->walking_aio_notifiers); 3703 bs->walking_aio_notifiers = true; 3704 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { 3705 if (ban->deleted) { 3706 bdrv_do_remove_aio_context_notifier(ban); 3707 } else { 3708 ban->attached_aio_context(new_context, ban->opaque); 3709 } 3710 } 3711 bs->walking_aio_notifiers = false; 3712 } 3713 3714 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context) 3715 { 3716 bdrv_drain(bs); /* ensure there are no in-flight requests */ 3717 3718 bdrv_detach_aio_context(bs); 3719 3720 /* This function executes in the old AioContext so acquire the new one in 3721 * case it runs in a different thread. 3722 */ 3723 aio_context_acquire(new_context); 3724 bdrv_attach_aio_context(bs, new_context); 3725 aio_context_release(new_context); 3726 } 3727 3728 void bdrv_add_aio_context_notifier(BlockDriverState *bs, 3729 void (*attached_aio_context)(AioContext *new_context, void *opaque), 3730 void (*detach_aio_context)(void *opaque), void *opaque) 3731 { 3732 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); 3733 *ban = (BdrvAioNotifier){ 3734 .attached_aio_context = attached_aio_context, 3735 .detach_aio_context = detach_aio_context, 3736 .opaque = opaque 3737 }; 3738 3739 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); 3740 } 3741 3742 void bdrv_remove_aio_context_notifier(BlockDriverState *bs, 3743 void (*attached_aio_context)(AioContext *, 3744 void *), 3745 void (*detach_aio_context)(void *), 3746 void *opaque) 3747 { 3748 BdrvAioNotifier *ban, *ban_next; 3749 3750 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { 3751 if (ban->attached_aio_context == attached_aio_context && 3752 ban->detach_aio_context == detach_aio_context && 3753 ban->opaque == opaque && 3754 ban->deleted == false) 3755 { 3756 if (bs->walking_aio_notifiers) { 3757 ban->deleted = true; 3758 } else { 3759 bdrv_do_remove_aio_context_notifier(ban); 3760 } 3761 return; 3762 } 3763 } 3764 3765 abort(); 3766 } 3767 3768 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, 3769 BlockDriverAmendStatusCB *status_cb, void *cb_opaque) 3770 { 3771 if (!bs->drv->bdrv_amend_options) { 3772 return -ENOTSUP; 3773 } 3774 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque); 3775 } 3776 3777 /* This function will be called by the bdrv_recurse_is_first_non_filter method 3778 * of block filter and by bdrv_is_first_non_filter. 3779 * It is used to test if the given bs is the candidate or recurse more in the 3780 * node graph. 3781 */ 3782 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs, 3783 BlockDriverState *candidate) 3784 { 3785 /* return false if basic checks fails */ 3786 if (!bs || !bs->drv) { 3787 return false; 3788 } 3789 3790 /* the code reached a non block filter driver -> check if the bs is 3791 * the same as the candidate. It's the recursion termination condition. 3792 */ 3793 if (!bs->drv->is_filter) { 3794 return bs == candidate; 3795 } 3796 /* Down this path the driver is a block filter driver */ 3797 3798 /* If the block filter recursion method is defined use it to recurse down 3799 * the node graph. 3800 */ 3801 if (bs->drv->bdrv_recurse_is_first_non_filter) { 3802 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate); 3803 } 3804 3805 /* the driver is a block filter but don't allow to recurse -> return false 3806 */ 3807 return false; 3808 } 3809 3810 /* This function checks if the candidate is the first non filter bs down it's 3811 * bs chain. Since we don't have pointers to parents it explore all bs chains 3812 * from the top. Some filters can choose not to pass down the recursion. 3813 */ 3814 bool bdrv_is_first_non_filter(BlockDriverState *candidate) 3815 { 3816 BlockDriverState *bs; 3817 BdrvNextIterator it; 3818 3819 /* walk down the bs forest recursively */ 3820 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { 3821 bool perm; 3822 3823 /* try to recurse in this top level bs */ 3824 perm = bdrv_recurse_is_first_non_filter(bs, candidate); 3825 3826 /* candidate is the first non filter */ 3827 if (perm) { 3828 return true; 3829 } 3830 } 3831 3832 return false; 3833 } 3834 3835 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, 3836 const char *node_name, Error **errp) 3837 { 3838 BlockDriverState *to_replace_bs = bdrv_find_node(node_name); 3839 AioContext *aio_context; 3840 3841 if (!to_replace_bs) { 3842 error_setg(errp, "Node name '%s' not found", node_name); 3843 return NULL; 3844 } 3845 3846 aio_context = bdrv_get_aio_context(to_replace_bs); 3847 aio_context_acquire(aio_context); 3848 3849 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { 3850 to_replace_bs = NULL; 3851 goto out; 3852 } 3853 3854 /* We don't want arbitrary node of the BDS chain to be replaced only the top 3855 * most non filter in order to prevent data corruption. 3856 * Another benefit is that this tests exclude backing files which are 3857 * blocked by the backing blockers. 3858 */ 3859 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) { 3860 error_setg(errp, "Only top most non filter can be replaced"); 3861 to_replace_bs = NULL; 3862 goto out; 3863 } 3864 3865 out: 3866 aio_context_release(aio_context); 3867 return to_replace_bs; 3868 } 3869 3870 static bool append_open_options(QDict *d, BlockDriverState *bs) 3871 { 3872 const QDictEntry *entry; 3873 QemuOptDesc *desc; 3874 BdrvChild *child; 3875 bool found_any = false; 3876 const char *p; 3877 3878 for (entry = qdict_first(bs->options); entry; 3879 entry = qdict_next(bs->options, entry)) 3880 { 3881 /* Exclude options for children */ 3882 QLIST_FOREACH(child, &bs->children, next) { 3883 if (strstart(qdict_entry_key(entry), child->name, &p) 3884 && (!*p || *p == '.')) 3885 { 3886 break; 3887 } 3888 } 3889 if (child) { 3890 continue; 3891 } 3892 3893 /* And exclude all non-driver-specific options */ 3894 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) { 3895 if (!strcmp(qdict_entry_key(entry), desc->name)) { 3896 break; 3897 } 3898 } 3899 if (desc->name) { 3900 continue; 3901 } 3902 3903 qobject_incref(qdict_entry_value(entry)); 3904 qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry)); 3905 found_any = true; 3906 } 3907 3908 return found_any; 3909 } 3910 3911 /* Updates the following BDS fields: 3912 * - exact_filename: A filename which may be used for opening a block device 3913 * which (mostly) equals the given BDS (even without any 3914 * other options; so reading and writing must return the same 3915 * results, but caching etc. may be different) 3916 * - full_open_options: Options which, when given when opening a block device 3917 * (without a filename), result in a BDS (mostly) 3918 * equalling the given one 3919 * - filename: If exact_filename is set, it is copied here. Otherwise, 3920 * full_open_options is converted to a JSON object, prefixed with 3921 * "json:" (for use through the JSON pseudo protocol) and put here. 3922 */ 3923 void bdrv_refresh_filename(BlockDriverState *bs) 3924 { 3925 BlockDriver *drv = bs->drv; 3926 QDict *opts; 3927 3928 if (!drv) { 3929 return; 3930 } 3931 3932 /* This BDS's file name will most probably depend on its file's name, so 3933 * refresh that first */ 3934 if (bs->file) { 3935 bdrv_refresh_filename(bs->file->bs); 3936 } 3937 3938 if (drv->bdrv_refresh_filename) { 3939 /* Obsolete information is of no use here, so drop the old file name 3940 * information before refreshing it */ 3941 bs->exact_filename[0] = '\0'; 3942 if (bs->full_open_options) { 3943 QDECREF(bs->full_open_options); 3944 bs->full_open_options = NULL; 3945 } 3946 3947 opts = qdict_new(); 3948 append_open_options(opts, bs); 3949 drv->bdrv_refresh_filename(bs, opts); 3950 QDECREF(opts); 3951 } else if (bs->file) { 3952 /* Try to reconstruct valid information from the underlying file */ 3953 bool has_open_options; 3954 3955 bs->exact_filename[0] = '\0'; 3956 if (bs->full_open_options) { 3957 QDECREF(bs->full_open_options); 3958 bs->full_open_options = NULL; 3959 } 3960 3961 opts = qdict_new(); 3962 has_open_options = append_open_options(opts, bs); 3963 3964 /* If no specific options have been given for this BDS, the filename of 3965 * the underlying file should suffice for this one as well */ 3966 if (bs->file->bs->exact_filename[0] && !has_open_options) { 3967 strcpy(bs->exact_filename, bs->file->bs->exact_filename); 3968 } 3969 /* Reconstructing the full options QDict is simple for most format block 3970 * drivers, as long as the full options are known for the underlying 3971 * file BDS. The full options QDict of that file BDS should somehow 3972 * contain a representation of the filename, therefore the following 3973 * suffices without querying the (exact_)filename of this BDS. */ 3974 if (bs->file->bs->full_open_options) { 3975 qdict_put_obj(opts, "driver", 3976 QOBJECT(qstring_from_str(drv->format_name))); 3977 QINCREF(bs->file->bs->full_open_options); 3978 qdict_put_obj(opts, "file", 3979 QOBJECT(bs->file->bs->full_open_options)); 3980 3981 bs->full_open_options = opts; 3982 } else { 3983 QDECREF(opts); 3984 } 3985 } else if (!bs->full_open_options && qdict_size(bs->options)) { 3986 /* There is no underlying file BDS (at least referenced by BDS.file), 3987 * so the full options QDict should be equal to the options given 3988 * specifically for this block device when it was opened (plus the 3989 * driver specification). 3990 * Because those options don't change, there is no need to update 3991 * full_open_options when it's already set. */ 3992 3993 opts = qdict_new(); 3994 append_open_options(opts, bs); 3995 qdict_put_obj(opts, "driver", 3996 QOBJECT(qstring_from_str(drv->format_name))); 3997 3998 if (bs->exact_filename[0]) { 3999 /* This may not work for all block protocol drivers (some may 4000 * require this filename to be parsed), but we have to find some 4001 * default solution here, so just include it. If some block driver 4002 * does not support pure options without any filename at all or 4003 * needs some special format of the options QDict, it needs to 4004 * implement the driver-specific bdrv_refresh_filename() function. 4005 */ 4006 qdict_put_obj(opts, "filename", 4007 QOBJECT(qstring_from_str(bs->exact_filename))); 4008 } 4009 4010 bs->full_open_options = opts; 4011 } 4012 4013 if (bs->exact_filename[0]) { 4014 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); 4015 } else if (bs->full_open_options) { 4016 QString *json = qobject_to_json(QOBJECT(bs->full_open_options)); 4017 snprintf(bs->filename, sizeof(bs->filename), "json:%s", 4018 qstring_get_str(json)); 4019 QDECREF(json); 4020 } 4021 } 4022 4023 /* 4024 * Hot add/remove a BDS's child. So the user can take a child offline when 4025 * it is broken and take a new child online 4026 */ 4027 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, 4028 Error **errp) 4029 { 4030 4031 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { 4032 error_setg(errp, "The node %s does not support adding a child", 4033 bdrv_get_device_or_node_name(parent_bs)); 4034 return; 4035 } 4036 4037 if (!QLIST_EMPTY(&child_bs->parents)) { 4038 error_setg(errp, "The node %s already has a parent", 4039 child_bs->node_name); 4040 return; 4041 } 4042 4043 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); 4044 } 4045 4046 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) 4047 { 4048 BdrvChild *tmp; 4049 4050 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { 4051 error_setg(errp, "The node %s does not support removing a child", 4052 bdrv_get_device_or_node_name(parent_bs)); 4053 return; 4054 } 4055 4056 QLIST_FOREACH(tmp, &parent_bs->children, next) { 4057 if (tmp == child) { 4058 break; 4059 } 4060 } 4061 4062 if (!tmp) { 4063 error_setg(errp, "The node %s does not have a child named %s", 4064 bdrv_get_device_or_node_name(parent_bs), 4065 bdrv_get_device_or_node_name(child->bs)); 4066 return; 4067 } 4068 4069 parent_bs->drv->bdrv_del_child(parent_bs, child, errp); 4070 } 4071