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