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