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 "config-host.h" 25 #include "qemu-common.h" 26 #include "trace.h" 27 #include "monitor/monitor.h" 28 #include "block/block_int.h" 29 #include "block/blockjob.h" 30 #include "qemu/module.h" 31 #include "qapi/qmp/qjson.h" 32 #include "sysemu/sysemu.h" 33 #include "qemu/notify.h" 34 #include "block/coroutine.h" 35 #include "block/qapi.h" 36 #include "qmp-commands.h" 37 #include "qemu/timer.h" 38 39 #ifdef CONFIG_BSD 40 #include <sys/types.h> 41 #include <sys/stat.h> 42 #include <sys/ioctl.h> 43 #include <sys/queue.h> 44 #ifndef __DragonFly__ 45 #include <sys/disk.h> 46 #endif 47 #endif 48 49 #ifdef _WIN32 50 #include <windows.h> 51 #endif 52 53 struct BdrvDirtyBitmap { 54 HBitmap *bitmap; 55 QLIST_ENTRY(BdrvDirtyBitmap) list; 56 }; 57 58 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ 59 60 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load); 61 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs, 62 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, 63 BlockDriverCompletionFunc *cb, void *opaque); 64 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs, 65 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, 66 BlockDriverCompletionFunc *cb, void *opaque); 67 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, 68 int64_t sector_num, int nb_sectors, 69 QEMUIOVector *iov); 70 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, 71 int64_t sector_num, int nb_sectors, 72 QEMUIOVector *iov); 73 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, 74 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, 75 BdrvRequestFlags flags); 76 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, 77 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, 78 BdrvRequestFlags flags); 79 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, 80 int64_t sector_num, 81 QEMUIOVector *qiov, 82 int nb_sectors, 83 BdrvRequestFlags flags, 84 BlockDriverCompletionFunc *cb, 85 void *opaque, 86 bool is_write); 87 static void coroutine_fn bdrv_co_do_rw(void *opaque); 88 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, 89 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags); 90 91 static QTAILQ_HEAD(, BlockDriverState) bdrv_states = 92 QTAILQ_HEAD_INITIALIZER(bdrv_states); 93 94 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states = 95 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states); 96 97 static QLIST_HEAD(, BlockDriver) bdrv_drivers = 98 QLIST_HEAD_INITIALIZER(bdrv_drivers); 99 100 /* If non-zero, use only whitelisted block drivers */ 101 static int use_bdrv_whitelist; 102 103 #ifdef _WIN32 104 static int is_windows_drive_prefix(const char *filename) 105 { 106 return (((filename[0] >= 'a' && filename[0] <= 'z') || 107 (filename[0] >= 'A' && filename[0] <= 'Z')) && 108 filename[1] == ':'); 109 } 110 111 int is_windows_drive(const char *filename) 112 { 113 if (is_windows_drive_prefix(filename) && 114 filename[2] == '\0') 115 return 1; 116 if (strstart(filename, "\\\\.\\", NULL) || 117 strstart(filename, "//./", NULL)) 118 return 1; 119 return 0; 120 } 121 #endif 122 123 /* throttling disk I/O limits */ 124 void bdrv_set_io_limits(BlockDriverState *bs, 125 ThrottleConfig *cfg) 126 { 127 int i; 128 129 throttle_config(&bs->throttle_state, cfg); 130 131 for (i = 0; i < 2; i++) { 132 qemu_co_enter_next(&bs->throttled_reqs[i]); 133 } 134 } 135 136 /* this function drain all the throttled IOs */ 137 static bool bdrv_start_throttled_reqs(BlockDriverState *bs) 138 { 139 bool drained = false; 140 bool enabled = bs->io_limits_enabled; 141 int i; 142 143 bs->io_limits_enabled = false; 144 145 for (i = 0; i < 2; i++) { 146 while (qemu_co_enter_next(&bs->throttled_reqs[i])) { 147 drained = true; 148 } 149 } 150 151 bs->io_limits_enabled = enabled; 152 153 return drained; 154 } 155 156 void bdrv_io_limits_disable(BlockDriverState *bs) 157 { 158 bs->io_limits_enabled = false; 159 160 bdrv_start_throttled_reqs(bs); 161 162 throttle_destroy(&bs->throttle_state); 163 } 164 165 static void bdrv_throttle_read_timer_cb(void *opaque) 166 { 167 BlockDriverState *bs = opaque; 168 qemu_co_enter_next(&bs->throttled_reqs[0]); 169 } 170 171 static void bdrv_throttle_write_timer_cb(void *opaque) 172 { 173 BlockDriverState *bs = opaque; 174 qemu_co_enter_next(&bs->throttled_reqs[1]); 175 } 176 177 /* should be called before bdrv_set_io_limits if a limit is set */ 178 void bdrv_io_limits_enable(BlockDriverState *bs) 179 { 180 assert(!bs->io_limits_enabled); 181 throttle_init(&bs->throttle_state, 182 QEMU_CLOCK_VIRTUAL, 183 bdrv_throttle_read_timer_cb, 184 bdrv_throttle_write_timer_cb, 185 bs); 186 bs->io_limits_enabled = true; 187 } 188 189 /* This function makes an IO wait if needed 190 * 191 * @nb_sectors: the number of sectors of the IO 192 * @is_write: is the IO a write 193 */ 194 static void bdrv_io_limits_intercept(BlockDriverState *bs, 195 int nb_sectors, 196 bool is_write) 197 { 198 /* does this io must wait */ 199 bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write); 200 201 /* if must wait or any request of this type throttled queue the IO */ 202 if (must_wait || 203 !qemu_co_queue_empty(&bs->throttled_reqs[is_write])) { 204 qemu_co_queue_wait(&bs->throttled_reqs[is_write]); 205 } 206 207 /* the IO will be executed, do the accounting */ 208 throttle_account(&bs->throttle_state, 209 is_write, 210 nb_sectors * BDRV_SECTOR_SIZE); 211 212 /* if the next request must wait -> do nothing */ 213 if (throttle_schedule_timer(&bs->throttle_state, is_write)) { 214 return; 215 } 216 217 /* else queue next request for execution */ 218 qemu_co_queue_next(&bs->throttled_reqs[is_write]); 219 } 220 221 /* check if the path starts with "<protocol>:" */ 222 static int path_has_protocol(const char *path) 223 { 224 const char *p; 225 226 #ifdef _WIN32 227 if (is_windows_drive(path) || 228 is_windows_drive_prefix(path)) { 229 return 0; 230 } 231 p = path + strcspn(path, ":/\\"); 232 #else 233 p = path + strcspn(path, ":/"); 234 #endif 235 236 return *p == ':'; 237 } 238 239 int path_is_absolute(const char *path) 240 { 241 #ifdef _WIN32 242 /* specific case for names like: "\\.\d:" */ 243 if (is_windows_drive(path) || is_windows_drive_prefix(path)) { 244 return 1; 245 } 246 return (*path == '/' || *path == '\\'); 247 #else 248 return (*path == '/'); 249 #endif 250 } 251 252 /* if filename is absolute, just copy it to dest. Otherwise, build a 253 path to it by considering it is relative to base_path. URL are 254 supported. */ 255 void path_combine(char *dest, int dest_size, 256 const char *base_path, 257 const char *filename) 258 { 259 const char *p, *p1; 260 int len; 261 262 if (dest_size <= 0) 263 return; 264 if (path_is_absolute(filename)) { 265 pstrcpy(dest, dest_size, filename); 266 } else { 267 p = strchr(base_path, ':'); 268 if (p) 269 p++; 270 else 271 p = base_path; 272 p1 = strrchr(base_path, '/'); 273 #ifdef _WIN32 274 { 275 const char *p2; 276 p2 = strrchr(base_path, '\\'); 277 if (!p1 || p2 > p1) 278 p1 = p2; 279 } 280 #endif 281 if (p1) 282 p1++; 283 else 284 p1 = base_path; 285 if (p1 > p) 286 p = p1; 287 len = p - base_path; 288 if (len > dest_size - 1) 289 len = dest_size - 1; 290 memcpy(dest, base_path, len); 291 dest[len] = '\0'; 292 pstrcat(dest, dest_size, filename); 293 } 294 } 295 296 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz) 297 { 298 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) { 299 pstrcpy(dest, sz, bs->backing_file); 300 } else { 301 path_combine(dest, sz, bs->filename, bs->backing_file); 302 } 303 } 304 305 void bdrv_register(BlockDriver *bdrv) 306 { 307 /* Block drivers without coroutine functions need emulation */ 308 if (!bdrv->bdrv_co_readv) { 309 bdrv->bdrv_co_readv = bdrv_co_readv_em; 310 bdrv->bdrv_co_writev = bdrv_co_writev_em; 311 312 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if 313 * the block driver lacks aio we need to emulate that too. 314 */ 315 if (!bdrv->bdrv_aio_readv) { 316 /* add AIO emulation layer */ 317 bdrv->bdrv_aio_readv = bdrv_aio_readv_em; 318 bdrv->bdrv_aio_writev = bdrv_aio_writev_em; 319 } 320 } 321 322 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); 323 } 324 325 /* create a new block device (by default it is empty) */ 326 BlockDriverState *bdrv_new(const char *device_name) 327 { 328 BlockDriverState *bs; 329 330 bs = g_malloc0(sizeof(BlockDriverState)); 331 QLIST_INIT(&bs->dirty_bitmaps); 332 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name); 333 if (device_name[0] != '\0') { 334 QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list); 335 } 336 bdrv_iostatus_disable(bs); 337 notifier_list_init(&bs->close_notifiers); 338 notifier_with_return_list_init(&bs->before_write_notifiers); 339 qemu_co_queue_init(&bs->throttled_reqs[0]); 340 qemu_co_queue_init(&bs->throttled_reqs[1]); 341 bs->refcnt = 1; 342 343 return bs; 344 } 345 346 void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify) 347 { 348 notifier_list_add(&bs->close_notifiers, notify); 349 } 350 351 BlockDriver *bdrv_find_format(const char *format_name) 352 { 353 BlockDriver *drv1; 354 QLIST_FOREACH(drv1, &bdrv_drivers, list) { 355 if (!strcmp(drv1->format_name, format_name)) { 356 return drv1; 357 } 358 } 359 return NULL; 360 } 361 362 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) 363 { 364 static const char *whitelist_rw[] = { 365 CONFIG_BDRV_RW_WHITELIST 366 }; 367 static const char *whitelist_ro[] = { 368 CONFIG_BDRV_RO_WHITELIST 369 }; 370 const char **p; 371 372 if (!whitelist_rw[0] && !whitelist_ro[0]) { 373 return 1; /* no whitelist, anything goes */ 374 } 375 376 for (p = whitelist_rw; *p; p++) { 377 if (!strcmp(drv->format_name, *p)) { 378 return 1; 379 } 380 } 381 if (read_only) { 382 for (p = whitelist_ro; *p; p++) { 383 if (!strcmp(drv->format_name, *p)) { 384 return 1; 385 } 386 } 387 } 388 return 0; 389 } 390 391 BlockDriver *bdrv_find_whitelisted_format(const char *format_name, 392 bool read_only) 393 { 394 BlockDriver *drv = bdrv_find_format(format_name); 395 return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL; 396 } 397 398 typedef struct CreateCo { 399 BlockDriver *drv; 400 char *filename; 401 QEMUOptionParameter *options; 402 int ret; 403 Error *err; 404 } CreateCo; 405 406 static void coroutine_fn bdrv_create_co_entry(void *opaque) 407 { 408 Error *local_err = NULL; 409 int ret; 410 411 CreateCo *cco = opaque; 412 assert(cco->drv); 413 414 ret = cco->drv->bdrv_create(cco->filename, cco->options, &local_err); 415 if (error_is_set(&local_err)) { 416 error_propagate(&cco->err, local_err); 417 } 418 cco->ret = ret; 419 } 420 421 int bdrv_create(BlockDriver *drv, const char* filename, 422 QEMUOptionParameter *options, Error **errp) 423 { 424 int ret; 425 426 Coroutine *co; 427 CreateCo cco = { 428 .drv = drv, 429 .filename = g_strdup(filename), 430 .options = options, 431 .ret = NOT_DONE, 432 .err = NULL, 433 }; 434 435 if (!drv->bdrv_create) { 436 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name); 437 ret = -ENOTSUP; 438 goto out; 439 } 440 441 if (qemu_in_coroutine()) { 442 /* Fast-path if already in coroutine context */ 443 bdrv_create_co_entry(&cco); 444 } else { 445 co = qemu_coroutine_create(bdrv_create_co_entry); 446 qemu_coroutine_enter(co, &cco); 447 while (cco.ret == NOT_DONE) { 448 qemu_aio_wait(); 449 } 450 } 451 452 ret = cco.ret; 453 if (ret < 0) { 454 if (error_is_set(&cco.err)) { 455 error_propagate(errp, cco.err); 456 } else { 457 error_setg_errno(errp, -ret, "Could not create image"); 458 } 459 } 460 461 out: 462 g_free(cco.filename); 463 return ret; 464 } 465 466 int bdrv_create_file(const char* filename, QEMUOptionParameter *options, 467 Error **errp) 468 { 469 BlockDriver *drv; 470 Error *local_err = NULL; 471 int ret; 472 473 drv = bdrv_find_protocol(filename, true); 474 if (drv == NULL) { 475 error_setg(errp, "Could not find protocol for file '%s'", filename); 476 return -ENOENT; 477 } 478 479 ret = bdrv_create(drv, filename, options, &local_err); 480 if (error_is_set(&local_err)) { 481 error_propagate(errp, local_err); 482 } 483 return ret; 484 } 485 486 /* 487 * Create a uniquely-named empty temporary file. 488 * Return 0 upon success, otherwise a negative errno value. 489 */ 490 int get_tmp_filename(char *filename, int size) 491 { 492 #ifdef _WIN32 493 char temp_dir[MAX_PATH]; 494 /* GetTempFileName requires that its output buffer (4th param) 495 have length MAX_PATH or greater. */ 496 assert(size >= MAX_PATH); 497 return (GetTempPath(MAX_PATH, temp_dir) 498 && GetTempFileName(temp_dir, "qem", 0, filename) 499 ? 0 : -GetLastError()); 500 #else 501 int fd; 502 const char *tmpdir; 503 tmpdir = getenv("TMPDIR"); 504 if (!tmpdir) 505 tmpdir = "/tmp"; 506 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) { 507 return -EOVERFLOW; 508 } 509 fd = mkstemp(filename); 510 if (fd < 0) { 511 return -errno; 512 } 513 if (close(fd) != 0) { 514 unlink(filename); 515 return -errno; 516 } 517 return 0; 518 #endif 519 } 520 521 /* 522 * Detect host devices. By convention, /dev/cdrom[N] is always 523 * recognized as a host CDROM. 524 */ 525 static BlockDriver *find_hdev_driver(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_device) { 532 score = d->bdrv_probe_device(filename); 533 if (score > score_max) { 534 score_max = score; 535 drv = d; 536 } 537 } 538 } 539 540 return drv; 541 } 542 543 BlockDriver *bdrv_find_protocol(const char *filename, 544 bool allow_protocol_prefix) 545 { 546 BlockDriver *drv1; 547 char protocol[128]; 548 int len; 549 const char *p; 550 551 /* TODO Drivers without bdrv_file_open must be specified explicitly */ 552 553 /* 554 * XXX(hch): we really should not let host device detection 555 * override an explicit protocol specification, but moving this 556 * later breaks access to device names with colons in them. 557 * Thanks to the brain-dead persistent naming schemes on udev- 558 * based Linux systems those actually are quite common. 559 */ 560 drv1 = find_hdev_driver(filename); 561 if (drv1) { 562 return drv1; 563 } 564 565 if (!path_has_protocol(filename) || !allow_protocol_prefix) { 566 return bdrv_find_format("file"); 567 } 568 569 p = strchr(filename, ':'); 570 assert(p != NULL); 571 len = p - filename; 572 if (len > sizeof(protocol) - 1) 573 len = sizeof(protocol) - 1; 574 memcpy(protocol, filename, len); 575 protocol[len] = '\0'; 576 QLIST_FOREACH(drv1, &bdrv_drivers, list) { 577 if (drv1->protocol_name && 578 !strcmp(drv1->protocol_name, protocol)) { 579 return drv1; 580 } 581 } 582 return NULL; 583 } 584 585 static int find_image_format(BlockDriverState *bs, const char *filename, 586 BlockDriver **pdrv, Error **errp) 587 { 588 int score, score_max; 589 BlockDriver *drv1, *drv; 590 uint8_t buf[2048]; 591 int ret = 0; 592 593 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ 594 if (bs->sg || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) { 595 drv = bdrv_find_format("raw"); 596 if (!drv) { 597 error_setg(errp, "Could not find raw image format"); 598 ret = -ENOENT; 599 } 600 *pdrv = drv; 601 return ret; 602 } 603 604 ret = bdrv_pread(bs, 0, buf, sizeof(buf)); 605 if (ret < 0) { 606 error_setg_errno(errp, -ret, "Could not read image for determining its " 607 "format"); 608 *pdrv = NULL; 609 return ret; 610 } 611 612 score_max = 0; 613 drv = NULL; 614 QLIST_FOREACH(drv1, &bdrv_drivers, list) { 615 if (drv1->bdrv_probe) { 616 score = drv1->bdrv_probe(buf, ret, filename); 617 if (score > score_max) { 618 score_max = score; 619 drv = drv1; 620 } 621 } 622 } 623 if (!drv) { 624 error_setg(errp, "Could not determine image format: No compatible " 625 "driver found"); 626 ret = -ENOENT; 627 } 628 *pdrv = drv; 629 return ret; 630 } 631 632 /** 633 * Set the current 'total_sectors' value 634 */ 635 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint) 636 { 637 BlockDriver *drv = bs->drv; 638 639 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ 640 if (bs->sg) 641 return 0; 642 643 /* query actual device if possible, otherwise just trust the hint */ 644 if (drv->bdrv_getlength) { 645 int64_t length = drv->bdrv_getlength(bs); 646 if (length < 0) { 647 return length; 648 } 649 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); 650 } 651 652 bs->total_sectors = hint; 653 return 0; 654 } 655 656 /** 657 * Set open flags for a given discard mode 658 * 659 * Return 0 on success, -1 if the discard mode was invalid. 660 */ 661 int bdrv_parse_discard_flags(const char *mode, int *flags) 662 { 663 *flags &= ~BDRV_O_UNMAP; 664 665 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { 666 /* do nothing */ 667 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { 668 *flags |= BDRV_O_UNMAP; 669 } else { 670 return -1; 671 } 672 673 return 0; 674 } 675 676 /** 677 * Set open flags for a given cache mode 678 * 679 * Return 0 on success, -1 if the cache mode was invalid. 680 */ 681 int bdrv_parse_cache_flags(const char *mode, int *flags) 682 { 683 *flags &= ~BDRV_O_CACHE_MASK; 684 685 if (!strcmp(mode, "off") || !strcmp(mode, "none")) { 686 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB; 687 } else if (!strcmp(mode, "directsync")) { 688 *flags |= BDRV_O_NOCACHE; 689 } else if (!strcmp(mode, "writeback")) { 690 *flags |= BDRV_O_CACHE_WB; 691 } else if (!strcmp(mode, "unsafe")) { 692 *flags |= BDRV_O_CACHE_WB; 693 *flags |= BDRV_O_NO_FLUSH; 694 } else if (!strcmp(mode, "writethrough")) { 695 /* this is the default */ 696 } else { 697 return -1; 698 } 699 700 return 0; 701 } 702 703 /** 704 * The copy-on-read flag is actually a reference count so multiple users may 705 * use the feature without worrying about clobbering its previous state. 706 * Copy-on-read stays enabled until all users have called to disable it. 707 */ 708 void bdrv_enable_copy_on_read(BlockDriverState *bs) 709 { 710 bs->copy_on_read++; 711 } 712 713 void bdrv_disable_copy_on_read(BlockDriverState *bs) 714 { 715 assert(bs->copy_on_read > 0); 716 bs->copy_on_read--; 717 } 718 719 static int bdrv_open_flags(BlockDriverState *bs, int flags) 720 { 721 int open_flags = flags | BDRV_O_CACHE_WB; 722 723 /* 724 * Clear flags that are internal to the block layer before opening the 725 * image. 726 */ 727 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); 728 729 /* 730 * Snapshots should be writable. 731 */ 732 if (bs->is_temporary) { 733 open_flags |= BDRV_O_RDWR; 734 } 735 736 return open_flags; 737 } 738 739 static int bdrv_assign_node_name(BlockDriverState *bs, 740 const char *node_name, 741 Error **errp) 742 { 743 if (!node_name) { 744 return 0; 745 } 746 747 /* empty string node name is invalid */ 748 if (node_name[0] == '\0') { 749 error_setg(errp, "Empty node name"); 750 return -EINVAL; 751 } 752 753 /* takes care of avoiding duplicates node names */ 754 if (bdrv_find_node(node_name)) { 755 error_setg(errp, "Duplicate node name"); 756 return -EINVAL; 757 } 758 759 /* copy node name into the bs and insert it into the graph list */ 760 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); 761 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); 762 763 return 0; 764 } 765 766 /* 767 * Common part for opening disk images and files 768 * 769 * Removes all processed options from *options. 770 */ 771 static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file, 772 QDict *options, int flags, BlockDriver *drv, Error **errp) 773 { 774 int ret, open_flags; 775 const char *filename; 776 const char *node_name = NULL; 777 Error *local_err = NULL; 778 779 assert(drv != NULL); 780 assert(bs->file == NULL); 781 assert(options != NULL && bs->options != options); 782 783 if (file != NULL) { 784 filename = file->filename; 785 } else { 786 filename = qdict_get_try_str(options, "filename"); 787 } 788 789 trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name); 790 791 node_name = qdict_get_try_str(options, "node-name"); 792 ret = bdrv_assign_node_name(bs, node_name, errp); 793 if (ret < 0) { 794 return ret; 795 } 796 qdict_del(options, "node-name"); 797 798 /* bdrv_open() with directly using a protocol as drv. This layer is already 799 * opened, so assign it to bs (while file becomes a closed BlockDriverState) 800 * and return immediately. */ 801 if (file != NULL && drv->bdrv_file_open) { 802 bdrv_swap(file, bs); 803 return 0; 804 } 805 806 bs->open_flags = flags; 807 bs->buffer_alignment = 512; 808 bs->zero_beyond_eof = true; 809 open_flags = bdrv_open_flags(bs, flags); 810 bs->read_only = !(open_flags & BDRV_O_RDWR); 811 812 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) { 813 error_setg(errp, 814 !bs->read_only && bdrv_is_whitelisted(drv, true) 815 ? "Driver '%s' can only be used for read-only devices" 816 : "Driver '%s' is not whitelisted", 817 drv->format_name); 818 return -ENOTSUP; 819 } 820 821 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */ 822 if (flags & BDRV_O_COPY_ON_READ) { 823 if (!bs->read_only) { 824 bdrv_enable_copy_on_read(bs); 825 } else { 826 error_setg(errp, "Can't use copy-on-read on read-only device"); 827 return -EINVAL; 828 } 829 } 830 831 if (filename != NULL) { 832 pstrcpy(bs->filename, sizeof(bs->filename), filename); 833 } else { 834 bs->filename[0] = '\0'; 835 } 836 837 bs->drv = drv; 838 bs->opaque = g_malloc0(drv->instance_size); 839 840 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB); 841 842 /* Open the image, either directly or using a protocol */ 843 if (drv->bdrv_file_open) { 844 assert(file == NULL); 845 assert(!drv->bdrv_needs_filename || filename != NULL); 846 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); 847 } else { 848 if (file == NULL) { 849 error_setg(errp, "Can't use '%s' as a block driver for the " 850 "protocol level", drv->format_name); 851 ret = -EINVAL; 852 goto free_and_fail; 853 } 854 bs->file = file; 855 ret = drv->bdrv_open(bs, options, open_flags, &local_err); 856 } 857 858 if (ret < 0) { 859 if (error_is_set(&local_err)) { 860 error_propagate(errp, local_err); 861 } else if (bs->filename[0]) { 862 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); 863 } else { 864 error_setg_errno(errp, -ret, "Could not open image"); 865 } 866 goto free_and_fail; 867 } 868 869 ret = refresh_total_sectors(bs, bs->total_sectors); 870 if (ret < 0) { 871 error_setg_errno(errp, -ret, "Could not refresh total sector count"); 872 goto free_and_fail; 873 } 874 875 #ifndef _WIN32 876 if (bs->is_temporary) { 877 assert(bs->filename[0] != '\0'); 878 unlink(bs->filename); 879 } 880 #endif 881 return 0; 882 883 free_and_fail: 884 bs->file = NULL; 885 g_free(bs->opaque); 886 bs->opaque = NULL; 887 bs->drv = NULL; 888 return ret; 889 } 890 891 /* 892 * Opens a file using a protocol (file, host_device, nbd, ...) 893 * 894 * options is a QDict of options to pass to the block drivers, or NULL for an 895 * empty set of options. The reference to the QDict belongs to the block layer 896 * after the call (even on failure), so if the caller intends to reuse the 897 * dictionary, it needs to use QINCREF() before calling bdrv_file_open. 898 */ 899 int bdrv_file_open(BlockDriverState **pbs, const char *filename, 900 const char *reference, QDict *options, int flags, 901 Error **errp) 902 { 903 BlockDriverState *bs = NULL; 904 BlockDriver *drv; 905 const char *drvname; 906 bool allow_protocol_prefix = false; 907 Error *local_err = NULL; 908 int ret; 909 910 /* NULL means an empty set of options */ 911 if (options == NULL) { 912 options = qdict_new(); 913 } 914 915 if (reference) { 916 if (filename || qdict_size(options)) { 917 error_setg(errp, "Cannot reference an existing block device with " 918 "additional options or a new filename"); 919 return -EINVAL; 920 } 921 QDECREF(options); 922 923 bs = bdrv_find(reference); 924 if (!bs) { 925 error_setg(errp, "Cannot find block device '%s'", reference); 926 return -ENODEV; 927 } 928 bdrv_ref(bs); 929 *pbs = bs; 930 return 0; 931 } 932 933 bs = bdrv_new(""); 934 bs->options = options; 935 options = qdict_clone_shallow(options); 936 937 /* Fetch the file name from the options QDict if necessary */ 938 if (!filename) { 939 filename = qdict_get_try_str(options, "filename"); 940 } else if (filename && !qdict_haskey(options, "filename")) { 941 qdict_put(options, "filename", qstring_from_str(filename)); 942 allow_protocol_prefix = true; 943 } else { 944 error_setg(errp, "Can't specify 'file' and 'filename' options at the " 945 "same time"); 946 ret = -EINVAL; 947 goto fail; 948 } 949 950 /* Find the right block driver */ 951 drvname = qdict_get_try_str(options, "driver"); 952 if (drvname) { 953 drv = bdrv_find_format(drvname); 954 if (!drv) { 955 error_setg(errp, "Unknown driver '%s'", drvname); 956 } 957 qdict_del(options, "driver"); 958 } else if (filename) { 959 drv = bdrv_find_protocol(filename, allow_protocol_prefix); 960 if (!drv) { 961 error_setg(errp, "Unknown protocol"); 962 } 963 } else { 964 error_setg(errp, "Must specify either driver or file"); 965 drv = NULL; 966 } 967 968 if (!drv) { 969 /* errp has been set already */ 970 ret = -ENOENT; 971 goto fail; 972 } 973 974 /* Parse the filename and open it */ 975 if (drv->bdrv_parse_filename && filename) { 976 drv->bdrv_parse_filename(filename, options, &local_err); 977 if (error_is_set(&local_err)) { 978 error_propagate(errp, local_err); 979 ret = -EINVAL; 980 goto fail; 981 } 982 qdict_del(options, "filename"); 983 } else if (drv->bdrv_needs_filename && !filename) { 984 error_setg(errp, "The '%s' block driver requires a file name", 985 drv->format_name); 986 ret = -EINVAL; 987 goto fail; 988 } 989 990 if (!drv->bdrv_file_open) { 991 ret = bdrv_open(bs, filename, options, flags, drv, &local_err); 992 options = NULL; 993 } else { 994 ret = bdrv_open_common(bs, NULL, options, flags, drv, &local_err); 995 } 996 if (ret < 0) { 997 error_propagate(errp, local_err); 998 goto fail; 999 } 1000 1001 /* Check if any unknown options were used */ 1002 if (options && (qdict_size(options) != 0)) { 1003 const QDictEntry *entry = qdict_first(options); 1004 error_setg(errp, "Block protocol '%s' doesn't support the option '%s'", 1005 drv->format_name, entry->key); 1006 ret = -EINVAL; 1007 goto fail; 1008 } 1009 QDECREF(options); 1010 1011 bs->growable = 1; 1012 *pbs = bs; 1013 return 0; 1014 1015 fail: 1016 QDECREF(options); 1017 if (!bs->drv) { 1018 QDECREF(bs->options); 1019 } 1020 bdrv_unref(bs); 1021 return ret; 1022 } 1023 1024 /* 1025 * Opens the backing file for a BlockDriverState if not yet open 1026 * 1027 * options is a QDict of options to pass to the block drivers, or NULL for an 1028 * empty set of options. The reference to the QDict is transferred to this 1029 * function (even on failure), so if the caller intends to reuse the dictionary, 1030 * it needs to use QINCREF() before calling bdrv_file_open. 1031 */ 1032 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp) 1033 { 1034 char backing_filename[PATH_MAX]; 1035 int back_flags, ret; 1036 BlockDriver *back_drv = NULL; 1037 Error *local_err = NULL; 1038 1039 if (bs->backing_hd != NULL) { 1040 QDECREF(options); 1041 return 0; 1042 } 1043 1044 /* NULL means an empty set of options */ 1045 if (options == NULL) { 1046 options = qdict_new(); 1047 } 1048 1049 bs->open_flags &= ~BDRV_O_NO_BACKING; 1050 if (qdict_haskey(options, "file.filename")) { 1051 backing_filename[0] = '\0'; 1052 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { 1053 QDECREF(options); 1054 return 0; 1055 } else { 1056 bdrv_get_full_backing_filename(bs, backing_filename, 1057 sizeof(backing_filename)); 1058 } 1059 1060 bs->backing_hd = bdrv_new(""); 1061 1062 if (bs->backing_format[0] != '\0') { 1063 back_drv = bdrv_find_format(bs->backing_format); 1064 } 1065 1066 /* backing files always opened read-only */ 1067 back_flags = bs->open_flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | 1068 BDRV_O_COPY_ON_READ); 1069 1070 ret = bdrv_open(bs->backing_hd, 1071 *backing_filename ? backing_filename : NULL, options, 1072 back_flags, back_drv, &local_err); 1073 if (ret < 0) { 1074 bdrv_unref(bs->backing_hd); 1075 bs->backing_hd = NULL; 1076 bs->open_flags |= BDRV_O_NO_BACKING; 1077 error_setg(errp, "Could not open backing file: %s", 1078 error_get_pretty(local_err)); 1079 error_free(local_err); 1080 return ret; 1081 } 1082 1083 if (bs->backing_hd->file) { 1084 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 1085 bs->backing_hd->file->filename); 1086 } 1087 1088 return 0; 1089 } 1090 1091 /* 1092 * Opens a disk image whose options are given as BlockdevRef in another block 1093 * device's options. 1094 * 1095 * If force_raw is true, bdrv_file_open() will be used, thereby preventing any 1096 * image format auto-detection. If it is false and a filename is given, 1097 * bdrv_open() will be used for auto-detection. 1098 * 1099 * If allow_none is true, no image will be opened if filename is false and no 1100 * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned. 1101 * 1102 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. 1103 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict 1104 * itself, all options starting with "${bdref_key}." are considered part of the 1105 * BlockdevRef. 1106 * 1107 * The BlockdevRef will be removed from the options QDict. 1108 */ 1109 int bdrv_open_image(BlockDriverState **pbs, const char *filename, 1110 QDict *options, const char *bdref_key, int flags, 1111 bool force_raw, bool allow_none, Error **errp) 1112 { 1113 QDict *image_options; 1114 int ret; 1115 char *bdref_key_dot; 1116 const char *reference; 1117 1118 bdref_key_dot = g_strdup_printf("%s.", bdref_key); 1119 qdict_extract_subqdict(options, &image_options, bdref_key_dot); 1120 g_free(bdref_key_dot); 1121 1122 reference = qdict_get_try_str(options, bdref_key); 1123 if (!filename && !reference && !qdict_size(image_options)) { 1124 if (allow_none) { 1125 ret = 0; 1126 } else { 1127 error_setg(errp, "A block device must be specified for \"%s\"", 1128 bdref_key); 1129 ret = -EINVAL; 1130 } 1131 goto done; 1132 } 1133 1134 if (filename && !force_raw) { 1135 /* If a filename is given and the block driver should be detected 1136 automatically (instead of using none), use bdrv_open() in order to do 1137 that auto-detection. */ 1138 BlockDriverState *bs; 1139 1140 if (reference) { 1141 error_setg(errp, "Cannot reference an existing block device while " 1142 "giving a filename"); 1143 ret = -EINVAL; 1144 goto done; 1145 } 1146 1147 bs = bdrv_new(""); 1148 ret = bdrv_open(bs, filename, image_options, flags, NULL, errp); 1149 if (ret < 0) { 1150 bdrv_unref(bs); 1151 } else { 1152 *pbs = bs; 1153 } 1154 } else { 1155 ret = bdrv_file_open(pbs, filename, reference, image_options, flags, 1156 errp); 1157 } 1158 1159 done: 1160 qdict_del(options, bdref_key); 1161 return ret; 1162 } 1163 1164 /* 1165 * Opens a disk image (raw, qcow2, vmdk, ...) 1166 * 1167 * options is a QDict of options to pass to the block drivers, or NULL for an 1168 * empty set of options. The reference to the QDict belongs to the block layer 1169 * after the call (even on failure), so if the caller intends to reuse the 1170 * dictionary, it needs to use QINCREF() before calling bdrv_open. 1171 */ 1172 int bdrv_open(BlockDriverState *bs, const char *filename, QDict *options, 1173 int flags, BlockDriver *drv, Error **errp) 1174 { 1175 int ret; 1176 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */ 1177 char tmp_filename[PATH_MAX + 1]; 1178 BlockDriverState *file = NULL; 1179 const char *drvname; 1180 Error *local_err = NULL; 1181 1182 /* NULL means an empty set of options */ 1183 if (options == NULL) { 1184 options = qdict_new(); 1185 } 1186 1187 bs->options = options; 1188 options = qdict_clone_shallow(options); 1189 1190 /* For snapshot=on, create a temporary qcow2 overlay */ 1191 if (flags & BDRV_O_SNAPSHOT) { 1192 BlockDriverState *bs1; 1193 int64_t total_size; 1194 BlockDriver *bdrv_qcow2; 1195 QEMUOptionParameter *create_options; 1196 QDict *snapshot_options; 1197 1198 /* if snapshot, we create a temporary backing file and open it 1199 instead of opening 'filename' directly */ 1200 1201 /* Get the required size from the image */ 1202 bs1 = bdrv_new(""); 1203 QINCREF(options); 1204 ret = bdrv_open(bs1, filename, options, BDRV_O_NO_BACKING, 1205 drv, &local_err); 1206 if (ret < 0) { 1207 bdrv_unref(bs1); 1208 goto fail; 1209 } 1210 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK; 1211 1212 bdrv_unref(bs1); 1213 1214 /* Create the temporary image */ 1215 ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename)); 1216 if (ret < 0) { 1217 error_setg_errno(errp, -ret, "Could not get temporary filename"); 1218 goto fail; 1219 } 1220 1221 bdrv_qcow2 = bdrv_find_format("qcow2"); 1222 create_options = parse_option_parameters("", bdrv_qcow2->create_options, 1223 NULL); 1224 1225 set_option_parameter_int(create_options, BLOCK_OPT_SIZE, total_size); 1226 1227 ret = bdrv_create(bdrv_qcow2, tmp_filename, create_options, &local_err); 1228 free_option_parameters(create_options); 1229 if (ret < 0) { 1230 error_setg_errno(errp, -ret, "Could not create temporary overlay " 1231 "'%s': %s", tmp_filename, 1232 error_get_pretty(local_err)); 1233 error_free(local_err); 1234 local_err = NULL; 1235 goto fail; 1236 } 1237 1238 /* Prepare a new options QDict for the temporary file, where user 1239 * options refer to the backing file */ 1240 if (filename) { 1241 qdict_put(options, "file.filename", qstring_from_str(filename)); 1242 } 1243 if (drv) { 1244 qdict_put(options, "driver", qstring_from_str(drv->format_name)); 1245 } 1246 1247 snapshot_options = qdict_new(); 1248 qdict_put(snapshot_options, "backing", options); 1249 qdict_flatten(snapshot_options); 1250 1251 bs->options = snapshot_options; 1252 options = qdict_clone_shallow(bs->options); 1253 1254 filename = tmp_filename; 1255 drv = bdrv_qcow2; 1256 bs->is_temporary = 1; 1257 } 1258 1259 /* Open image file without format layer */ 1260 if (flags & BDRV_O_RDWR) { 1261 flags |= BDRV_O_ALLOW_RDWR; 1262 } 1263 1264 ret = bdrv_open_image(&file, filename, options, "file", 1265 bdrv_open_flags(bs, flags | BDRV_O_UNMAP), true, true, 1266 &local_err); 1267 if (ret < 0) { 1268 goto fail; 1269 } 1270 1271 /* Find the right image format driver */ 1272 drvname = qdict_get_try_str(options, "driver"); 1273 if (drvname) { 1274 drv = bdrv_find_format(drvname); 1275 qdict_del(options, "driver"); 1276 if (!drv) { 1277 error_setg(errp, "Invalid driver: '%s'", drvname); 1278 ret = -EINVAL; 1279 goto unlink_and_fail; 1280 } 1281 } 1282 1283 if (!drv) { 1284 if (file) { 1285 ret = find_image_format(file, filename, &drv, &local_err); 1286 } else { 1287 error_setg(errp, "Must specify either driver or file"); 1288 ret = -EINVAL; 1289 goto unlink_and_fail; 1290 } 1291 } 1292 1293 if (!drv) { 1294 goto unlink_and_fail; 1295 } 1296 1297 /* Open the image */ 1298 ret = bdrv_open_common(bs, file, options, flags, drv, &local_err); 1299 if (ret < 0) { 1300 goto unlink_and_fail; 1301 } 1302 1303 if (file && (bs->file != file)) { 1304 bdrv_unref(file); 1305 file = NULL; 1306 } 1307 1308 /* If there is a backing file, use it */ 1309 if ((flags & BDRV_O_NO_BACKING) == 0) { 1310 QDict *backing_options; 1311 1312 qdict_extract_subqdict(options, &backing_options, "backing."); 1313 ret = bdrv_open_backing_file(bs, backing_options, &local_err); 1314 if (ret < 0) { 1315 goto close_and_fail; 1316 } 1317 } 1318 1319 /* Check if any unknown options were used */ 1320 if (qdict_size(options) != 0) { 1321 const QDictEntry *entry = qdict_first(options); 1322 error_setg(errp, "Block format '%s' used by device '%s' doesn't " 1323 "support the option '%s'", drv->format_name, bs->device_name, 1324 entry->key); 1325 1326 ret = -EINVAL; 1327 goto close_and_fail; 1328 } 1329 QDECREF(options); 1330 1331 if (!bdrv_key_required(bs)) { 1332 bdrv_dev_change_media_cb(bs, true); 1333 } 1334 1335 return 0; 1336 1337 unlink_and_fail: 1338 if (file != NULL) { 1339 bdrv_unref(file); 1340 } 1341 if (bs->is_temporary) { 1342 unlink(filename); 1343 } 1344 fail: 1345 QDECREF(bs->options); 1346 QDECREF(options); 1347 bs->options = NULL; 1348 if (error_is_set(&local_err)) { 1349 error_propagate(errp, local_err); 1350 } 1351 return ret; 1352 1353 close_and_fail: 1354 bdrv_close(bs); 1355 QDECREF(options); 1356 if (error_is_set(&local_err)) { 1357 error_propagate(errp, local_err); 1358 } 1359 return ret; 1360 } 1361 1362 typedef struct BlockReopenQueueEntry { 1363 bool prepared; 1364 BDRVReopenState state; 1365 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry; 1366 } BlockReopenQueueEntry; 1367 1368 /* 1369 * Adds a BlockDriverState to a simple queue for an atomic, transactional 1370 * reopen of multiple devices. 1371 * 1372 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT 1373 * already performed, or alternatively may be NULL a new BlockReopenQueue will 1374 * be created and initialized. This newly created BlockReopenQueue should be 1375 * passed back in for subsequent calls that are intended to be of the same 1376 * atomic 'set'. 1377 * 1378 * bs is the BlockDriverState to add to the reopen queue. 1379 * 1380 * flags contains the open flags for the associated bs 1381 * 1382 * returns a pointer to bs_queue, which is either the newly allocated 1383 * bs_queue, or the existing bs_queue being used. 1384 * 1385 */ 1386 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, 1387 BlockDriverState *bs, int flags) 1388 { 1389 assert(bs != NULL); 1390 1391 BlockReopenQueueEntry *bs_entry; 1392 if (bs_queue == NULL) { 1393 bs_queue = g_new0(BlockReopenQueue, 1); 1394 QSIMPLEQ_INIT(bs_queue); 1395 } 1396 1397 if (bs->file) { 1398 bdrv_reopen_queue(bs_queue, bs->file, flags); 1399 } 1400 1401 bs_entry = g_new0(BlockReopenQueueEntry, 1); 1402 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry); 1403 1404 bs_entry->state.bs = bs; 1405 bs_entry->state.flags = flags; 1406 1407 return bs_queue; 1408 } 1409 1410 /* 1411 * Reopen multiple BlockDriverStates atomically & transactionally. 1412 * 1413 * The queue passed in (bs_queue) must have been built up previous 1414 * via bdrv_reopen_queue(). 1415 * 1416 * Reopens all BDS specified in the queue, with the appropriate 1417 * flags. All devices are prepared for reopen, and failure of any 1418 * device will cause all device changes to be abandonded, and intermediate 1419 * data cleaned up. 1420 * 1421 * If all devices prepare successfully, then the changes are committed 1422 * to all devices. 1423 * 1424 */ 1425 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) 1426 { 1427 int ret = -1; 1428 BlockReopenQueueEntry *bs_entry, *next; 1429 Error *local_err = NULL; 1430 1431 assert(bs_queue != NULL); 1432 1433 bdrv_drain_all(); 1434 1435 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { 1436 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) { 1437 error_propagate(errp, local_err); 1438 goto cleanup; 1439 } 1440 bs_entry->prepared = true; 1441 } 1442 1443 /* If we reach this point, we have success and just need to apply the 1444 * changes 1445 */ 1446 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { 1447 bdrv_reopen_commit(&bs_entry->state); 1448 } 1449 1450 ret = 0; 1451 1452 cleanup: 1453 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { 1454 if (ret && bs_entry->prepared) { 1455 bdrv_reopen_abort(&bs_entry->state); 1456 } 1457 g_free(bs_entry); 1458 } 1459 g_free(bs_queue); 1460 return ret; 1461 } 1462 1463 1464 /* Reopen a single BlockDriverState with the specified flags. */ 1465 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp) 1466 { 1467 int ret = -1; 1468 Error *local_err = NULL; 1469 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags); 1470 1471 ret = bdrv_reopen_multiple(queue, &local_err); 1472 if (local_err != NULL) { 1473 error_propagate(errp, local_err); 1474 } 1475 return ret; 1476 } 1477 1478 1479 /* 1480 * Prepares a BlockDriverState for reopen. All changes are staged in the 1481 * 'opaque' field of the BDRVReopenState, which is used and allocated by 1482 * the block driver layer .bdrv_reopen_prepare() 1483 * 1484 * bs is the BlockDriverState to reopen 1485 * flags are the new open flags 1486 * queue is the reopen queue 1487 * 1488 * Returns 0 on success, non-zero on error. On error errp will be set 1489 * as well. 1490 * 1491 * On failure, bdrv_reopen_abort() will be called to clean up any data. 1492 * It is the responsibility of the caller to then call the abort() or 1493 * commit() for any other BDS that have been left in a prepare() state 1494 * 1495 */ 1496 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue, 1497 Error **errp) 1498 { 1499 int ret = -1; 1500 Error *local_err = NULL; 1501 BlockDriver *drv; 1502 1503 assert(reopen_state != NULL); 1504 assert(reopen_state->bs->drv != NULL); 1505 drv = reopen_state->bs->drv; 1506 1507 /* if we are to stay read-only, do not allow permission change 1508 * to r/w */ 1509 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) && 1510 reopen_state->flags & BDRV_O_RDWR) { 1511 error_set(errp, QERR_DEVICE_IS_READ_ONLY, 1512 reopen_state->bs->device_name); 1513 goto error; 1514 } 1515 1516 1517 ret = bdrv_flush(reopen_state->bs); 1518 if (ret) { 1519 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive", 1520 strerror(-ret)); 1521 goto error; 1522 } 1523 1524 if (drv->bdrv_reopen_prepare) { 1525 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); 1526 if (ret) { 1527 if (local_err != NULL) { 1528 error_propagate(errp, local_err); 1529 } else { 1530 error_setg(errp, "failed while preparing to reopen image '%s'", 1531 reopen_state->bs->filename); 1532 } 1533 goto error; 1534 } 1535 } else { 1536 /* It is currently mandatory to have a bdrv_reopen_prepare() 1537 * handler for each supported drv. */ 1538 error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, 1539 drv->format_name, reopen_state->bs->device_name, 1540 "reopening of file"); 1541 ret = -1; 1542 goto error; 1543 } 1544 1545 ret = 0; 1546 1547 error: 1548 return ret; 1549 } 1550 1551 /* 1552 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and 1553 * makes them final by swapping the staging BlockDriverState contents into 1554 * the active BlockDriverState contents. 1555 */ 1556 void bdrv_reopen_commit(BDRVReopenState *reopen_state) 1557 { 1558 BlockDriver *drv; 1559 1560 assert(reopen_state != NULL); 1561 drv = reopen_state->bs->drv; 1562 assert(drv != NULL); 1563 1564 /* If there are any driver level actions to take */ 1565 if (drv->bdrv_reopen_commit) { 1566 drv->bdrv_reopen_commit(reopen_state); 1567 } 1568 1569 /* set BDS specific flags now */ 1570 reopen_state->bs->open_flags = reopen_state->flags; 1571 reopen_state->bs->enable_write_cache = !!(reopen_state->flags & 1572 BDRV_O_CACHE_WB); 1573 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR); 1574 } 1575 1576 /* 1577 * Abort the reopen, and delete and free the staged changes in 1578 * reopen_state 1579 */ 1580 void bdrv_reopen_abort(BDRVReopenState *reopen_state) 1581 { 1582 BlockDriver *drv; 1583 1584 assert(reopen_state != NULL); 1585 drv = reopen_state->bs->drv; 1586 assert(drv != NULL); 1587 1588 if (drv->bdrv_reopen_abort) { 1589 drv->bdrv_reopen_abort(reopen_state); 1590 } 1591 } 1592 1593 1594 void bdrv_close(BlockDriverState *bs) 1595 { 1596 if (bs->job) { 1597 block_job_cancel_sync(bs->job); 1598 } 1599 bdrv_drain_all(); /* complete I/O */ 1600 bdrv_flush(bs); 1601 bdrv_drain_all(); /* in case flush left pending I/O */ 1602 notifier_list_notify(&bs->close_notifiers, bs); 1603 1604 if (bs->drv) { 1605 if (bs->backing_hd) { 1606 bdrv_unref(bs->backing_hd); 1607 bs->backing_hd = NULL; 1608 } 1609 bs->drv->bdrv_close(bs); 1610 g_free(bs->opaque); 1611 #ifdef _WIN32 1612 if (bs->is_temporary) { 1613 unlink(bs->filename); 1614 } 1615 #endif 1616 bs->opaque = NULL; 1617 bs->drv = NULL; 1618 bs->copy_on_read = 0; 1619 bs->backing_file[0] = '\0'; 1620 bs->backing_format[0] = '\0'; 1621 bs->total_sectors = 0; 1622 bs->encrypted = 0; 1623 bs->valid_key = 0; 1624 bs->sg = 0; 1625 bs->growable = 0; 1626 bs->zero_beyond_eof = false; 1627 QDECREF(bs->options); 1628 bs->options = NULL; 1629 1630 if (bs->file != NULL) { 1631 bdrv_unref(bs->file); 1632 bs->file = NULL; 1633 } 1634 } 1635 1636 bdrv_dev_change_media_cb(bs, false); 1637 1638 /*throttling disk I/O limits*/ 1639 if (bs->io_limits_enabled) { 1640 bdrv_io_limits_disable(bs); 1641 } 1642 } 1643 1644 void bdrv_close_all(void) 1645 { 1646 BlockDriverState *bs; 1647 1648 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 1649 bdrv_close(bs); 1650 } 1651 } 1652 1653 /* Check if any requests are in-flight (including throttled requests) */ 1654 static bool bdrv_requests_pending(BlockDriverState *bs) 1655 { 1656 if (!QLIST_EMPTY(&bs->tracked_requests)) { 1657 return true; 1658 } 1659 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) { 1660 return true; 1661 } 1662 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) { 1663 return true; 1664 } 1665 if (bs->file && bdrv_requests_pending(bs->file)) { 1666 return true; 1667 } 1668 if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) { 1669 return true; 1670 } 1671 return false; 1672 } 1673 1674 static bool bdrv_requests_pending_all(void) 1675 { 1676 BlockDriverState *bs; 1677 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 1678 if (bdrv_requests_pending(bs)) { 1679 return true; 1680 } 1681 } 1682 return false; 1683 } 1684 1685 /* 1686 * Wait for pending requests to complete across all BlockDriverStates 1687 * 1688 * This function does not flush data to disk, use bdrv_flush_all() for that 1689 * after calling this function. 1690 * 1691 * Note that completion of an asynchronous I/O operation can trigger any 1692 * number of other I/O operations on other devices---for example a coroutine 1693 * can be arbitrarily complex and a constant flow of I/O can come until the 1694 * coroutine is complete. Because of this, it is not possible to have a 1695 * function to drain a single device's I/O queue. 1696 */ 1697 void bdrv_drain_all(void) 1698 { 1699 /* Always run first iteration so any pending completion BHs run */ 1700 bool busy = true; 1701 BlockDriverState *bs; 1702 1703 while (busy) { 1704 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 1705 bdrv_start_throttled_reqs(bs); 1706 } 1707 1708 busy = bdrv_requests_pending_all(); 1709 busy |= aio_poll(qemu_get_aio_context(), busy); 1710 } 1711 } 1712 1713 /* make a BlockDriverState anonymous by removing from bdrv_state and 1714 * graph_bdrv_state list. 1715 Also, NULL terminate the device_name to prevent double remove */ 1716 void bdrv_make_anon(BlockDriverState *bs) 1717 { 1718 if (bs->device_name[0] != '\0') { 1719 QTAILQ_REMOVE(&bdrv_states, bs, device_list); 1720 } 1721 bs->device_name[0] = '\0'; 1722 if (bs->node_name[0] != '\0') { 1723 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); 1724 } 1725 bs->node_name[0] = '\0'; 1726 } 1727 1728 static void bdrv_rebind(BlockDriverState *bs) 1729 { 1730 if (bs->drv && bs->drv->bdrv_rebind) { 1731 bs->drv->bdrv_rebind(bs); 1732 } 1733 } 1734 1735 static void bdrv_move_feature_fields(BlockDriverState *bs_dest, 1736 BlockDriverState *bs_src) 1737 { 1738 /* move some fields that need to stay attached to the device */ 1739 bs_dest->open_flags = bs_src->open_flags; 1740 1741 /* dev info */ 1742 bs_dest->dev_ops = bs_src->dev_ops; 1743 bs_dest->dev_opaque = bs_src->dev_opaque; 1744 bs_dest->dev = bs_src->dev; 1745 bs_dest->buffer_alignment = bs_src->buffer_alignment; 1746 bs_dest->copy_on_read = bs_src->copy_on_read; 1747 1748 bs_dest->enable_write_cache = bs_src->enable_write_cache; 1749 1750 /* i/o throttled req */ 1751 memcpy(&bs_dest->throttle_state, 1752 &bs_src->throttle_state, 1753 sizeof(ThrottleState)); 1754 bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0]; 1755 bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1]; 1756 bs_dest->io_limits_enabled = bs_src->io_limits_enabled; 1757 1758 /* r/w error */ 1759 bs_dest->on_read_error = bs_src->on_read_error; 1760 bs_dest->on_write_error = bs_src->on_write_error; 1761 1762 /* i/o status */ 1763 bs_dest->iostatus_enabled = bs_src->iostatus_enabled; 1764 bs_dest->iostatus = bs_src->iostatus; 1765 1766 /* dirty bitmap */ 1767 bs_dest->dirty_bitmaps = bs_src->dirty_bitmaps; 1768 1769 /* reference count */ 1770 bs_dest->refcnt = bs_src->refcnt; 1771 1772 /* job */ 1773 bs_dest->in_use = bs_src->in_use; 1774 bs_dest->job = bs_src->job; 1775 1776 /* keep the same entry in bdrv_states */ 1777 pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name), 1778 bs_src->device_name); 1779 bs_dest->device_list = bs_src->device_list; 1780 1781 /* keep the same entry in graph_bdrv_states 1782 * We do want to swap name but don't want to swap linked list entries 1783 */ 1784 bs_dest->node_list = bs_src->node_list; 1785 } 1786 1787 /* 1788 * Swap bs contents for two image chains while they are live, 1789 * while keeping required fields on the BlockDriverState that is 1790 * actually attached to a device. 1791 * 1792 * This will modify the BlockDriverState fields, and swap contents 1793 * between bs_new and bs_old. Both bs_new and bs_old are modified. 1794 * 1795 * bs_new is required to be anonymous. 1796 * 1797 * This function does not create any image files. 1798 */ 1799 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old) 1800 { 1801 BlockDriverState tmp; 1802 1803 /* bs_new must be anonymous and shouldn't have anything fancy enabled */ 1804 assert(bs_new->device_name[0] == '\0'); 1805 assert(QLIST_EMPTY(&bs_new->dirty_bitmaps)); 1806 assert(bs_new->job == NULL); 1807 assert(bs_new->dev == NULL); 1808 assert(bs_new->in_use == 0); 1809 assert(bs_new->io_limits_enabled == false); 1810 assert(!throttle_have_timer(&bs_new->throttle_state)); 1811 1812 tmp = *bs_new; 1813 *bs_new = *bs_old; 1814 *bs_old = tmp; 1815 1816 /* there are some fields that should not be swapped, move them back */ 1817 bdrv_move_feature_fields(&tmp, bs_old); 1818 bdrv_move_feature_fields(bs_old, bs_new); 1819 bdrv_move_feature_fields(bs_new, &tmp); 1820 1821 /* bs_new shouldn't be in bdrv_states even after the swap! */ 1822 assert(bs_new->device_name[0] == '\0'); 1823 1824 /* Check a few fields that should remain attached to the device */ 1825 assert(bs_new->dev == NULL); 1826 assert(bs_new->job == NULL); 1827 assert(bs_new->in_use == 0); 1828 assert(bs_new->io_limits_enabled == false); 1829 assert(!throttle_have_timer(&bs_new->throttle_state)); 1830 1831 bdrv_rebind(bs_new); 1832 bdrv_rebind(bs_old); 1833 } 1834 1835 /* 1836 * Add new bs contents at the top of an image chain while the chain is 1837 * live, while keeping required fields on the top layer. 1838 * 1839 * This will modify the BlockDriverState fields, and swap contents 1840 * between bs_new and bs_top. Both bs_new and bs_top are modified. 1841 * 1842 * bs_new is required to be anonymous. 1843 * 1844 * This function does not create any image files. 1845 */ 1846 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top) 1847 { 1848 bdrv_swap(bs_new, bs_top); 1849 1850 /* The contents of 'tmp' will become bs_top, as we are 1851 * swapping bs_new and bs_top contents. */ 1852 bs_top->backing_hd = bs_new; 1853 bs_top->open_flags &= ~BDRV_O_NO_BACKING; 1854 pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file), 1855 bs_new->filename); 1856 pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format), 1857 bs_new->drv ? bs_new->drv->format_name : ""); 1858 } 1859 1860 static void bdrv_delete(BlockDriverState *bs) 1861 { 1862 assert(!bs->dev); 1863 assert(!bs->job); 1864 assert(!bs->in_use); 1865 assert(!bs->refcnt); 1866 assert(QLIST_EMPTY(&bs->dirty_bitmaps)); 1867 1868 bdrv_close(bs); 1869 1870 /* remove from list, if necessary */ 1871 bdrv_make_anon(bs); 1872 1873 g_free(bs); 1874 } 1875 1876 int bdrv_attach_dev(BlockDriverState *bs, void *dev) 1877 /* TODO change to DeviceState *dev when all users are qdevified */ 1878 { 1879 if (bs->dev) { 1880 return -EBUSY; 1881 } 1882 bs->dev = dev; 1883 bdrv_iostatus_reset(bs); 1884 return 0; 1885 } 1886 1887 /* TODO qdevified devices don't use this, remove when devices are qdevified */ 1888 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev) 1889 { 1890 if (bdrv_attach_dev(bs, dev) < 0) { 1891 abort(); 1892 } 1893 } 1894 1895 void bdrv_detach_dev(BlockDriverState *bs, void *dev) 1896 /* TODO change to DeviceState *dev when all users are qdevified */ 1897 { 1898 assert(bs->dev == dev); 1899 bs->dev = NULL; 1900 bs->dev_ops = NULL; 1901 bs->dev_opaque = NULL; 1902 bs->buffer_alignment = 512; 1903 } 1904 1905 /* TODO change to return DeviceState * when all users are qdevified */ 1906 void *bdrv_get_attached_dev(BlockDriverState *bs) 1907 { 1908 return bs->dev; 1909 } 1910 1911 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops, 1912 void *opaque) 1913 { 1914 bs->dev_ops = ops; 1915 bs->dev_opaque = opaque; 1916 } 1917 1918 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv, 1919 enum MonitorEvent ev, 1920 BlockErrorAction action, bool is_read) 1921 { 1922 QObject *data; 1923 const char *action_str; 1924 1925 switch (action) { 1926 case BDRV_ACTION_REPORT: 1927 action_str = "report"; 1928 break; 1929 case BDRV_ACTION_IGNORE: 1930 action_str = "ignore"; 1931 break; 1932 case BDRV_ACTION_STOP: 1933 action_str = "stop"; 1934 break; 1935 default: 1936 abort(); 1937 } 1938 1939 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }", 1940 bdrv->device_name, 1941 action_str, 1942 is_read ? "read" : "write"); 1943 monitor_protocol_event(ev, data); 1944 1945 qobject_decref(data); 1946 } 1947 1948 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected) 1949 { 1950 QObject *data; 1951 1952 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }", 1953 bdrv_get_device_name(bs), ejected); 1954 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data); 1955 1956 qobject_decref(data); 1957 } 1958 1959 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load) 1960 { 1961 if (bs->dev_ops && bs->dev_ops->change_media_cb) { 1962 bool tray_was_closed = !bdrv_dev_is_tray_open(bs); 1963 bs->dev_ops->change_media_cb(bs->dev_opaque, load); 1964 if (tray_was_closed) { 1965 /* tray open */ 1966 bdrv_emit_qmp_eject_event(bs, true); 1967 } 1968 if (load) { 1969 /* tray close */ 1970 bdrv_emit_qmp_eject_event(bs, false); 1971 } 1972 } 1973 } 1974 1975 bool bdrv_dev_has_removable_media(BlockDriverState *bs) 1976 { 1977 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb); 1978 } 1979 1980 void bdrv_dev_eject_request(BlockDriverState *bs, bool force) 1981 { 1982 if (bs->dev_ops && bs->dev_ops->eject_request_cb) { 1983 bs->dev_ops->eject_request_cb(bs->dev_opaque, force); 1984 } 1985 } 1986 1987 bool bdrv_dev_is_tray_open(BlockDriverState *bs) 1988 { 1989 if (bs->dev_ops && bs->dev_ops->is_tray_open) { 1990 return bs->dev_ops->is_tray_open(bs->dev_opaque); 1991 } 1992 return false; 1993 } 1994 1995 static void bdrv_dev_resize_cb(BlockDriverState *bs) 1996 { 1997 if (bs->dev_ops && bs->dev_ops->resize_cb) { 1998 bs->dev_ops->resize_cb(bs->dev_opaque); 1999 } 2000 } 2001 2002 bool bdrv_dev_is_medium_locked(BlockDriverState *bs) 2003 { 2004 if (bs->dev_ops && bs->dev_ops->is_medium_locked) { 2005 return bs->dev_ops->is_medium_locked(bs->dev_opaque); 2006 } 2007 return false; 2008 } 2009 2010 /* 2011 * Run consistency checks on an image 2012 * 2013 * Returns 0 if the check could be completed (it doesn't mean that the image is 2014 * free of errors) or -errno when an internal error occurred. The results of the 2015 * check are stored in res. 2016 */ 2017 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix) 2018 { 2019 if (bs->drv->bdrv_check == NULL) { 2020 return -ENOTSUP; 2021 } 2022 2023 memset(res, 0, sizeof(*res)); 2024 return bs->drv->bdrv_check(bs, res, fix); 2025 } 2026 2027 #define COMMIT_BUF_SECTORS 2048 2028 2029 /* commit COW file into the raw image */ 2030 int bdrv_commit(BlockDriverState *bs) 2031 { 2032 BlockDriver *drv = bs->drv; 2033 int64_t sector, total_sectors, length, backing_length; 2034 int n, ro, open_flags; 2035 int ret = 0; 2036 uint8_t *buf = NULL; 2037 char filename[PATH_MAX]; 2038 2039 if (!drv) 2040 return -ENOMEDIUM; 2041 2042 if (!bs->backing_hd) { 2043 return -ENOTSUP; 2044 } 2045 2046 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) { 2047 return -EBUSY; 2048 } 2049 2050 ro = bs->backing_hd->read_only; 2051 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */ 2052 pstrcpy(filename, sizeof(filename), bs->backing_hd->filename); 2053 open_flags = bs->backing_hd->open_flags; 2054 2055 if (ro) { 2056 if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) { 2057 return -EACCES; 2058 } 2059 } 2060 2061 length = bdrv_getlength(bs); 2062 if (length < 0) { 2063 ret = length; 2064 goto ro_cleanup; 2065 } 2066 2067 backing_length = bdrv_getlength(bs->backing_hd); 2068 if (backing_length < 0) { 2069 ret = backing_length; 2070 goto ro_cleanup; 2071 } 2072 2073 /* If our top snapshot is larger than the backing file image, 2074 * grow the backing file image if possible. If not possible, 2075 * we must return an error */ 2076 if (length > backing_length) { 2077 ret = bdrv_truncate(bs->backing_hd, length); 2078 if (ret < 0) { 2079 goto ro_cleanup; 2080 } 2081 } 2082 2083 total_sectors = length >> BDRV_SECTOR_BITS; 2084 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE); 2085 2086 for (sector = 0; sector < total_sectors; sector += n) { 2087 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n); 2088 if (ret < 0) { 2089 goto ro_cleanup; 2090 } 2091 if (ret) { 2092 ret = bdrv_read(bs, sector, buf, n); 2093 if (ret < 0) { 2094 goto ro_cleanup; 2095 } 2096 2097 ret = bdrv_write(bs->backing_hd, sector, buf, n); 2098 if (ret < 0) { 2099 goto ro_cleanup; 2100 } 2101 } 2102 } 2103 2104 if (drv->bdrv_make_empty) { 2105 ret = drv->bdrv_make_empty(bs); 2106 if (ret < 0) { 2107 goto ro_cleanup; 2108 } 2109 bdrv_flush(bs); 2110 } 2111 2112 /* 2113 * Make sure all data we wrote to the backing device is actually 2114 * stable on disk. 2115 */ 2116 if (bs->backing_hd) { 2117 bdrv_flush(bs->backing_hd); 2118 } 2119 2120 ret = 0; 2121 ro_cleanup: 2122 g_free(buf); 2123 2124 if (ro) { 2125 /* ignoring error return here */ 2126 bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL); 2127 } 2128 2129 return ret; 2130 } 2131 2132 int bdrv_commit_all(void) 2133 { 2134 BlockDriverState *bs; 2135 2136 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 2137 if (bs->drv && bs->backing_hd) { 2138 int ret = bdrv_commit(bs); 2139 if (ret < 0) { 2140 return ret; 2141 } 2142 } 2143 } 2144 return 0; 2145 } 2146 2147 /** 2148 * Remove an active request from the tracked requests list 2149 * 2150 * This function should be called when a tracked request is completing. 2151 */ 2152 static void tracked_request_end(BdrvTrackedRequest *req) 2153 { 2154 QLIST_REMOVE(req, list); 2155 qemu_co_queue_restart_all(&req->wait_queue); 2156 } 2157 2158 /** 2159 * Add an active request to the tracked requests list 2160 */ 2161 static void tracked_request_begin(BdrvTrackedRequest *req, 2162 BlockDriverState *bs, 2163 int64_t sector_num, 2164 int nb_sectors, bool is_write) 2165 { 2166 *req = (BdrvTrackedRequest){ 2167 .bs = bs, 2168 .sector_num = sector_num, 2169 .nb_sectors = nb_sectors, 2170 .is_write = is_write, 2171 .co = qemu_coroutine_self(), 2172 }; 2173 2174 qemu_co_queue_init(&req->wait_queue); 2175 2176 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); 2177 } 2178 2179 /** 2180 * Round a region to cluster boundaries 2181 */ 2182 void bdrv_round_to_clusters(BlockDriverState *bs, 2183 int64_t sector_num, int nb_sectors, 2184 int64_t *cluster_sector_num, 2185 int *cluster_nb_sectors) 2186 { 2187 BlockDriverInfo bdi; 2188 2189 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { 2190 *cluster_sector_num = sector_num; 2191 *cluster_nb_sectors = nb_sectors; 2192 } else { 2193 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; 2194 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); 2195 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + 2196 nb_sectors, c); 2197 } 2198 } 2199 2200 static bool tracked_request_overlaps(BdrvTrackedRequest *req, 2201 int64_t sector_num, int nb_sectors) { 2202 /* aaaa bbbb */ 2203 if (sector_num >= req->sector_num + req->nb_sectors) { 2204 return false; 2205 } 2206 /* bbbb aaaa */ 2207 if (req->sector_num >= sector_num + nb_sectors) { 2208 return false; 2209 } 2210 return true; 2211 } 2212 2213 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs, 2214 int64_t sector_num, int nb_sectors) 2215 { 2216 BdrvTrackedRequest *req; 2217 int64_t cluster_sector_num; 2218 int cluster_nb_sectors; 2219 bool retry; 2220 2221 /* If we touch the same cluster it counts as an overlap. This guarantees 2222 * that allocating writes will be serialized and not race with each other 2223 * for the same cluster. For example, in copy-on-read it ensures that the 2224 * CoR read and write operations are atomic and guest writes cannot 2225 * interleave between them. 2226 */ 2227 bdrv_round_to_clusters(bs, sector_num, nb_sectors, 2228 &cluster_sector_num, &cluster_nb_sectors); 2229 2230 do { 2231 retry = false; 2232 QLIST_FOREACH(req, &bs->tracked_requests, list) { 2233 if (tracked_request_overlaps(req, cluster_sector_num, 2234 cluster_nb_sectors)) { 2235 /* Hitting this means there was a reentrant request, for 2236 * example, a block driver issuing nested requests. This must 2237 * never happen since it means deadlock. 2238 */ 2239 assert(qemu_coroutine_self() != req->co); 2240 2241 qemu_co_queue_wait(&req->wait_queue); 2242 retry = true; 2243 break; 2244 } 2245 } 2246 } while (retry); 2247 } 2248 2249 /* 2250 * Return values: 2251 * 0 - success 2252 * -EINVAL - backing format specified, but no file 2253 * -ENOSPC - can't update the backing file because no space is left in the 2254 * image file header 2255 * -ENOTSUP - format driver doesn't support changing the backing file 2256 */ 2257 int bdrv_change_backing_file(BlockDriverState *bs, 2258 const char *backing_file, const char *backing_fmt) 2259 { 2260 BlockDriver *drv = bs->drv; 2261 int ret; 2262 2263 /* Backing file format doesn't make sense without a backing file */ 2264 if (backing_fmt && !backing_file) { 2265 return -EINVAL; 2266 } 2267 2268 if (drv->bdrv_change_backing_file != NULL) { 2269 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); 2270 } else { 2271 ret = -ENOTSUP; 2272 } 2273 2274 if (ret == 0) { 2275 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 2276 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2277 } 2278 return ret; 2279 } 2280 2281 /* 2282 * Finds the image layer in the chain that has 'bs' as its backing file. 2283 * 2284 * active is the current topmost image. 2285 * 2286 * Returns NULL if bs is not found in active's image chain, 2287 * or if active == bs. 2288 */ 2289 BlockDriverState *bdrv_find_overlay(BlockDriverState *active, 2290 BlockDriverState *bs) 2291 { 2292 BlockDriverState *overlay = NULL; 2293 BlockDriverState *intermediate; 2294 2295 assert(active != NULL); 2296 assert(bs != NULL); 2297 2298 /* if bs is the same as active, then by definition it has no overlay 2299 */ 2300 if (active == bs) { 2301 return NULL; 2302 } 2303 2304 intermediate = active; 2305 while (intermediate->backing_hd) { 2306 if (intermediate->backing_hd == bs) { 2307 overlay = intermediate; 2308 break; 2309 } 2310 intermediate = intermediate->backing_hd; 2311 } 2312 2313 return overlay; 2314 } 2315 2316 typedef struct BlkIntermediateStates { 2317 BlockDriverState *bs; 2318 QSIMPLEQ_ENTRY(BlkIntermediateStates) entry; 2319 } BlkIntermediateStates; 2320 2321 2322 /* 2323 * Drops images above 'base' up to and including 'top', and sets the image 2324 * above 'top' to have base as its backing file. 2325 * 2326 * Requires that the overlay to 'top' is opened r/w, so that the backing file 2327 * information in 'bs' can be properly updated. 2328 * 2329 * E.g., this will convert the following chain: 2330 * bottom <- base <- intermediate <- top <- active 2331 * 2332 * to 2333 * 2334 * bottom <- base <- active 2335 * 2336 * It is allowed for bottom==base, in which case it converts: 2337 * 2338 * base <- intermediate <- top <- active 2339 * 2340 * to 2341 * 2342 * base <- active 2343 * 2344 * Error conditions: 2345 * if active == top, that is considered an error 2346 * 2347 */ 2348 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top, 2349 BlockDriverState *base) 2350 { 2351 BlockDriverState *intermediate; 2352 BlockDriverState *base_bs = NULL; 2353 BlockDriverState *new_top_bs = NULL; 2354 BlkIntermediateStates *intermediate_state, *next; 2355 int ret = -EIO; 2356 2357 QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete; 2358 QSIMPLEQ_INIT(&states_to_delete); 2359 2360 if (!top->drv || !base->drv) { 2361 goto exit; 2362 } 2363 2364 new_top_bs = bdrv_find_overlay(active, top); 2365 2366 if (new_top_bs == NULL) { 2367 /* we could not find the image above 'top', this is an error */ 2368 goto exit; 2369 } 2370 2371 /* special case of new_top_bs->backing_hd already pointing to base - nothing 2372 * to do, no intermediate images */ 2373 if (new_top_bs->backing_hd == base) { 2374 ret = 0; 2375 goto exit; 2376 } 2377 2378 intermediate = top; 2379 2380 /* now we will go down through the list, and add each BDS we find 2381 * into our deletion queue, until we hit the 'base' 2382 */ 2383 while (intermediate) { 2384 intermediate_state = g_malloc0(sizeof(BlkIntermediateStates)); 2385 intermediate_state->bs = intermediate; 2386 QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry); 2387 2388 if (intermediate->backing_hd == base) { 2389 base_bs = intermediate->backing_hd; 2390 break; 2391 } 2392 intermediate = intermediate->backing_hd; 2393 } 2394 if (base_bs == NULL) { 2395 /* something went wrong, we did not end at the base. safely 2396 * unravel everything, and exit with error */ 2397 goto exit; 2398 } 2399 2400 /* success - we can delete the intermediate states, and link top->base */ 2401 ret = bdrv_change_backing_file(new_top_bs, base_bs->filename, 2402 base_bs->drv ? base_bs->drv->format_name : ""); 2403 if (ret) { 2404 goto exit; 2405 } 2406 new_top_bs->backing_hd = base_bs; 2407 2408 2409 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) { 2410 /* so that bdrv_close() does not recursively close the chain */ 2411 intermediate_state->bs->backing_hd = NULL; 2412 bdrv_unref(intermediate_state->bs); 2413 } 2414 ret = 0; 2415 2416 exit: 2417 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) { 2418 g_free(intermediate_state); 2419 } 2420 return ret; 2421 } 2422 2423 2424 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, 2425 size_t size) 2426 { 2427 int64_t len; 2428 2429 if (!bdrv_is_inserted(bs)) 2430 return -ENOMEDIUM; 2431 2432 if (bs->growable) 2433 return 0; 2434 2435 len = bdrv_getlength(bs); 2436 2437 if (offset < 0) 2438 return -EIO; 2439 2440 if ((offset > len) || (len - offset < size)) 2441 return -EIO; 2442 2443 return 0; 2444 } 2445 2446 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num, 2447 int nb_sectors) 2448 { 2449 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE, 2450 nb_sectors * BDRV_SECTOR_SIZE); 2451 } 2452 2453 typedef struct RwCo { 2454 BlockDriverState *bs; 2455 int64_t sector_num; 2456 int nb_sectors; 2457 QEMUIOVector *qiov; 2458 bool is_write; 2459 int ret; 2460 BdrvRequestFlags flags; 2461 } RwCo; 2462 2463 static void coroutine_fn bdrv_rw_co_entry(void *opaque) 2464 { 2465 RwCo *rwco = opaque; 2466 2467 if (!rwco->is_write) { 2468 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num, 2469 rwco->nb_sectors, rwco->qiov, 2470 rwco->flags); 2471 } else { 2472 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num, 2473 rwco->nb_sectors, rwco->qiov, 2474 rwco->flags); 2475 } 2476 } 2477 2478 /* 2479 * Process a vectored synchronous request using coroutines 2480 */ 2481 static int bdrv_rwv_co(BlockDriverState *bs, int64_t sector_num, 2482 QEMUIOVector *qiov, bool is_write, 2483 BdrvRequestFlags flags) 2484 { 2485 Coroutine *co; 2486 RwCo rwco = { 2487 .bs = bs, 2488 .sector_num = sector_num, 2489 .nb_sectors = qiov->size >> BDRV_SECTOR_BITS, 2490 .qiov = qiov, 2491 .is_write = is_write, 2492 .ret = NOT_DONE, 2493 .flags = flags, 2494 }; 2495 assert((qiov->size & (BDRV_SECTOR_SIZE - 1)) == 0); 2496 2497 /** 2498 * In sync call context, when the vcpu is blocked, this throttling timer 2499 * will not fire; so the I/O throttling function has to be disabled here 2500 * if it has been enabled. 2501 */ 2502 if (bs->io_limits_enabled) { 2503 fprintf(stderr, "Disabling I/O throttling on '%s' due " 2504 "to synchronous I/O.\n", bdrv_get_device_name(bs)); 2505 bdrv_io_limits_disable(bs); 2506 } 2507 2508 if (qemu_in_coroutine()) { 2509 /* Fast-path if already in coroutine context */ 2510 bdrv_rw_co_entry(&rwco); 2511 } else { 2512 co = qemu_coroutine_create(bdrv_rw_co_entry); 2513 qemu_coroutine_enter(co, &rwco); 2514 while (rwco.ret == NOT_DONE) { 2515 qemu_aio_wait(); 2516 } 2517 } 2518 return rwco.ret; 2519 } 2520 2521 /* 2522 * Process a synchronous request using coroutines 2523 */ 2524 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf, 2525 int nb_sectors, bool is_write, BdrvRequestFlags flags) 2526 { 2527 QEMUIOVector qiov; 2528 struct iovec iov = { 2529 .iov_base = (void *)buf, 2530 .iov_len = nb_sectors * BDRV_SECTOR_SIZE, 2531 }; 2532 2533 qemu_iovec_init_external(&qiov, &iov, 1); 2534 return bdrv_rwv_co(bs, sector_num, &qiov, is_write, flags); 2535 } 2536 2537 /* return < 0 if error. See bdrv_write() for the return codes */ 2538 int bdrv_read(BlockDriverState *bs, int64_t sector_num, 2539 uint8_t *buf, int nb_sectors) 2540 { 2541 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0); 2542 } 2543 2544 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */ 2545 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num, 2546 uint8_t *buf, int nb_sectors) 2547 { 2548 bool enabled; 2549 int ret; 2550 2551 enabled = bs->io_limits_enabled; 2552 bs->io_limits_enabled = false; 2553 ret = bdrv_read(bs, sector_num, buf, nb_sectors); 2554 bs->io_limits_enabled = enabled; 2555 return ret; 2556 } 2557 2558 /* Return < 0 if error. Important errors are: 2559 -EIO generic I/O error (may happen for all errors) 2560 -ENOMEDIUM No media inserted. 2561 -EINVAL Invalid sector number or nb_sectors 2562 -EACCES Trying to write a read-only device 2563 */ 2564 int bdrv_write(BlockDriverState *bs, int64_t sector_num, 2565 const uint8_t *buf, int nb_sectors) 2566 { 2567 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0); 2568 } 2569 2570 int bdrv_writev(BlockDriverState *bs, int64_t sector_num, QEMUIOVector *qiov) 2571 { 2572 return bdrv_rwv_co(bs, sector_num, qiov, true, 0); 2573 } 2574 2575 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num, 2576 int nb_sectors, BdrvRequestFlags flags) 2577 { 2578 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true, 2579 BDRV_REQ_ZERO_WRITE | flags); 2580 } 2581 2582 /* 2583 * Completely zero out a block device with the help of bdrv_write_zeroes. 2584 * The operation is sped up by checking the block status and only writing 2585 * zeroes to the device if they currently do not return zeroes. Optional 2586 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP). 2587 * 2588 * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). 2589 */ 2590 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags) 2591 { 2592 int64_t target_size = bdrv_getlength(bs) / BDRV_SECTOR_SIZE; 2593 int64_t ret, nb_sectors, sector_num = 0; 2594 int n; 2595 2596 for (;;) { 2597 nb_sectors = target_size - sector_num; 2598 if (nb_sectors <= 0) { 2599 return 0; 2600 } 2601 if (nb_sectors > INT_MAX) { 2602 nb_sectors = INT_MAX; 2603 } 2604 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n); 2605 if (ret < 0) { 2606 error_report("error getting block status at sector %" PRId64 ": %s", 2607 sector_num, strerror(-ret)); 2608 return ret; 2609 } 2610 if (ret & BDRV_BLOCK_ZERO) { 2611 sector_num += n; 2612 continue; 2613 } 2614 ret = bdrv_write_zeroes(bs, sector_num, n, flags); 2615 if (ret < 0) { 2616 error_report("error writing zeroes at sector %" PRId64 ": %s", 2617 sector_num, strerror(-ret)); 2618 return ret; 2619 } 2620 sector_num += n; 2621 } 2622 } 2623 2624 int bdrv_pread(BlockDriverState *bs, int64_t offset, 2625 void *buf, int count1) 2626 { 2627 uint8_t tmp_buf[BDRV_SECTOR_SIZE]; 2628 int len, nb_sectors, count; 2629 int64_t sector_num; 2630 int ret; 2631 2632 count = count1; 2633 /* first read to align to sector start */ 2634 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1); 2635 if (len > count) 2636 len = count; 2637 sector_num = offset >> BDRV_SECTOR_BITS; 2638 if (len > 0) { 2639 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) 2640 return ret; 2641 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len); 2642 count -= len; 2643 if (count == 0) 2644 return count1; 2645 sector_num++; 2646 buf += len; 2647 } 2648 2649 /* read the sectors "in place" */ 2650 nb_sectors = count >> BDRV_SECTOR_BITS; 2651 if (nb_sectors > 0) { 2652 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0) 2653 return ret; 2654 sector_num += nb_sectors; 2655 len = nb_sectors << BDRV_SECTOR_BITS; 2656 buf += len; 2657 count -= len; 2658 } 2659 2660 /* add data from the last sector */ 2661 if (count > 0) { 2662 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) 2663 return ret; 2664 memcpy(buf, tmp_buf, count); 2665 } 2666 return count1; 2667 } 2668 2669 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov) 2670 { 2671 uint8_t tmp_buf[BDRV_SECTOR_SIZE]; 2672 int len, nb_sectors, count; 2673 int64_t sector_num; 2674 int ret; 2675 2676 count = qiov->size; 2677 2678 /* first write to align to sector start */ 2679 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1); 2680 if (len > count) 2681 len = count; 2682 sector_num = offset >> BDRV_SECTOR_BITS; 2683 if (len > 0) { 2684 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) 2685 return ret; 2686 qemu_iovec_to_buf(qiov, 0, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), 2687 len); 2688 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) 2689 return ret; 2690 count -= len; 2691 if (count == 0) 2692 return qiov->size; 2693 sector_num++; 2694 } 2695 2696 /* write the sectors "in place" */ 2697 nb_sectors = count >> BDRV_SECTOR_BITS; 2698 if (nb_sectors > 0) { 2699 QEMUIOVector qiov_inplace; 2700 2701 qemu_iovec_init(&qiov_inplace, qiov->niov); 2702 qemu_iovec_concat(&qiov_inplace, qiov, len, 2703 nb_sectors << BDRV_SECTOR_BITS); 2704 ret = bdrv_writev(bs, sector_num, &qiov_inplace); 2705 qemu_iovec_destroy(&qiov_inplace); 2706 if (ret < 0) { 2707 return ret; 2708 } 2709 2710 sector_num += nb_sectors; 2711 len = nb_sectors << BDRV_SECTOR_BITS; 2712 count -= len; 2713 } 2714 2715 /* add data from the last sector */ 2716 if (count > 0) { 2717 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) 2718 return ret; 2719 qemu_iovec_to_buf(qiov, qiov->size - count, tmp_buf, count); 2720 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) 2721 return ret; 2722 } 2723 return qiov->size; 2724 } 2725 2726 int bdrv_pwrite(BlockDriverState *bs, int64_t offset, 2727 const void *buf, int count1) 2728 { 2729 QEMUIOVector qiov; 2730 struct iovec iov = { 2731 .iov_base = (void *) buf, 2732 .iov_len = count1, 2733 }; 2734 2735 qemu_iovec_init_external(&qiov, &iov, 1); 2736 return bdrv_pwritev(bs, offset, &qiov); 2737 } 2738 2739 /* 2740 * Writes to the file and ensures that no writes are reordered across this 2741 * request (acts as a barrier) 2742 * 2743 * Returns 0 on success, -errno in error cases. 2744 */ 2745 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset, 2746 const void *buf, int count) 2747 { 2748 int ret; 2749 2750 ret = bdrv_pwrite(bs, offset, buf, count); 2751 if (ret < 0) { 2752 return ret; 2753 } 2754 2755 /* No flush needed for cache modes that already do it */ 2756 if (bs->enable_write_cache) { 2757 bdrv_flush(bs); 2758 } 2759 2760 return 0; 2761 } 2762 2763 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, 2764 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) 2765 { 2766 /* Perform I/O through a temporary buffer so that users who scribble over 2767 * their read buffer while the operation is in progress do not end up 2768 * modifying the image file. This is critical for zero-copy guest I/O 2769 * where anything might happen inside guest memory. 2770 */ 2771 void *bounce_buffer; 2772 2773 BlockDriver *drv = bs->drv; 2774 struct iovec iov; 2775 QEMUIOVector bounce_qiov; 2776 int64_t cluster_sector_num; 2777 int cluster_nb_sectors; 2778 size_t skip_bytes; 2779 int ret; 2780 2781 /* Cover entire cluster so no additional backing file I/O is required when 2782 * allocating cluster in the image file. 2783 */ 2784 bdrv_round_to_clusters(bs, sector_num, nb_sectors, 2785 &cluster_sector_num, &cluster_nb_sectors); 2786 2787 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, 2788 cluster_sector_num, cluster_nb_sectors); 2789 2790 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE; 2791 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len); 2792 qemu_iovec_init_external(&bounce_qiov, &iov, 1); 2793 2794 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors, 2795 &bounce_qiov); 2796 if (ret < 0) { 2797 goto err; 2798 } 2799 2800 if (drv->bdrv_co_write_zeroes && 2801 buffer_is_zero(bounce_buffer, iov.iov_len)) { 2802 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num, 2803 cluster_nb_sectors, 0); 2804 } else { 2805 /* This does not change the data on the disk, it is not necessary 2806 * to flush even in cache=writethrough mode. 2807 */ 2808 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors, 2809 &bounce_qiov); 2810 } 2811 2812 if (ret < 0) { 2813 /* It might be okay to ignore write errors for guest requests. If this 2814 * is a deliberate copy-on-read then we don't want to ignore the error. 2815 * Simply report it in all cases. 2816 */ 2817 goto err; 2818 } 2819 2820 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE; 2821 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, 2822 nb_sectors * BDRV_SECTOR_SIZE); 2823 2824 err: 2825 qemu_vfree(bounce_buffer); 2826 return ret; 2827 } 2828 2829 /* 2830 * Handle a read request in coroutine context 2831 */ 2832 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, 2833 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, 2834 BdrvRequestFlags flags) 2835 { 2836 BlockDriver *drv = bs->drv; 2837 BdrvTrackedRequest req; 2838 int ret; 2839 2840 if (!drv) { 2841 return -ENOMEDIUM; 2842 } 2843 if (bdrv_check_request(bs, sector_num, nb_sectors)) { 2844 return -EIO; 2845 } 2846 2847 if (bs->copy_on_read) { 2848 flags |= BDRV_REQ_COPY_ON_READ; 2849 } 2850 if (flags & BDRV_REQ_COPY_ON_READ) { 2851 bs->copy_on_read_in_flight++; 2852 } 2853 2854 if (bs->copy_on_read_in_flight) { 2855 wait_for_overlapping_requests(bs, sector_num, nb_sectors); 2856 } 2857 2858 /* throttling disk I/O */ 2859 if (bs->io_limits_enabled) { 2860 bdrv_io_limits_intercept(bs, nb_sectors, false); 2861 } 2862 2863 tracked_request_begin(&req, bs, sector_num, nb_sectors, false); 2864 2865 if (flags & BDRV_REQ_COPY_ON_READ) { 2866 int pnum; 2867 2868 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum); 2869 if (ret < 0) { 2870 goto out; 2871 } 2872 2873 if (!ret || pnum != nb_sectors) { 2874 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov); 2875 goto out; 2876 } 2877 } 2878 2879 if (!(bs->zero_beyond_eof && bs->growable)) { 2880 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); 2881 } else { 2882 /* Read zeros after EOF of growable BDSes */ 2883 int64_t len, total_sectors, max_nb_sectors; 2884 2885 len = bdrv_getlength(bs); 2886 if (len < 0) { 2887 ret = len; 2888 goto out; 2889 } 2890 2891 total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE); 2892 max_nb_sectors = MAX(0, total_sectors - sector_num); 2893 if (max_nb_sectors > 0) { 2894 ret = drv->bdrv_co_readv(bs, sector_num, 2895 MIN(nb_sectors, max_nb_sectors), qiov); 2896 } else { 2897 ret = 0; 2898 } 2899 2900 /* Reading beyond end of file is supposed to produce zeroes */ 2901 if (ret == 0 && total_sectors < sector_num + nb_sectors) { 2902 uint64_t offset = MAX(0, total_sectors - sector_num); 2903 uint64_t bytes = (sector_num + nb_sectors - offset) * 2904 BDRV_SECTOR_SIZE; 2905 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes); 2906 } 2907 } 2908 2909 out: 2910 tracked_request_end(&req); 2911 2912 if (flags & BDRV_REQ_COPY_ON_READ) { 2913 bs->copy_on_read_in_flight--; 2914 } 2915 2916 return ret; 2917 } 2918 2919 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num, 2920 int nb_sectors, QEMUIOVector *qiov) 2921 { 2922 trace_bdrv_co_readv(bs, sector_num, nb_sectors); 2923 2924 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0); 2925 } 2926 2927 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs, 2928 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) 2929 { 2930 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors); 2931 2932 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 2933 BDRV_REQ_COPY_ON_READ); 2934 } 2935 2936 /* if no limit is specified in the BlockLimits use a default 2937 * of 32768 512-byte sectors (16 MiB) per request. 2938 */ 2939 #define MAX_WRITE_ZEROES_DEFAULT 32768 2940 2941 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, 2942 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags) 2943 { 2944 BlockDriver *drv = bs->drv; 2945 QEMUIOVector qiov; 2946 struct iovec iov = {0}; 2947 int ret = 0; 2948 2949 int max_write_zeroes = bs->bl.max_write_zeroes ? 2950 bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT; 2951 2952 while (nb_sectors > 0 && !ret) { 2953 int num = nb_sectors; 2954 2955 /* Align request. Block drivers can expect the "bulk" of the request 2956 * to be aligned. 2957 */ 2958 if (bs->bl.write_zeroes_alignment 2959 && num > bs->bl.write_zeroes_alignment) { 2960 if (sector_num % bs->bl.write_zeroes_alignment != 0) { 2961 /* Make a small request up to the first aligned sector. */ 2962 num = bs->bl.write_zeroes_alignment; 2963 num -= sector_num % bs->bl.write_zeroes_alignment; 2964 } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) { 2965 /* Shorten the request to the last aligned sector. num cannot 2966 * underflow because num > bs->bl.write_zeroes_alignment. 2967 */ 2968 num -= (sector_num + num) % bs->bl.write_zeroes_alignment; 2969 } 2970 } 2971 2972 /* limit request size */ 2973 if (num > max_write_zeroes) { 2974 num = max_write_zeroes; 2975 } 2976 2977 ret = -ENOTSUP; 2978 /* First try the efficient write zeroes operation */ 2979 if (drv->bdrv_co_write_zeroes) { 2980 ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags); 2981 } 2982 2983 if (ret == -ENOTSUP) { 2984 /* Fall back to bounce buffer if write zeroes is unsupported */ 2985 iov.iov_len = num * BDRV_SECTOR_SIZE; 2986 if (iov.iov_base == NULL) { 2987 iov.iov_base = qemu_blockalign(bs, num * BDRV_SECTOR_SIZE); 2988 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE); 2989 } 2990 qemu_iovec_init_external(&qiov, &iov, 1); 2991 2992 ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov); 2993 2994 /* Keep bounce buffer around if it is big enough for all 2995 * all future requests. 2996 */ 2997 if (num < max_write_zeroes) { 2998 qemu_vfree(iov.iov_base); 2999 iov.iov_base = NULL; 3000 } 3001 } 3002 3003 sector_num += num; 3004 nb_sectors -= num; 3005 } 3006 3007 qemu_vfree(iov.iov_base); 3008 return ret; 3009 } 3010 3011 /* 3012 * Handle a write request in coroutine context 3013 */ 3014 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, 3015 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, 3016 BdrvRequestFlags flags) 3017 { 3018 BlockDriver *drv = bs->drv; 3019 BdrvTrackedRequest req; 3020 int ret; 3021 3022 if (!bs->drv) { 3023 return -ENOMEDIUM; 3024 } 3025 if (bs->read_only) { 3026 return -EACCES; 3027 } 3028 if (bdrv_check_request(bs, sector_num, nb_sectors)) { 3029 return -EIO; 3030 } 3031 3032 if (bs->copy_on_read_in_flight) { 3033 wait_for_overlapping_requests(bs, sector_num, nb_sectors); 3034 } 3035 3036 /* throttling disk I/O */ 3037 if (bs->io_limits_enabled) { 3038 bdrv_io_limits_intercept(bs, nb_sectors, true); 3039 } 3040 3041 tracked_request_begin(&req, bs, sector_num, nb_sectors, true); 3042 3043 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req); 3044 3045 if (ret < 0) { 3046 /* Do nothing, write notifier decided to fail this request */ 3047 } else if (flags & BDRV_REQ_ZERO_WRITE) { 3048 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags); 3049 } else { 3050 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); 3051 } 3052 3053 if (ret == 0 && !bs->enable_write_cache) { 3054 ret = bdrv_co_flush(bs); 3055 } 3056 3057 bdrv_set_dirty(bs, sector_num, nb_sectors); 3058 3059 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) { 3060 bs->wr_highest_sector = sector_num + nb_sectors - 1; 3061 } 3062 if (bs->growable && ret >= 0) { 3063 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors); 3064 } 3065 3066 tracked_request_end(&req); 3067 3068 return ret; 3069 } 3070 3071 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num, 3072 int nb_sectors, QEMUIOVector *qiov) 3073 { 3074 trace_bdrv_co_writev(bs, sector_num, nb_sectors); 3075 3076 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0); 3077 } 3078 3079 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs, 3080 int64_t sector_num, int nb_sectors, 3081 BdrvRequestFlags flags) 3082 { 3083 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags); 3084 3085 if (!(bs->open_flags & BDRV_O_UNMAP)) { 3086 flags &= ~BDRV_REQ_MAY_UNMAP; 3087 } 3088 3089 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL, 3090 BDRV_REQ_ZERO_WRITE | flags); 3091 } 3092 3093 /** 3094 * Truncate file to 'offset' bytes (needed only for file protocols) 3095 */ 3096 int bdrv_truncate(BlockDriverState *bs, int64_t offset) 3097 { 3098 BlockDriver *drv = bs->drv; 3099 int ret; 3100 if (!drv) 3101 return -ENOMEDIUM; 3102 if (!drv->bdrv_truncate) 3103 return -ENOTSUP; 3104 if (bs->read_only) 3105 return -EACCES; 3106 if (bdrv_in_use(bs)) 3107 return -EBUSY; 3108 ret = drv->bdrv_truncate(bs, offset); 3109 if (ret == 0) { 3110 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); 3111 bdrv_dev_resize_cb(bs); 3112 } 3113 return ret; 3114 } 3115 3116 /** 3117 * Length of a allocated file in bytes. Sparse files are counted by actual 3118 * allocated space. Return < 0 if error or unknown. 3119 */ 3120 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) 3121 { 3122 BlockDriver *drv = bs->drv; 3123 if (!drv) { 3124 return -ENOMEDIUM; 3125 } 3126 if (drv->bdrv_get_allocated_file_size) { 3127 return drv->bdrv_get_allocated_file_size(bs); 3128 } 3129 if (bs->file) { 3130 return bdrv_get_allocated_file_size(bs->file); 3131 } 3132 return -ENOTSUP; 3133 } 3134 3135 /** 3136 * Length of a file in bytes. Return < 0 if error or unknown. 3137 */ 3138 int64_t bdrv_getlength(BlockDriverState *bs) 3139 { 3140 BlockDriver *drv = bs->drv; 3141 if (!drv) 3142 return -ENOMEDIUM; 3143 3144 if (drv->has_variable_length) { 3145 int ret = refresh_total_sectors(bs, bs->total_sectors); 3146 if (ret < 0) { 3147 return ret; 3148 } 3149 } 3150 return bs->total_sectors * BDRV_SECTOR_SIZE; 3151 } 3152 3153 /* return 0 as number of sectors if no device present or error */ 3154 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) 3155 { 3156 int64_t length; 3157 length = bdrv_getlength(bs); 3158 if (length < 0) 3159 length = 0; 3160 else 3161 length = length >> BDRV_SECTOR_BITS; 3162 *nb_sectors_ptr = length; 3163 } 3164 3165 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error, 3166 BlockdevOnError on_write_error) 3167 { 3168 bs->on_read_error = on_read_error; 3169 bs->on_write_error = on_write_error; 3170 } 3171 3172 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read) 3173 { 3174 return is_read ? bs->on_read_error : bs->on_write_error; 3175 } 3176 3177 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error) 3178 { 3179 BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error; 3180 3181 switch (on_err) { 3182 case BLOCKDEV_ON_ERROR_ENOSPC: 3183 return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT; 3184 case BLOCKDEV_ON_ERROR_STOP: 3185 return BDRV_ACTION_STOP; 3186 case BLOCKDEV_ON_ERROR_REPORT: 3187 return BDRV_ACTION_REPORT; 3188 case BLOCKDEV_ON_ERROR_IGNORE: 3189 return BDRV_ACTION_IGNORE; 3190 default: 3191 abort(); 3192 } 3193 } 3194 3195 /* This is done by device models because, while the block layer knows 3196 * about the error, it does not know whether an operation comes from 3197 * the device or the block layer (from a job, for example). 3198 */ 3199 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action, 3200 bool is_read, int error) 3201 { 3202 assert(error >= 0); 3203 bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read); 3204 if (action == BDRV_ACTION_STOP) { 3205 vm_stop(RUN_STATE_IO_ERROR); 3206 bdrv_iostatus_set_err(bs, error); 3207 } 3208 } 3209 3210 int bdrv_is_read_only(BlockDriverState *bs) 3211 { 3212 return bs->read_only; 3213 } 3214 3215 int bdrv_is_sg(BlockDriverState *bs) 3216 { 3217 return bs->sg; 3218 } 3219 3220 int bdrv_enable_write_cache(BlockDriverState *bs) 3221 { 3222 return bs->enable_write_cache; 3223 } 3224 3225 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce) 3226 { 3227 bs->enable_write_cache = wce; 3228 3229 /* so a reopen() will preserve wce */ 3230 if (wce) { 3231 bs->open_flags |= BDRV_O_CACHE_WB; 3232 } else { 3233 bs->open_flags &= ~BDRV_O_CACHE_WB; 3234 } 3235 } 3236 3237 int bdrv_is_encrypted(BlockDriverState *bs) 3238 { 3239 if (bs->backing_hd && bs->backing_hd->encrypted) 3240 return 1; 3241 return bs->encrypted; 3242 } 3243 3244 int bdrv_key_required(BlockDriverState *bs) 3245 { 3246 BlockDriverState *backing_hd = bs->backing_hd; 3247 3248 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key) 3249 return 1; 3250 return (bs->encrypted && !bs->valid_key); 3251 } 3252 3253 int bdrv_set_key(BlockDriverState *bs, const char *key) 3254 { 3255 int ret; 3256 if (bs->backing_hd && bs->backing_hd->encrypted) { 3257 ret = bdrv_set_key(bs->backing_hd, key); 3258 if (ret < 0) 3259 return ret; 3260 if (!bs->encrypted) 3261 return 0; 3262 } 3263 if (!bs->encrypted) { 3264 return -EINVAL; 3265 } else if (!bs->drv || !bs->drv->bdrv_set_key) { 3266 return -ENOMEDIUM; 3267 } 3268 ret = bs->drv->bdrv_set_key(bs, key); 3269 if (ret < 0) { 3270 bs->valid_key = 0; 3271 } else if (!bs->valid_key) { 3272 bs->valid_key = 1; 3273 /* call the change callback now, we skipped it on open */ 3274 bdrv_dev_change_media_cb(bs, true); 3275 } 3276 return ret; 3277 } 3278 3279 const char *bdrv_get_format_name(BlockDriverState *bs) 3280 { 3281 return bs->drv ? bs->drv->format_name : NULL; 3282 } 3283 3284 void bdrv_iterate_format(void (*it)(void *opaque, const char *name), 3285 void *opaque) 3286 { 3287 BlockDriver *drv; 3288 3289 QLIST_FOREACH(drv, &bdrv_drivers, list) { 3290 it(opaque, drv->format_name); 3291 } 3292 } 3293 3294 /* This function is to find block backend bs */ 3295 BlockDriverState *bdrv_find(const char *name) 3296 { 3297 BlockDriverState *bs; 3298 3299 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 3300 if (!strcmp(name, bs->device_name)) { 3301 return bs; 3302 } 3303 } 3304 return NULL; 3305 } 3306 3307 /* This function is to find a node in the bs graph */ 3308 BlockDriverState *bdrv_find_node(const char *node_name) 3309 { 3310 BlockDriverState *bs; 3311 3312 assert(node_name); 3313 3314 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { 3315 if (!strcmp(node_name, bs->node_name)) { 3316 return bs; 3317 } 3318 } 3319 return NULL; 3320 } 3321 3322 /* Put this QMP function here so it can access the static graph_bdrv_states. */ 3323 BlockDeviceInfoList *bdrv_named_nodes_list(void) 3324 { 3325 BlockDeviceInfoList *list, *entry; 3326 BlockDriverState *bs; 3327 3328 list = NULL; 3329 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { 3330 entry = g_malloc0(sizeof(*entry)); 3331 entry->value = bdrv_block_device_info(bs); 3332 entry->next = list; 3333 list = entry; 3334 } 3335 3336 return list; 3337 } 3338 3339 BlockDriverState *bdrv_lookup_bs(const char *device, 3340 const char *node_name, 3341 Error **errp) 3342 { 3343 BlockDriverState *bs = NULL; 3344 3345 if ((!device && !node_name) || (device && node_name)) { 3346 error_setg(errp, "Use either device or node-name but not both"); 3347 return NULL; 3348 } 3349 3350 if (device) { 3351 bs = bdrv_find(device); 3352 3353 if (!bs) { 3354 error_set(errp, QERR_DEVICE_NOT_FOUND, device); 3355 return NULL; 3356 } 3357 3358 return bs; 3359 } 3360 3361 bs = bdrv_find_node(node_name); 3362 3363 if (!bs) { 3364 error_set(errp, QERR_DEVICE_NOT_FOUND, node_name); 3365 return NULL; 3366 } 3367 3368 return bs; 3369 } 3370 3371 BlockDriverState *bdrv_next(BlockDriverState *bs) 3372 { 3373 if (!bs) { 3374 return QTAILQ_FIRST(&bdrv_states); 3375 } 3376 return QTAILQ_NEXT(bs, device_list); 3377 } 3378 3379 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque) 3380 { 3381 BlockDriverState *bs; 3382 3383 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 3384 it(opaque, bs); 3385 } 3386 } 3387 3388 const char *bdrv_get_device_name(BlockDriverState *bs) 3389 { 3390 return bs->device_name; 3391 } 3392 3393 int bdrv_get_flags(BlockDriverState *bs) 3394 { 3395 return bs->open_flags; 3396 } 3397 3398 int bdrv_flush_all(void) 3399 { 3400 BlockDriverState *bs; 3401 int result = 0; 3402 3403 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 3404 int ret = bdrv_flush(bs); 3405 if (ret < 0 && !result) { 3406 result = ret; 3407 } 3408 } 3409 3410 return result; 3411 } 3412 3413 int bdrv_has_zero_init_1(BlockDriverState *bs) 3414 { 3415 return 1; 3416 } 3417 3418 int bdrv_has_zero_init(BlockDriverState *bs) 3419 { 3420 assert(bs->drv); 3421 3422 /* If BS is a copy on write image, it is initialized to 3423 the contents of the base image, which may not be zeroes. */ 3424 if (bs->backing_hd) { 3425 return 0; 3426 } 3427 if (bs->drv->bdrv_has_zero_init) { 3428 return bs->drv->bdrv_has_zero_init(bs); 3429 } 3430 3431 /* safe default */ 3432 return 0; 3433 } 3434 3435 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs) 3436 { 3437 BlockDriverInfo bdi; 3438 3439 if (bs->backing_hd) { 3440 return false; 3441 } 3442 3443 if (bdrv_get_info(bs, &bdi) == 0) { 3444 return bdi.unallocated_blocks_are_zero; 3445 } 3446 3447 return false; 3448 } 3449 3450 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) 3451 { 3452 BlockDriverInfo bdi; 3453 3454 if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) { 3455 return false; 3456 } 3457 3458 if (bdrv_get_info(bs, &bdi) == 0) { 3459 return bdi.can_write_zeroes_with_unmap; 3460 } 3461 3462 return false; 3463 } 3464 3465 typedef struct BdrvCoGetBlockStatusData { 3466 BlockDriverState *bs; 3467 BlockDriverState *base; 3468 int64_t sector_num; 3469 int nb_sectors; 3470 int *pnum; 3471 int64_t ret; 3472 bool done; 3473 } BdrvCoGetBlockStatusData; 3474 3475 /* 3476 * Returns true iff the specified sector is present in the disk image. Drivers 3477 * not implementing the functionality are assumed to not support backing files, 3478 * hence all their sectors are reported as allocated. 3479 * 3480 * If 'sector_num' is beyond the end of the disk image the return value is 0 3481 * and 'pnum' is set to 0. 3482 * 3483 * 'pnum' is set to the number of sectors (including and immediately following 3484 * the specified sector) that are known to be in the same 3485 * allocated/unallocated state. 3486 * 3487 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes 3488 * beyond the end of the disk image it will be clamped. 3489 */ 3490 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, 3491 int64_t sector_num, 3492 int nb_sectors, int *pnum) 3493 { 3494 int64_t length; 3495 int64_t n; 3496 int64_t ret, ret2; 3497 3498 length = bdrv_getlength(bs); 3499 if (length < 0) { 3500 return length; 3501 } 3502 3503 if (sector_num >= (length >> BDRV_SECTOR_BITS)) { 3504 *pnum = 0; 3505 return 0; 3506 } 3507 3508 n = bs->total_sectors - sector_num; 3509 if (n < nb_sectors) { 3510 nb_sectors = n; 3511 } 3512 3513 if (!bs->drv->bdrv_co_get_block_status) { 3514 *pnum = nb_sectors; 3515 ret = BDRV_BLOCK_DATA; 3516 if (bs->drv->protocol_name) { 3517 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE); 3518 } 3519 return ret; 3520 } 3521 3522 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum); 3523 if (ret < 0) { 3524 *pnum = 0; 3525 return ret; 3526 } 3527 3528 if (ret & BDRV_BLOCK_RAW) { 3529 assert(ret & BDRV_BLOCK_OFFSET_VALID); 3530 return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS, 3531 *pnum, pnum); 3532 } 3533 3534 if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) { 3535 if (bdrv_unallocated_blocks_are_zero(bs)) { 3536 ret |= BDRV_BLOCK_ZERO; 3537 } else if (bs->backing_hd) { 3538 BlockDriverState *bs2 = bs->backing_hd; 3539 int64_t length2 = bdrv_getlength(bs2); 3540 if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) { 3541 ret |= BDRV_BLOCK_ZERO; 3542 } 3543 } 3544 } 3545 3546 if (bs->file && 3547 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && 3548 (ret & BDRV_BLOCK_OFFSET_VALID)) { 3549 ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS, 3550 *pnum, pnum); 3551 if (ret2 >= 0) { 3552 /* Ignore errors. This is just providing extra information, it 3553 * is useful but not necessary. 3554 */ 3555 ret |= (ret2 & BDRV_BLOCK_ZERO); 3556 } 3557 } 3558 3559 return ret; 3560 } 3561 3562 /* Coroutine wrapper for bdrv_get_block_status() */ 3563 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque) 3564 { 3565 BdrvCoGetBlockStatusData *data = opaque; 3566 BlockDriverState *bs = data->bs; 3567 3568 data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors, 3569 data->pnum); 3570 data->done = true; 3571 } 3572 3573 /* 3574 * Synchronous wrapper around bdrv_co_get_block_status(). 3575 * 3576 * See bdrv_co_get_block_status() for details. 3577 */ 3578 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num, 3579 int nb_sectors, int *pnum) 3580 { 3581 Coroutine *co; 3582 BdrvCoGetBlockStatusData data = { 3583 .bs = bs, 3584 .sector_num = sector_num, 3585 .nb_sectors = nb_sectors, 3586 .pnum = pnum, 3587 .done = false, 3588 }; 3589 3590 if (qemu_in_coroutine()) { 3591 /* Fast-path if already in coroutine context */ 3592 bdrv_get_block_status_co_entry(&data); 3593 } else { 3594 co = qemu_coroutine_create(bdrv_get_block_status_co_entry); 3595 qemu_coroutine_enter(co, &data); 3596 while (!data.done) { 3597 qemu_aio_wait(); 3598 } 3599 } 3600 return data.ret; 3601 } 3602 3603 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, 3604 int nb_sectors, int *pnum) 3605 { 3606 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum); 3607 if (ret < 0) { 3608 return ret; 3609 } 3610 return 3611 (ret & BDRV_BLOCK_DATA) || 3612 ((ret & BDRV_BLOCK_ZERO) && !bdrv_has_zero_init(bs)); 3613 } 3614 3615 /* 3616 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] 3617 * 3618 * Return true if the given sector is allocated in any image between 3619 * BASE and TOP (inclusive). BASE can be NULL to check if the given 3620 * sector is allocated in any image of the chain. Return false otherwise. 3621 * 3622 * 'pnum' is set to the number of sectors (including and immediately following 3623 * the specified sector) that are known to be in the same 3624 * allocated/unallocated state. 3625 * 3626 */ 3627 int bdrv_is_allocated_above(BlockDriverState *top, 3628 BlockDriverState *base, 3629 int64_t sector_num, 3630 int nb_sectors, int *pnum) 3631 { 3632 BlockDriverState *intermediate; 3633 int ret, n = nb_sectors; 3634 3635 intermediate = top; 3636 while (intermediate && intermediate != base) { 3637 int pnum_inter; 3638 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors, 3639 &pnum_inter); 3640 if (ret < 0) { 3641 return ret; 3642 } else if (ret) { 3643 *pnum = pnum_inter; 3644 return 1; 3645 } 3646 3647 /* 3648 * [sector_num, nb_sectors] is unallocated on top but intermediate 3649 * might have 3650 * 3651 * [sector_num+x, nr_sectors] allocated. 3652 */ 3653 if (n > pnum_inter && 3654 (intermediate == top || 3655 sector_num + pnum_inter < intermediate->total_sectors)) { 3656 n = pnum_inter; 3657 } 3658 3659 intermediate = intermediate->backing_hd; 3660 } 3661 3662 *pnum = n; 3663 return 0; 3664 } 3665 3666 const char *bdrv_get_encrypted_filename(BlockDriverState *bs) 3667 { 3668 if (bs->backing_hd && bs->backing_hd->encrypted) 3669 return bs->backing_file; 3670 else if (bs->encrypted) 3671 return bs->filename; 3672 else 3673 return NULL; 3674 } 3675 3676 void bdrv_get_backing_filename(BlockDriverState *bs, 3677 char *filename, int filename_size) 3678 { 3679 pstrcpy(filename, filename_size, bs->backing_file); 3680 } 3681 3682 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num, 3683 const uint8_t *buf, int nb_sectors) 3684 { 3685 BlockDriver *drv = bs->drv; 3686 if (!drv) 3687 return -ENOMEDIUM; 3688 if (!drv->bdrv_write_compressed) 3689 return -ENOTSUP; 3690 if (bdrv_check_request(bs, sector_num, nb_sectors)) 3691 return -EIO; 3692 3693 assert(QLIST_EMPTY(&bs->dirty_bitmaps)); 3694 3695 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors); 3696 } 3697 3698 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 3699 { 3700 BlockDriver *drv = bs->drv; 3701 if (!drv) 3702 return -ENOMEDIUM; 3703 if (!drv->bdrv_get_info) 3704 return -ENOTSUP; 3705 memset(bdi, 0, sizeof(*bdi)); 3706 return drv->bdrv_get_info(bs, bdi); 3707 } 3708 3709 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs) 3710 { 3711 BlockDriver *drv = bs->drv; 3712 if (drv && drv->bdrv_get_specific_info) { 3713 return drv->bdrv_get_specific_info(bs); 3714 } 3715 return NULL; 3716 } 3717 3718 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, 3719 int64_t pos, int size) 3720 { 3721 QEMUIOVector qiov; 3722 struct iovec iov = { 3723 .iov_base = (void *) buf, 3724 .iov_len = size, 3725 }; 3726 3727 qemu_iovec_init_external(&qiov, &iov, 1); 3728 return bdrv_writev_vmstate(bs, &qiov, pos); 3729 } 3730 3731 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) 3732 { 3733 BlockDriver *drv = bs->drv; 3734 3735 if (!drv) { 3736 return -ENOMEDIUM; 3737 } else if (drv->bdrv_save_vmstate) { 3738 return drv->bdrv_save_vmstate(bs, qiov, pos); 3739 } else if (bs->file) { 3740 return bdrv_writev_vmstate(bs->file, qiov, pos); 3741 } 3742 3743 return -ENOTSUP; 3744 } 3745 3746 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, 3747 int64_t pos, int size) 3748 { 3749 BlockDriver *drv = bs->drv; 3750 if (!drv) 3751 return -ENOMEDIUM; 3752 if (drv->bdrv_load_vmstate) 3753 return drv->bdrv_load_vmstate(bs, buf, pos, size); 3754 if (bs->file) 3755 return bdrv_load_vmstate(bs->file, buf, pos, size); 3756 return -ENOTSUP; 3757 } 3758 3759 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event) 3760 { 3761 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) { 3762 return; 3763 } 3764 3765 bs->drv->bdrv_debug_event(bs, event); 3766 } 3767 3768 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, 3769 const char *tag) 3770 { 3771 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { 3772 bs = bs->file; 3773 } 3774 3775 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { 3776 return bs->drv->bdrv_debug_breakpoint(bs, event, tag); 3777 } 3778 3779 return -ENOTSUP; 3780 } 3781 3782 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) 3783 { 3784 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) { 3785 bs = bs->file; 3786 } 3787 3788 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) { 3789 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); 3790 } 3791 3792 return -ENOTSUP; 3793 } 3794 3795 int bdrv_debug_resume(BlockDriverState *bs, const char *tag) 3796 { 3797 while (bs && bs->drv && !bs->drv->bdrv_debug_resume) { 3798 bs = bs->file; 3799 } 3800 3801 if (bs && bs->drv && bs->drv->bdrv_debug_resume) { 3802 return bs->drv->bdrv_debug_resume(bs, tag); 3803 } 3804 3805 return -ENOTSUP; 3806 } 3807 3808 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) 3809 { 3810 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { 3811 bs = bs->file; 3812 } 3813 3814 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { 3815 return bs->drv->bdrv_debug_is_suspended(bs, tag); 3816 } 3817 3818 return false; 3819 } 3820 3821 int bdrv_is_snapshot(BlockDriverState *bs) 3822 { 3823 return !!(bs->open_flags & BDRV_O_SNAPSHOT); 3824 } 3825 3826 /* backing_file can either be relative, or absolute, or a protocol. If it is 3827 * relative, it must be relative to the chain. So, passing in bs->filename 3828 * from a BDS as backing_file should not be done, as that may be relative to 3829 * the CWD rather than the chain. */ 3830 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, 3831 const char *backing_file) 3832 { 3833 char *filename_full = NULL; 3834 char *backing_file_full = NULL; 3835 char *filename_tmp = NULL; 3836 int is_protocol = 0; 3837 BlockDriverState *curr_bs = NULL; 3838 BlockDriverState *retval = NULL; 3839 3840 if (!bs || !bs->drv || !backing_file) { 3841 return NULL; 3842 } 3843 3844 filename_full = g_malloc(PATH_MAX); 3845 backing_file_full = g_malloc(PATH_MAX); 3846 filename_tmp = g_malloc(PATH_MAX); 3847 3848 is_protocol = path_has_protocol(backing_file); 3849 3850 for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) { 3851 3852 /* If either of the filename paths is actually a protocol, then 3853 * compare unmodified paths; otherwise make paths relative */ 3854 if (is_protocol || path_has_protocol(curr_bs->backing_file)) { 3855 if (strcmp(backing_file, curr_bs->backing_file) == 0) { 3856 retval = curr_bs->backing_hd; 3857 break; 3858 } 3859 } else { 3860 /* If not an absolute filename path, make it relative to the current 3861 * image's filename path */ 3862 path_combine(filename_tmp, PATH_MAX, curr_bs->filename, 3863 backing_file); 3864 3865 /* We are going to compare absolute pathnames */ 3866 if (!realpath(filename_tmp, filename_full)) { 3867 continue; 3868 } 3869 3870 /* We need to make sure the backing filename we are comparing against 3871 * is relative to the current image filename (or absolute) */ 3872 path_combine(filename_tmp, PATH_MAX, curr_bs->filename, 3873 curr_bs->backing_file); 3874 3875 if (!realpath(filename_tmp, backing_file_full)) { 3876 continue; 3877 } 3878 3879 if (strcmp(backing_file_full, filename_full) == 0) { 3880 retval = curr_bs->backing_hd; 3881 break; 3882 } 3883 } 3884 } 3885 3886 g_free(filename_full); 3887 g_free(backing_file_full); 3888 g_free(filename_tmp); 3889 return retval; 3890 } 3891 3892 int bdrv_get_backing_file_depth(BlockDriverState *bs) 3893 { 3894 if (!bs->drv) { 3895 return 0; 3896 } 3897 3898 if (!bs->backing_hd) { 3899 return 0; 3900 } 3901 3902 return 1 + bdrv_get_backing_file_depth(bs->backing_hd); 3903 } 3904 3905 BlockDriverState *bdrv_find_base(BlockDriverState *bs) 3906 { 3907 BlockDriverState *curr_bs = NULL; 3908 3909 if (!bs) { 3910 return NULL; 3911 } 3912 3913 curr_bs = bs; 3914 3915 while (curr_bs->backing_hd) { 3916 curr_bs = curr_bs->backing_hd; 3917 } 3918 return curr_bs; 3919 } 3920 3921 /**************************************************************/ 3922 /* async I/Os */ 3923 3924 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num, 3925 QEMUIOVector *qiov, int nb_sectors, 3926 BlockDriverCompletionFunc *cb, void *opaque) 3927 { 3928 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque); 3929 3930 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0, 3931 cb, opaque, false); 3932 } 3933 3934 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num, 3935 QEMUIOVector *qiov, int nb_sectors, 3936 BlockDriverCompletionFunc *cb, void *opaque) 3937 { 3938 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque); 3939 3940 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0, 3941 cb, opaque, true); 3942 } 3943 3944 BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs, 3945 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags, 3946 BlockDriverCompletionFunc *cb, void *opaque) 3947 { 3948 trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque); 3949 3950 return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors, 3951 BDRV_REQ_ZERO_WRITE | flags, 3952 cb, opaque, true); 3953 } 3954 3955 3956 typedef struct MultiwriteCB { 3957 int error; 3958 int num_requests; 3959 int num_callbacks; 3960 struct { 3961 BlockDriverCompletionFunc *cb; 3962 void *opaque; 3963 QEMUIOVector *free_qiov; 3964 } callbacks[]; 3965 } MultiwriteCB; 3966 3967 static void multiwrite_user_cb(MultiwriteCB *mcb) 3968 { 3969 int i; 3970 3971 for (i = 0; i < mcb->num_callbacks; i++) { 3972 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error); 3973 if (mcb->callbacks[i].free_qiov) { 3974 qemu_iovec_destroy(mcb->callbacks[i].free_qiov); 3975 } 3976 g_free(mcb->callbacks[i].free_qiov); 3977 } 3978 } 3979 3980 static void multiwrite_cb(void *opaque, int ret) 3981 { 3982 MultiwriteCB *mcb = opaque; 3983 3984 trace_multiwrite_cb(mcb, ret); 3985 3986 if (ret < 0 && !mcb->error) { 3987 mcb->error = ret; 3988 } 3989 3990 mcb->num_requests--; 3991 if (mcb->num_requests == 0) { 3992 multiwrite_user_cb(mcb); 3993 g_free(mcb); 3994 } 3995 } 3996 3997 static int multiwrite_req_compare(const void *a, const void *b) 3998 { 3999 const BlockRequest *req1 = a, *req2 = b; 4000 4001 /* 4002 * Note that we can't simply subtract req2->sector from req1->sector 4003 * here as that could overflow the return value. 4004 */ 4005 if (req1->sector > req2->sector) { 4006 return 1; 4007 } else if (req1->sector < req2->sector) { 4008 return -1; 4009 } else { 4010 return 0; 4011 } 4012 } 4013 4014 /* 4015 * Takes a bunch of requests and tries to merge them. Returns the number of 4016 * requests that remain after merging. 4017 */ 4018 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs, 4019 int num_reqs, MultiwriteCB *mcb) 4020 { 4021 int i, outidx; 4022 4023 // Sort requests by start sector 4024 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare); 4025 4026 // Check if adjacent requests touch the same clusters. If so, combine them, 4027 // filling up gaps with zero sectors. 4028 outidx = 0; 4029 for (i = 1; i < num_reqs; i++) { 4030 int merge = 0; 4031 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors; 4032 4033 // Handle exactly sequential writes and overlapping writes. 4034 if (reqs[i].sector <= oldreq_last) { 4035 merge = 1; 4036 } 4037 4038 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) { 4039 merge = 0; 4040 } 4041 4042 if (merge) { 4043 size_t size; 4044 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov)); 4045 qemu_iovec_init(qiov, 4046 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1); 4047 4048 // Add the first request to the merged one. If the requests are 4049 // overlapping, drop the last sectors of the first request. 4050 size = (reqs[i].sector - reqs[outidx].sector) << 9; 4051 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size); 4052 4053 // We should need to add any zeros between the two requests 4054 assert (reqs[i].sector <= oldreq_last); 4055 4056 // Add the second request 4057 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size); 4058 4059 reqs[outidx].nb_sectors = qiov->size >> 9; 4060 reqs[outidx].qiov = qiov; 4061 4062 mcb->callbacks[i].free_qiov = reqs[outidx].qiov; 4063 } else { 4064 outidx++; 4065 reqs[outidx].sector = reqs[i].sector; 4066 reqs[outidx].nb_sectors = reqs[i].nb_sectors; 4067 reqs[outidx].qiov = reqs[i].qiov; 4068 } 4069 } 4070 4071 return outidx + 1; 4072 } 4073 4074 /* 4075 * Submit multiple AIO write requests at once. 4076 * 4077 * On success, the function returns 0 and all requests in the reqs array have 4078 * been submitted. In error case this function returns -1, and any of the 4079 * requests may or may not be submitted yet. In particular, this means that the 4080 * callback will be called for some of the requests, for others it won't. The 4081 * caller must check the error field of the BlockRequest to wait for the right 4082 * callbacks (if error != 0, no callback will be called). 4083 * 4084 * The implementation may modify the contents of the reqs array, e.g. to merge 4085 * requests. However, the fields opaque and error are left unmodified as they 4086 * are used to signal failure for a single request to the caller. 4087 */ 4088 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs) 4089 { 4090 MultiwriteCB *mcb; 4091 int i; 4092 4093 /* don't submit writes if we don't have a medium */ 4094 if (bs->drv == NULL) { 4095 for (i = 0; i < num_reqs; i++) { 4096 reqs[i].error = -ENOMEDIUM; 4097 } 4098 return -1; 4099 } 4100 4101 if (num_reqs == 0) { 4102 return 0; 4103 } 4104 4105 // Create MultiwriteCB structure 4106 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks)); 4107 mcb->num_requests = 0; 4108 mcb->num_callbacks = num_reqs; 4109 4110 for (i = 0; i < num_reqs; i++) { 4111 mcb->callbacks[i].cb = reqs[i].cb; 4112 mcb->callbacks[i].opaque = reqs[i].opaque; 4113 } 4114 4115 // Check for mergable requests 4116 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb); 4117 4118 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs); 4119 4120 /* Run the aio requests. */ 4121 mcb->num_requests = num_reqs; 4122 for (i = 0; i < num_reqs; i++) { 4123 bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov, 4124 reqs[i].nb_sectors, reqs[i].flags, 4125 multiwrite_cb, mcb, 4126 true); 4127 } 4128 4129 return 0; 4130 } 4131 4132 void bdrv_aio_cancel(BlockDriverAIOCB *acb) 4133 { 4134 acb->aiocb_info->cancel(acb); 4135 } 4136 4137 /**************************************************************/ 4138 /* async block device emulation */ 4139 4140 typedef struct BlockDriverAIOCBSync { 4141 BlockDriverAIOCB common; 4142 QEMUBH *bh; 4143 int ret; 4144 /* vector translation state */ 4145 QEMUIOVector *qiov; 4146 uint8_t *bounce; 4147 int is_write; 4148 } BlockDriverAIOCBSync; 4149 4150 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb) 4151 { 4152 BlockDriverAIOCBSync *acb = 4153 container_of(blockacb, BlockDriverAIOCBSync, common); 4154 qemu_bh_delete(acb->bh); 4155 acb->bh = NULL; 4156 qemu_aio_release(acb); 4157 } 4158 4159 static const AIOCBInfo bdrv_em_aiocb_info = { 4160 .aiocb_size = sizeof(BlockDriverAIOCBSync), 4161 .cancel = bdrv_aio_cancel_em, 4162 }; 4163 4164 static void bdrv_aio_bh_cb(void *opaque) 4165 { 4166 BlockDriverAIOCBSync *acb = opaque; 4167 4168 if (!acb->is_write) 4169 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size); 4170 qemu_vfree(acb->bounce); 4171 acb->common.cb(acb->common.opaque, acb->ret); 4172 qemu_bh_delete(acb->bh); 4173 acb->bh = NULL; 4174 qemu_aio_release(acb); 4175 } 4176 4177 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs, 4178 int64_t sector_num, 4179 QEMUIOVector *qiov, 4180 int nb_sectors, 4181 BlockDriverCompletionFunc *cb, 4182 void *opaque, 4183 int is_write) 4184 4185 { 4186 BlockDriverAIOCBSync *acb; 4187 4188 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque); 4189 acb->is_write = is_write; 4190 acb->qiov = qiov; 4191 acb->bounce = qemu_blockalign(bs, qiov->size); 4192 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb); 4193 4194 if (is_write) { 4195 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size); 4196 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors); 4197 } else { 4198 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors); 4199 } 4200 4201 qemu_bh_schedule(acb->bh); 4202 4203 return &acb->common; 4204 } 4205 4206 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs, 4207 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, 4208 BlockDriverCompletionFunc *cb, void *opaque) 4209 { 4210 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); 4211 } 4212 4213 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs, 4214 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, 4215 BlockDriverCompletionFunc *cb, void *opaque) 4216 { 4217 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); 4218 } 4219 4220 4221 typedef struct BlockDriverAIOCBCoroutine { 4222 BlockDriverAIOCB common; 4223 BlockRequest req; 4224 bool is_write; 4225 bool *done; 4226 QEMUBH* bh; 4227 } BlockDriverAIOCBCoroutine; 4228 4229 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb) 4230 { 4231 BlockDriverAIOCBCoroutine *acb = 4232 container_of(blockacb, BlockDriverAIOCBCoroutine, common); 4233 bool done = false; 4234 4235 acb->done = &done; 4236 while (!done) { 4237 qemu_aio_wait(); 4238 } 4239 } 4240 4241 static const AIOCBInfo bdrv_em_co_aiocb_info = { 4242 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine), 4243 .cancel = bdrv_aio_co_cancel_em, 4244 }; 4245 4246 static void bdrv_co_em_bh(void *opaque) 4247 { 4248 BlockDriverAIOCBCoroutine *acb = opaque; 4249 4250 acb->common.cb(acb->common.opaque, acb->req.error); 4251 4252 if (acb->done) { 4253 *acb->done = true; 4254 } 4255 4256 qemu_bh_delete(acb->bh); 4257 qemu_aio_release(acb); 4258 } 4259 4260 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */ 4261 static void coroutine_fn bdrv_co_do_rw(void *opaque) 4262 { 4263 BlockDriverAIOCBCoroutine *acb = opaque; 4264 BlockDriverState *bs = acb->common.bs; 4265 4266 if (!acb->is_write) { 4267 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector, 4268 acb->req.nb_sectors, acb->req.qiov, acb->req.flags); 4269 } else { 4270 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector, 4271 acb->req.nb_sectors, acb->req.qiov, acb->req.flags); 4272 } 4273 4274 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); 4275 qemu_bh_schedule(acb->bh); 4276 } 4277 4278 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, 4279 int64_t sector_num, 4280 QEMUIOVector *qiov, 4281 int nb_sectors, 4282 BdrvRequestFlags flags, 4283 BlockDriverCompletionFunc *cb, 4284 void *opaque, 4285 bool is_write) 4286 { 4287 Coroutine *co; 4288 BlockDriverAIOCBCoroutine *acb; 4289 4290 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); 4291 acb->req.sector = sector_num; 4292 acb->req.nb_sectors = nb_sectors; 4293 acb->req.qiov = qiov; 4294 acb->req.flags = flags; 4295 acb->is_write = is_write; 4296 acb->done = NULL; 4297 4298 co = qemu_coroutine_create(bdrv_co_do_rw); 4299 qemu_coroutine_enter(co, acb); 4300 4301 return &acb->common; 4302 } 4303 4304 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) 4305 { 4306 BlockDriverAIOCBCoroutine *acb = opaque; 4307 BlockDriverState *bs = acb->common.bs; 4308 4309 acb->req.error = bdrv_co_flush(bs); 4310 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); 4311 qemu_bh_schedule(acb->bh); 4312 } 4313 4314 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs, 4315 BlockDriverCompletionFunc *cb, void *opaque) 4316 { 4317 trace_bdrv_aio_flush(bs, opaque); 4318 4319 Coroutine *co; 4320 BlockDriverAIOCBCoroutine *acb; 4321 4322 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); 4323 acb->done = NULL; 4324 4325 co = qemu_coroutine_create(bdrv_aio_flush_co_entry); 4326 qemu_coroutine_enter(co, acb); 4327 4328 return &acb->common; 4329 } 4330 4331 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque) 4332 { 4333 BlockDriverAIOCBCoroutine *acb = opaque; 4334 BlockDriverState *bs = acb->common.bs; 4335 4336 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors); 4337 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); 4338 qemu_bh_schedule(acb->bh); 4339 } 4340 4341 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs, 4342 int64_t sector_num, int nb_sectors, 4343 BlockDriverCompletionFunc *cb, void *opaque) 4344 { 4345 Coroutine *co; 4346 BlockDriverAIOCBCoroutine *acb; 4347 4348 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque); 4349 4350 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); 4351 acb->req.sector = sector_num; 4352 acb->req.nb_sectors = nb_sectors; 4353 acb->done = NULL; 4354 co = qemu_coroutine_create(bdrv_aio_discard_co_entry); 4355 qemu_coroutine_enter(co, acb); 4356 4357 return &acb->common; 4358 } 4359 4360 void bdrv_init(void) 4361 { 4362 module_call_init(MODULE_INIT_BLOCK); 4363 } 4364 4365 void bdrv_init_with_whitelist(void) 4366 { 4367 use_bdrv_whitelist = 1; 4368 bdrv_init(); 4369 } 4370 4371 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, 4372 BlockDriverCompletionFunc *cb, void *opaque) 4373 { 4374 BlockDriverAIOCB *acb; 4375 4376 acb = g_slice_alloc(aiocb_info->aiocb_size); 4377 acb->aiocb_info = aiocb_info; 4378 acb->bs = bs; 4379 acb->cb = cb; 4380 acb->opaque = opaque; 4381 return acb; 4382 } 4383 4384 void qemu_aio_release(void *p) 4385 { 4386 BlockDriverAIOCB *acb = p; 4387 g_slice_free1(acb->aiocb_info->aiocb_size, acb); 4388 } 4389 4390 /**************************************************************/ 4391 /* Coroutine block device emulation */ 4392 4393 typedef struct CoroutineIOCompletion { 4394 Coroutine *coroutine; 4395 int ret; 4396 } CoroutineIOCompletion; 4397 4398 static void bdrv_co_io_em_complete(void *opaque, int ret) 4399 { 4400 CoroutineIOCompletion *co = opaque; 4401 4402 co->ret = ret; 4403 qemu_coroutine_enter(co->coroutine, NULL); 4404 } 4405 4406 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num, 4407 int nb_sectors, QEMUIOVector *iov, 4408 bool is_write) 4409 { 4410 CoroutineIOCompletion co = { 4411 .coroutine = qemu_coroutine_self(), 4412 }; 4413 BlockDriverAIOCB *acb; 4414 4415 if (is_write) { 4416 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors, 4417 bdrv_co_io_em_complete, &co); 4418 } else { 4419 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors, 4420 bdrv_co_io_em_complete, &co); 4421 } 4422 4423 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb); 4424 if (!acb) { 4425 return -EIO; 4426 } 4427 qemu_coroutine_yield(); 4428 4429 return co.ret; 4430 } 4431 4432 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, 4433 int64_t sector_num, int nb_sectors, 4434 QEMUIOVector *iov) 4435 { 4436 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false); 4437 } 4438 4439 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, 4440 int64_t sector_num, int nb_sectors, 4441 QEMUIOVector *iov) 4442 { 4443 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true); 4444 } 4445 4446 static void coroutine_fn bdrv_flush_co_entry(void *opaque) 4447 { 4448 RwCo *rwco = opaque; 4449 4450 rwco->ret = bdrv_co_flush(rwco->bs); 4451 } 4452 4453 int coroutine_fn bdrv_co_flush(BlockDriverState *bs) 4454 { 4455 int ret; 4456 4457 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) { 4458 return 0; 4459 } 4460 4461 /* Write back cached data to the OS even with cache=unsafe */ 4462 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); 4463 if (bs->drv->bdrv_co_flush_to_os) { 4464 ret = bs->drv->bdrv_co_flush_to_os(bs); 4465 if (ret < 0) { 4466 return ret; 4467 } 4468 } 4469 4470 /* But don't actually force it to the disk with cache=unsafe */ 4471 if (bs->open_flags & BDRV_O_NO_FLUSH) { 4472 goto flush_parent; 4473 } 4474 4475 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); 4476 if (bs->drv->bdrv_co_flush_to_disk) { 4477 ret = bs->drv->bdrv_co_flush_to_disk(bs); 4478 } else if (bs->drv->bdrv_aio_flush) { 4479 BlockDriverAIOCB *acb; 4480 CoroutineIOCompletion co = { 4481 .coroutine = qemu_coroutine_self(), 4482 }; 4483 4484 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); 4485 if (acb == NULL) { 4486 ret = -EIO; 4487 } else { 4488 qemu_coroutine_yield(); 4489 ret = co.ret; 4490 } 4491 } else { 4492 /* 4493 * Some block drivers always operate in either writethrough or unsafe 4494 * mode and don't support bdrv_flush therefore. Usually qemu doesn't 4495 * know how the server works (because the behaviour is hardcoded or 4496 * depends on server-side configuration), so we can't ensure that 4497 * everything is safe on disk. Returning an error doesn't work because 4498 * that would break guests even if the server operates in writethrough 4499 * mode. 4500 * 4501 * Let's hope the user knows what he's doing. 4502 */ 4503 ret = 0; 4504 } 4505 if (ret < 0) { 4506 return ret; 4507 } 4508 4509 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH 4510 * in the case of cache=unsafe, so there are no useless flushes. 4511 */ 4512 flush_parent: 4513 return bdrv_co_flush(bs->file); 4514 } 4515 4516 void bdrv_invalidate_cache(BlockDriverState *bs) 4517 { 4518 if (bs->drv && bs->drv->bdrv_invalidate_cache) { 4519 bs->drv->bdrv_invalidate_cache(bs); 4520 } 4521 } 4522 4523 void bdrv_invalidate_cache_all(void) 4524 { 4525 BlockDriverState *bs; 4526 4527 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 4528 bdrv_invalidate_cache(bs); 4529 } 4530 } 4531 4532 void bdrv_clear_incoming_migration_all(void) 4533 { 4534 BlockDriverState *bs; 4535 4536 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 4537 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING); 4538 } 4539 } 4540 4541 int bdrv_flush(BlockDriverState *bs) 4542 { 4543 Coroutine *co; 4544 RwCo rwco = { 4545 .bs = bs, 4546 .ret = NOT_DONE, 4547 }; 4548 4549 if (qemu_in_coroutine()) { 4550 /* Fast-path if already in coroutine context */ 4551 bdrv_flush_co_entry(&rwco); 4552 } else { 4553 co = qemu_coroutine_create(bdrv_flush_co_entry); 4554 qemu_coroutine_enter(co, &rwco); 4555 while (rwco.ret == NOT_DONE) { 4556 qemu_aio_wait(); 4557 } 4558 } 4559 4560 return rwco.ret; 4561 } 4562 4563 static void coroutine_fn bdrv_discard_co_entry(void *opaque) 4564 { 4565 RwCo *rwco = opaque; 4566 4567 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors); 4568 } 4569 4570 /* if no limit is specified in the BlockLimits use a default 4571 * of 32768 512-byte sectors (16 MiB) per request. 4572 */ 4573 #define MAX_DISCARD_DEFAULT 32768 4574 4575 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num, 4576 int nb_sectors) 4577 { 4578 int max_discard; 4579 4580 if (!bs->drv) { 4581 return -ENOMEDIUM; 4582 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) { 4583 return -EIO; 4584 } else if (bs->read_only) { 4585 return -EROFS; 4586 } 4587 4588 bdrv_reset_dirty(bs, sector_num, nb_sectors); 4589 4590 /* Do nothing if disabled. */ 4591 if (!(bs->open_flags & BDRV_O_UNMAP)) { 4592 return 0; 4593 } 4594 4595 if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) { 4596 return 0; 4597 } 4598 4599 max_discard = bs->bl.max_discard ? bs->bl.max_discard : MAX_DISCARD_DEFAULT; 4600 while (nb_sectors > 0) { 4601 int ret; 4602 int num = nb_sectors; 4603 4604 /* align request */ 4605 if (bs->bl.discard_alignment && 4606 num >= bs->bl.discard_alignment && 4607 sector_num % bs->bl.discard_alignment) { 4608 if (num > bs->bl.discard_alignment) { 4609 num = bs->bl.discard_alignment; 4610 } 4611 num -= sector_num % bs->bl.discard_alignment; 4612 } 4613 4614 /* limit request size */ 4615 if (num > max_discard) { 4616 num = max_discard; 4617 } 4618 4619 if (bs->drv->bdrv_co_discard) { 4620 ret = bs->drv->bdrv_co_discard(bs, sector_num, num); 4621 } else { 4622 BlockDriverAIOCB *acb; 4623 CoroutineIOCompletion co = { 4624 .coroutine = qemu_coroutine_self(), 4625 }; 4626 4627 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors, 4628 bdrv_co_io_em_complete, &co); 4629 if (acb == NULL) { 4630 return -EIO; 4631 } else { 4632 qemu_coroutine_yield(); 4633 ret = co.ret; 4634 } 4635 } 4636 if (ret && ret != -ENOTSUP) { 4637 return ret; 4638 } 4639 4640 sector_num += num; 4641 nb_sectors -= num; 4642 } 4643 return 0; 4644 } 4645 4646 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors) 4647 { 4648 Coroutine *co; 4649 RwCo rwco = { 4650 .bs = bs, 4651 .sector_num = sector_num, 4652 .nb_sectors = nb_sectors, 4653 .ret = NOT_DONE, 4654 }; 4655 4656 if (qemu_in_coroutine()) { 4657 /* Fast-path if already in coroutine context */ 4658 bdrv_discard_co_entry(&rwco); 4659 } else { 4660 co = qemu_coroutine_create(bdrv_discard_co_entry); 4661 qemu_coroutine_enter(co, &rwco); 4662 while (rwco.ret == NOT_DONE) { 4663 qemu_aio_wait(); 4664 } 4665 } 4666 4667 return rwco.ret; 4668 } 4669 4670 /**************************************************************/ 4671 /* removable device support */ 4672 4673 /** 4674 * Return TRUE if the media is present 4675 */ 4676 int bdrv_is_inserted(BlockDriverState *bs) 4677 { 4678 BlockDriver *drv = bs->drv; 4679 4680 if (!drv) 4681 return 0; 4682 if (!drv->bdrv_is_inserted) 4683 return 1; 4684 return drv->bdrv_is_inserted(bs); 4685 } 4686 4687 /** 4688 * Return whether the media changed since the last call to this 4689 * function, or -ENOTSUP if we don't know. Most drivers don't know. 4690 */ 4691 int bdrv_media_changed(BlockDriverState *bs) 4692 { 4693 BlockDriver *drv = bs->drv; 4694 4695 if (drv && drv->bdrv_media_changed) { 4696 return drv->bdrv_media_changed(bs); 4697 } 4698 return -ENOTSUP; 4699 } 4700 4701 /** 4702 * If eject_flag is TRUE, eject the media. Otherwise, close the tray 4703 */ 4704 void bdrv_eject(BlockDriverState *bs, bool eject_flag) 4705 { 4706 BlockDriver *drv = bs->drv; 4707 4708 if (drv && drv->bdrv_eject) { 4709 drv->bdrv_eject(bs, eject_flag); 4710 } 4711 4712 if (bs->device_name[0] != '\0') { 4713 bdrv_emit_qmp_eject_event(bs, eject_flag); 4714 } 4715 } 4716 4717 /** 4718 * Lock or unlock the media (if it is locked, the user won't be able 4719 * to eject it manually). 4720 */ 4721 void bdrv_lock_medium(BlockDriverState *bs, bool locked) 4722 { 4723 BlockDriver *drv = bs->drv; 4724 4725 trace_bdrv_lock_medium(bs, locked); 4726 4727 if (drv && drv->bdrv_lock_medium) { 4728 drv->bdrv_lock_medium(bs, locked); 4729 } 4730 } 4731 4732 /* needed for generic scsi interface */ 4733 4734 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 4735 { 4736 BlockDriver *drv = bs->drv; 4737 4738 if (drv && drv->bdrv_ioctl) 4739 return drv->bdrv_ioctl(bs, req, buf); 4740 return -ENOTSUP; 4741 } 4742 4743 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs, 4744 unsigned long int req, void *buf, 4745 BlockDriverCompletionFunc *cb, void *opaque) 4746 { 4747 BlockDriver *drv = bs->drv; 4748 4749 if (drv && drv->bdrv_aio_ioctl) 4750 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque); 4751 return NULL; 4752 } 4753 4754 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align) 4755 { 4756 bs->buffer_alignment = align; 4757 } 4758 4759 void *qemu_blockalign(BlockDriverState *bs, size_t size) 4760 { 4761 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size); 4762 } 4763 4764 /* 4765 * Check if all memory in this vector is sector aligned. 4766 */ 4767 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) 4768 { 4769 int i; 4770 4771 for (i = 0; i < qiov->niov; i++) { 4772 if ((uintptr_t) qiov->iov[i].iov_base % bs->buffer_alignment) { 4773 return false; 4774 } 4775 } 4776 4777 return true; 4778 } 4779 4780 BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity) 4781 { 4782 int64_t bitmap_size; 4783 BdrvDirtyBitmap *bitmap; 4784 4785 assert((granularity & (granularity - 1)) == 0); 4786 4787 granularity >>= BDRV_SECTOR_BITS; 4788 assert(granularity); 4789 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS); 4790 bitmap = g_malloc0(sizeof(BdrvDirtyBitmap)); 4791 bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1); 4792 QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list); 4793 return bitmap; 4794 } 4795 4796 void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap) 4797 { 4798 BdrvDirtyBitmap *bm, *next; 4799 QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) { 4800 if (bm == bitmap) { 4801 QLIST_REMOVE(bitmap, list); 4802 hbitmap_free(bitmap->bitmap); 4803 g_free(bitmap); 4804 return; 4805 } 4806 } 4807 } 4808 4809 BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs) 4810 { 4811 BdrvDirtyBitmap *bm; 4812 BlockDirtyInfoList *list = NULL; 4813 BlockDirtyInfoList **plist = &list; 4814 4815 QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) { 4816 BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo)); 4817 BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList)); 4818 info->count = bdrv_get_dirty_count(bs, bm); 4819 info->granularity = 4820 ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap)); 4821 entry->value = info; 4822 *plist = entry; 4823 plist = &entry->next; 4824 } 4825 4826 return list; 4827 } 4828 4829 int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector) 4830 { 4831 if (bitmap) { 4832 return hbitmap_get(bitmap->bitmap, sector); 4833 } else { 4834 return 0; 4835 } 4836 } 4837 4838 void bdrv_dirty_iter_init(BlockDriverState *bs, 4839 BdrvDirtyBitmap *bitmap, HBitmapIter *hbi) 4840 { 4841 hbitmap_iter_init(hbi, bitmap->bitmap, 0); 4842 } 4843 4844 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, 4845 int nr_sectors) 4846 { 4847 BdrvDirtyBitmap *bitmap; 4848 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) { 4849 hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors); 4850 } 4851 } 4852 4853 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors) 4854 { 4855 BdrvDirtyBitmap *bitmap; 4856 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) { 4857 hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors); 4858 } 4859 } 4860 4861 int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap) 4862 { 4863 return hbitmap_count(bitmap->bitmap); 4864 } 4865 4866 /* Get a reference to bs */ 4867 void bdrv_ref(BlockDriverState *bs) 4868 { 4869 bs->refcnt++; 4870 } 4871 4872 /* Release a previously grabbed reference to bs. 4873 * If after releasing, reference count is zero, the BlockDriverState is 4874 * deleted. */ 4875 void bdrv_unref(BlockDriverState *bs) 4876 { 4877 assert(bs->refcnt > 0); 4878 if (--bs->refcnt == 0) { 4879 bdrv_delete(bs); 4880 } 4881 } 4882 4883 void bdrv_set_in_use(BlockDriverState *bs, int in_use) 4884 { 4885 assert(bs->in_use != in_use); 4886 bs->in_use = in_use; 4887 } 4888 4889 int bdrv_in_use(BlockDriverState *bs) 4890 { 4891 return bs->in_use; 4892 } 4893 4894 void bdrv_iostatus_enable(BlockDriverState *bs) 4895 { 4896 bs->iostatus_enabled = true; 4897 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; 4898 } 4899 4900 /* The I/O status is only enabled if the drive explicitly 4901 * enables it _and_ the VM is configured to stop on errors */ 4902 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs) 4903 { 4904 return (bs->iostatus_enabled && 4905 (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC || 4906 bs->on_write_error == BLOCKDEV_ON_ERROR_STOP || 4907 bs->on_read_error == BLOCKDEV_ON_ERROR_STOP)); 4908 } 4909 4910 void bdrv_iostatus_disable(BlockDriverState *bs) 4911 { 4912 bs->iostatus_enabled = false; 4913 } 4914 4915 void bdrv_iostatus_reset(BlockDriverState *bs) 4916 { 4917 if (bdrv_iostatus_is_enabled(bs)) { 4918 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; 4919 if (bs->job) { 4920 block_job_iostatus_reset(bs->job); 4921 } 4922 } 4923 } 4924 4925 void bdrv_iostatus_set_err(BlockDriverState *bs, int error) 4926 { 4927 assert(bdrv_iostatus_is_enabled(bs)); 4928 if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 4929 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : 4930 BLOCK_DEVICE_IO_STATUS_FAILED; 4931 } 4932 } 4933 4934 void 4935 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes, 4936 enum BlockAcctType type) 4937 { 4938 assert(type < BDRV_MAX_IOTYPE); 4939 4940 cookie->bytes = bytes; 4941 cookie->start_time_ns = get_clock(); 4942 cookie->type = type; 4943 } 4944 4945 void 4946 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie) 4947 { 4948 assert(cookie->type < BDRV_MAX_IOTYPE); 4949 4950 bs->nr_bytes[cookie->type] += cookie->bytes; 4951 bs->nr_ops[cookie->type]++; 4952 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns; 4953 } 4954 4955 void bdrv_img_create(const char *filename, const char *fmt, 4956 const char *base_filename, const char *base_fmt, 4957 char *options, uint64_t img_size, int flags, 4958 Error **errp, bool quiet) 4959 { 4960 QEMUOptionParameter *param = NULL, *create_options = NULL; 4961 QEMUOptionParameter *backing_fmt, *backing_file, *size; 4962 BlockDriver *drv, *proto_drv; 4963 BlockDriver *backing_drv = NULL; 4964 Error *local_err = NULL; 4965 int ret = 0; 4966 4967 /* Find driver and parse its options */ 4968 drv = bdrv_find_format(fmt); 4969 if (!drv) { 4970 error_setg(errp, "Unknown file format '%s'", fmt); 4971 return; 4972 } 4973 4974 proto_drv = bdrv_find_protocol(filename, true); 4975 if (!proto_drv) { 4976 error_setg(errp, "Unknown protocol '%s'", filename); 4977 return; 4978 } 4979 4980 create_options = append_option_parameters(create_options, 4981 drv->create_options); 4982 create_options = append_option_parameters(create_options, 4983 proto_drv->create_options); 4984 4985 /* Create parameter list with default values */ 4986 param = parse_option_parameters("", create_options, param); 4987 4988 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size); 4989 4990 /* Parse -o options */ 4991 if (options) { 4992 param = parse_option_parameters(options, create_options, param); 4993 if (param == NULL) { 4994 error_setg(errp, "Invalid options for file format '%s'.", fmt); 4995 goto out; 4996 } 4997 } 4998 4999 if (base_filename) { 5000 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE, 5001 base_filename)) { 5002 error_setg(errp, "Backing file not supported for file format '%s'", 5003 fmt); 5004 goto out; 5005 } 5006 } 5007 5008 if (base_fmt) { 5009 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) { 5010 error_setg(errp, "Backing file format not supported for file " 5011 "format '%s'", fmt); 5012 goto out; 5013 } 5014 } 5015 5016 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE); 5017 if (backing_file && backing_file->value.s) { 5018 if (!strcmp(filename, backing_file->value.s)) { 5019 error_setg(errp, "Error: Trying to create an image with the " 5020 "same filename as the backing file"); 5021 goto out; 5022 } 5023 } 5024 5025 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT); 5026 if (backing_fmt && backing_fmt->value.s) { 5027 backing_drv = bdrv_find_format(backing_fmt->value.s); 5028 if (!backing_drv) { 5029 error_setg(errp, "Unknown backing file format '%s'", 5030 backing_fmt->value.s); 5031 goto out; 5032 } 5033 } 5034 5035 // The size for the image must always be specified, with one exception: 5036 // If we are using a backing file, we can obtain the size from there 5037 size = get_option_parameter(param, BLOCK_OPT_SIZE); 5038 if (size && size->value.n == -1) { 5039 if (backing_file && backing_file->value.s) { 5040 BlockDriverState *bs; 5041 uint64_t size; 5042 char buf[32]; 5043 int back_flags; 5044 5045 /* backing files always opened read-only */ 5046 back_flags = 5047 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); 5048 5049 bs = bdrv_new(""); 5050 5051 ret = bdrv_open(bs, backing_file->value.s, NULL, back_flags, 5052 backing_drv, &local_err); 5053 if (ret < 0) { 5054 error_setg_errno(errp, -ret, "Could not open '%s': %s", 5055 backing_file->value.s, 5056 error_get_pretty(local_err)); 5057 error_free(local_err); 5058 local_err = NULL; 5059 bdrv_unref(bs); 5060 goto out; 5061 } 5062 bdrv_get_geometry(bs, &size); 5063 size *= 512; 5064 5065 snprintf(buf, sizeof(buf), "%" PRId64, size); 5066 set_option_parameter(param, BLOCK_OPT_SIZE, buf); 5067 5068 bdrv_unref(bs); 5069 } else { 5070 error_setg(errp, "Image creation needs a size parameter"); 5071 goto out; 5072 } 5073 } 5074 5075 if (!quiet) { 5076 printf("Formatting '%s', fmt=%s ", filename, fmt); 5077 print_option_parameters(param); 5078 puts(""); 5079 } 5080 ret = bdrv_create(drv, filename, param, &local_err); 5081 if (ret == -EFBIG) { 5082 /* This is generally a better message than whatever the driver would 5083 * deliver (especially because of the cluster_size_hint), since that 5084 * is most probably not much different from "image too large". */ 5085 const char *cluster_size_hint = ""; 5086 if (get_option_parameter(create_options, BLOCK_OPT_CLUSTER_SIZE)) { 5087 cluster_size_hint = " (try using a larger cluster size)"; 5088 } 5089 error_setg(errp, "The image size is too large for file format '%s'" 5090 "%s", fmt, cluster_size_hint); 5091 error_free(local_err); 5092 local_err = NULL; 5093 } 5094 5095 out: 5096 free_option_parameters(create_options); 5097 free_option_parameters(param); 5098 5099 if (error_is_set(&local_err)) { 5100 error_propagate(errp, local_err); 5101 } 5102 } 5103 5104 AioContext *bdrv_get_aio_context(BlockDriverState *bs) 5105 { 5106 /* Currently BlockDriverState always uses the main loop AioContext */ 5107 return qemu_get_aio_context(); 5108 } 5109 5110 void bdrv_add_before_write_notifier(BlockDriverState *bs, 5111 NotifierWithReturn *notifier) 5112 { 5113 notifier_with_return_list_add(&bs->before_write_notifiers, notifier); 5114 } 5115 5116 int bdrv_amend_options(BlockDriverState *bs, QEMUOptionParameter *options) 5117 { 5118 if (bs->drv->bdrv_amend_options == NULL) { 5119 return -ENOTSUP; 5120 } 5121 return bs->drv->bdrv_amend_options(bs, options); 5122 } 5123 5124 /* Used to recurse on single child block filters. 5125 * Single child block filter will store their child in bs->file. 5126 */ 5127 bool bdrv_generic_is_first_non_filter(BlockDriverState *bs, 5128 BlockDriverState *candidate) 5129 { 5130 if (!bs->drv) { 5131 return false; 5132 } 5133 5134 if (!bs->drv->authorizations[BS_IS_A_FILTER]) { 5135 if (bs == candidate) { 5136 return true; 5137 } else { 5138 return false; 5139 } 5140 } 5141 5142 if (!bs->drv->authorizations[BS_FILTER_PASS_DOWN]) { 5143 return false; 5144 } 5145 5146 if (!bs->file) { 5147 return false; 5148 } 5149 5150 return bdrv_recurse_is_first_non_filter(bs->file, candidate); 5151 } 5152 5153 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs, 5154 BlockDriverState *candidate) 5155 { 5156 if (bs->drv && bs->drv->bdrv_recurse_is_first_non_filter) { 5157 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate); 5158 } 5159 5160 return bdrv_generic_is_first_non_filter(bs, candidate); 5161 } 5162 5163 /* This function checks if the candidate is the first non filter bs down it's 5164 * bs chain. Since we don't have pointers to parents it explore all bs chains 5165 * from the top. Some filters can choose not to pass down the recursion. 5166 */ 5167 bool bdrv_is_first_non_filter(BlockDriverState *candidate) 5168 { 5169 BlockDriverState *bs; 5170 5171 /* walk down the bs forest recursively */ 5172 QTAILQ_FOREACH(bs, &bdrv_states, device_list) { 5173 bool perm; 5174 5175 if (!bs->file) { 5176 continue; 5177 } 5178 5179 perm = bdrv_recurse_is_first_non_filter(bs->file, candidate); 5180 5181 /* candidate is the first non filter */ 5182 if (perm) { 5183 return true; 5184 } 5185 } 5186 5187 return false; 5188 } 5189