1 /* 2 * Block driver for Parallels disk image format 3 * 4 * Copyright (c) 2007 Alex Beregszaszi 5 * Copyright (c) 2015 Denis V. Lunev <den@openvz.org> 6 * 7 * This code was originally based on comparing different disk images created 8 * by Parallels. Currently it is based on opened OpenVZ sources 9 * available at 10 * http://git.openvz.org/?p=ploop;a=summary 11 * 12 * Permission is hereby granted, free of charge, to any person obtaining a copy 13 * of this software and associated documentation files (the "Software"), to deal 14 * in the Software without restriction, including without limitation the rights 15 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 16 * copies of the Software, and to permit persons to whom the Software is 17 * furnished to do so, subject to the following conditions: 18 * 19 * The above copyright notice and this permission notice shall be included in 20 * all copies or substantial portions of the Software. 21 * 22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 25 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 27 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 28 * THE SOFTWARE. 29 */ 30 31 #include "qemu/osdep.h" 32 #include "qapi/error.h" 33 #include "block/block_int.h" 34 #include "sysemu/block-backend.h" 35 #include "qemu/module.h" 36 #include "qemu/option.h" 37 #include "qapi/qmp/qdict.h" 38 #include "qapi/qobject-input-visitor.h" 39 #include "qapi/qapi-visit-block-core.h" 40 #include "qemu/bswap.h" 41 #include "qemu/bitmap.h" 42 #include "migration/blocker.h" 43 #include "parallels.h" 44 45 /**************************************************************/ 46 47 #define HEADER_MAGIC "WithoutFreeSpace" 48 #define HEADER_MAGIC2 "WithouFreSpacExt" 49 #define HEADER_VERSION 2 50 #define HEADER_INUSE_MAGIC (0x746F6E59) 51 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32) 52 53 static QEnumLookup prealloc_mode_lookup = { 54 .array = (const char *const[]) { 55 "falloc", 56 "truncate", 57 }, 58 .size = PRL_PREALLOC_MODE__MAX 59 }; 60 61 #define PARALLELS_OPT_PREALLOC_MODE "prealloc-mode" 62 #define PARALLELS_OPT_PREALLOC_SIZE "prealloc-size" 63 64 static QemuOptsList parallels_runtime_opts = { 65 .name = "parallels", 66 .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head), 67 .desc = { 68 { 69 .name = PARALLELS_OPT_PREALLOC_SIZE, 70 .type = QEMU_OPT_SIZE, 71 .help = "Preallocation size on image expansion", 72 .def_value_str = "128M", 73 }, 74 { 75 .name = PARALLELS_OPT_PREALLOC_MODE, 76 .type = QEMU_OPT_STRING, 77 .help = "Preallocation mode on image expansion " 78 "(allowed values: falloc, truncate)", 79 .def_value_str = "falloc", 80 }, 81 { /* end of list */ }, 82 }, 83 }; 84 85 static QemuOptsList parallels_create_opts = { 86 .name = "parallels-create-opts", 87 .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head), 88 .desc = { 89 { 90 .name = BLOCK_OPT_SIZE, 91 .type = QEMU_OPT_SIZE, 92 .help = "Virtual disk size", 93 }, 94 { 95 .name = BLOCK_OPT_CLUSTER_SIZE, 96 .type = QEMU_OPT_SIZE, 97 .help = "Parallels image cluster size", 98 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE), 99 }, 100 { /* end of list */ } 101 } 102 }; 103 104 105 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx) 106 { 107 return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier; 108 } 109 110 static uint32_t bat_entry_off(uint32_t idx) 111 { 112 return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx; 113 } 114 115 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num) 116 { 117 uint32_t index, offset; 118 119 index = sector_num / s->tracks; 120 offset = sector_num % s->tracks; 121 122 /* not allocated */ 123 if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) { 124 return -1; 125 } 126 return bat2sect(s, index) + offset; 127 } 128 129 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num, 130 int nb_sectors) 131 { 132 int ret = s->tracks - sector_num % s->tracks; 133 return MIN(nb_sectors, ret); 134 } 135 136 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num, 137 int nb_sectors, int *pnum) 138 { 139 int64_t start_off = -2, prev_end_off = -2; 140 141 *pnum = 0; 142 while (nb_sectors > 0 || start_off == -2) { 143 int64_t offset = seek_to_sector(s, sector_num); 144 int to_end; 145 146 if (start_off == -2) { 147 start_off = offset; 148 prev_end_off = offset; 149 } else if (offset != prev_end_off) { 150 break; 151 } 152 153 to_end = cluster_remainder(s, sector_num, nb_sectors); 154 nb_sectors -= to_end; 155 sector_num += to_end; 156 *pnum += to_end; 157 158 if (offset > 0) { 159 prev_end_off += to_end; 160 } 161 } 162 return start_off; 163 } 164 165 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num, 166 int nb_sectors, int *pnum) 167 { 168 int ret; 169 BDRVParallelsState *s = bs->opaque; 170 int64_t pos, space, idx, to_allocate, i, len; 171 172 pos = block_status(s, sector_num, nb_sectors, pnum); 173 if (pos > 0) { 174 return pos; 175 } 176 177 idx = sector_num / s->tracks; 178 to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx; 179 180 /* This function is called only by parallels_co_writev(), which will never 181 * pass a sector_num at or beyond the end of the image (because the block 182 * layer never passes such a sector_num to that function). Therefore, idx 183 * is always below s->bat_size. 184 * block_status() will limit *pnum so that sector_num + *pnum will not 185 * exceed the image end. Therefore, idx + to_allocate cannot exceed 186 * s->bat_size. 187 * Note that s->bat_size is an unsigned int, therefore idx + to_allocate 188 * will always fit into a uint32_t. */ 189 assert(idx < s->bat_size && idx + to_allocate <= s->bat_size); 190 191 space = to_allocate * s->tracks; 192 len = bdrv_getlength(bs->file->bs); 193 if (len < 0) { 194 return len; 195 } 196 if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) { 197 space += s->prealloc_size; 198 if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) { 199 ret = bdrv_pwrite_zeroes(bs->file, 200 s->data_end << BDRV_SECTOR_BITS, 201 space << BDRV_SECTOR_BITS, 0); 202 } else { 203 ret = bdrv_truncate(bs->file, 204 (s->data_end + space) << BDRV_SECTOR_BITS, 205 PREALLOC_MODE_OFF, NULL); 206 } 207 if (ret < 0) { 208 return ret; 209 } 210 } 211 212 /* Try to read from backing to fill empty clusters 213 * FIXME: 1. previous write_zeroes may be redundant 214 * 2. most of data we read from backing will be rewritten by 215 * parallels_co_writev. On aligned-to-cluster write we do not need 216 * this read at all. 217 * 3. it would be good to combine write of data from backing and new 218 * data into one write call */ 219 if (bs->backing) { 220 int64_t nb_cow_sectors = to_allocate * s->tracks; 221 int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS; 222 QEMUIOVector qiov; 223 struct iovec iov = { 224 .iov_len = nb_cow_bytes, 225 .iov_base = qemu_blockalign(bs, nb_cow_bytes) 226 }; 227 qemu_iovec_init_external(&qiov, &iov, 1); 228 229 ret = bdrv_co_readv(bs->backing, idx * s->tracks, nb_cow_sectors, 230 &qiov); 231 if (ret < 0) { 232 qemu_vfree(iov.iov_base); 233 return ret; 234 } 235 236 ret = bdrv_co_writev(bs->file, s->data_end, nb_cow_sectors, &qiov); 237 qemu_vfree(iov.iov_base); 238 if (ret < 0) { 239 return ret; 240 } 241 } 242 243 for (i = 0; i < to_allocate; i++) { 244 s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier); 245 s->data_end += s->tracks; 246 bitmap_set(s->bat_dirty_bmap, 247 bat_entry_off(idx + i) / s->bat_dirty_block, 1); 248 } 249 250 return bat2sect(s, idx) + sector_num % s->tracks; 251 } 252 253 254 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs) 255 { 256 BDRVParallelsState *s = bs->opaque; 257 unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block); 258 unsigned long bit; 259 260 qemu_co_mutex_lock(&s->lock); 261 262 bit = find_first_bit(s->bat_dirty_bmap, size); 263 while (bit < size) { 264 uint32_t off = bit * s->bat_dirty_block; 265 uint32_t to_write = s->bat_dirty_block; 266 int ret; 267 268 if (off + to_write > s->header_size) { 269 to_write = s->header_size - off; 270 } 271 ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off, 272 to_write); 273 if (ret < 0) { 274 qemu_co_mutex_unlock(&s->lock); 275 return ret; 276 } 277 bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1); 278 } 279 bitmap_zero(s->bat_dirty_bmap, size); 280 281 qemu_co_mutex_unlock(&s->lock); 282 return 0; 283 } 284 285 286 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs, 287 bool want_zero, 288 int64_t offset, 289 int64_t bytes, 290 int64_t *pnum, 291 int64_t *map, 292 BlockDriverState **file) 293 { 294 BDRVParallelsState *s = bs->opaque; 295 int count; 296 297 assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE)); 298 qemu_co_mutex_lock(&s->lock); 299 offset = block_status(s, offset >> BDRV_SECTOR_BITS, 300 bytes >> BDRV_SECTOR_BITS, &count); 301 qemu_co_mutex_unlock(&s->lock); 302 303 *pnum = count * BDRV_SECTOR_SIZE; 304 if (offset < 0) { 305 return 0; 306 } 307 308 *map = offset * BDRV_SECTOR_SIZE; 309 *file = bs->file->bs; 310 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 311 } 312 313 static coroutine_fn int parallels_co_writev(BlockDriverState *bs, 314 int64_t sector_num, int nb_sectors, 315 QEMUIOVector *qiov, int flags) 316 { 317 BDRVParallelsState *s = bs->opaque; 318 uint64_t bytes_done = 0; 319 QEMUIOVector hd_qiov; 320 int ret = 0; 321 322 assert(!flags); 323 qemu_iovec_init(&hd_qiov, qiov->niov); 324 325 while (nb_sectors > 0) { 326 int64_t position; 327 int n, nbytes; 328 329 qemu_co_mutex_lock(&s->lock); 330 position = allocate_clusters(bs, sector_num, nb_sectors, &n); 331 qemu_co_mutex_unlock(&s->lock); 332 if (position < 0) { 333 ret = (int)position; 334 break; 335 } 336 337 nbytes = n << BDRV_SECTOR_BITS; 338 339 qemu_iovec_reset(&hd_qiov); 340 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes); 341 342 ret = bdrv_co_writev(bs->file, position, n, &hd_qiov); 343 if (ret < 0) { 344 break; 345 } 346 347 nb_sectors -= n; 348 sector_num += n; 349 bytes_done += nbytes; 350 } 351 352 qemu_iovec_destroy(&hd_qiov); 353 return ret; 354 } 355 356 static coroutine_fn int parallels_co_readv(BlockDriverState *bs, 357 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) 358 { 359 BDRVParallelsState *s = bs->opaque; 360 uint64_t bytes_done = 0; 361 QEMUIOVector hd_qiov; 362 int ret = 0; 363 364 qemu_iovec_init(&hd_qiov, qiov->niov); 365 366 while (nb_sectors > 0) { 367 int64_t position; 368 int n, nbytes; 369 370 qemu_co_mutex_lock(&s->lock); 371 position = block_status(s, sector_num, nb_sectors, &n); 372 qemu_co_mutex_unlock(&s->lock); 373 374 nbytes = n << BDRV_SECTOR_BITS; 375 376 qemu_iovec_reset(&hd_qiov); 377 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes); 378 379 if (position < 0) { 380 if (bs->backing) { 381 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov); 382 if (ret < 0) { 383 break; 384 } 385 } else { 386 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes); 387 } 388 } else { 389 ret = bdrv_co_readv(bs->file, position, n, &hd_qiov); 390 if (ret < 0) { 391 break; 392 } 393 } 394 395 nb_sectors -= n; 396 sector_num += n; 397 bytes_done += nbytes; 398 } 399 400 qemu_iovec_destroy(&hd_qiov); 401 return ret; 402 } 403 404 405 static int coroutine_fn parallels_co_check(BlockDriverState *bs, 406 BdrvCheckResult *res, 407 BdrvCheckMode fix) 408 { 409 BDRVParallelsState *s = bs->opaque; 410 int64_t size, prev_off, high_off; 411 int ret; 412 uint32_t i; 413 bool flush_bat = false; 414 int cluster_size = s->tracks << BDRV_SECTOR_BITS; 415 416 size = bdrv_getlength(bs->file->bs); 417 if (size < 0) { 418 res->check_errors++; 419 return size; 420 } 421 422 qemu_co_mutex_lock(&s->lock); 423 if (s->header_unclean) { 424 fprintf(stderr, "%s image was not closed correctly\n", 425 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR"); 426 res->corruptions++; 427 if (fix & BDRV_FIX_ERRORS) { 428 /* parallels_close will do the job right */ 429 res->corruptions_fixed++; 430 s->header_unclean = false; 431 } 432 } 433 434 res->bfi.total_clusters = s->bat_size; 435 res->bfi.compressed_clusters = 0; /* compression is not supported */ 436 437 high_off = 0; 438 prev_off = 0; 439 for (i = 0; i < s->bat_size; i++) { 440 int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS; 441 if (off == 0) { 442 prev_off = 0; 443 continue; 444 } 445 446 /* cluster outside the image */ 447 if (off > size) { 448 fprintf(stderr, "%s cluster %u is outside image\n", 449 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i); 450 res->corruptions++; 451 if (fix & BDRV_FIX_ERRORS) { 452 prev_off = 0; 453 s->bat_bitmap[i] = 0; 454 res->corruptions_fixed++; 455 flush_bat = true; 456 continue; 457 } 458 } 459 460 res->bfi.allocated_clusters++; 461 if (off > high_off) { 462 high_off = off; 463 } 464 465 if (prev_off != 0 && (prev_off + cluster_size) != off) { 466 res->bfi.fragmented_clusters++; 467 } 468 prev_off = off; 469 } 470 471 ret = 0; 472 if (flush_bat) { 473 ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size); 474 if (ret < 0) { 475 res->check_errors++; 476 goto out; 477 } 478 } 479 480 res->image_end_offset = high_off + cluster_size; 481 if (size > res->image_end_offset) { 482 int64_t count; 483 count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size); 484 fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n", 485 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR", 486 size - res->image_end_offset); 487 res->leaks += count; 488 if (fix & BDRV_FIX_LEAKS) { 489 Error *local_err = NULL; 490 ret = bdrv_truncate(bs->file, res->image_end_offset, 491 PREALLOC_MODE_OFF, &local_err); 492 if (ret < 0) { 493 error_report_err(local_err); 494 res->check_errors++; 495 goto out; 496 } 497 res->leaks_fixed += count; 498 } 499 } 500 501 out: 502 qemu_co_mutex_unlock(&s->lock); 503 return ret; 504 } 505 506 507 static int coroutine_fn parallels_co_create(BlockdevCreateOptions* opts, 508 Error **errp) 509 { 510 BlockdevCreateOptionsParallels *parallels_opts; 511 BlockDriverState *bs; 512 BlockBackend *blk; 513 int64_t total_size, cl_size; 514 uint32_t bat_entries, bat_sectors; 515 ParallelsHeader header; 516 uint8_t tmp[BDRV_SECTOR_SIZE]; 517 int ret; 518 519 assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS); 520 parallels_opts = &opts->u.parallels; 521 522 /* Sanity checks */ 523 total_size = parallels_opts->size; 524 525 if (parallels_opts->has_cluster_size) { 526 cl_size = parallels_opts->cluster_size; 527 } else { 528 cl_size = DEFAULT_CLUSTER_SIZE; 529 } 530 531 /* XXX What is the real limit here? This is an insanely large maximum. */ 532 if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) { 533 error_setg(errp, "Cluster size is too large"); 534 return -EINVAL; 535 } 536 if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) { 537 error_setg(errp, "Image size is too large for this cluster size"); 538 return -E2BIG; 539 } 540 541 if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) { 542 error_setg(errp, "Image size must be a multiple of 512 bytes"); 543 return -EINVAL; 544 } 545 546 if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) { 547 error_setg(errp, "Cluster size must be a multiple of 512 bytes"); 548 return -EINVAL; 549 } 550 551 /* Create BlockBackend to write to the image */ 552 bs = bdrv_open_blockdev_ref(parallels_opts->file, errp); 553 if (bs == NULL) { 554 return -EIO; 555 } 556 557 blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); 558 ret = blk_insert_bs(blk, bs, errp); 559 if (ret < 0) { 560 goto out; 561 } 562 blk_set_allow_write_beyond_eof(blk, true); 563 564 /* Create image format */ 565 ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp); 566 if (ret < 0) { 567 goto out; 568 } 569 570 bat_entries = DIV_ROUND_UP(total_size, cl_size); 571 bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size); 572 bat_sectors = (bat_sectors * cl_size) >> BDRV_SECTOR_BITS; 573 574 memset(&header, 0, sizeof(header)); 575 memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic)); 576 header.version = cpu_to_le32(HEADER_VERSION); 577 /* don't care much about geometry, it is not used on image level */ 578 header.heads = cpu_to_le32(HEADS_NUMBER); 579 header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE 580 / HEADS_NUMBER / SEC_IN_CYL); 581 header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS); 582 header.bat_entries = cpu_to_le32(bat_entries); 583 header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE)); 584 header.data_off = cpu_to_le32(bat_sectors); 585 586 /* write all the data */ 587 memset(tmp, 0, sizeof(tmp)); 588 memcpy(tmp, &header, sizeof(header)); 589 590 ret = blk_pwrite(blk, 0, tmp, BDRV_SECTOR_SIZE, 0); 591 if (ret < 0) { 592 goto exit; 593 } 594 ret = blk_pwrite_zeroes(blk, BDRV_SECTOR_SIZE, 595 (bat_sectors - 1) << BDRV_SECTOR_BITS, 0); 596 if (ret < 0) { 597 goto exit; 598 } 599 600 ret = 0; 601 out: 602 blk_unref(blk); 603 bdrv_unref(bs); 604 return ret; 605 606 exit: 607 error_setg_errno(errp, -ret, "Failed to create Parallels image"); 608 goto out; 609 } 610 611 static int coroutine_fn parallels_co_create_opts(const char *filename, 612 QemuOpts *opts, 613 Error **errp) 614 { 615 BlockdevCreateOptions *create_options = NULL; 616 Error *local_err = NULL; 617 BlockDriverState *bs = NULL; 618 QDict *qdict = NULL; 619 QObject *qobj; 620 Visitor *v; 621 int ret; 622 623 static const QDictRenames opt_renames[] = { 624 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, 625 { NULL, NULL }, 626 }; 627 628 /* Parse options and convert legacy syntax */ 629 qdict = qemu_opts_to_qdict_filtered(opts, NULL, ¶llels_create_opts, 630 true); 631 632 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 633 ret = -EINVAL; 634 goto done; 635 } 636 637 /* Create and open the file (protocol layer) */ 638 ret = bdrv_create_file(filename, opts, &local_err); 639 if (ret < 0) { 640 error_propagate(errp, local_err); 641 goto done; 642 } 643 644 bs = bdrv_open(filename, NULL, NULL, 645 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 646 if (bs == NULL) { 647 ret = -EIO; 648 goto done; 649 } 650 651 /* Now get the QAPI type BlockdevCreateOptions */ 652 qdict_put_str(qdict, "driver", "parallels"); 653 qdict_put_str(qdict, "file", bs->node_name); 654 655 qobj = qdict_crumple(qdict, errp); 656 qobject_unref(qdict); 657 qdict = qobject_to(QDict, qobj); 658 if (qdict == NULL) { 659 ret = -EINVAL; 660 goto done; 661 } 662 663 v = qobject_input_visitor_new_keyval(QOBJECT(qdict)); 664 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); 665 visit_free(v); 666 667 if (local_err) { 668 error_propagate(errp, local_err); 669 ret = -EINVAL; 670 goto done; 671 } 672 673 /* Silently round up sizes */ 674 create_options->u.parallels.size = 675 ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE); 676 create_options->u.parallels.cluster_size = 677 ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE); 678 679 /* Create the Parallels image (format layer) */ 680 ret = parallels_co_create(create_options, errp); 681 if (ret < 0) { 682 goto done; 683 } 684 ret = 0; 685 686 done: 687 qobject_unref(qdict); 688 bdrv_unref(bs); 689 qapi_free_BlockdevCreateOptions(create_options); 690 return ret; 691 } 692 693 694 static int parallels_probe(const uint8_t *buf, int buf_size, 695 const char *filename) 696 { 697 const ParallelsHeader *ph = (const void *)buf; 698 699 if (buf_size < sizeof(ParallelsHeader)) { 700 return 0; 701 } 702 703 if ((!memcmp(ph->magic, HEADER_MAGIC, 16) || 704 !memcmp(ph->magic, HEADER_MAGIC2, 16)) && 705 (le32_to_cpu(ph->version) == HEADER_VERSION)) { 706 return 100; 707 } 708 709 return 0; 710 } 711 712 static int parallels_update_header(BlockDriverState *bs) 713 { 714 BDRVParallelsState *s = bs->opaque; 715 unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs), 716 sizeof(ParallelsHeader)); 717 718 if (size > s->header_size) { 719 size = s->header_size; 720 } 721 return bdrv_pwrite_sync(bs->file, 0, s->header, size); 722 } 723 724 static int parallels_open(BlockDriverState *bs, QDict *options, int flags, 725 Error **errp) 726 { 727 BDRVParallelsState *s = bs->opaque; 728 ParallelsHeader ph; 729 int ret, size, i; 730 QemuOpts *opts = NULL; 731 Error *local_err = NULL; 732 char *buf; 733 734 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 735 false, errp); 736 if (!bs->file) { 737 return -EINVAL; 738 } 739 740 ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph)); 741 if (ret < 0) { 742 goto fail; 743 } 744 745 bs->total_sectors = le64_to_cpu(ph.nb_sectors); 746 747 if (le32_to_cpu(ph.version) != HEADER_VERSION) { 748 goto fail_format; 749 } 750 if (!memcmp(ph.magic, HEADER_MAGIC, 16)) { 751 s->off_multiplier = 1; 752 bs->total_sectors = 0xffffffff & bs->total_sectors; 753 } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) { 754 s->off_multiplier = le32_to_cpu(ph.tracks); 755 } else { 756 goto fail_format; 757 } 758 759 s->tracks = le32_to_cpu(ph.tracks); 760 if (s->tracks == 0) { 761 error_setg(errp, "Invalid image: Zero sectors per track"); 762 ret = -EINVAL; 763 goto fail; 764 } 765 if (s->tracks > INT32_MAX/513) { 766 error_setg(errp, "Invalid image: Too big cluster"); 767 ret = -EFBIG; 768 goto fail; 769 } 770 771 s->bat_size = le32_to_cpu(ph.bat_entries); 772 if (s->bat_size > INT_MAX / sizeof(uint32_t)) { 773 error_setg(errp, "Catalog too large"); 774 ret = -EFBIG; 775 goto fail; 776 } 777 778 size = bat_entry_off(s->bat_size); 779 s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs)); 780 s->header = qemu_try_blockalign(bs->file->bs, s->header_size); 781 if (s->header == NULL) { 782 ret = -ENOMEM; 783 goto fail; 784 } 785 s->data_end = le32_to_cpu(ph.data_off); 786 if (s->data_end == 0) { 787 s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE); 788 } 789 if (s->data_end < s->header_size) { 790 /* there is not enough unused space to fit to block align between BAT 791 and actual data. We can't avoid read-modify-write... */ 792 s->header_size = size; 793 } 794 795 ret = bdrv_pread(bs->file, 0, s->header, s->header_size); 796 if (ret < 0) { 797 goto fail; 798 } 799 s->bat_bitmap = (uint32_t *)(s->header + 1); 800 801 for (i = 0; i < s->bat_size; i++) { 802 int64_t off = bat2sect(s, i); 803 if (off >= s->data_end) { 804 s->data_end = off + s->tracks; 805 } 806 } 807 808 if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) { 809 /* Image was not closed correctly. The check is mandatory */ 810 s->header_unclean = true; 811 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 812 error_setg(errp, "parallels: Image was not closed correctly; " 813 "cannot be opened read/write"); 814 ret = -EACCES; 815 goto fail; 816 } 817 } 818 819 opts = qemu_opts_create(¶llels_runtime_opts, NULL, 0, &local_err); 820 if (local_err != NULL) { 821 goto fail_options; 822 } 823 824 qemu_opts_absorb_qdict(opts, options, &local_err); 825 if (local_err != NULL) { 826 goto fail_options; 827 } 828 829 s->prealloc_size = 830 qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0); 831 s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS); 832 buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE); 833 s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf, 834 PRL_PREALLOC_MODE_FALLOCATE, 835 &local_err); 836 g_free(buf); 837 if (local_err != NULL) { 838 goto fail_options; 839 } 840 841 if (!bdrv_has_zero_init(bs->file->bs)) { 842 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE; 843 } 844 845 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) { 846 s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC); 847 ret = parallels_update_header(bs); 848 if (ret < 0) { 849 goto fail; 850 } 851 } 852 853 s->bat_dirty_block = 4 * getpagesize(); 854 s->bat_dirty_bmap = 855 bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block)); 856 857 /* Disable migration until bdrv_invalidate_cache method is added */ 858 error_setg(&s->migration_blocker, "The Parallels format used by node '%s' " 859 "does not support live migration", 860 bdrv_get_device_or_node_name(bs)); 861 ret = migrate_add_blocker(s->migration_blocker, &local_err); 862 if (local_err) { 863 error_propagate(errp, local_err); 864 error_free(s->migration_blocker); 865 goto fail; 866 } 867 qemu_co_mutex_init(&s->lock); 868 return 0; 869 870 fail_format: 871 error_setg(errp, "Image not in Parallels format"); 872 ret = -EINVAL; 873 fail: 874 qemu_vfree(s->header); 875 return ret; 876 877 fail_options: 878 error_propagate(errp, local_err); 879 ret = -EINVAL; 880 goto fail; 881 } 882 883 884 static void parallels_close(BlockDriverState *bs) 885 { 886 BDRVParallelsState *s = bs->opaque; 887 888 if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) { 889 s->header->inuse = 0; 890 parallels_update_header(bs); 891 bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, 892 PREALLOC_MODE_OFF, NULL); 893 } 894 895 g_free(s->bat_dirty_bmap); 896 qemu_vfree(s->header); 897 898 migrate_del_blocker(s->migration_blocker); 899 error_free(s->migration_blocker); 900 } 901 902 static BlockDriver bdrv_parallels = { 903 .format_name = "parallels", 904 .instance_size = sizeof(BDRVParallelsState), 905 .bdrv_probe = parallels_probe, 906 .bdrv_open = parallels_open, 907 .bdrv_close = parallels_close, 908 .bdrv_child_perm = bdrv_format_default_perms, 909 .bdrv_co_block_status = parallels_co_block_status, 910 .bdrv_has_zero_init = bdrv_has_zero_init_1, 911 .bdrv_co_flush_to_os = parallels_co_flush_to_os, 912 .bdrv_co_readv = parallels_co_readv, 913 .bdrv_co_writev = parallels_co_writev, 914 .supports_backing = true, 915 .bdrv_co_create = parallels_co_create, 916 .bdrv_co_create_opts = parallels_co_create_opts, 917 .bdrv_co_check = parallels_co_check, 918 .create_opts = ¶llels_create_opts, 919 }; 920 921 static void bdrv_parallels_init(void) 922 { 923 bdrv_register(&bdrv_parallels); 924 } 925 926 block_init(bdrv_parallels_init); 927