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 "qemu/error-report.h" 33 #include "qapi/error.h" 34 #include "block/block_int.h" 35 #include "block/qdict.h" 36 #include "sysemu/block-backend.h" 37 #include "qemu/module.h" 38 #include "qemu/option.h" 39 #include "qapi/qmp/qdict.h" 40 #include "qapi/qobject-input-visitor.h" 41 #include "qapi/qapi-visit-block-core.h" 42 #include "qemu/bswap.h" 43 #include "qemu/bitmap.h" 44 #include "qemu/memalign.h" 45 #include "migration/blocker.h" 46 #include "parallels.h" 47 48 /**************************************************************/ 49 50 #define HEADER_MAGIC "WithoutFreeSpace" 51 #define HEADER_MAGIC2 "WithouFreSpacExt" 52 #define HEADER_VERSION 2 53 #define HEADER_INUSE_MAGIC (0x746F6E59) 54 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32) 55 56 static QEnumLookup prealloc_mode_lookup = { 57 .array = (const char *const[]) { 58 "falloc", 59 "truncate", 60 }, 61 .size = PRL_PREALLOC_MODE__MAX 62 }; 63 64 #define PARALLELS_OPT_PREALLOC_MODE "prealloc-mode" 65 #define PARALLELS_OPT_PREALLOC_SIZE "prealloc-size" 66 67 static QemuOptsList parallels_runtime_opts = { 68 .name = "parallels", 69 .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head), 70 .desc = { 71 { 72 .name = PARALLELS_OPT_PREALLOC_SIZE, 73 .type = QEMU_OPT_SIZE, 74 .help = "Preallocation size on image expansion", 75 .def_value_str = "128M", 76 }, 77 { 78 .name = PARALLELS_OPT_PREALLOC_MODE, 79 .type = QEMU_OPT_STRING, 80 .help = "Preallocation mode on image expansion " 81 "(allowed values: falloc, truncate)", 82 .def_value_str = "falloc", 83 }, 84 { /* end of list */ }, 85 }, 86 }; 87 88 static QemuOptsList parallels_create_opts = { 89 .name = "parallels-create-opts", 90 .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head), 91 .desc = { 92 { 93 .name = BLOCK_OPT_SIZE, 94 .type = QEMU_OPT_SIZE, 95 .help = "Virtual disk size", 96 }, 97 { 98 .name = BLOCK_OPT_CLUSTER_SIZE, 99 .type = QEMU_OPT_SIZE, 100 .help = "Parallels image cluster size", 101 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE), 102 }, 103 { /* end of list */ } 104 } 105 }; 106 107 108 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx) 109 { 110 return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier; 111 } 112 113 static uint32_t bat_entry_off(uint32_t idx) 114 { 115 return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx; 116 } 117 118 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num) 119 { 120 uint32_t index, offset; 121 122 index = sector_num / s->tracks; 123 offset = sector_num % s->tracks; 124 125 /* not allocated */ 126 if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) { 127 return -1; 128 } 129 return bat2sect(s, index) + offset; 130 } 131 132 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num, 133 int nb_sectors) 134 { 135 int ret = s->tracks - sector_num % s->tracks; 136 return MIN(nb_sectors, ret); 137 } 138 139 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num, 140 int nb_sectors, int *pnum) 141 { 142 int64_t start_off = -2, prev_end_off = -2; 143 144 *pnum = 0; 145 while (nb_sectors > 0 || start_off == -2) { 146 int64_t offset = seek_to_sector(s, sector_num); 147 int to_end; 148 149 if (start_off == -2) { 150 start_off = offset; 151 prev_end_off = offset; 152 } else if (offset != prev_end_off) { 153 break; 154 } 155 156 to_end = cluster_remainder(s, sector_num, nb_sectors); 157 nb_sectors -= to_end; 158 sector_num += to_end; 159 *pnum += to_end; 160 161 if (offset > 0) { 162 prev_end_off += to_end; 163 } 164 } 165 return start_off; 166 } 167 168 static coroutine_fn int64_t allocate_clusters(BlockDriverState *bs, 169 int64_t sector_num, 170 int nb_sectors, int *pnum) 171 { 172 int ret = 0; 173 BDRVParallelsState *s = bs->opaque; 174 int64_t pos, space, idx, to_allocate, i, len; 175 176 pos = block_status(s, sector_num, nb_sectors, pnum); 177 if (pos > 0) { 178 return pos; 179 } 180 181 idx = sector_num / s->tracks; 182 to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx; 183 184 /* This function is called only by parallels_co_writev(), which will never 185 * pass a sector_num at or beyond the end of the image (because the block 186 * layer never passes such a sector_num to that function). Therefore, idx 187 * is always below s->bat_size. 188 * block_status() will limit *pnum so that sector_num + *pnum will not 189 * exceed the image end. Therefore, idx + to_allocate cannot exceed 190 * s->bat_size. 191 * Note that s->bat_size is an unsigned int, therefore idx + to_allocate 192 * will always fit into a uint32_t. */ 193 assert(idx < s->bat_size && idx + to_allocate <= s->bat_size); 194 195 space = to_allocate * s->tracks; 196 len = bdrv_getlength(bs->file->bs); 197 if (len < 0) { 198 return len; 199 } 200 if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) { 201 space += s->prealloc_size; 202 /* 203 * We require the expanded size to read back as zero. If the 204 * user permitted truncation, we try that; but if it fails, we 205 * force the safer-but-slower fallocate. 206 */ 207 if (s->prealloc_mode == PRL_PREALLOC_MODE_TRUNCATE) { 208 ret = bdrv_truncate(bs->file, 209 (s->data_end + space) << BDRV_SECTOR_BITS, 210 false, PREALLOC_MODE_OFF, BDRV_REQ_ZERO_WRITE, 211 NULL); 212 if (ret == -ENOTSUP) { 213 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE; 214 } 215 } 216 if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) { 217 ret = bdrv_pwrite_zeroes(bs->file, 218 s->data_end << BDRV_SECTOR_BITS, 219 space << BDRV_SECTOR_BITS, 0); 220 } 221 if (ret < 0) { 222 return ret; 223 } 224 } 225 226 /* Try to read from backing to fill empty clusters 227 * FIXME: 1. previous write_zeroes may be redundant 228 * 2. most of data we read from backing will be rewritten by 229 * parallels_co_writev. On aligned-to-cluster write we do not need 230 * this read at all. 231 * 3. it would be good to combine write of data from backing and new 232 * data into one write call */ 233 if (bs->backing) { 234 int64_t nb_cow_sectors = to_allocate * s->tracks; 235 int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS; 236 void *buf = qemu_blockalign(bs, nb_cow_bytes); 237 238 ret = bdrv_co_pread(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE, 239 nb_cow_bytes, buf, 0); 240 if (ret < 0) { 241 qemu_vfree(buf); 242 return ret; 243 } 244 245 ret = bdrv_co_pwrite(bs->file, s->data_end * BDRV_SECTOR_SIZE, 246 nb_cow_bytes, buf, 0); 247 qemu_vfree(buf); 248 if (ret < 0) { 249 return ret; 250 } 251 } 252 253 for (i = 0; i < to_allocate; i++) { 254 s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier); 255 s->data_end += s->tracks; 256 bitmap_set(s->bat_dirty_bmap, 257 bat_entry_off(idx + i) / s->bat_dirty_block, 1); 258 } 259 260 return bat2sect(s, idx) + sector_num % s->tracks; 261 } 262 263 264 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs) 265 { 266 BDRVParallelsState *s = bs->opaque; 267 unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block); 268 unsigned long bit; 269 270 qemu_co_mutex_lock(&s->lock); 271 272 bit = find_first_bit(s->bat_dirty_bmap, size); 273 while (bit < size) { 274 uint32_t off = bit * s->bat_dirty_block; 275 uint32_t to_write = s->bat_dirty_block; 276 int ret; 277 278 if (off + to_write > s->header_size) { 279 to_write = s->header_size - off; 280 } 281 ret = bdrv_pwrite(bs->file, off, to_write, (uint8_t *)s->header + off, 282 0); 283 if (ret < 0) { 284 qemu_co_mutex_unlock(&s->lock); 285 return ret; 286 } 287 bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1); 288 } 289 bitmap_zero(s->bat_dirty_bmap, size); 290 291 qemu_co_mutex_unlock(&s->lock); 292 return 0; 293 } 294 295 296 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs, 297 bool want_zero, 298 int64_t offset, 299 int64_t bytes, 300 int64_t *pnum, 301 int64_t *map, 302 BlockDriverState **file) 303 { 304 BDRVParallelsState *s = bs->opaque; 305 int count; 306 307 assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE)); 308 qemu_co_mutex_lock(&s->lock); 309 offset = block_status(s, offset >> BDRV_SECTOR_BITS, 310 bytes >> BDRV_SECTOR_BITS, &count); 311 qemu_co_mutex_unlock(&s->lock); 312 313 *pnum = count * BDRV_SECTOR_SIZE; 314 if (offset < 0) { 315 return 0; 316 } 317 318 *map = offset * BDRV_SECTOR_SIZE; 319 *file = bs->file->bs; 320 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 321 } 322 323 static coroutine_fn int parallels_co_writev(BlockDriverState *bs, 324 int64_t sector_num, int nb_sectors, 325 QEMUIOVector *qiov, int flags) 326 { 327 BDRVParallelsState *s = bs->opaque; 328 uint64_t bytes_done = 0; 329 QEMUIOVector hd_qiov; 330 int ret = 0; 331 332 assert(!flags); 333 qemu_iovec_init(&hd_qiov, qiov->niov); 334 335 while (nb_sectors > 0) { 336 int64_t position; 337 int n, nbytes; 338 339 qemu_co_mutex_lock(&s->lock); 340 position = allocate_clusters(bs, sector_num, nb_sectors, &n); 341 qemu_co_mutex_unlock(&s->lock); 342 if (position < 0) { 343 ret = (int)position; 344 break; 345 } 346 347 nbytes = n << BDRV_SECTOR_BITS; 348 349 qemu_iovec_reset(&hd_qiov); 350 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes); 351 352 ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes, 353 &hd_qiov, 0); 354 if (ret < 0) { 355 break; 356 } 357 358 nb_sectors -= n; 359 sector_num += n; 360 bytes_done += nbytes; 361 } 362 363 qemu_iovec_destroy(&hd_qiov); 364 return ret; 365 } 366 367 static coroutine_fn int parallels_co_readv(BlockDriverState *bs, 368 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) 369 { 370 BDRVParallelsState *s = bs->opaque; 371 uint64_t bytes_done = 0; 372 QEMUIOVector hd_qiov; 373 int ret = 0; 374 375 qemu_iovec_init(&hd_qiov, qiov->niov); 376 377 while (nb_sectors > 0) { 378 int64_t position; 379 int n, nbytes; 380 381 qemu_co_mutex_lock(&s->lock); 382 position = block_status(s, sector_num, nb_sectors, &n); 383 qemu_co_mutex_unlock(&s->lock); 384 385 nbytes = n << BDRV_SECTOR_BITS; 386 387 qemu_iovec_reset(&hd_qiov); 388 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes); 389 390 if (position < 0) { 391 if (bs->backing) { 392 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE, 393 nbytes, &hd_qiov, 0); 394 if (ret < 0) { 395 break; 396 } 397 } else { 398 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes); 399 } 400 } else { 401 ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes, 402 &hd_qiov, 0); 403 if (ret < 0) { 404 break; 405 } 406 } 407 408 nb_sectors -= n; 409 sector_num += n; 410 bytes_done += nbytes; 411 } 412 413 qemu_iovec_destroy(&hd_qiov); 414 return ret; 415 } 416 417 418 static int coroutine_fn parallels_co_check(BlockDriverState *bs, 419 BdrvCheckResult *res, 420 BdrvCheckMode fix) 421 { 422 BDRVParallelsState *s = bs->opaque; 423 int64_t size, prev_off, high_off; 424 int ret; 425 uint32_t i; 426 bool flush_bat = false; 427 428 size = bdrv_getlength(bs->file->bs); 429 if (size < 0) { 430 res->check_errors++; 431 return size; 432 } 433 434 qemu_co_mutex_lock(&s->lock); 435 if (s->header_unclean) { 436 fprintf(stderr, "%s image was not closed correctly\n", 437 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR"); 438 res->corruptions++; 439 if (fix & BDRV_FIX_ERRORS) { 440 /* parallels_close will do the job right */ 441 res->corruptions_fixed++; 442 s->header_unclean = false; 443 } 444 } 445 446 res->bfi.total_clusters = s->bat_size; 447 res->bfi.compressed_clusters = 0; /* compression is not supported */ 448 449 high_off = 0; 450 prev_off = 0; 451 for (i = 0; i < s->bat_size; i++) { 452 int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS; 453 if (off == 0) { 454 prev_off = 0; 455 continue; 456 } 457 458 /* cluster outside the image */ 459 if (off > size) { 460 fprintf(stderr, "%s cluster %u is outside image\n", 461 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i); 462 res->corruptions++; 463 if (fix & BDRV_FIX_ERRORS) { 464 prev_off = 0; 465 s->bat_bitmap[i] = 0; 466 res->corruptions_fixed++; 467 flush_bat = true; 468 continue; 469 } 470 } 471 472 res->bfi.allocated_clusters++; 473 if (off > high_off) { 474 high_off = off; 475 } 476 477 if (prev_off != 0 && (prev_off + s->cluster_size) != off) { 478 res->bfi.fragmented_clusters++; 479 } 480 prev_off = off; 481 } 482 483 ret = 0; 484 if (flush_bat) { 485 ret = bdrv_co_pwrite_sync(bs->file, 0, s->header_size, s->header, 0); 486 if (ret < 0) { 487 res->check_errors++; 488 goto out; 489 } 490 } 491 492 res->image_end_offset = high_off + s->cluster_size; 493 if (size > res->image_end_offset) { 494 int64_t count; 495 count = DIV_ROUND_UP(size - res->image_end_offset, s->cluster_size); 496 fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n", 497 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR", 498 size - res->image_end_offset); 499 res->leaks += count; 500 if (fix & BDRV_FIX_LEAKS) { 501 Error *local_err = NULL; 502 503 /* 504 * In order to really repair the image, we must shrink it. 505 * That means we have to pass exact=true. 506 */ 507 ret = bdrv_truncate(bs->file, res->image_end_offset, true, 508 PREALLOC_MODE_OFF, 0, &local_err); 509 if (ret < 0) { 510 error_report_err(local_err); 511 res->check_errors++; 512 goto out; 513 } 514 res->leaks_fixed += count; 515 } 516 } 517 518 out: 519 qemu_co_mutex_unlock(&s->lock); 520 return ret; 521 } 522 523 524 static int coroutine_fn parallels_co_create(BlockdevCreateOptions* opts, 525 Error **errp) 526 { 527 BlockdevCreateOptionsParallels *parallels_opts; 528 BlockDriverState *bs; 529 BlockBackend *blk; 530 int64_t total_size, cl_size; 531 uint32_t bat_entries, bat_sectors; 532 ParallelsHeader header; 533 uint8_t tmp[BDRV_SECTOR_SIZE]; 534 int ret; 535 536 assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS); 537 parallels_opts = &opts->u.parallels; 538 539 /* Sanity checks */ 540 total_size = parallels_opts->size; 541 542 if (parallels_opts->has_cluster_size) { 543 cl_size = parallels_opts->cluster_size; 544 } else { 545 cl_size = DEFAULT_CLUSTER_SIZE; 546 } 547 548 /* XXX What is the real limit here? This is an insanely large maximum. */ 549 if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) { 550 error_setg(errp, "Cluster size is too large"); 551 return -EINVAL; 552 } 553 if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) { 554 error_setg(errp, "Image size is too large for this cluster size"); 555 return -E2BIG; 556 } 557 558 if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) { 559 error_setg(errp, "Image size must be a multiple of 512 bytes"); 560 return -EINVAL; 561 } 562 563 if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) { 564 error_setg(errp, "Cluster size must be a multiple of 512 bytes"); 565 return -EINVAL; 566 } 567 568 /* Create BlockBackend to write to the image */ 569 bs = bdrv_open_blockdev_ref(parallels_opts->file, errp); 570 if (bs == NULL) { 571 return -EIO; 572 } 573 574 blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, 575 errp); 576 if (!blk) { 577 ret = -EPERM; 578 goto out; 579 } 580 blk_set_allow_write_beyond_eof(blk, true); 581 582 /* Create image format */ 583 bat_entries = DIV_ROUND_UP(total_size, cl_size); 584 bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size); 585 bat_sectors = (bat_sectors * cl_size) >> BDRV_SECTOR_BITS; 586 587 memset(&header, 0, sizeof(header)); 588 memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic)); 589 header.version = cpu_to_le32(HEADER_VERSION); 590 /* don't care much about geometry, it is not used on image level */ 591 header.heads = cpu_to_le32(HEADS_NUMBER); 592 header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE 593 / HEADS_NUMBER / SEC_IN_CYL); 594 header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS); 595 header.bat_entries = cpu_to_le32(bat_entries); 596 header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE)); 597 header.data_off = cpu_to_le32(bat_sectors); 598 599 /* write all the data */ 600 memset(tmp, 0, sizeof(tmp)); 601 memcpy(tmp, &header, sizeof(header)); 602 603 ret = blk_pwrite(blk, 0, BDRV_SECTOR_SIZE, tmp, 0); 604 if (ret < 0) { 605 goto exit; 606 } 607 ret = blk_pwrite_zeroes(blk, BDRV_SECTOR_SIZE, 608 (bat_sectors - 1) << BDRV_SECTOR_BITS, 0); 609 if (ret < 0) { 610 goto exit; 611 } 612 613 ret = 0; 614 out: 615 blk_unref(blk); 616 bdrv_unref(bs); 617 return ret; 618 619 exit: 620 error_setg_errno(errp, -ret, "Failed to create Parallels image"); 621 goto out; 622 } 623 624 static int coroutine_fn parallels_co_create_opts(BlockDriver *drv, 625 const char *filename, 626 QemuOpts *opts, 627 Error **errp) 628 { 629 BlockdevCreateOptions *create_options = NULL; 630 BlockDriverState *bs = NULL; 631 QDict *qdict; 632 Visitor *v; 633 int ret; 634 635 static const QDictRenames opt_renames[] = { 636 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, 637 { NULL, NULL }, 638 }; 639 640 /* Parse options and convert legacy syntax */ 641 qdict = qemu_opts_to_qdict_filtered(opts, NULL, ¶llels_create_opts, 642 true); 643 644 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 645 ret = -EINVAL; 646 goto done; 647 } 648 649 /* Create and open the file (protocol layer) */ 650 ret = bdrv_create_file(filename, opts, errp); 651 if (ret < 0) { 652 goto done; 653 } 654 655 bs = bdrv_open(filename, NULL, NULL, 656 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 657 if (bs == NULL) { 658 ret = -EIO; 659 goto done; 660 } 661 662 /* Now get the QAPI type BlockdevCreateOptions */ 663 qdict_put_str(qdict, "driver", "parallels"); 664 qdict_put_str(qdict, "file", bs->node_name); 665 666 v = qobject_input_visitor_new_flat_confused(qdict, errp); 667 if (!v) { 668 ret = -EINVAL; 669 goto done; 670 } 671 672 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 673 visit_free(v); 674 if (!create_options) { 675 ret = -EINVAL; 676 goto done; 677 } 678 679 /* Silently round up sizes */ 680 create_options->u.parallels.size = 681 ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE); 682 create_options->u.parallels.cluster_size = 683 ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE); 684 685 /* Create the Parallels image (format layer) */ 686 ret = parallels_co_create(create_options, errp); 687 if (ret < 0) { 688 goto done; 689 } 690 ret = 0; 691 692 done: 693 qobject_unref(qdict); 694 bdrv_unref(bs); 695 qapi_free_BlockdevCreateOptions(create_options); 696 return ret; 697 } 698 699 700 static int parallels_probe(const uint8_t *buf, int buf_size, 701 const char *filename) 702 { 703 const ParallelsHeader *ph = (const void *)buf; 704 705 if (buf_size < sizeof(ParallelsHeader)) { 706 return 0; 707 } 708 709 if ((!memcmp(ph->magic, HEADER_MAGIC, 16) || 710 !memcmp(ph->magic, HEADER_MAGIC2, 16)) && 711 (le32_to_cpu(ph->version) == HEADER_VERSION)) { 712 return 100; 713 } 714 715 return 0; 716 } 717 718 static int parallels_update_header(BlockDriverState *bs) 719 { 720 BDRVParallelsState *s = bs->opaque; 721 unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs), 722 sizeof(ParallelsHeader)); 723 724 if (size > s->header_size) { 725 size = s->header_size; 726 } 727 return bdrv_pwrite_sync(bs->file, 0, size, s->header, 0); 728 } 729 730 static int parallels_open(BlockDriverState *bs, QDict *options, int flags, 731 Error **errp) 732 { 733 BDRVParallelsState *s = bs->opaque; 734 ParallelsHeader ph; 735 int ret, size, i; 736 QemuOpts *opts = NULL; 737 Error *local_err = NULL; 738 char *buf; 739 740 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, 741 BDRV_CHILD_IMAGE, false, errp); 742 if (!bs->file) { 743 return -EINVAL; 744 } 745 746 ret = bdrv_pread(bs->file, 0, sizeof(ph), &ph, 0); 747 if (ret < 0) { 748 goto fail; 749 } 750 751 bs->total_sectors = le64_to_cpu(ph.nb_sectors); 752 753 if (le32_to_cpu(ph.version) != HEADER_VERSION) { 754 goto fail_format; 755 } 756 if (!memcmp(ph.magic, HEADER_MAGIC, 16)) { 757 s->off_multiplier = 1; 758 bs->total_sectors = 0xffffffff & bs->total_sectors; 759 } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) { 760 s->off_multiplier = le32_to_cpu(ph.tracks); 761 } else { 762 goto fail_format; 763 } 764 765 s->tracks = le32_to_cpu(ph.tracks); 766 if (s->tracks == 0) { 767 error_setg(errp, "Invalid image: Zero sectors per track"); 768 ret = -EINVAL; 769 goto fail; 770 } 771 if (s->tracks > INT32_MAX/513) { 772 error_setg(errp, "Invalid image: Too big cluster"); 773 ret = -EFBIG; 774 goto fail; 775 } 776 s->cluster_size = s->tracks << BDRV_SECTOR_BITS; 777 778 s->bat_size = le32_to_cpu(ph.bat_entries); 779 if (s->bat_size > INT_MAX / sizeof(uint32_t)) { 780 error_setg(errp, "Catalog too large"); 781 ret = -EFBIG; 782 goto fail; 783 } 784 785 size = bat_entry_off(s->bat_size); 786 s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs)); 787 s->header = qemu_try_blockalign(bs->file->bs, s->header_size); 788 if (s->header == NULL) { 789 ret = -ENOMEM; 790 goto fail; 791 } 792 s->data_end = le32_to_cpu(ph.data_off); 793 if (s->data_end == 0) { 794 s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE); 795 } 796 if (s->data_end < s->header_size) { 797 /* there is not enough unused space to fit to block align between BAT 798 and actual data. We can't avoid read-modify-write... */ 799 s->header_size = size; 800 } 801 802 ret = bdrv_pread(bs->file, 0, s->header_size, s->header, 0); 803 if (ret < 0) { 804 goto fail; 805 } 806 s->bat_bitmap = (uint32_t *)(s->header + 1); 807 808 for (i = 0; i < s->bat_size; i++) { 809 int64_t off = bat2sect(s, i); 810 if (off >= s->data_end) { 811 s->data_end = off + s->tracks; 812 } 813 } 814 815 if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) { 816 /* Image was not closed correctly. The check is mandatory */ 817 s->header_unclean = true; 818 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 819 error_setg(errp, "parallels: Image was not closed correctly; " 820 "cannot be opened read/write"); 821 ret = -EACCES; 822 goto fail; 823 } 824 } 825 826 opts = qemu_opts_create(¶llels_runtime_opts, NULL, 0, errp); 827 if (!opts) { 828 goto fail_options; 829 } 830 831 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 832 goto fail_options; 833 } 834 835 s->prealloc_size = 836 qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0); 837 s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS); 838 buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE); 839 /* prealloc_mode can be downgraded later during allocate_clusters */ 840 s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf, 841 PRL_PREALLOC_MODE_FALLOCATE, 842 &local_err); 843 g_free(buf); 844 if (local_err != NULL) { 845 error_propagate(errp, local_err); 846 goto fail_options; 847 } 848 849 if (ph.ext_off) { 850 if (flags & BDRV_O_RDWR) { 851 /* 852 * It's unsafe to open image RW if there is an extension (as we 853 * don't support it). But parallels driver in QEMU historically 854 * ignores the extension, so print warning and don't care. 855 */ 856 warn_report("Format Extension ignored in RW mode"); 857 } else { 858 ret = parallels_read_format_extension( 859 bs, le64_to_cpu(ph.ext_off) << BDRV_SECTOR_BITS, errp); 860 if (ret < 0) { 861 goto fail; 862 } 863 } 864 } 865 866 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) { 867 s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC); 868 ret = parallels_update_header(bs); 869 if (ret < 0) { 870 goto fail; 871 } 872 } 873 874 s->bat_dirty_block = 4 * qemu_real_host_page_size(); 875 s->bat_dirty_bmap = 876 bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block)); 877 878 /* Disable migration until bdrv_activate method is added */ 879 error_setg(&s->migration_blocker, "The Parallels format used by node '%s' " 880 "does not support live migration", 881 bdrv_get_device_or_node_name(bs)); 882 ret = migrate_add_blocker(s->migration_blocker, errp); 883 if (ret < 0) { 884 error_free(s->migration_blocker); 885 goto fail; 886 } 887 qemu_co_mutex_init(&s->lock); 888 return 0; 889 890 fail_format: 891 error_setg(errp, "Image not in Parallels format"); 892 fail_options: 893 ret = -EINVAL; 894 fail: 895 qemu_vfree(s->header); 896 return ret; 897 } 898 899 900 static void parallels_close(BlockDriverState *bs) 901 { 902 BDRVParallelsState *s = bs->opaque; 903 904 if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) { 905 s->header->inuse = 0; 906 parallels_update_header(bs); 907 908 /* errors are ignored, so we might as well pass exact=true */ 909 bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true, 910 PREALLOC_MODE_OFF, 0, NULL); 911 } 912 913 g_free(s->bat_dirty_bmap); 914 qemu_vfree(s->header); 915 916 migrate_del_blocker(s->migration_blocker); 917 error_free(s->migration_blocker); 918 } 919 920 static BlockDriver bdrv_parallels = { 921 .format_name = "parallels", 922 .instance_size = sizeof(BDRVParallelsState), 923 .bdrv_probe = parallels_probe, 924 .bdrv_open = parallels_open, 925 .bdrv_close = parallels_close, 926 .bdrv_child_perm = bdrv_default_perms, 927 .bdrv_co_block_status = parallels_co_block_status, 928 .bdrv_has_zero_init = bdrv_has_zero_init_1, 929 .bdrv_co_flush_to_os = parallels_co_flush_to_os, 930 .bdrv_co_readv = parallels_co_readv, 931 .bdrv_co_writev = parallels_co_writev, 932 .is_format = true, 933 .supports_backing = true, 934 .bdrv_co_create = parallels_co_create, 935 .bdrv_co_create_opts = parallels_co_create_opts, 936 .bdrv_co_check = parallels_co_check, 937 .create_opts = ¶llels_create_opts, 938 }; 939 940 static void bdrv_parallels_init(void) 941 { 942 bdrv_register(&bdrv_parallels); 943 } 944 945 block_init(bdrv_parallels_init); 946