qemu/block/vhdx-log.c
<<
>>
Prefs
   1/*
   2 * Block driver for Hyper-V VHDX Images
   3 *
   4 * Copyright (c) 2013 Red Hat, Inc.,
   5 *
   6 * Authors:
   7 *  Jeff Cody <jcody@redhat.com>
   8 *
   9 *  This is based on the "VHDX Format Specification v1.00", published 8/25/2012
  10 *  by Microsoft:
  11 *      https://www.microsoft.com/en-us/download/details.aspx?id=34750
  12 *
  13 * This file covers the functionality of the metadata log writing, parsing, and
  14 * replay.
  15 *
  16 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
  17 * See the COPYING.LIB file in the top-level directory.
  18 *
  19 */
  20#include "qemu/osdep.h"
  21#include "qapi/error.h"
  22#include "qemu-common.h"
  23#include "block/block_int.h"
  24#include "qemu/error-report.h"
  25#include "qemu/module.h"
  26#include "qemu/bswap.h"
  27#include "block/vhdx.h"
  28
  29
  30typedef struct VHDXLogSequence {
  31    bool valid;
  32    uint32_t count;
  33    VHDXLogEntries log;
  34    VHDXLogEntryHeader hdr;
  35} VHDXLogSequence;
  36
  37typedef struct VHDXLogDescEntries {
  38    VHDXLogEntryHeader hdr;
  39    VHDXLogDescriptor desc[];
  40} VHDXLogDescEntries;
  41
  42static const MSGUID zero_guid = { 0 };
  43
  44/* The log located on the disk is circular buffer containing
  45 * sectors of 4096 bytes each.
  46 *
  47 * It is assumed for the read/write functions below that the
  48 * circular buffer scheme uses a 'one sector open' to indicate
  49 * the buffer is full.  Given the validation methods used for each
  50 * sector, this method should be compatible with other methods that
  51 * do not waste a sector.
  52 */
  53
  54
  55/* Allow peeking at the hdr entry at the beginning of the current
  56 * read index, without advancing the read index */
  57static int vhdx_log_peek_hdr(BlockDriverState *bs, VHDXLogEntries *log,
  58                             VHDXLogEntryHeader *hdr)
  59{
  60    int ret = 0;
  61    uint64_t offset;
  62    uint32_t read;
  63
  64    assert(hdr != NULL);
  65
  66    /* peek is only supported on sector boundaries */
  67    if (log->read % VHDX_LOG_SECTOR_SIZE) {
  68        ret = -EFAULT;
  69        goto exit;
  70    }
  71
  72    read = log->read;
  73    /* we are guaranteed that a) log sectors are 4096 bytes,
  74     * and b) the log length is a multiple of 1MB. So, there
  75     * is always a round number of sectors in the buffer */
  76    if ((read + sizeof(VHDXLogEntryHeader)) > log->length) {
  77        read = 0;
  78    }
  79
  80    if (read == log->write) {
  81        ret = -EINVAL;
  82        goto exit;
  83    }
  84
  85    offset = log->offset + read;
  86
  87    ret = bdrv_pread(bs->file, offset, hdr, sizeof(VHDXLogEntryHeader));
  88    if (ret < 0) {
  89        goto exit;
  90    }
  91    vhdx_log_entry_hdr_le_import(hdr);
  92
  93exit:
  94    return ret;
  95}
  96
  97/* Index increment for log, based on sector boundaries */
  98static int vhdx_log_inc_idx(uint32_t idx, uint64_t length)
  99{
 100    idx += VHDX_LOG_SECTOR_SIZE;
 101    /* we are guaranteed that a) log sectors are 4096 bytes,
 102     * and b) the log length is a multiple of 1MB. So, there
 103     * is always a round number of sectors in the buffer */
 104    return idx >= length ? 0 : idx;
 105}
 106
 107
 108/* Reset the log to empty */
 109static void vhdx_log_reset(BlockDriverState *bs, BDRVVHDXState *s)
 110{
 111    MSGUID guid = { 0 };
 112    s->log.read = s->log.write = 0;
 113    /* a log guid of 0 indicates an empty log to any parser of v0
 114     * VHDX logs */
 115    vhdx_update_headers(bs, s, false, &guid);
 116}
 117
 118/* Reads num_sectors from the log (all log sectors are 4096 bytes),
 119 * into buffer 'buffer'.  Upon return, *sectors_read will contain
 120 * the number of sectors successfully read.
 121 *
 122 * It is assumed that 'buffer' is already allocated, and of sufficient
 123 * size (i.e. >= 4096*num_sectors).
 124 *
 125 * If 'peek' is true, then the tail (read) pointer for the circular buffer is
 126 * not modified.
 127 *
 128 * 0 is returned on success, -errno otherwise.  */
 129static int vhdx_log_read_sectors(BlockDriverState *bs, VHDXLogEntries *log,
 130                                 uint32_t *sectors_read, void *buffer,
 131                                 uint32_t num_sectors, bool peek)
 132{
 133    int ret = 0;
 134    uint64_t offset;
 135    uint32_t read;
 136
 137    read = log->read;
 138
 139    *sectors_read = 0;
 140    while (num_sectors) {
 141        if (read == log->write) {
 142            /* empty */
 143            break;
 144        }
 145        offset = log->offset + read;
 146
 147        ret = bdrv_pread(bs->file, offset, buffer, VHDX_LOG_SECTOR_SIZE);
 148        if (ret < 0) {
 149            goto exit;
 150        }
 151        read = vhdx_log_inc_idx(read, log->length);
 152
 153        *sectors_read = *sectors_read + 1;
 154        num_sectors--;
 155    }
 156
 157exit:
 158    if (!peek) {
 159        log->read = read;
 160    }
 161    return ret;
 162}
 163
 164/* Writes num_sectors to the log (all log sectors are 4096 bytes),
 165 * from buffer 'buffer'.  Upon return, *sectors_written will contain
 166 * the number of sectors successfully written.
 167 *
 168 * It is assumed that 'buffer' is at least 4096*num_sectors large.
 169 *
 170 * 0 is returned on success, -errno otherwise */
 171static int vhdx_log_write_sectors(BlockDriverState *bs, VHDXLogEntries *log,
 172                                  uint32_t *sectors_written, void *buffer,
 173                                  uint32_t num_sectors)
 174{
 175    int ret = 0;
 176    uint64_t offset;
 177    uint32_t write;
 178    void *buffer_tmp;
 179    BDRVVHDXState *s = bs->opaque;
 180
 181    ret = vhdx_user_visible_write(bs, s);
 182    if (ret < 0) {
 183        goto exit;
 184    }
 185
 186    write = log->write;
 187
 188    buffer_tmp = buffer;
 189    while (num_sectors) {
 190
 191        offset = log->offset + write;
 192        write = vhdx_log_inc_idx(write, log->length);
 193        if (write == log->read) {
 194            /* full */
 195            break;
 196        }
 197        ret = bdrv_pwrite(bs->file, offset, buffer_tmp,
 198                          VHDX_LOG_SECTOR_SIZE);
 199        if (ret < 0) {
 200            goto exit;
 201        }
 202        buffer_tmp += VHDX_LOG_SECTOR_SIZE;
 203
 204        log->write = write;
 205        *sectors_written = *sectors_written + 1;
 206        num_sectors--;
 207    }
 208
 209exit:
 210    return ret;
 211}
 212
 213
 214/* Validates a log entry header */
 215static bool vhdx_log_hdr_is_valid(VHDXLogEntries *log, VHDXLogEntryHeader *hdr,
 216                                  BDRVVHDXState *s)
 217{
 218    int valid = false;
 219
 220    if (hdr->signature != VHDX_LOG_SIGNATURE) {
 221        goto exit;
 222    }
 223
 224    /* if the individual entry length is larger than the whole log
 225     * buffer, that is obviously invalid */
 226    if (log->length < hdr->entry_length) {
 227        goto exit;
 228    }
 229
 230    /* length of entire entry must be in units of 4KB (log sector size) */
 231    if (hdr->entry_length % (VHDX_LOG_SECTOR_SIZE)) {
 232        goto exit;
 233    }
 234
 235    /* per spec, sequence # must be > 0 */
 236    if (hdr->sequence_number == 0) {
 237        goto exit;
 238    }
 239
 240    /* log entries are only valid if they match the file-wide log guid
 241     * found in the active header */
 242    if (!guid_eq(hdr->log_guid, s->headers[s->curr_header]->log_guid)) {
 243        goto exit;
 244    }
 245
 246    if (hdr->descriptor_count * sizeof(VHDXLogDescriptor) > hdr->entry_length) {
 247        goto exit;
 248    }
 249
 250    valid = true;
 251
 252exit:
 253    return valid;
 254}
 255
 256/*
 257 * Given a log header, this will validate that the descriptors and the
 258 * corresponding data sectors (if applicable)
 259 *
 260 * Validation consists of:
 261 *      1. Making sure the sequence numbers matches the entry header
 262 *      2. Verifying a valid signature ('zero' or 'desc' for descriptors)
 263 *      3. File offset field is a multiple of 4KB
 264 *      4. If a data descriptor, the corresponding data sector
 265 *         has its signature ('data') and matching sequence number
 266 *
 267 * @desc: the data buffer containing the descriptor
 268 * @hdr:  the log entry header
 269 *
 270 * Returns true if valid
 271 */
 272static bool vhdx_log_desc_is_valid(VHDXLogDescriptor *desc,
 273                                   VHDXLogEntryHeader *hdr)
 274{
 275    bool ret = false;
 276
 277    if (desc->sequence_number != hdr->sequence_number) {
 278        goto exit;
 279    }
 280    if (desc->file_offset % VHDX_LOG_SECTOR_SIZE) {
 281        goto exit;
 282    }
 283
 284    if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) {
 285        if (desc->zero_length % VHDX_LOG_SECTOR_SIZE == 0) {
 286            /* valid */
 287            ret = true;
 288        }
 289    } else if (desc->signature == VHDX_LOG_DESC_SIGNATURE) {
 290            /* valid */
 291            ret = true;
 292    }
 293
 294exit:
 295    return ret;
 296}
 297
 298
 299/* Prior to sector data for a log entry, there is the header
 300 * and the descriptors referenced in the header:
 301 *
 302 * [] = 4KB sector
 303 *
 304 * [ hdr, desc ][   desc   ][ ... ][ data ][ ... ]
 305 *
 306 * The first sector in a log entry has a 64 byte header, and
 307 * up to 126 32-byte descriptors.  If more descriptors than
 308 * 126 are required, then subsequent sectors can have up to 128
 309 * descriptors.  Each sector is 4KB.  Data follows the descriptor
 310 * sectors.
 311 *
 312 * This will return the number of sectors needed to encompass
 313 * the passed number of descriptors in desc_cnt.
 314 *
 315 * This will never return 0, even if desc_cnt is 0.
 316 */
 317static int vhdx_compute_desc_sectors(uint32_t desc_cnt)
 318{
 319    uint32_t desc_sectors;
 320
 321    desc_cnt += 2; /* account for header in first sector */
 322    desc_sectors = desc_cnt / 128;
 323    if (desc_cnt % 128) {
 324        desc_sectors++;
 325    }
 326
 327    return desc_sectors;
 328}
 329
 330
 331/* Reads the log header, and subsequent descriptors (if any).  This
 332 * will allocate all the space for buffer, which must be NULL when
 333 * passed into this function. Each descriptor will also be validated,
 334 * and error returned if any are invalid. */
 335static int vhdx_log_read_desc(BlockDriverState *bs, BDRVVHDXState *s,
 336                              VHDXLogEntries *log, VHDXLogDescEntries **buffer,
 337                              bool convert_endian)
 338{
 339    int ret = 0;
 340    uint32_t desc_sectors;
 341    uint32_t sectors_read;
 342    VHDXLogEntryHeader hdr;
 343    VHDXLogDescEntries *desc_entries = NULL;
 344    VHDXLogDescriptor desc;
 345    int i;
 346
 347    assert(*buffer == NULL);
 348
 349    ret = vhdx_log_peek_hdr(bs, log, &hdr);
 350    if (ret < 0) {
 351        goto exit;
 352    }
 353
 354    if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) {
 355        ret = -EINVAL;
 356        goto exit;
 357    }
 358
 359    desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count);
 360    desc_entries = qemu_try_blockalign(bs->file->bs,
 361                                       desc_sectors * VHDX_LOG_SECTOR_SIZE);
 362    if (desc_entries == NULL) {
 363        ret = -ENOMEM;
 364        goto exit;
 365    }
 366
 367    ret = vhdx_log_read_sectors(bs, log, &sectors_read, desc_entries,
 368                                desc_sectors, false);
 369    if (ret < 0) {
 370        goto free_and_exit;
 371    }
 372    if (sectors_read != desc_sectors) {
 373        ret = -EINVAL;
 374        goto free_and_exit;
 375    }
 376
 377    /* put in proper endianness, and validate each desc */
 378    for (i = 0; i < hdr.descriptor_count; i++) {
 379        desc = desc_entries->desc[i];
 380        vhdx_log_desc_le_import(&desc);
 381        if (convert_endian) {
 382            desc_entries->desc[i] = desc;
 383        }
 384        if (vhdx_log_desc_is_valid(&desc, &hdr) == false) {
 385            ret = -EINVAL;
 386            goto free_and_exit;
 387        }
 388    }
 389    if (convert_endian) {
 390        desc_entries->hdr = hdr;
 391    }
 392
 393    *buffer = desc_entries;
 394    goto exit;
 395
 396free_and_exit:
 397    qemu_vfree(desc_entries);
 398exit:
 399    return ret;
 400}
 401
 402
 403/* Flushes the descriptor described by desc to the VHDX image file.
 404 * If the descriptor is a data descriptor, than 'data' must be non-NULL,
 405 * and >= 4096 bytes (VHDX_LOG_SECTOR_SIZE), containing the data to be
 406 * written.
 407 *
 408 * Verification is performed to make sure the sequence numbers of a data
 409 * descriptor match the sequence number in the desc.
 410 *
 411 * For a zero descriptor, it may describe multiple sectors to fill with zeroes.
 412 * In this case, it should be noted that zeroes are written to disk, and the
 413 * image file is not extended as a sparse file.  */
 414static int vhdx_log_flush_desc(BlockDriverState *bs, VHDXLogDescriptor *desc,
 415                               VHDXLogDataSector *data)
 416{
 417    int ret = 0;
 418    uint64_t seq, file_offset;
 419    uint32_t offset = 0;
 420    void *buffer = NULL;
 421    uint64_t count = 1;
 422    int i;
 423
 424    buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
 425
 426    if (desc->signature == VHDX_LOG_DESC_SIGNATURE) {
 427        /* data sector */
 428        if (data == NULL) {
 429            ret = -EFAULT;
 430            goto exit;
 431        }
 432
 433        /* The sequence number of the data sector must match that
 434         * in the descriptor */
 435        seq = data->sequence_high;
 436        seq <<= 32;
 437        seq |= data->sequence_low & 0xffffffff;
 438
 439        if (seq != desc->sequence_number) {
 440            ret = -EINVAL;
 441            goto exit;
 442        }
 443
 444        /* Each data sector is in total 4096 bytes, however the first
 445         * 8 bytes, and last 4 bytes, are located in the descriptor */
 446        memcpy(buffer, &desc->leading_bytes, 8);
 447        offset += 8;
 448
 449        memcpy(buffer+offset, data->data, 4084);
 450        offset += 4084;
 451
 452        memcpy(buffer+offset, &desc->trailing_bytes, 4);
 453
 454    } else if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) {
 455        /* write 'count' sectors of sector */
 456        memset(buffer, 0, VHDX_LOG_SECTOR_SIZE);
 457        count = desc->zero_length / VHDX_LOG_SECTOR_SIZE;
 458    } else {
 459        error_report("Invalid VHDX log descriptor entry signature 0x%" PRIx32,
 460                      desc->signature);
 461        ret = -EINVAL;
 462        goto exit;
 463    }
 464
 465    file_offset = desc->file_offset;
 466
 467    /* count is only > 1 if we are writing zeroes */
 468    for (i = 0; i < count; i++) {
 469        ret = bdrv_pwrite_sync(bs->file, file_offset, buffer,
 470                               VHDX_LOG_SECTOR_SIZE);
 471        if (ret < 0) {
 472            goto exit;
 473        }
 474        file_offset += VHDX_LOG_SECTOR_SIZE;
 475    }
 476
 477exit:
 478    qemu_vfree(buffer);
 479    return ret;
 480}
 481
 482/* Flush the entire log (as described by 'logs') to the VHDX image
 483 * file, and then set the log to 'empty' status once complete.
 484 *
 485 * The log entries should be validate prior to flushing */
 486static int vhdx_log_flush(BlockDriverState *bs, BDRVVHDXState *s,
 487                          VHDXLogSequence *logs)
 488{
 489    int ret = 0;
 490    int i;
 491    uint32_t cnt, sectors_read;
 492    uint64_t new_file_size;
 493    void *data = NULL;
 494    VHDXLogDescEntries *desc_entries = NULL;
 495    VHDXLogEntryHeader hdr_tmp = { 0 };
 496
 497    cnt = logs->count;
 498
 499    data = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
 500
 501    ret = vhdx_user_visible_write(bs, s);
 502    if (ret < 0) {
 503        goto exit;
 504    }
 505
 506    /* each iteration represents one log sequence, which may span multiple
 507     * sectors */
 508    while (cnt--) {
 509        ret = vhdx_log_peek_hdr(bs, &logs->log, &hdr_tmp);
 510        if (ret < 0) {
 511            goto exit;
 512        }
 513        /* if the log shows a FlushedFileOffset larger than our current file
 514         * size, then that means the file has been truncated / corrupted, and
 515         * we must refused to open it / use it */
 516        if (hdr_tmp.flushed_file_offset > bdrv_getlength(bs->file->bs)) {
 517            ret = -EINVAL;
 518            goto exit;
 519        }
 520
 521        ret = vhdx_log_read_desc(bs, s, &logs->log, &desc_entries, true);
 522        if (ret < 0) {
 523            goto exit;
 524        }
 525
 526        for (i = 0; i < desc_entries->hdr.descriptor_count; i++) {
 527            if (desc_entries->desc[i].signature == VHDX_LOG_DESC_SIGNATURE) {
 528                /* data sector, so read a sector to flush */
 529                ret = vhdx_log_read_sectors(bs, &logs->log, &sectors_read,
 530                                            data, 1, false);
 531                if (ret < 0) {
 532                    goto exit;
 533                }
 534                if (sectors_read != 1) {
 535                    ret = -EINVAL;
 536                    goto exit;
 537                }
 538                vhdx_log_data_le_import(data);
 539            }
 540
 541            ret = vhdx_log_flush_desc(bs, &desc_entries->desc[i], data);
 542            if (ret < 0) {
 543                goto exit;
 544            }
 545        }
 546        if (bdrv_getlength(bs->file->bs) < desc_entries->hdr.last_file_offset) {
 547            new_file_size = desc_entries->hdr.last_file_offset;
 548            if (new_file_size % (1024*1024)) {
 549                /* round up to nearest 1MB boundary */
 550                new_file_size = ((new_file_size >> 20) + 1) << 20;
 551                bdrv_truncate(bs->file, new_file_size);
 552            }
 553        }
 554        qemu_vfree(desc_entries);
 555        desc_entries = NULL;
 556    }
 557
 558    bdrv_flush(bs);
 559    /* once the log is fully flushed, indicate that we have an empty log
 560     * now.  This also sets the log guid to 0, to indicate an empty log */
 561    vhdx_log_reset(bs, s);
 562
 563exit:
 564    qemu_vfree(data);
 565    qemu_vfree(desc_entries);
 566    return ret;
 567}
 568
 569static int vhdx_validate_log_entry(BlockDriverState *bs, BDRVVHDXState *s,
 570                                   VHDXLogEntries *log, uint64_t seq,
 571                                   bool *valid, VHDXLogEntryHeader *entry)
 572{
 573    int ret = 0;
 574    VHDXLogEntryHeader hdr;
 575    void *buffer = NULL;
 576    uint32_t i, desc_sectors, total_sectors, crc;
 577    uint32_t sectors_read = 0;
 578    VHDXLogDescEntries *desc_buffer = NULL;
 579
 580    *valid = false;
 581
 582    ret = vhdx_log_peek_hdr(bs, log, &hdr);
 583    if (ret < 0) {
 584        goto inc_and_exit;
 585    }
 586
 587    if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) {
 588        goto inc_and_exit;
 589    }
 590
 591    if (seq > 0) {
 592        if (hdr.sequence_number != seq + 1) {
 593            goto inc_and_exit;
 594        }
 595    }
 596
 597    desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count);
 598
 599    /* Read all log sectors, and calculate log checksum */
 600
 601    total_sectors = hdr.entry_length / VHDX_LOG_SECTOR_SIZE;
 602
 603
 604    /* read_desc() will increment the read idx */
 605    ret = vhdx_log_read_desc(bs, s, log, &desc_buffer, false);
 606    if (ret < 0) {
 607        goto free_and_exit;
 608    }
 609
 610    crc = vhdx_checksum_calc(0xffffffff, (void *)desc_buffer,
 611                            desc_sectors * VHDX_LOG_SECTOR_SIZE, 4);
 612    crc ^= 0xffffffff;
 613
 614    buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
 615    if (total_sectors > desc_sectors) {
 616        for (i = 0; i < total_sectors - desc_sectors; i++) {
 617            sectors_read = 0;
 618            ret = vhdx_log_read_sectors(bs, log, &sectors_read, buffer,
 619                                        1, false);
 620            if (ret < 0 || sectors_read != 1) {
 621                goto free_and_exit;
 622            }
 623            crc = vhdx_checksum_calc(crc, buffer, VHDX_LOG_SECTOR_SIZE, -1);
 624            crc ^= 0xffffffff;
 625        }
 626    }
 627    crc ^= 0xffffffff;
 628    if (crc != hdr.checksum) {
 629        goto free_and_exit;
 630    }
 631
 632    *valid = true;
 633    *entry = hdr;
 634    goto free_and_exit;
 635
 636inc_and_exit:
 637    log->read = vhdx_log_inc_idx(log->read, log->length);
 638
 639free_and_exit:
 640    qemu_vfree(buffer);
 641    qemu_vfree(desc_buffer);
 642    return ret;
 643}
 644
 645/* Search through the log circular buffer, and find the valid, active
 646 * log sequence, if any exists
 647 * */
 648static int vhdx_log_search(BlockDriverState *bs, BDRVVHDXState *s,
 649                           VHDXLogSequence *logs)
 650{
 651    int ret = 0;
 652    uint32_t tail;
 653    bool seq_valid = false;
 654    VHDXLogSequence candidate = { 0 };
 655    VHDXLogEntryHeader hdr = { 0 };
 656    VHDXLogEntries curr_log;
 657
 658    memcpy(&curr_log, &s->log, sizeof(VHDXLogEntries));
 659    curr_log.write = curr_log.length;   /* assume log is full */
 660    curr_log.read = 0;
 661
 662
 663    /* now we will go through the whole log sector by sector, until
 664     * we find a valid, active log sequence, or reach the end of the
 665     * log buffer */
 666    for (;;) {
 667        uint64_t curr_seq = 0;
 668        VHDXLogSequence current = { 0 };
 669
 670        tail = curr_log.read;
 671
 672        ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq,
 673                                      &seq_valid, &hdr);
 674        if (ret < 0) {
 675            goto exit;
 676        }
 677
 678        if (seq_valid) {
 679            current.valid     = true;
 680            current.log       = curr_log;
 681            current.log.read  = tail;
 682            current.log.write = curr_log.read;
 683            current.count     = 1;
 684            current.hdr       = hdr;
 685
 686
 687            for (;;) {
 688                ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq,
 689                                              &seq_valid, &hdr);
 690                if (ret < 0) {
 691                    goto exit;
 692                }
 693                if (seq_valid == false) {
 694                    break;
 695                }
 696                current.log.write = curr_log.read;
 697                current.count++;
 698
 699                curr_seq = hdr.sequence_number;
 700            }
 701        }
 702
 703        if (current.valid) {
 704            if (candidate.valid == false ||
 705                current.hdr.sequence_number > candidate.hdr.sequence_number) {
 706                candidate = current;
 707            }
 708        }
 709
 710        if (curr_log.read < tail) {
 711            break;
 712        }
 713    }
 714
 715    *logs = candidate;
 716
 717    if (candidate.valid) {
 718        /* this is the next sequence number, for writes */
 719        s->log.sequence = candidate.hdr.sequence_number + 1;
 720    }
 721
 722
 723exit:
 724    return ret;
 725}
 726
 727/* Parse the replay log.  Per the VHDX spec, if the log is present
 728 * it must be replayed prior to opening the file, even read-only.
 729 *
 730 * If read-only, we must replay the log in RAM (or refuse to open
 731 * a dirty VHDX file read-only) */
 732int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s, bool *flushed,
 733                   Error **errp)
 734{
 735    int ret = 0;
 736    VHDXHeader *hdr;
 737    VHDXLogSequence logs = { 0 };
 738
 739    hdr = s->headers[s->curr_header];
 740
 741    *flushed = false;
 742
 743    /* s->log.hdr is freed in vhdx_close() */
 744    if (s->log.hdr == NULL) {
 745        s->log.hdr = qemu_blockalign(bs, sizeof(VHDXLogEntryHeader));
 746    }
 747
 748    s->log.offset = hdr->log_offset;
 749    s->log.length = hdr->log_length;
 750
 751    if (s->log.offset < VHDX_LOG_MIN_SIZE ||
 752        s->log.offset % VHDX_LOG_MIN_SIZE) {
 753        ret = -EINVAL;
 754        goto exit;
 755    }
 756
 757    /* per spec, only log version of 0 is supported */
 758    if (hdr->log_version != 0) {
 759        ret = -EINVAL;
 760        goto exit;
 761    }
 762
 763    /* If either the log guid, or log length is zero,
 764     * then a replay log is not present */
 765    if (guid_eq(hdr->log_guid, zero_guid)) {
 766        goto exit;
 767    }
 768
 769    if (hdr->log_length == 0) {
 770        goto exit;
 771    }
 772
 773    if (hdr->log_length % VHDX_LOG_MIN_SIZE) {
 774        ret = -EINVAL;
 775        goto exit;
 776    }
 777
 778
 779    /* The log is present, we need to find if and where there is an active
 780     * sequence of valid entries present in the log.  */
 781
 782    ret = vhdx_log_search(bs, s, &logs);
 783    if (ret < 0) {
 784        goto exit;
 785    }
 786
 787    if (logs.valid) {
 788        if (bs->read_only) {
 789            ret = -EPERM;
 790            error_setg(errp,
 791                       "VHDX image file '%s' opened read-only, but "
 792                       "contains a log that needs to be replayed",
 793                       bs->filename);
 794            error_append_hint(errp,  "To replay the log, run:\n"
 795                              "qemu-img check -r all '%s'\n",
 796                              bs->filename);
 797            goto exit;
 798        }
 799        /* now flush the log */
 800        ret = vhdx_log_flush(bs, s, &logs);
 801        if (ret < 0) {
 802            goto exit;
 803        }
 804        *flushed = true;
 805    }
 806
 807
 808exit:
 809    return ret;
 810}
 811
 812
 813
 814static void vhdx_log_raw_to_le_sector(VHDXLogDescriptor *desc,
 815                                      VHDXLogDataSector *sector, void *data,
 816                                      uint64_t seq)
 817{
 818    /* 8 + 4084 + 4 = 4096, 1 log sector */
 819    memcpy(&desc->leading_bytes, data, 8);
 820    data += 8;
 821    cpu_to_le64s(&desc->leading_bytes);
 822    memcpy(sector->data, data, 4084);
 823    data += 4084;
 824    memcpy(&desc->trailing_bytes, data, 4);
 825    cpu_to_le32s(&desc->trailing_bytes);
 826    data += 4;
 827
 828    sector->sequence_high  = (uint32_t) (seq >> 32);
 829    sector->sequence_low   = (uint32_t) (seq & 0xffffffff);
 830    sector->data_signature = VHDX_LOG_DATA_SIGNATURE;
 831
 832    vhdx_log_desc_le_export(desc);
 833    vhdx_log_data_le_export(sector);
 834}
 835
 836
 837static int vhdx_log_write(BlockDriverState *bs, BDRVVHDXState *s,
 838                          void *data, uint32_t length, uint64_t offset)
 839{
 840    int ret = 0;
 841    void *buffer = NULL;
 842    void *merged_sector = NULL;
 843    void *data_tmp, *sector_write;
 844    unsigned int i;
 845    int sector_offset;
 846    uint32_t desc_sectors, sectors, total_length;
 847    uint32_t sectors_written = 0;
 848    uint32_t aligned_length;
 849    uint32_t leading_length = 0;
 850    uint32_t trailing_length = 0;
 851    uint32_t partial_sectors = 0;
 852    uint32_t bytes_written = 0;
 853    uint64_t file_offset;
 854    VHDXHeader *header;
 855    VHDXLogEntryHeader new_hdr;
 856    VHDXLogDescriptor *new_desc = NULL;
 857    VHDXLogDataSector *data_sector = NULL;
 858    MSGUID new_guid = { 0 };
 859
 860    header = s->headers[s->curr_header];
 861
 862    /* need to have offset read data, and be on 4096 byte boundary */
 863
 864    if (length > header->log_length) {
 865        /* no log present.  we could create a log here instead of failing */
 866        ret = -EINVAL;
 867        goto exit;
 868    }
 869
 870    if (guid_eq(header->log_guid, zero_guid)) {
 871        vhdx_guid_generate(&new_guid);
 872        vhdx_update_headers(bs, s, false, &new_guid);
 873    } else {
 874        /* currently, we require that the log be flushed after
 875         * every write. */
 876        ret = -ENOTSUP;
 877        goto exit;
 878    }
 879
 880    /* 0 is an invalid sequence number, but may also represent the first
 881     * log write (or a wrapped seq) */
 882    if (s->log.sequence == 0) {
 883        s->log.sequence = 1;
 884    }
 885
 886    sector_offset = offset % VHDX_LOG_SECTOR_SIZE;
 887    file_offset = (offset / VHDX_LOG_SECTOR_SIZE) * VHDX_LOG_SECTOR_SIZE;
 888
 889    aligned_length = length;
 890
 891    /* add in the unaligned head and tail bytes */
 892    if (sector_offset) {
 893        leading_length = (VHDX_LOG_SECTOR_SIZE - sector_offset);
 894        leading_length = leading_length > length ? length : leading_length;
 895        aligned_length -= leading_length;
 896        partial_sectors++;
 897    }
 898
 899    sectors = aligned_length / VHDX_LOG_SECTOR_SIZE;
 900    trailing_length = aligned_length - (sectors * VHDX_LOG_SECTOR_SIZE);
 901    if (trailing_length) {
 902        partial_sectors++;
 903    }
 904
 905    sectors += partial_sectors;
 906
 907    /* sectors is now how many sectors the data itself takes, not
 908     * including the header and descriptor metadata */
 909
 910    new_hdr = (VHDXLogEntryHeader) {
 911                .signature           = VHDX_LOG_SIGNATURE,
 912                .tail                = s->log.tail,
 913                .sequence_number     = s->log.sequence,
 914                .descriptor_count    = sectors,
 915                .reserved            = 0,
 916                .flushed_file_offset = bdrv_getlength(bs->file->bs),
 917                .last_file_offset    = bdrv_getlength(bs->file->bs),
 918              };
 919
 920    new_hdr.log_guid = header->log_guid;
 921
 922    desc_sectors = vhdx_compute_desc_sectors(new_hdr.descriptor_count);
 923
 924    total_length = (desc_sectors + sectors) * VHDX_LOG_SECTOR_SIZE;
 925    new_hdr.entry_length = total_length;
 926
 927    vhdx_log_entry_hdr_le_export(&new_hdr);
 928
 929    buffer = qemu_blockalign(bs, total_length);
 930    memcpy(buffer, &new_hdr, sizeof(new_hdr));
 931
 932    new_desc = buffer + sizeof(new_hdr);
 933    data_sector = buffer + (desc_sectors * VHDX_LOG_SECTOR_SIZE);
 934    data_tmp = data;
 935
 936    /* All log sectors are 4KB, so for any partial sectors we must
 937     * merge the data with preexisting data from the final file
 938     * destination */
 939    merged_sector = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
 940
 941    for (i = 0; i < sectors; i++) {
 942        new_desc->signature       = VHDX_LOG_DESC_SIGNATURE;
 943        new_desc->sequence_number = s->log.sequence;
 944        new_desc->file_offset     = file_offset;
 945
 946        if (i == 0 && leading_length) {
 947            /* partial sector at the front of the buffer */
 948            ret = bdrv_pread(bs->file, file_offset, merged_sector,
 949                             VHDX_LOG_SECTOR_SIZE);
 950            if (ret < 0) {
 951                goto exit;
 952            }
 953            memcpy(merged_sector + sector_offset, data_tmp, leading_length);
 954            bytes_written = leading_length;
 955            sector_write = merged_sector;
 956        } else if (i == sectors - 1 && trailing_length) {
 957            /* partial sector at the end of the buffer */
 958            ret = bdrv_pread(bs->file,
 959                            file_offset,
 960                            merged_sector + trailing_length,
 961                            VHDX_LOG_SECTOR_SIZE - trailing_length);
 962            if (ret < 0) {
 963                goto exit;
 964            }
 965            memcpy(merged_sector, data_tmp, trailing_length);
 966            bytes_written = trailing_length;
 967            sector_write = merged_sector;
 968        } else {
 969            bytes_written = VHDX_LOG_SECTOR_SIZE;
 970            sector_write = data_tmp;
 971        }
 972
 973        /* populate the raw sector data into the proper structures,
 974         * as well as update the descriptor, and convert to proper
 975         * endianness */
 976        vhdx_log_raw_to_le_sector(new_desc, data_sector, sector_write,
 977                                  s->log.sequence);
 978
 979        data_tmp += bytes_written;
 980        data_sector++;
 981        new_desc++;
 982        file_offset += VHDX_LOG_SECTOR_SIZE;
 983    }
 984
 985    /* checksum covers entire entry, from the log header through the
 986     * last data sector */
 987    vhdx_update_checksum(buffer, total_length,
 988                         offsetof(VHDXLogEntryHeader, checksum));
 989
 990    /* now write to the log */
 991    ret = vhdx_log_write_sectors(bs, &s->log, &sectors_written, buffer,
 992                                 desc_sectors + sectors);
 993    if (ret < 0) {
 994        goto exit;
 995    }
 996
 997    if (sectors_written != desc_sectors + sectors) {
 998        /* instead of failing, we could flush the log here */
 999        ret = -EINVAL;
1000        goto exit;
1001    }
1002
1003    s->log.sequence++;
1004    /* write new tail */
1005    s->log.tail = s->log.write;
1006
1007exit:
1008    qemu_vfree(buffer);
1009    qemu_vfree(merged_sector);
1010    return ret;
1011}
1012
1013/* Perform a log write, and then immediately flush the entire log */
1014int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
1015                             void *data, uint32_t length, uint64_t offset)
1016{
1017    int ret = 0;
1018    VHDXLogSequence logs = { .valid = true,
1019                             .count = 1,
1020                             .hdr = { 0 } };
1021
1022
1023    /* Make sure data written (new and/or changed blocks) is stable
1024     * on disk, before creating log entry */
1025    bdrv_flush(bs);
1026    ret = vhdx_log_write(bs, s, data, length, offset);
1027    if (ret < 0) {
1028        goto exit;
1029    }
1030    logs.log = s->log;
1031
1032    /* Make sure log is stable on disk */
1033    bdrv_flush(bs);
1034    ret = vhdx_log_flush(bs, s, &logs);
1035    if (ret < 0) {
1036        goto exit;
1037    }
1038
1039    s->log = logs.log;
1040
1041exit:
1042    return ret;
1043}
1044
1045