qemu/block/qcow2-cluster.c
<<
>>
Prefs
   1/*
   2 * Block driver for the QCOW version 2 format
   3 *
   4 * Copyright (c) 2004-2006 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
  25#include "qemu/osdep.h"
  26#include <zlib.h>
  27
  28#include "qapi/error.h"
  29#include "qemu-common.h"
  30#include "block/block_int.h"
  31#include "block/qcow2.h"
  32#include "qemu/bswap.h"
  33#include "trace.h"
  34
  35int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t exact_size)
  36{
  37    BDRVQcow2State *s = bs->opaque;
  38    int new_l1_size, i, ret;
  39
  40    if (exact_size >= s->l1_size) {
  41        return 0;
  42    }
  43
  44    new_l1_size = exact_size;
  45
  46#ifdef DEBUG_ALLOC2
  47    fprintf(stderr, "shrink l1_table from %d to %d\n", s->l1_size, new_l1_size);
  48#endif
  49
  50    BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_WRITE_TABLE);
  51    ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset +
  52                                       new_l1_size * sizeof(uint64_t),
  53                             (s->l1_size - new_l1_size) * sizeof(uint64_t), 0);
  54    if (ret < 0) {
  55        goto fail;
  56    }
  57
  58    ret = bdrv_flush(bs->file->bs);
  59    if (ret < 0) {
  60        goto fail;
  61    }
  62
  63    BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_FREE_L2_CLUSTERS);
  64    for (i = s->l1_size - 1; i > new_l1_size - 1; i--) {
  65        if ((s->l1_table[i] & L1E_OFFSET_MASK) == 0) {
  66            continue;
  67        }
  68        qcow2_free_clusters(bs, s->l1_table[i] & L1E_OFFSET_MASK,
  69                            s->cluster_size, QCOW2_DISCARD_ALWAYS);
  70        s->l1_table[i] = 0;
  71    }
  72    return 0;
  73
  74fail:
  75    /*
  76     * If the write in the l1_table failed the image may contain a partially
  77     * overwritten l1_table. In this case it would be better to clear the
  78     * l1_table in memory to avoid possible image corruption.
  79     */
  80    memset(s->l1_table + new_l1_size, 0,
  81           (s->l1_size - new_l1_size) * sizeof(uint64_t));
  82    return ret;
  83}
  84
  85int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
  86                        bool exact_size)
  87{
  88    BDRVQcow2State *s = bs->opaque;
  89    int new_l1_size2, ret, i;
  90    uint64_t *new_l1_table;
  91    int64_t old_l1_table_offset, old_l1_size;
  92    int64_t new_l1_table_offset, new_l1_size;
  93    uint8_t data[12];
  94
  95    if (min_size <= s->l1_size)
  96        return 0;
  97
  98    /* Do a sanity check on min_size before trying to calculate new_l1_size
  99     * (this prevents overflows during the while loop for the calculation of
 100     * new_l1_size) */
 101    if (min_size > INT_MAX / sizeof(uint64_t)) {
 102        return -EFBIG;
 103    }
 104
 105    if (exact_size) {
 106        new_l1_size = min_size;
 107    } else {
 108        /* Bump size up to reduce the number of times we have to grow */
 109        new_l1_size = s->l1_size;
 110        if (new_l1_size == 0) {
 111            new_l1_size = 1;
 112        }
 113        while (min_size > new_l1_size) {
 114            new_l1_size = DIV_ROUND_UP(new_l1_size * 3, 2);
 115        }
 116    }
 117
 118    QEMU_BUILD_BUG_ON(QCOW_MAX_L1_SIZE > INT_MAX);
 119    if (new_l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
 120        return -EFBIG;
 121    }
 122
 123#ifdef DEBUG_ALLOC2
 124    fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n",
 125            s->l1_size, new_l1_size);
 126#endif
 127
 128    new_l1_size2 = sizeof(uint64_t) * new_l1_size;
 129    new_l1_table = qemu_try_blockalign(bs->file->bs,
 130                                       align_offset(new_l1_size2, 512));
 131    if (new_l1_table == NULL) {
 132        return -ENOMEM;
 133    }
 134    memset(new_l1_table, 0, align_offset(new_l1_size2, 512));
 135
 136    if (s->l1_size) {
 137        memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
 138    }
 139
 140    /* write new table (align to cluster) */
 141    BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
 142    new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
 143    if (new_l1_table_offset < 0) {
 144        qemu_vfree(new_l1_table);
 145        return new_l1_table_offset;
 146    }
 147
 148    ret = qcow2_cache_flush(bs, s->refcount_block_cache);
 149    if (ret < 0) {
 150        goto fail;
 151    }
 152
 153    /* the L1 position has not yet been updated, so these clusters must
 154     * indeed be completely free */
 155    ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset,
 156                                        new_l1_size2);
 157    if (ret < 0) {
 158        goto fail;
 159    }
 160
 161    BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
 162    for(i = 0; i < s->l1_size; i++)
 163        new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
 164    ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset,
 165                           new_l1_table, new_l1_size2);
 166    if (ret < 0)
 167        goto fail;
 168    for(i = 0; i < s->l1_size; i++)
 169        new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
 170
 171    /* set new table */
 172    BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
 173    stl_be_p(data, new_l1_size);
 174    stq_be_p(data + 4, new_l1_table_offset);
 175    ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size),
 176                           data, sizeof(data));
 177    if (ret < 0) {
 178        goto fail;
 179    }
 180    qemu_vfree(s->l1_table);
 181    old_l1_table_offset = s->l1_table_offset;
 182    s->l1_table_offset = new_l1_table_offset;
 183    s->l1_table = new_l1_table;
 184    old_l1_size = s->l1_size;
 185    s->l1_size = new_l1_size;
 186    qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t),
 187                        QCOW2_DISCARD_OTHER);
 188    return 0;
 189 fail:
 190    qemu_vfree(new_l1_table);
 191    qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2,
 192                        QCOW2_DISCARD_OTHER);
 193    return ret;
 194}
 195
 196/*
 197 * l2_load
 198 *
 199 * Loads a L2 table into memory. If the table is in the cache, the cache
 200 * is used; otherwise the L2 table is loaded from the image file.
 201 *
 202 * Returns a pointer to the L2 table on success, or NULL if the read from
 203 * the image file failed.
 204 */
 205
 206static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
 207    uint64_t **l2_table)
 208{
 209    BDRVQcow2State *s = bs->opaque;
 210
 211    return qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
 212                           (void **)l2_table);
 213}
 214
 215/*
 216 * Writes one sector of the L1 table to the disk (can't update single entries
 217 * and we really don't want bdrv_pread to perform a read-modify-write)
 218 */
 219#define L1_ENTRIES_PER_SECTOR (512 / 8)
 220int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
 221{
 222    BDRVQcow2State *s = bs->opaque;
 223    uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 };
 224    int l1_start_index;
 225    int i, ret;
 226
 227    l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
 228    for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size;
 229         i++)
 230    {
 231        buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
 232    }
 233
 234    ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
 235            s->l1_table_offset + 8 * l1_start_index, sizeof(buf));
 236    if (ret < 0) {
 237        return ret;
 238    }
 239
 240    BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
 241    ret = bdrv_pwrite_sync(bs->file,
 242                           s->l1_table_offset + 8 * l1_start_index,
 243                           buf, sizeof(buf));
 244    if (ret < 0) {
 245        return ret;
 246    }
 247
 248    return 0;
 249}
 250
 251/*
 252 * l2_allocate
 253 *
 254 * Allocate a new l2 entry in the file. If l1_index points to an already
 255 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
 256 * table) copy the contents of the old L2 table into the newly allocated one.
 257 * Otherwise the new table is initialized with zeros.
 258 *
 259 */
 260
 261static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
 262{
 263    BDRVQcow2State *s = bs->opaque;
 264    uint64_t old_l2_offset;
 265    uint64_t *l2_table = NULL;
 266    int64_t l2_offset;
 267    int ret;
 268
 269    old_l2_offset = s->l1_table[l1_index];
 270
 271    trace_qcow2_l2_allocate(bs, l1_index);
 272
 273    /* allocate a new l2 entry */
 274
 275    l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
 276    if (l2_offset < 0) {
 277        ret = l2_offset;
 278        goto fail;
 279    }
 280
 281    /* If we're allocating the table at offset 0 then something is wrong */
 282    if (l2_offset == 0) {
 283        qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid "
 284                                "allocation of L2 table at offset 0");
 285        ret = -EIO;
 286        goto fail;
 287    }
 288
 289    ret = qcow2_cache_flush(bs, s->refcount_block_cache);
 290    if (ret < 0) {
 291        goto fail;
 292    }
 293
 294    /* allocate a new entry in the l2 cache */
 295
 296    trace_qcow2_l2_allocate_get_empty(bs, l1_index);
 297    ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
 298    if (ret < 0) {
 299        goto fail;
 300    }
 301
 302    l2_table = *table;
 303
 304    if ((old_l2_offset & L1E_OFFSET_MASK) == 0) {
 305        /* if there was no old l2 table, clear the new table */
 306        memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 307    } else {
 308        uint64_t* old_table;
 309
 310        /* if there was an old l2 table, read it from the disk */
 311        BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
 312        ret = qcow2_cache_get(bs, s->l2_table_cache,
 313            old_l2_offset & L1E_OFFSET_MASK,
 314            (void**) &old_table);
 315        if (ret < 0) {
 316            goto fail;
 317        }
 318
 319        memcpy(l2_table, old_table, s->cluster_size);
 320
 321        qcow2_cache_put(bs, s->l2_table_cache, (void **) &old_table);
 322    }
 323
 324    /* write the l2 table to the file */
 325    BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
 326
 327    trace_qcow2_l2_allocate_write_l2(bs, l1_index);
 328    qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
 329    ret = qcow2_cache_flush(bs, s->l2_table_cache);
 330    if (ret < 0) {
 331        goto fail;
 332    }
 333
 334    /* update the L1 entry */
 335    trace_qcow2_l2_allocate_write_l1(bs, l1_index);
 336    s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
 337    ret = qcow2_write_l1_entry(bs, l1_index);
 338    if (ret < 0) {
 339        goto fail;
 340    }
 341
 342    *table = l2_table;
 343    trace_qcow2_l2_allocate_done(bs, l1_index, 0);
 344    return 0;
 345
 346fail:
 347    trace_qcow2_l2_allocate_done(bs, l1_index, ret);
 348    if (l2_table != NULL) {
 349        qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
 350    }
 351    s->l1_table[l1_index] = old_l2_offset;
 352    if (l2_offset > 0) {
 353        qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
 354                            QCOW2_DISCARD_ALWAYS);
 355    }
 356    return ret;
 357}
 358
 359/*
 360 * Checks how many clusters in a given L2 table are contiguous in the image
 361 * file. As soon as one of the flags in the bitmask stop_flags changes compared
 362 * to the first cluster, the search is stopped and the cluster is not counted
 363 * as contiguous. (This allows it, for example, to stop at the first compressed
 364 * cluster which may require a different handling)
 365 */
 366static int count_contiguous_clusters(int nb_clusters, int cluster_size,
 367        uint64_t *l2_table, uint64_t stop_flags)
 368{
 369    int i;
 370    QCow2ClusterType first_cluster_type;
 371    uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
 372    uint64_t first_entry = be64_to_cpu(l2_table[0]);
 373    uint64_t offset = first_entry & mask;
 374
 375    if (!offset) {
 376        return 0;
 377    }
 378
 379    /* must be allocated */
 380    first_cluster_type = qcow2_get_cluster_type(first_entry);
 381    assert(first_cluster_type == QCOW2_CLUSTER_NORMAL ||
 382           first_cluster_type == QCOW2_CLUSTER_ZERO_ALLOC);
 383
 384    for (i = 0; i < nb_clusters; i++) {
 385        uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask;
 386        if (offset + (uint64_t) i * cluster_size != l2_entry) {
 387            break;
 388        }
 389    }
 390
 391        return i;
 392}
 393
 394/*
 395 * Checks how many consecutive unallocated clusters in a given L2
 396 * table have the same cluster type.
 397 */
 398static int count_contiguous_clusters_unallocated(int nb_clusters,
 399                                                 uint64_t *l2_table,
 400                                                 QCow2ClusterType wanted_type)
 401{
 402    int i;
 403
 404    assert(wanted_type == QCOW2_CLUSTER_ZERO_PLAIN ||
 405           wanted_type == QCOW2_CLUSTER_UNALLOCATED);
 406    for (i = 0; i < nb_clusters; i++) {
 407        uint64_t entry = be64_to_cpu(l2_table[i]);
 408        QCow2ClusterType type = qcow2_get_cluster_type(entry);
 409
 410        if (type != wanted_type) {
 411            break;
 412        }
 413    }
 414
 415    return i;
 416}
 417
 418static int coroutine_fn do_perform_cow_read(BlockDriverState *bs,
 419                                            uint64_t src_cluster_offset,
 420                                            unsigned offset_in_cluster,
 421                                            QEMUIOVector *qiov)
 422{
 423    int ret;
 424
 425    if (qiov->size == 0) {
 426        return 0;
 427    }
 428
 429    BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
 430
 431    if (!bs->drv) {
 432        return -ENOMEDIUM;
 433    }
 434
 435    /* Call .bdrv_co_readv() directly instead of using the public block-layer
 436     * interface.  This avoids double I/O throttling and request tracking,
 437     * which can lead to deadlock when block layer copy-on-read is enabled.
 438     */
 439    ret = bs->drv->bdrv_co_preadv(bs, src_cluster_offset + offset_in_cluster,
 440                                  qiov->size, qiov, 0);
 441    if (ret < 0) {
 442        return ret;
 443    }
 444
 445    return 0;
 446}
 447
 448static bool coroutine_fn do_perform_cow_encrypt(BlockDriverState *bs,
 449                                                uint64_t src_cluster_offset,
 450                                                uint64_t cluster_offset,
 451                                                unsigned offset_in_cluster,
 452                                                uint8_t *buffer,
 453                                                unsigned bytes)
 454{
 455    if (bytes && bs->encrypted) {
 456        BDRVQcow2State *s = bs->opaque;
 457        int64_t offset = (s->crypt_physical_offset ?
 458                          (cluster_offset + offset_in_cluster) :
 459                          (src_cluster_offset + offset_in_cluster));
 460        assert((offset_in_cluster & ~BDRV_SECTOR_MASK) == 0);
 461        assert((bytes & ~BDRV_SECTOR_MASK) == 0);
 462        assert(s->crypto);
 463        if (qcrypto_block_encrypt(s->crypto, offset, buffer, bytes, NULL) < 0) {
 464            return false;
 465        }
 466    }
 467    return true;
 468}
 469
 470static int coroutine_fn do_perform_cow_write(BlockDriverState *bs,
 471                                             uint64_t cluster_offset,
 472                                             unsigned offset_in_cluster,
 473                                             QEMUIOVector *qiov)
 474{
 475    int ret;
 476
 477    if (qiov->size == 0) {
 478        return 0;
 479    }
 480
 481    ret = qcow2_pre_write_overlap_check(bs, 0,
 482            cluster_offset + offset_in_cluster, qiov->size);
 483    if (ret < 0) {
 484        return ret;
 485    }
 486
 487    BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
 488    ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster,
 489                          qiov->size, qiov, 0);
 490    if (ret < 0) {
 491        return ret;
 492    }
 493
 494    return 0;
 495}
 496
 497
 498/*
 499 * get_cluster_offset
 500 *
 501 * For a given offset of the virtual disk, find the cluster type and offset in
 502 * the qcow2 file. The offset is stored in *cluster_offset.
 503 *
 504 * On entry, *bytes is the maximum number of contiguous bytes starting at
 505 * offset that we are interested in.
 506 *
 507 * On exit, *bytes is the number of bytes starting at offset that have the same
 508 * cluster type and (if applicable) are stored contiguously in the image file.
 509 * Compressed clusters are always returned one by one.
 510 *
 511 * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
 512 * cases.
 513 */
 514int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
 515                             unsigned int *bytes, uint64_t *cluster_offset)
 516{
 517    BDRVQcow2State *s = bs->opaque;
 518    unsigned int l2_index;
 519    uint64_t l1_index, l2_offset, *l2_table;
 520    int l1_bits, c;
 521    unsigned int offset_in_cluster;
 522    uint64_t bytes_available, bytes_needed, nb_clusters;
 523    QCow2ClusterType type;
 524    int ret;
 525
 526    offset_in_cluster = offset_into_cluster(s, offset);
 527    bytes_needed = (uint64_t) *bytes + offset_in_cluster;
 528
 529    l1_bits = s->l2_bits + s->cluster_bits;
 530
 531    /* compute how many bytes there are between the start of the cluster
 532     * containing offset and the end of the l1 entry */
 533    bytes_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1))
 534                    + offset_in_cluster;
 535
 536    if (bytes_needed > bytes_available) {
 537        bytes_needed = bytes_available;
 538    }
 539
 540    *cluster_offset = 0;
 541
 542    /* seek to the l2 offset in the l1 table */
 543
 544    l1_index = offset >> l1_bits;
 545    if (l1_index >= s->l1_size) {
 546        type = QCOW2_CLUSTER_UNALLOCATED;
 547        goto out;
 548    }
 549
 550    l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
 551    if (!l2_offset) {
 552        type = QCOW2_CLUSTER_UNALLOCATED;
 553        goto out;
 554    }
 555
 556    if (offset_into_cluster(s, l2_offset)) {
 557        qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
 558                                " unaligned (L1 index: %#" PRIx64 ")",
 559                                l2_offset, l1_index);
 560        return -EIO;
 561    }
 562
 563    /* load the l2 table in memory */
 564
 565    ret = l2_load(bs, l2_offset, &l2_table);
 566    if (ret < 0) {
 567        return ret;
 568    }
 569
 570    /* find the cluster offset for the given disk offset */
 571
 572    l2_index = offset_to_l2_index(s, offset);
 573    *cluster_offset = be64_to_cpu(l2_table[l2_index]);
 574
 575    nb_clusters = size_to_clusters(s, bytes_needed);
 576    /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned
 577     * integers; the minimum cluster size is 512, so this assertion is always
 578     * true */
 579    assert(nb_clusters <= INT_MAX);
 580
 581    type = qcow2_get_cluster_type(*cluster_offset);
 582    if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN ||
 583                                type == QCOW2_CLUSTER_ZERO_ALLOC)) {
 584        qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found"
 585                                " in pre-v3 image (L2 offset: %#" PRIx64
 586                                ", L2 index: %#x)", l2_offset, l2_index);
 587        ret = -EIO;
 588        goto fail;
 589    }
 590    switch (type) {
 591    case QCOW2_CLUSTER_COMPRESSED:
 592        /* Compressed clusters can only be processed one by one */
 593        c = 1;
 594        *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK;
 595        break;
 596    case QCOW2_CLUSTER_ZERO_PLAIN:
 597    case QCOW2_CLUSTER_UNALLOCATED:
 598        /* how many empty clusters ? */
 599        c = count_contiguous_clusters_unallocated(nb_clusters,
 600                                                  &l2_table[l2_index], type);
 601        *cluster_offset = 0;
 602        break;
 603    case QCOW2_CLUSTER_ZERO_ALLOC:
 604    case QCOW2_CLUSTER_NORMAL:
 605        /* how many allocated clusters ? */
 606        c = count_contiguous_clusters(nb_clusters, s->cluster_size,
 607                                      &l2_table[l2_index], QCOW_OFLAG_ZERO);
 608        *cluster_offset &= L2E_OFFSET_MASK;
 609        if (offset_into_cluster(s, *cluster_offset)) {
 610            qcow2_signal_corruption(bs, true, -1, -1,
 611                                    "Cluster allocation offset %#"
 612                                    PRIx64 " unaligned (L2 offset: %#" PRIx64
 613                                    ", L2 index: %#x)", *cluster_offset,
 614                                    l2_offset, l2_index);
 615            ret = -EIO;
 616            goto fail;
 617        }
 618        break;
 619    default:
 620        abort();
 621    }
 622
 623    qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
 624
 625    bytes_available = (int64_t)c * s->cluster_size;
 626
 627out:
 628    if (bytes_available > bytes_needed) {
 629        bytes_available = bytes_needed;
 630    }
 631
 632    /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster;
 633     * subtracting offset_in_cluster will therefore definitely yield something
 634     * not exceeding UINT_MAX */
 635    assert(bytes_available - offset_in_cluster <= UINT_MAX);
 636    *bytes = bytes_available - offset_in_cluster;
 637
 638    return type;
 639
 640fail:
 641    qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
 642    return ret;
 643}
 644
 645/*
 646 * get_cluster_table
 647 *
 648 * for a given disk offset, load (and allocate if needed)
 649 * the l2 table.
 650 *
 651 * the l2 table offset in the qcow2 file and the cluster index
 652 * in the l2 table are given to the caller.
 653 *
 654 * Returns 0 on success, -errno in failure case
 655 */
 656static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
 657                             uint64_t **new_l2_table,
 658                             int *new_l2_index)
 659{
 660    BDRVQcow2State *s = bs->opaque;
 661    unsigned int l2_index;
 662    uint64_t l1_index, l2_offset;
 663    uint64_t *l2_table = NULL;
 664    int ret;
 665
 666    /* seek to the l2 offset in the l1 table */
 667
 668    l1_index = offset >> (s->l2_bits + s->cluster_bits);
 669    if (l1_index >= s->l1_size) {
 670        ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
 671        if (ret < 0) {
 672            return ret;
 673        }
 674    }
 675
 676    assert(l1_index < s->l1_size);
 677    l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
 678    if (offset_into_cluster(s, l2_offset)) {
 679        qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
 680                                " unaligned (L1 index: %#" PRIx64 ")",
 681                                l2_offset, l1_index);
 682        return -EIO;
 683    }
 684
 685    /* seek the l2 table of the given l2 offset */
 686
 687    if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
 688        /* load the l2 table in memory */
 689        ret = l2_load(bs, l2_offset, &l2_table);
 690        if (ret < 0) {
 691            return ret;
 692        }
 693    } else {
 694        /* First allocate a new L2 table (and do COW if needed) */
 695        ret = l2_allocate(bs, l1_index, &l2_table);
 696        if (ret < 0) {
 697            return ret;
 698        }
 699
 700        /* Then decrease the refcount of the old table */
 701        if (l2_offset) {
 702            qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
 703                                QCOW2_DISCARD_OTHER);
 704        }
 705    }
 706
 707    /* find the cluster offset for the given disk offset */
 708
 709    l2_index = offset_to_l2_index(s, offset);
 710
 711    *new_l2_table = l2_table;
 712    *new_l2_index = l2_index;
 713
 714    return 0;
 715}
 716
 717/*
 718 * alloc_compressed_cluster_offset
 719 *
 720 * For a given offset of the disk image, return cluster offset in
 721 * qcow2 file.
 722 *
 723 * If the offset is not found, allocate a new compressed cluster.
 724 *
 725 * Return the cluster offset if successful,
 726 * Return 0, otherwise.
 727 *
 728 */
 729
 730uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
 731                                               uint64_t offset,
 732                                               int compressed_size)
 733{
 734    BDRVQcow2State *s = bs->opaque;
 735    int l2_index, ret;
 736    uint64_t *l2_table;
 737    int64_t cluster_offset;
 738    int nb_csectors;
 739
 740    ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
 741    if (ret < 0) {
 742        return 0;
 743    }
 744
 745    /* Compression can't overwrite anything. Fail if the cluster was already
 746     * allocated. */
 747    cluster_offset = be64_to_cpu(l2_table[l2_index]);
 748    if (cluster_offset & L2E_OFFSET_MASK) {
 749        qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
 750        return 0;
 751    }
 752
 753    cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
 754    if (cluster_offset < 0) {
 755        qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
 756        return 0;
 757    }
 758
 759    nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
 760                  (cluster_offset >> 9);
 761
 762    cluster_offset |= QCOW_OFLAG_COMPRESSED |
 763                      ((uint64_t)nb_csectors << s->csize_shift);
 764
 765    /* update L2 table */
 766
 767    /* compressed clusters never have the copied flag */
 768
 769    BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
 770    qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
 771    l2_table[l2_index] = cpu_to_be64(cluster_offset);
 772    qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
 773
 774    return cluster_offset;
 775}
 776
 777static int perform_cow(BlockDriverState *bs, QCowL2Meta *m)
 778{
 779    BDRVQcow2State *s = bs->opaque;
 780    Qcow2COWRegion *start = &m->cow_start;
 781    Qcow2COWRegion *end = &m->cow_end;
 782    unsigned buffer_size;
 783    unsigned data_bytes = end->offset - (start->offset + start->nb_bytes);
 784    bool merge_reads;
 785    uint8_t *start_buffer, *end_buffer;
 786    QEMUIOVector qiov;
 787    int ret;
 788
 789    assert(start->nb_bytes <= UINT_MAX - end->nb_bytes);
 790    assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes);
 791    assert(start->offset + start->nb_bytes <= end->offset);
 792    assert(!m->data_qiov || m->data_qiov->size == data_bytes);
 793
 794    if (start->nb_bytes == 0 && end->nb_bytes == 0) {
 795        return 0;
 796    }
 797
 798    /* If we have to read both the start and end COW regions and the
 799     * middle region is not too large then perform just one read
 800     * operation */
 801    merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384;
 802    if (merge_reads) {
 803        buffer_size = start->nb_bytes + data_bytes + end->nb_bytes;
 804    } else {
 805        /* If we have to do two reads, add some padding in the middle
 806         * if necessary to make sure that the end region is optimally
 807         * aligned. */
 808        size_t align = bdrv_opt_mem_align(bs);
 809        assert(align > 0 && align <= UINT_MAX);
 810        assert(QEMU_ALIGN_UP(start->nb_bytes, align) <=
 811               UINT_MAX - end->nb_bytes);
 812        buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes;
 813    }
 814
 815    /* Reserve a buffer large enough to store all the data that we're
 816     * going to read */
 817    start_buffer = qemu_try_blockalign(bs, buffer_size);
 818    if (start_buffer == NULL) {
 819        return -ENOMEM;
 820    }
 821    /* The part of the buffer where the end region is located */
 822    end_buffer = start_buffer + buffer_size - end->nb_bytes;
 823
 824    qemu_iovec_init(&qiov, 2 + (m->data_qiov ? m->data_qiov->niov : 0));
 825
 826    qemu_co_mutex_unlock(&s->lock);
 827    /* First we read the existing data from both COW regions. We
 828     * either read the whole region in one go, or the start and end
 829     * regions separately. */
 830    if (merge_reads) {
 831        qemu_iovec_add(&qiov, start_buffer, buffer_size);
 832        ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov);
 833    } else {
 834        qemu_iovec_add(&qiov, start_buffer, start->nb_bytes);
 835        ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov);
 836        if (ret < 0) {
 837            goto fail;
 838        }
 839
 840        qemu_iovec_reset(&qiov);
 841        qemu_iovec_add(&qiov, end_buffer, end->nb_bytes);
 842        ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov);
 843    }
 844    if (ret < 0) {
 845        goto fail;
 846    }
 847
 848    /* Encrypt the data if necessary before writing it */
 849    if (bs->encrypted) {
 850        if (!do_perform_cow_encrypt(bs, m->offset, m->alloc_offset,
 851                                    start->offset, start_buffer,
 852                                    start->nb_bytes) ||
 853            !do_perform_cow_encrypt(bs, m->offset, m->alloc_offset,
 854                                    end->offset, end_buffer, end->nb_bytes)) {
 855            ret = -EIO;
 856            goto fail;
 857        }
 858    }
 859
 860    /* And now we can write everything. If we have the guest data we
 861     * can write everything in one single operation */
 862    if (m->data_qiov) {
 863        qemu_iovec_reset(&qiov);
 864        if (start->nb_bytes) {
 865            qemu_iovec_add(&qiov, start_buffer, start->nb_bytes);
 866        }
 867        qemu_iovec_concat(&qiov, m->data_qiov, 0, data_bytes);
 868        if (end->nb_bytes) {
 869            qemu_iovec_add(&qiov, end_buffer, end->nb_bytes);
 870        }
 871        /* NOTE: we have a write_aio blkdebug event here followed by
 872         * a cow_write one in do_perform_cow_write(), but there's only
 873         * one single I/O operation */
 874        BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
 875        ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov);
 876    } else {
 877        /* If there's no guest data then write both COW regions separately */
 878        qemu_iovec_reset(&qiov);
 879        qemu_iovec_add(&qiov, start_buffer, start->nb_bytes);
 880        ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov);
 881        if (ret < 0) {
 882            goto fail;
 883        }
 884
 885        qemu_iovec_reset(&qiov);
 886        qemu_iovec_add(&qiov, end_buffer, end->nb_bytes);
 887        ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov);
 888    }
 889
 890fail:
 891    qemu_co_mutex_lock(&s->lock);
 892
 893    /*
 894     * Before we update the L2 table to actually point to the new cluster, we
 895     * need to be sure that the refcounts have been increased and COW was
 896     * handled.
 897     */
 898    if (ret == 0) {
 899        qcow2_cache_depends_on_flush(s->l2_table_cache);
 900    }
 901
 902    qemu_vfree(start_buffer);
 903    qemu_iovec_destroy(&qiov);
 904    return ret;
 905}
 906
 907int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
 908{
 909    BDRVQcow2State *s = bs->opaque;
 910    int i, j = 0, l2_index, ret;
 911    uint64_t *old_cluster, *l2_table;
 912    uint64_t cluster_offset = m->alloc_offset;
 913
 914    trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
 915    assert(m->nb_clusters > 0);
 916
 917    old_cluster = g_try_new(uint64_t, m->nb_clusters);
 918    if (old_cluster == NULL) {
 919        ret = -ENOMEM;
 920        goto err;
 921    }
 922
 923    /* copy content of unmodified sectors */
 924    ret = perform_cow(bs, m);
 925    if (ret < 0) {
 926        goto err;
 927    }
 928
 929    /* Update L2 table. */
 930    if (s->use_lazy_refcounts) {
 931        qcow2_mark_dirty(bs);
 932    }
 933    if (qcow2_need_accurate_refcounts(s)) {
 934        qcow2_cache_set_dependency(bs, s->l2_table_cache,
 935                                   s->refcount_block_cache);
 936    }
 937
 938    ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index);
 939    if (ret < 0) {
 940        goto err;
 941    }
 942    qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
 943
 944    assert(l2_index + m->nb_clusters <= s->l2_size);
 945    for (i = 0; i < m->nb_clusters; i++) {
 946        /* if two concurrent writes happen to the same unallocated cluster
 947         * each write allocates separate cluster and writes data concurrently.
 948         * The first one to complete updates l2 table with pointer to its
 949         * cluster the second one has to do RMW (which is done above by
 950         * perform_cow()), update l2 table with its cluster pointer and free
 951         * old cluster. This is what this loop does */
 952        if (l2_table[l2_index + i] != 0) {
 953            old_cluster[j++] = l2_table[l2_index + i];
 954        }
 955
 956        l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
 957                    (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
 958     }
 959
 960
 961    qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
 962
 963    /*
 964     * If this was a COW, we need to decrease the refcount of the old cluster.
 965     *
 966     * Don't discard clusters that reach a refcount of 0 (e.g. compressed
 967     * clusters), the next write will reuse them anyway.
 968     */
 969    if (!m->keep_old_clusters && j != 0) {
 970        for (i = 0; i < j; i++) {
 971            qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1,
 972                                    QCOW2_DISCARD_NEVER);
 973        }
 974    }
 975
 976    ret = 0;
 977err:
 978    g_free(old_cluster);
 979    return ret;
 980 }
 981
 982/*
 983 * Returns the number of contiguous clusters that can be used for an allocating
 984 * write, but require COW to be performed (this includes yet unallocated space,
 985 * which must copy from the backing file)
 986 */
 987static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters,
 988    uint64_t *l2_table, int l2_index)
 989{
 990    int i;
 991
 992    for (i = 0; i < nb_clusters; i++) {
 993        uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]);
 994        QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry);
 995
 996        switch(cluster_type) {
 997        case QCOW2_CLUSTER_NORMAL:
 998            if (l2_entry & QCOW_OFLAG_COPIED) {
 999                goto out;
1000            }
1001            break;
1002        case QCOW2_CLUSTER_UNALLOCATED:
1003        case QCOW2_CLUSTER_COMPRESSED:
1004        case QCOW2_CLUSTER_ZERO_PLAIN:
1005        case QCOW2_CLUSTER_ZERO_ALLOC:
1006            break;
1007        default:
1008            abort();
1009        }
1010    }
1011
1012out:
1013    assert(i <= nb_clusters);
1014    return i;
1015}
1016
1017/*
1018 * Check if there already is an AIO write request in flight which allocates
1019 * the same cluster. In this case we need to wait until the previous
1020 * request has completed and updated the L2 table accordingly.
1021 *
1022 * Returns:
1023 *   0       if there was no dependency. *cur_bytes indicates the number of
1024 *           bytes from guest_offset that can be read before the next
1025 *           dependency must be processed (or the request is complete)
1026 *
1027 *   -EAGAIN if we had to wait for another request, previously gathered
1028 *           information on cluster allocation may be invalid now. The caller
1029 *           must start over anyway, so consider *cur_bytes undefined.
1030 */
1031static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,
1032    uint64_t *cur_bytes, QCowL2Meta **m)
1033{
1034    BDRVQcow2State *s = bs->opaque;
1035    QCowL2Meta *old_alloc;
1036    uint64_t bytes = *cur_bytes;
1037
1038    QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
1039
1040        uint64_t start = guest_offset;
1041        uint64_t end = start + bytes;
1042        uint64_t old_start = l2meta_cow_start(old_alloc);
1043        uint64_t old_end = l2meta_cow_end(old_alloc);
1044
1045        if (end <= old_start || start >= old_end) {
1046            /* No intersection */
1047        } else {
1048            if (start < old_start) {
1049                /* Stop at the start of a running allocation */
1050                bytes = old_start - start;
1051            } else {
1052                bytes = 0;
1053            }
1054
1055            /* Stop if already an l2meta exists. After yielding, it wouldn't
1056             * be valid any more, so we'd have to clean up the old L2Metas
1057             * and deal with requests depending on them before starting to
1058             * gather new ones. Not worth the trouble. */
1059            if (bytes == 0 && *m) {
1060                *cur_bytes = 0;
1061                return 0;
1062            }
1063
1064            if (bytes == 0) {
1065                /* Wait for the dependency to complete. We need to recheck
1066                 * the free/allocated clusters when we continue. */
1067                qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock);
1068                return -EAGAIN;
1069            }
1070        }
1071    }
1072
1073    /* Make sure that existing clusters and new allocations are only used up to
1074     * the next dependency if we shortened the request above */
1075    *cur_bytes = bytes;
1076
1077    return 0;
1078}
1079
1080/*
1081 * Checks how many already allocated clusters that don't require a copy on
1082 * write there are at the given guest_offset (up to *bytes). If
1083 * *host_offset is not zero, only physically contiguous clusters beginning at
1084 * this host offset are counted.
1085 *
1086 * Note that guest_offset may not be cluster aligned. In this case, the
1087 * returned *host_offset points to exact byte referenced by guest_offset and
1088 * therefore isn't cluster aligned as well.
1089 *
1090 * Returns:
1091 *   0:     if no allocated clusters are available at the given offset.
1092 *          *bytes is normally unchanged. It is set to 0 if the cluster
1093 *          is allocated and doesn't need COW, but doesn't have the right
1094 *          physical offset.
1095 *
1096 *   1:     if allocated clusters that don't require a COW are available at
1097 *          the requested offset. *bytes may have decreased and describes
1098 *          the length of the area that can be written to.
1099 *
1100 *  -errno: in error cases
1101 */
1102static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
1103    uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
1104{
1105    BDRVQcow2State *s = bs->opaque;
1106    int l2_index;
1107    uint64_t cluster_offset;
1108    uint64_t *l2_table;
1109    uint64_t nb_clusters;
1110    unsigned int keep_clusters;
1111    int ret;
1112
1113    trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset,
1114                              *bytes);
1115
1116    assert(*host_offset == 0 ||    offset_into_cluster(s, guest_offset)
1117                                == offset_into_cluster(s, *host_offset));
1118
1119    /*
1120     * Calculate the number of clusters to look for. We stop at L2 table
1121     * boundaries to keep things simple.
1122     */
1123    nb_clusters =
1124        size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1125
1126    l2_index = offset_to_l2_index(s, guest_offset);
1127    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1128    assert(nb_clusters <= INT_MAX);
1129
1130    /* Find L2 entry for the first involved cluster */
1131    ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1132    if (ret < 0) {
1133        return ret;
1134    }
1135
1136    cluster_offset = be64_to_cpu(l2_table[l2_index]);
1137
1138    /* Check how many clusters are already allocated and don't need COW */
1139    if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
1140        && (cluster_offset & QCOW_OFLAG_COPIED))
1141    {
1142        /* If a specific host_offset is required, check it */
1143        bool offset_matches =
1144            (cluster_offset & L2E_OFFSET_MASK) == *host_offset;
1145
1146        if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) {
1147            qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1148                                    "%#llx unaligned (guest offset: %#" PRIx64
1149                                    ")", cluster_offset & L2E_OFFSET_MASK,
1150                                    guest_offset);
1151            ret = -EIO;
1152            goto out;
1153        }
1154
1155        if (*host_offset != 0 && !offset_matches) {
1156            *bytes = 0;
1157            ret = 0;
1158            goto out;
1159        }
1160
1161        /* We keep all QCOW_OFLAG_COPIED clusters */
1162        keep_clusters =
1163            count_contiguous_clusters(nb_clusters, s->cluster_size,
1164                                      &l2_table[l2_index],
1165                                      QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
1166        assert(keep_clusters <= nb_clusters);
1167
1168        *bytes = MIN(*bytes,
1169                 keep_clusters * s->cluster_size
1170                 - offset_into_cluster(s, guest_offset));
1171
1172        ret = 1;
1173    } else {
1174        ret = 0;
1175    }
1176
1177    /* Cleanup */
1178out:
1179    qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1180
1181    /* Only return a host offset if we actually made progress. Otherwise we
1182     * would make requirements for handle_alloc() that it can't fulfill */
1183    if (ret > 0) {
1184        *host_offset = (cluster_offset & L2E_OFFSET_MASK)
1185                     + offset_into_cluster(s, guest_offset);
1186    }
1187
1188    return ret;
1189}
1190
1191/*
1192 * Allocates new clusters for the given guest_offset.
1193 *
1194 * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
1195 * contain the number of clusters that have been allocated and are contiguous
1196 * in the image file.
1197 *
1198 * If *host_offset is non-zero, it specifies the offset in the image file at
1199 * which the new clusters must start. *nb_clusters can be 0 on return in this
1200 * case if the cluster at host_offset is already in use. If *host_offset is
1201 * zero, the clusters can be allocated anywhere in the image file.
1202 *
1203 * *host_offset is updated to contain the offset into the image file at which
1204 * the first allocated cluster starts.
1205 *
1206 * Return 0 on success and -errno in error cases. -EAGAIN means that the
1207 * function has been waiting for another request and the allocation must be
1208 * restarted, but the whole request should not be failed.
1209 */
1210static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset,
1211                                   uint64_t *host_offset, uint64_t *nb_clusters)
1212{
1213    BDRVQcow2State *s = bs->opaque;
1214
1215    trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset,
1216                                         *host_offset, *nb_clusters);
1217
1218    /* Allocate new clusters */
1219    trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
1220    if (*host_offset == 0) {
1221        int64_t cluster_offset =
1222            qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size);
1223        if (cluster_offset < 0) {
1224            return cluster_offset;
1225        }
1226        *host_offset = cluster_offset;
1227        return 0;
1228    } else {
1229        int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
1230        if (ret < 0) {
1231            return ret;
1232        }
1233        *nb_clusters = ret;
1234        return 0;
1235    }
1236}
1237
1238/*
1239 * Allocates new clusters for an area that either is yet unallocated or needs a
1240 * copy on write. If *host_offset is non-zero, clusters are only allocated if
1241 * the new allocation can match the specified host offset.
1242 *
1243 * Note that guest_offset may not be cluster aligned. In this case, the
1244 * returned *host_offset points to exact byte referenced by guest_offset and
1245 * therefore isn't cluster aligned as well.
1246 *
1247 * Returns:
1248 *   0:     if no clusters could be allocated. *bytes is set to 0,
1249 *          *host_offset is left unchanged.
1250 *
1251 *   1:     if new clusters were allocated. *bytes may be decreased if the
1252 *          new allocation doesn't cover all of the requested area.
1253 *          *host_offset is updated to contain the host offset of the first
1254 *          newly allocated cluster.
1255 *
1256 *  -errno: in error cases
1257 */
1258static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
1259    uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
1260{
1261    BDRVQcow2State *s = bs->opaque;
1262    int l2_index;
1263    uint64_t *l2_table;
1264    uint64_t entry;
1265    uint64_t nb_clusters;
1266    int ret;
1267    bool keep_old_clusters = false;
1268
1269    uint64_t alloc_cluster_offset = 0;
1270
1271    trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
1272                             *bytes);
1273    assert(*bytes > 0);
1274
1275    /*
1276     * Calculate the number of clusters to look for. We stop at L2 table
1277     * boundaries to keep things simple.
1278     */
1279    nb_clusters =
1280        size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1281
1282    l2_index = offset_to_l2_index(s, guest_offset);
1283    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1284    assert(nb_clusters <= INT_MAX);
1285
1286    /* Find L2 entry for the first involved cluster */
1287    ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1288    if (ret < 0) {
1289        return ret;
1290    }
1291
1292    entry = be64_to_cpu(l2_table[l2_index]);
1293
1294    /* For the moment, overwrite compressed clusters one by one */
1295    if (entry & QCOW_OFLAG_COMPRESSED) {
1296        nb_clusters = 1;
1297    } else {
1298        nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index);
1299    }
1300
1301    /* This function is only called when there were no non-COW clusters, so if
1302     * we can't find any unallocated or COW clusters either, something is
1303     * wrong with our code. */
1304    assert(nb_clusters > 0);
1305
1306    if (qcow2_get_cluster_type(entry) == QCOW2_CLUSTER_ZERO_ALLOC &&
1307        (entry & QCOW_OFLAG_COPIED) &&
1308        (!*host_offset ||
1309         start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK)))
1310    {
1311        int preallocated_nb_clusters;
1312
1313        if (offset_into_cluster(s, entry & L2E_OFFSET_MASK)) {
1314            qcow2_signal_corruption(bs, true, -1, -1, "Preallocated zero "
1315                                    "cluster offset %#llx unaligned (guest "
1316                                    "offset: %#" PRIx64 ")",
1317                                    entry & L2E_OFFSET_MASK, guest_offset);
1318            ret = -EIO;
1319            goto fail;
1320        }
1321
1322        /* Try to reuse preallocated zero clusters; contiguous normal clusters
1323         * would be fine, too, but count_cow_clusters() above has limited
1324         * nb_clusters already to a range of COW clusters */
1325        preallocated_nb_clusters =
1326            count_contiguous_clusters(nb_clusters, s->cluster_size,
1327                                      &l2_table[l2_index], QCOW_OFLAG_COPIED);
1328        assert(preallocated_nb_clusters > 0);
1329
1330        nb_clusters = preallocated_nb_clusters;
1331        alloc_cluster_offset = entry & L2E_OFFSET_MASK;
1332
1333        /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2()
1334         * should not free them. */
1335        keep_old_clusters = true;
1336    }
1337
1338    qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1339
1340    if (!alloc_cluster_offset) {
1341        /* Allocate, if necessary at a given offset in the image file */
1342        alloc_cluster_offset = start_of_cluster(s, *host_offset);
1343        ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
1344                                      &nb_clusters);
1345        if (ret < 0) {
1346            goto fail;
1347        }
1348
1349        /* Can't extend contiguous allocation */
1350        if (nb_clusters == 0) {
1351            *bytes = 0;
1352            return 0;
1353        }
1354
1355        /* !*host_offset would overwrite the image header and is reserved for
1356         * "no host offset preferred". If 0 was a valid host offset, it'd
1357         * trigger the following overlap check; do that now to avoid having an
1358         * invalid value in *host_offset. */
1359        if (!alloc_cluster_offset) {
1360            ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset,
1361                                                nb_clusters * s->cluster_size);
1362            assert(ret < 0);
1363            goto fail;
1364        }
1365    }
1366
1367    /*
1368     * Save info needed for meta data update.
1369     *
1370     * requested_bytes: Number of bytes from the start of the first
1371     * newly allocated cluster to the end of the (possibly shortened
1372     * before) write request.
1373     *
1374     * avail_bytes: Number of bytes from the start of the first
1375     * newly allocated to the end of the last newly allocated cluster.
1376     *
1377     * nb_bytes: The number of bytes from the start of the first
1378     * newly allocated cluster to the end of the area that the write
1379     * request actually writes to (excluding COW at the end)
1380     */
1381    uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset);
1382    int avail_bytes = MIN(INT_MAX, nb_clusters << s->cluster_bits);
1383    int nb_bytes = MIN(requested_bytes, avail_bytes);
1384    QCowL2Meta *old_m = *m;
1385
1386    *m = g_malloc0(sizeof(**m));
1387
1388    **m = (QCowL2Meta) {
1389        .next           = old_m,
1390
1391        .alloc_offset   = alloc_cluster_offset,
1392        .offset         = start_of_cluster(s, guest_offset),
1393        .nb_clusters    = nb_clusters,
1394
1395        .keep_old_clusters  = keep_old_clusters,
1396
1397        .cow_start = {
1398            .offset     = 0,
1399            .nb_bytes   = offset_into_cluster(s, guest_offset),
1400        },
1401        .cow_end = {
1402            .offset     = nb_bytes,
1403            .nb_bytes   = avail_bytes - nb_bytes,
1404        },
1405    };
1406    qemu_co_queue_init(&(*m)->dependent_requests);
1407    QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
1408
1409    *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
1410    *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
1411    assert(*bytes != 0);
1412
1413    return 1;
1414
1415fail:
1416    if (*m && (*m)->nb_clusters > 0) {
1417        QLIST_REMOVE(*m, next_in_flight);
1418    }
1419    return ret;
1420}
1421
1422/*
1423 * alloc_cluster_offset
1424 *
1425 * For a given offset on the virtual disk, find the cluster offset in qcow2
1426 * file. If the offset is not found, allocate a new cluster.
1427 *
1428 * If the cluster was already allocated, m->nb_clusters is set to 0 and
1429 * other fields in m are meaningless.
1430 *
1431 * If the cluster is newly allocated, m->nb_clusters is set to the number of
1432 * contiguous clusters that have been allocated. In this case, the other
1433 * fields of m are valid and contain information about the first allocated
1434 * cluster.
1435 *
1436 * If the request conflicts with another write request in flight, the coroutine
1437 * is queued and will be reentered when the dependency has completed.
1438 *
1439 * Return 0 on success and -errno in error cases
1440 */
1441int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
1442                               unsigned int *bytes, uint64_t *host_offset,
1443                               QCowL2Meta **m)
1444{
1445    BDRVQcow2State *s = bs->opaque;
1446    uint64_t start, remaining;
1447    uint64_t cluster_offset;
1448    uint64_t cur_bytes;
1449    int ret;
1450
1451    trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes);
1452
1453again:
1454    start = offset;
1455    remaining = *bytes;
1456    cluster_offset = 0;
1457    *host_offset = 0;
1458    cur_bytes = 0;
1459    *m = NULL;
1460
1461    while (true) {
1462
1463        if (!*host_offset) {
1464            *host_offset = start_of_cluster(s, cluster_offset);
1465        }
1466
1467        assert(remaining >= cur_bytes);
1468
1469        start           += cur_bytes;
1470        remaining       -= cur_bytes;
1471        cluster_offset  += cur_bytes;
1472
1473        if (remaining == 0) {
1474            break;
1475        }
1476
1477        cur_bytes = remaining;
1478
1479        /*
1480         * Now start gathering as many contiguous clusters as possible:
1481         *
1482         * 1. Check for overlaps with in-flight allocations
1483         *
1484         *      a) Overlap not in the first cluster -> shorten this request and
1485         *         let the caller handle the rest in its next loop iteration.
1486         *
1487         *      b) Real overlaps of two requests. Yield and restart the search
1488         *         for contiguous clusters (the situation could have changed
1489         *         while we were sleeping)
1490         *
1491         *      c) TODO: Request starts in the same cluster as the in-flight
1492         *         allocation ends. Shorten the COW of the in-fight allocation,
1493         *         set cluster_offset to write to the same cluster and set up
1494         *         the right synchronisation between the in-flight request and
1495         *         the new one.
1496         */
1497        ret = handle_dependencies(bs, start, &cur_bytes, m);
1498        if (ret == -EAGAIN) {
1499            /* Currently handle_dependencies() doesn't yield if we already had
1500             * an allocation. If it did, we would have to clean up the L2Meta
1501             * structs before starting over. */
1502            assert(*m == NULL);
1503            goto again;
1504        } else if (ret < 0) {
1505            return ret;
1506        } else if (cur_bytes == 0) {
1507            break;
1508        } else {
1509            /* handle_dependencies() may have decreased cur_bytes (shortened
1510             * the allocations below) so that the next dependency is processed
1511             * correctly during the next loop iteration. */
1512        }
1513
1514        /*
1515         * 2. Count contiguous COPIED clusters.
1516         */
1517        ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m);
1518        if (ret < 0) {
1519            return ret;
1520        } else if (ret) {
1521            continue;
1522        } else if (cur_bytes == 0) {
1523            break;
1524        }
1525
1526        /*
1527         * 3. If the request still hasn't completed, allocate new clusters,
1528         *    considering any cluster_offset of steps 1c or 2.
1529         */
1530        ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m);
1531        if (ret < 0) {
1532            return ret;
1533        } else if (ret) {
1534            continue;
1535        } else {
1536            assert(cur_bytes == 0);
1537            break;
1538        }
1539    }
1540
1541    *bytes -= remaining;
1542    assert(*bytes > 0);
1543    assert(*host_offset != 0);
1544
1545    return 0;
1546}
1547
1548static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1549                             const uint8_t *buf, int buf_size)
1550{
1551    z_stream strm1, *strm = &strm1;
1552    int ret, out_len;
1553
1554    memset(strm, 0, sizeof(*strm));
1555
1556    strm->next_in = (uint8_t *)buf;
1557    strm->avail_in = buf_size;
1558    strm->next_out = out_buf;
1559    strm->avail_out = out_buf_size;
1560
1561    ret = inflateInit2(strm, -12);
1562    if (ret != Z_OK)
1563        return -1;
1564    ret = inflate(strm, Z_FINISH);
1565    out_len = strm->next_out - out_buf;
1566    if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1567        out_len != out_buf_size) {
1568        inflateEnd(strm);
1569        return -1;
1570    }
1571    inflateEnd(strm);
1572    return 0;
1573}
1574
1575int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
1576{
1577    BDRVQcow2State *s = bs->opaque;
1578    int ret, csize, nb_csectors, sector_offset;
1579    uint64_t coffset;
1580
1581    coffset = cluster_offset & s->cluster_offset_mask;
1582    if (s->cluster_cache_offset != coffset) {
1583        nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1584        sector_offset = coffset & 511;
1585        csize = nb_csectors * 512 - sector_offset;
1586
1587        /* Allocate buffers on first decompress operation, most images are
1588         * uncompressed and the memory overhead can be avoided.  The buffers
1589         * are freed in .bdrv_close().
1590         */
1591        if (!s->cluster_data) {
1592            /* one more sector for decompressed data alignment */
1593            s->cluster_data = qemu_try_blockalign(bs->file->bs,
1594                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size + 512);
1595            if (!s->cluster_data) {
1596                return -ENOMEM;
1597            }
1598        }
1599        if (!s->cluster_cache) {
1600            s->cluster_cache = g_malloc(s->cluster_size);
1601        }
1602
1603        BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
1604        ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data,
1605                        nb_csectors);
1606        if (ret < 0) {
1607            return ret;
1608        }
1609        if (decompress_buffer(s->cluster_cache, s->cluster_size,
1610                              s->cluster_data + sector_offset, csize) < 0) {
1611            return -EIO;
1612        }
1613        s->cluster_cache_offset = coffset;
1614    }
1615    return 0;
1616}
1617
1618/*
1619 * This discards as many clusters of nb_clusters as possible at once (i.e.
1620 * all clusters in the same L2 table) and returns the number of discarded
1621 * clusters.
1622 */
1623static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
1624                             uint64_t nb_clusters, enum qcow2_discard_type type,
1625                             bool full_discard)
1626{
1627    BDRVQcow2State *s = bs->opaque;
1628    uint64_t *l2_table;
1629    int l2_index;
1630    int ret;
1631    int i;
1632
1633    ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1634    if (ret < 0) {
1635        return ret;
1636    }
1637
1638    /* Limit nb_clusters to one L2 table */
1639    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1640    assert(nb_clusters <= INT_MAX);
1641
1642    for (i = 0; i < nb_clusters; i++) {
1643        uint64_t old_l2_entry;
1644
1645        old_l2_entry = be64_to_cpu(l2_table[l2_index + i]);
1646
1647        /*
1648         * If full_discard is false, make sure that a discarded area reads back
1649         * as zeroes for v3 images (we cannot do it for v2 without actually
1650         * writing a zero-filled buffer). We can skip the operation if the
1651         * cluster is already marked as zero, or if it's unallocated and we
1652         * don't have a backing file.
1653         *
1654         * TODO We might want to use bdrv_block_status(bs) here, but we're
1655         * holding s->lock, so that doesn't work today.
1656         *
1657         * If full_discard is true, the sector should not read back as zeroes,
1658         * but rather fall through to the backing file.
1659         */
1660        switch (qcow2_get_cluster_type(old_l2_entry)) {
1661        case QCOW2_CLUSTER_UNALLOCATED:
1662            if (full_discard || !bs->backing) {
1663                continue;
1664            }
1665            break;
1666
1667        case QCOW2_CLUSTER_ZERO_PLAIN:
1668            if (!full_discard) {
1669                continue;
1670            }
1671            break;
1672
1673        case QCOW2_CLUSTER_ZERO_ALLOC:
1674        case QCOW2_CLUSTER_NORMAL:
1675        case QCOW2_CLUSTER_COMPRESSED:
1676            break;
1677
1678        default:
1679            abort();
1680        }
1681
1682        /* First remove L2 entries */
1683        qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1684        if (!full_discard && s->qcow_version >= 3) {
1685            l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1686        } else {
1687            l2_table[l2_index + i] = cpu_to_be64(0);
1688        }
1689
1690        /* Then decrease the refcount */
1691        qcow2_free_any_clusters(bs, old_l2_entry, 1, type);
1692    }
1693
1694    qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1695
1696    return nb_clusters;
1697}
1698
1699int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset,
1700                          uint64_t bytes, enum qcow2_discard_type type,
1701                          bool full_discard)
1702{
1703    BDRVQcow2State *s = bs->opaque;
1704    uint64_t end_offset = offset + bytes;
1705    uint64_t nb_clusters;
1706    int64_t cleared;
1707    int ret;
1708
1709    /* Caller must pass aligned values, except at image end */
1710    assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
1711    assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) ||
1712           end_offset == bs->total_sectors << BDRV_SECTOR_BITS);
1713
1714    nb_clusters = size_to_clusters(s, bytes);
1715
1716    s->cache_discards = true;
1717
1718    /* Each L2 table is handled by its own loop iteration */
1719    while (nb_clusters > 0) {
1720        cleared = discard_single_l2(bs, offset, nb_clusters, type,
1721                                    full_discard);
1722        if (cleared < 0) {
1723            ret = cleared;
1724            goto fail;
1725        }
1726
1727        nb_clusters -= cleared;
1728        offset += (cleared * s->cluster_size);
1729    }
1730
1731    ret = 0;
1732fail:
1733    s->cache_discards = false;
1734    qcow2_process_discards(bs, ret);
1735
1736    return ret;
1737}
1738
1739/*
1740 * This zeroes as many clusters of nb_clusters as possible at once (i.e.
1741 * all clusters in the same L2 table) and returns the number of zeroed
1742 * clusters.
1743 */
1744static int zero_single_l2(BlockDriverState *bs, uint64_t offset,
1745                          uint64_t nb_clusters, int flags)
1746{
1747    BDRVQcow2State *s = bs->opaque;
1748    uint64_t *l2_table;
1749    int l2_index;
1750    int ret;
1751    int i;
1752    bool unmap = !!(flags & BDRV_REQ_MAY_UNMAP);
1753
1754    ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1755    if (ret < 0) {
1756        return ret;
1757    }
1758
1759    /* Limit nb_clusters to one L2 table */
1760    nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1761    assert(nb_clusters <= INT_MAX);
1762
1763    for (i = 0; i < nb_clusters; i++) {
1764        uint64_t old_offset;
1765        QCow2ClusterType cluster_type;
1766
1767        old_offset = be64_to_cpu(l2_table[l2_index + i]);
1768
1769        /*
1770         * Minimize L2 changes if the cluster already reads back as
1771         * zeroes with correct allocation.
1772         */
1773        cluster_type = qcow2_get_cluster_type(old_offset);
1774        if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN ||
1775            (cluster_type == QCOW2_CLUSTER_ZERO_ALLOC && !unmap)) {
1776            continue;
1777        }
1778
1779        qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1780        if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) {
1781            l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1782            qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
1783        } else {
1784            l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO);
1785        }
1786    }
1787
1788    qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1789
1790    return nb_clusters;
1791}
1792
1793int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset,
1794                          uint64_t bytes, int flags)
1795{
1796    BDRVQcow2State *s = bs->opaque;
1797    uint64_t end_offset = offset + bytes;
1798    uint64_t nb_clusters;
1799    int64_t cleared;
1800    int ret;
1801
1802    /* Caller must pass aligned values, except at image end */
1803    assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
1804    assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) ||
1805           end_offset == bs->total_sectors << BDRV_SECTOR_BITS);
1806
1807    /* The zero flag is only supported by version 3 and newer */
1808    if (s->qcow_version < 3) {
1809        return -ENOTSUP;
1810    }
1811
1812    /* Each L2 table is handled by its own loop iteration */
1813    nb_clusters = size_to_clusters(s, bytes);
1814
1815    s->cache_discards = true;
1816
1817    while (nb_clusters > 0) {
1818        cleared = zero_single_l2(bs, offset, nb_clusters, flags);
1819        if (cleared < 0) {
1820            ret = cleared;
1821            goto fail;
1822        }
1823
1824        nb_clusters -= cleared;
1825        offset += (cleared * s->cluster_size);
1826    }
1827
1828    ret = 0;
1829fail:
1830    s->cache_discards = false;
1831    qcow2_process_discards(bs, ret);
1832
1833    return ret;
1834}
1835
1836/*
1837 * Expands all zero clusters in a specific L1 table (or deallocates them, for
1838 * non-backed non-pre-allocated zero clusters).
1839 *
1840 * l1_entries and *visited_l1_entries are used to keep track of progress for
1841 * status_cb(). l1_entries contains the total number of L1 entries and
1842 * *visited_l1_entries counts all visited L1 entries.
1843 */
1844static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
1845                                      int l1_size, int64_t *visited_l1_entries,
1846                                      int64_t l1_entries,
1847                                      BlockDriverAmendStatusCB *status_cb,
1848                                      void *cb_opaque)
1849{
1850    BDRVQcow2State *s = bs->opaque;
1851    bool is_active_l1 = (l1_table == s->l1_table);
1852    uint64_t *l2_table = NULL;
1853    int ret;
1854    int i, j;
1855
1856    if (!is_active_l1) {
1857        /* inactive L2 tables require a buffer to be stored in when loading
1858         * them from disk */
1859        l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size);
1860        if (l2_table == NULL) {
1861            return -ENOMEM;
1862        }
1863    }
1864
1865    for (i = 0; i < l1_size; i++) {
1866        uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK;
1867        bool l2_dirty = false;
1868        uint64_t l2_refcount;
1869
1870        if (!l2_offset) {
1871            /* unallocated */
1872            (*visited_l1_entries)++;
1873            if (status_cb) {
1874                status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
1875            }
1876            continue;
1877        }
1878
1879        if (offset_into_cluster(s, l2_offset)) {
1880            qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1881                                    PRIx64 " unaligned (L1 index: %#x)",
1882                                    l2_offset, i);
1883            ret = -EIO;
1884            goto fail;
1885        }
1886
1887        if (is_active_l1) {
1888            /* get active L2 tables from cache */
1889            ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1890                    (void **)&l2_table);
1891        } else {
1892            /* load inactive L2 tables from disk */
1893            ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE,
1894                            (void *)l2_table, s->cluster_sectors);
1895        }
1896        if (ret < 0) {
1897            goto fail;
1898        }
1899
1900        ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1901                                 &l2_refcount);
1902        if (ret < 0) {
1903            goto fail;
1904        }
1905
1906        for (j = 0; j < s->l2_size; j++) {
1907            uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1908            int64_t offset = l2_entry & L2E_OFFSET_MASK;
1909            QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry);
1910
1911            if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN &&
1912                cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) {
1913                continue;
1914            }
1915
1916            if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) {
1917                if (!bs->backing) {
1918                    /* not backed; therefore we can simply deallocate the
1919                     * cluster */
1920                    l2_table[j] = 0;
1921                    l2_dirty = true;
1922                    continue;
1923                }
1924
1925                offset = qcow2_alloc_clusters(bs, s->cluster_size);
1926                if (offset < 0) {
1927                    ret = offset;
1928                    goto fail;
1929                }
1930
1931                if (l2_refcount > 1) {
1932                    /* For shared L2 tables, set the refcount accordingly (it is
1933                     * already 1 and needs to be l2_refcount) */
1934                    ret = qcow2_update_cluster_refcount(bs,
1935                            offset >> s->cluster_bits,
1936                            refcount_diff(1, l2_refcount), false,
1937                            QCOW2_DISCARD_OTHER);
1938                    if (ret < 0) {
1939                        qcow2_free_clusters(bs, offset, s->cluster_size,
1940                                            QCOW2_DISCARD_OTHER);
1941                        goto fail;
1942                    }
1943                }
1944            }
1945
1946            if (offset_into_cluster(s, offset)) {
1947                qcow2_signal_corruption(bs, true, -1, -1,
1948                                        "Cluster allocation offset "
1949                                        "%#" PRIx64 " unaligned (L2 offset: %#"
1950                                        PRIx64 ", L2 index: %#x)", offset,
1951                                        l2_offset, j);
1952                if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) {
1953                    qcow2_free_clusters(bs, offset, s->cluster_size,
1954                                        QCOW2_DISCARD_ALWAYS);
1955                }
1956                ret = -EIO;
1957                goto fail;
1958            }
1959
1960            ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
1961            if (ret < 0) {
1962                if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) {
1963                    qcow2_free_clusters(bs, offset, s->cluster_size,
1964                                        QCOW2_DISCARD_ALWAYS);
1965                }
1966                goto fail;
1967            }
1968
1969            ret = bdrv_pwrite_zeroes(bs->file, offset, s->cluster_size, 0);
1970            if (ret < 0) {
1971                if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) {
1972                    qcow2_free_clusters(bs, offset, s->cluster_size,
1973                                        QCOW2_DISCARD_ALWAYS);
1974                }
1975                goto fail;
1976            }
1977
1978            if (l2_refcount == 1) {
1979                l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED);
1980            } else {
1981                l2_table[j] = cpu_to_be64(offset);
1982            }
1983            l2_dirty = true;
1984        }
1985
1986        if (is_active_l1) {
1987            if (l2_dirty) {
1988                qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1989                qcow2_cache_depends_on_flush(s->l2_table_cache);
1990            }
1991            qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1992        } else {
1993            if (l2_dirty) {
1994                ret = qcow2_pre_write_overlap_check(bs,
1995                        QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset,
1996                        s->cluster_size);
1997                if (ret < 0) {
1998                    goto fail;
1999                }
2000
2001                ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE,
2002                                 (void *)l2_table, s->cluster_sectors);
2003                if (ret < 0) {
2004                    goto fail;
2005                }
2006            }
2007        }
2008
2009        (*visited_l1_entries)++;
2010        if (status_cb) {
2011            status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
2012        }
2013    }
2014
2015    ret = 0;
2016
2017fail:
2018    if (l2_table) {
2019        if (!is_active_l1) {
2020            qemu_vfree(l2_table);
2021        } else {
2022            qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
2023        }
2024    }
2025    return ret;
2026}
2027
2028/*
2029 * For backed images, expands all zero clusters on the image. For non-backed
2030 * images, deallocates all non-pre-allocated zero clusters (and claims the
2031 * allocation for pre-allocated ones). This is important for downgrading to a
2032 * qcow2 version which doesn't yet support metadata zero clusters.
2033 */
2034int qcow2_expand_zero_clusters(BlockDriverState *bs,
2035                               BlockDriverAmendStatusCB *status_cb,
2036                               void *cb_opaque)
2037{
2038    BDRVQcow2State *s = bs->opaque;
2039    uint64_t *l1_table = NULL;
2040    int64_t l1_entries = 0, visited_l1_entries = 0;
2041    int ret;
2042    int i, j;
2043
2044    if (status_cb) {
2045        l1_entries = s->l1_size;
2046        for (i = 0; i < s->nb_snapshots; i++) {
2047            l1_entries += s->snapshots[i].l1_size;
2048        }
2049    }
2050
2051    ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size,
2052                                     &visited_l1_entries, l1_entries,
2053                                     status_cb, cb_opaque);
2054    if (ret < 0) {
2055        goto fail;
2056    }
2057
2058    /* Inactive L1 tables may point to active L2 tables - therefore it is
2059     * necessary to flush the L2 table cache before trying to access the L2
2060     * tables pointed to by inactive L1 entries (else we might try to expand
2061     * zero clusters that have already been expanded); furthermore, it is also
2062     * necessary to empty the L2 table cache, since it may contain tables which
2063     * are now going to be modified directly on disk, bypassing the cache.
2064     * qcow2_cache_empty() does both for us. */
2065    ret = qcow2_cache_empty(bs, s->l2_table_cache);
2066    if (ret < 0) {
2067        goto fail;
2068    }
2069
2070    for (i = 0; i < s->nb_snapshots; i++) {
2071        int l1_sectors = DIV_ROUND_UP(s->snapshots[i].l1_size *
2072                                      sizeof(uint64_t), BDRV_SECTOR_SIZE);
2073
2074        l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE);
2075
2076        ret = bdrv_read(bs->file,
2077                        s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE,
2078                        (void *)l1_table, l1_sectors);
2079        if (ret < 0) {
2080            goto fail;
2081        }
2082
2083        for (j = 0; j < s->snapshots[i].l1_size; j++) {
2084            be64_to_cpus(&l1_table[j]);
2085        }
2086
2087        ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size,
2088                                         &visited_l1_entries, l1_entries,
2089                                         status_cb, cb_opaque);
2090        if (ret < 0) {
2091            goto fail;
2092        }
2093    }
2094
2095    ret = 0;
2096
2097fail:
2098    g_free(l1_table);
2099    return ret;
2100}
2101