block/qcow.c

   1 /*
   2  * Block driver for the QCOW 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 "qapi/error.h"
  27 #include "qemu/error-report.h"
  28 #include "block/block_int.h"
  29 #include "sysemu/block-backend.h"
  30 #include "qemu/module.h"
  31 #include "qemu/option.h"
  32 #include "qemu/bswap.h"
  33 #include <zlib.h>
  34 #include "qapi/qmp/qdict.h"
  35 #include "qapi/qmp/qstring.h"
  36 #include "qapi/qobject-input-visitor.h"
  37 #include "qapi/qapi-visit-block-core.h"
  38 #include "crypto/block.h"
  39 #include "migration/blocker.h"
  40 #include "block/crypto.h"
  41
  42 /**************************************************************/
  43 /* QEMU COW block driver with compression and encryption support */
  44
  45 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  46 #define QCOW_VERSION 1
  47
  48 #define QCOW_CRYPT_NONE 0
  49 #define QCOW_CRYPT_AES  1
  50
  51 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
  52
  53 typedef struct QCowHeader {
  54     uint32_t magic;
  55     uint32_t version;
  56     uint64_t backing_file_offset;
  57     uint32_t backing_file_size;
  58     uint32_t mtime;
  59     uint64_t size; /* in bytes */
  60     uint8_t cluster_bits;
  61     uint8_t l2_bits;
  62     uint16_t padding;
  63     uint32_t crypt_method;
  64     uint64_t l1_table_offset;
  65 } QEMU_PACKED QCowHeader;
  66
  67 #define L2_CACHE_SIZE 16
  68
  69 typedef struct BDRVQcowState {
  70     int cluster_bits;
  71     int cluster_size;
  72     int cluster_sectors;
  73     int l2_bits;
  74     int l2_size;
  75     unsigned int l1_size;
  76     uint64_t cluster_offset_mask;
  77     uint64_t l1_table_offset;
  78     uint64_t *l1_table;
  79     uint64_t *l2_cache;
  80     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  81     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  82     uint8_t *cluster_cache;
  83     uint8_t *cluster_data;
  84     uint64_t cluster_cache_offset;
  85     QCryptoBlock *crypto; /* Disk encryption format driver */
  86     uint32_t crypt_method_header;
  87     CoMutex lock;
  88     Error *migration_blocker;
  89 } BDRVQcowState;
  90
  91 static QemuOptsList qcow_create_opts;
  92
  93 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
  94
  95 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  96 {
  97     const QCowHeader *cow_header = (const void *)buf;
  98
  99     if (buf_size >= sizeof(QCowHeader) &&
 100         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
 101         be32_to_cpu(cow_header->version) == QCOW_VERSION)
 102         return 100;
 103     else
 104         return 0;
 105 }
 106
 107 static QemuOptsList qcow_runtime_opts = {
 108     .name = "qcow",
 109     .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head),
 110     .desc = {
 111         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
 112         { /* end of list */ }
 113     },
 114 };
 115
 116 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
 117                      Error **errp)
 118 {
 119     BDRVQcowState *s = bs->opaque;
 120     unsigned int len, i, shift;
 121     int ret;
 122     QCowHeader header;
 123     Error *local_err = NULL;
 124     QCryptoBlockOpenOptions *crypto_opts = NULL;
 125     unsigned int cflags = 0;
 126     QDict *encryptopts = NULL;
 127     const char *encryptfmt;
 128
 129     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
 130     encryptfmt = qdict_get_try_str(encryptopts, "format");
 131
 132     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
 133                                false, errp);
 134     if (!bs->file) {
 135         ret = -EINVAL;
 136         goto fail;
 137     }
 138
 139     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
 140     if (ret < 0) {
 141         goto fail;
 142     }
 143     be32_to_cpus(&header.magic);
 144     be32_to_cpus(&header.version);
 145     be64_to_cpus(&header.backing_file_offset);
 146     be32_to_cpus(&header.backing_file_size);
 147     be32_to_cpus(&header.mtime);
 148     be64_to_cpus(&header.size);
 149     be32_to_cpus(&header.crypt_method);
 150     be64_to_cpus(&header.l1_table_offset);
 151
 152     if (header.magic != QCOW_MAGIC) {
 153         error_setg(errp, "Image not in qcow format");
 154         ret = -EINVAL;
 155         goto fail;
 156     }
 157     if (header.version != QCOW_VERSION) {
 158         error_setg(errp, "Unsupported qcow version %" PRIu32, header.version);
 159         ret = -ENOTSUP;
 160         goto fail;
 161     }
 162
 163     if (header.size <= 1) {
 164         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
 165         ret = -EINVAL;
 166         goto fail;
 167     }
 168     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
 169         error_setg(errp, "Cluster size must be between 512 and 64k");
 170         ret = -EINVAL;
 171         goto fail;
 172     }
 173
 174     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
 175      * so bytes = num_entries << 3. */
 176     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
 177         error_setg(errp, "L2 table size must be between 512 and 64k");
 178         ret = -EINVAL;
 179         goto fail;
 180     }
 181
 182     s->crypt_method_header = header.crypt_method;
 183     if (s->crypt_method_header) {
 184         if (bdrv_uses_whitelist() &&
 185             s->crypt_method_header == QCOW_CRYPT_AES) {
 186             error_setg(errp,
 187                        "Use of AES-CBC encrypted qcow images is no longer "
 188                        "supported in system emulators");
 189             error_append_hint(errp,
 190                               "You can use 'qemu-img convert' to convert your "
 191                               "image to an alternative supported format, such "
 192                               "as unencrypted qcow, or raw with the LUKS "
 193                               "format instead.\n");
 194             ret = -ENOSYS;
 195             goto fail;
 196         }
 197         if (s->crypt_method_header == QCOW_CRYPT_AES) {
 198             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
 199                 error_setg(errp,
 200                            "Header reported 'aes' encryption format but "
 201                            "options specify '%s'", encryptfmt);
 202                 ret = -EINVAL;
 203                 goto fail;
 204             }
 205             qdict_del(encryptopts, "format");
 206             crypto_opts = block_crypto_open_opts_init(
 207                 Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
 208             if (!crypto_opts) {
 209                 ret = -EINVAL;
 210                 goto fail;
 211             }
 212
 213             if (flags & BDRV_O_NO_IO) {
 214                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
 215             }
 216             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
 217                                            NULL, NULL, cflags, errp);
 218             if (!s->crypto) {
 219                 ret = -EINVAL;
 220                 goto fail;
 221             }
 222         } else {
 223             error_setg(errp, "invalid encryption method in qcow header");
 224             ret = -EINVAL;
 225             goto fail;
 226         }
 227         bs->encrypted = true;
 228     } else {
 229         if (encryptfmt) {
 230             error_setg(errp, "No encryption in image header, but options "
 231                        "specified format '%s'", encryptfmt);
 232             ret = -EINVAL;
 233             goto fail;
 234         }
 235     }
 236     s->cluster_bits = header.cluster_bits;
 237     s->cluster_size = 1 << s->cluster_bits;
 238     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 239     s->l2_bits = header.l2_bits;
 240     s->l2_size = 1 << s->l2_bits;
 241     bs->total_sectors = header.size / 512;
 242     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 243
 244     /* read the level 1 table */
 245     shift = s->cluster_bits + s->l2_bits;
 246     if (header.size > UINT64_MAX - (1LL << shift)) {
 247         error_setg(errp, "Image too large");
 248         ret = -EINVAL;
 249         goto fail;
 250     } else {
 251         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
 252         if (l1_size > INT_MAX / sizeof(uint64_t)) {
 253             error_setg(errp, "Image too large");
 254             ret = -EINVAL;
 255             goto fail;
 256         }
 257         s->l1_size = l1_size;
 258     }
 259
 260     s->l1_table_offset = header.l1_table_offset;
 261     s->l1_table = g_try_new(uint64_t, s->l1_size);
 262     if (s->l1_table == NULL) {
 263         error_setg(errp, "Could not allocate memory for L1 table");
 264         ret = -ENOMEM;
 265         goto fail;
 266     }
 267
 268     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
 269                s->l1_size * sizeof(uint64_t));
 270     if (ret < 0) {
 271         goto fail;
 272     }
 273
 274     for(i = 0;i < s->l1_size; i++) {
 275         be64_to_cpus(&s->l1_table[i]);
 276     }
 277
 278     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
 279     s->l2_cache =
 280         qemu_try_blockalign(bs->file->bs,
 281                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 282     if (s->l2_cache == NULL) {
 283         error_setg(errp, "Could not allocate L2 table cache");
 284         ret = -ENOMEM;
 285         goto fail;
 286     }
 287     s->cluster_cache = g_malloc(s->cluster_size);
 288     s->cluster_data = g_malloc(s->cluster_size);
 289     s->cluster_cache_offset = -1;
 290
 291     /* read the backing file name */
 292     if (header.backing_file_offset != 0) {
 293         len = header.backing_file_size;
 294         if (len > 1023 || len >= sizeof(bs->backing_file)) {
 295             error_setg(errp, "Backing file name too long");
 296             ret = -EINVAL;
 297             goto fail;
 298         }
 299         ret = bdrv_pread(bs->file, header.backing_file_offset,
 300                    bs->backing_file, len);
 301         if (ret < 0) {
 302             goto fail;
 303         }
 304         bs->backing_file[len] = '\0';
 305     }
 306
 307     /* Disable migration when qcow images are used */
 308     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
 309                "does not support live migration",
 310                bdrv_get_device_or_node_name(bs));
 311     ret = migrate_add_blocker(s->migration_blocker, &local_err);
 312     if (local_err) {
 313         error_propagate(errp, local_err);
 314         error_free(s->migration_blocker);
 315         goto fail;
 316     }
 317
 318     qobject_unref(encryptopts);
 319     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
 320     qemu_co_mutex_init(&s->lock);
 321     return 0;
 322
 323  fail:
 324     g_free(s->l1_table);
 325     qemu_vfree(s->l2_cache);
 326     g_free(s->cluster_cache);
 327     g_free(s->cluster_data);
 328     qcrypto_block_free(s->crypto);
 329     qobject_unref(encryptopts);
 330     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
 331     return ret;
 332 }
 333
 334
 335 /* We have nothing to do for QCOW reopen, stubs just return
 336  * success */
 337 static int qcow_reopen_prepare(BDRVReopenState *state,
 338                                BlockReopenQueue *queue, Error **errp)
 339 {
 340     return 0;
 341 }
 342
 343
 344 /* 'allocate' is:
 345  *
 346  * 0 to not allocate.
 347  *
 348  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 349  * 'n_end')
 350  *
 351  * 2 to allocate a compressed cluster of size
 352  * 'compressed_size'. 'compressed_size' must be > 0 and <
 353  * cluster_size
 354  *
 355  * return 0 if not allocated, 1 if *result is assigned, and negative
 356  * errno on failure.
 357  */
 358 static int get_cluster_offset(BlockDriverState *bs,
 359                               uint64_t offset, int allocate,
 360                               int compressed_size,
 361                               int n_start, int n_end, uint64_t *result)
 362 {
 363     BDRVQcowState *s = bs->opaque;
 364     int min_index, i, j, l1_index, l2_index, ret;
 365     int64_t l2_offset;
 366     uint64_t *l2_table, cluster_offset, tmp;
 367     uint32_t min_count;
 368     int new_l2_table;
 369
 370     *result = 0;
 371     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 372     l2_offset = s->l1_table[l1_index];
 373     new_l2_table = 0;
 374     if (!l2_offset) {
 375         if (!allocate)
 376             return 0;
 377         /* allocate a new l2 entry */
 378         l2_offset = bdrv_getlength(bs->file->bs);
 379         if (l2_offset < 0) {
 380             return l2_offset;
 381         }
 382         /* round to cluster size */
 383         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
 384         /* update the L1 entry */
 385         s->l1_table[l1_index] = l2_offset;
 386         tmp = cpu_to_be64(l2_offset);
 387         BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
 388         ret = bdrv_pwrite_sync(bs->file,
 389                                s->l1_table_offset + l1_index * sizeof(tmp),
 390                                &tmp, sizeof(tmp));
 391         if (ret < 0) {
 392             return ret;
 393         }
 394         new_l2_table = 1;
 395     }
 396     for(i = 0; i < L2_CACHE_SIZE; i++) {
 397         if (l2_offset == s->l2_cache_offsets[i]) {
 398             /* increment the hit count */
 399             if (++s->l2_cache_counts[i] == 0xffffffff) {
 400                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 401                     s->l2_cache_counts[j] >>= 1;
 402                 }
 403             }
 404             l2_table = s->l2_cache + (i << s->l2_bits);
 405             goto found;
 406         }
 407     }
 408     /* not found: load a new entry in the least used one */
 409     min_index = 0;
 410     min_count = 0xffffffff;
 411     for(i = 0; i < L2_CACHE_SIZE; i++) {
 412         if (s->l2_cache_counts[i] < min_count) {
 413             min_count = s->l2_cache_counts[i];
 414             min_index = i;
 415         }
 416     }
 417     l2_table = s->l2_cache + (min_index << s->l2_bits);
 418     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
 419     if (new_l2_table) {
 420         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 421         ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
 422                                s->l2_size * sizeof(uint64_t));
 423         if (ret < 0) {
 424             return ret;
 425         }
 426     } else {
 427         ret = bdrv_pread(bs->file, l2_offset, l2_table,
 428                          s->l2_size * sizeof(uint64_t));
 429         if (ret < 0) {
 430             return ret;
 431         }
 432     }
 433     s->l2_cache_offsets[min_index] = l2_offset;
 434     s->l2_cache_counts[min_index] = 1;
 435  found:
 436     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 437     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 438     if (!cluster_offset ||
 439         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 440         if (!allocate)
 441             return 0;
 442         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
 443         /* allocate a new cluster */
 444         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 445             (n_end - n_start) < s->cluster_sectors) {
 446             /* if the cluster is already compressed, we must
 447                decompress it in the case it is not completely
 448                overwritten */
 449             if (decompress_cluster(bs, cluster_offset) < 0) {
 450                 return -EIO;
 451             }
 452             cluster_offset = bdrv_getlength(bs->file->bs);
 453             if ((int64_t) cluster_offset < 0) {
 454                 return cluster_offset;
 455             }
 456             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
 457             /* write the cluster content */
 458             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
 459             ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
 460                               s->cluster_size);
 461             if (ret < 0) {
 462                 return ret;
 463             }
 464         } else {
 465             cluster_offset = bdrv_getlength(bs->file->bs);
 466             if ((int64_t) cluster_offset < 0) {
 467                 return cluster_offset;
 468             }
 469             if (allocate == 1) {
 470                 /* round to cluster size */
 471                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
 472                 if (cluster_offset + s->cluster_size > INT64_MAX) {
 473                     return -E2BIG;
 474                 }
 475                 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
 476                                     PREALLOC_MODE_OFF, NULL);
 477                 if (ret < 0) {
 478                     return ret;
 479                 }
 480                 /* if encrypted, we must initialize the cluster
 481                    content which won't be written */
 482                 if (bs->encrypted &&
 483                     (n_end - n_start) < s->cluster_sectors) {
 484                     uint64_t start_sect;
 485                     assert(s->crypto);
 486                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 487                     for(i = 0; i < s->cluster_sectors; i++) {
 488                         if (i < n_start || i >= n_end) {
 489                             memset(s->cluster_data, 0x00, 512);
 490                             if (qcrypto_block_encrypt(s->crypto,
 491                                                       (start_sect + i) *
 492                                                       BDRV_SECTOR_SIZE,
 493                                                       s->cluster_data,
 494                                                       BDRV_SECTOR_SIZE,
 495                                                       NULL) < 0) {
 496                                 return -EIO;
 497                             }
 498                             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
 499                             ret = bdrv_pwrite(bs->file,
 500                                               cluster_offset + i * 512,
 501                                               s->cluster_data, 512);
 502                             if (ret < 0) {
 503                                 return ret;
 504                             }
 505                         }
 506                     }
 507                 }
 508             } else if (allocate == 2) {
 509                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 510                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 511             }
 512         }
 513         /* update L2 table */
 514         tmp = cpu_to_be64(cluster_offset);
 515         l2_table[l2_index] = tmp;
 516         if (allocate == 2) {
 517             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
 518         } else {
 519             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
 520         }
 521         ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
 522                                &tmp, sizeof(tmp));
 523         if (ret < 0) {
 524             return ret;
 525         }
 526     }
 527     *result = cluster_offset;
 528     return 1;
 529 }
 530
 531 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
 532                                              bool want_zero,
 533                                              int64_t offset, int64_t bytes,
 534                                              int64_t *pnum, int64_t *map,
 535                                              BlockDriverState **file)
 536 {
 537     BDRVQcowState *s = bs->opaque;
 538     int index_in_cluster, ret;
 539     int64_t n;
 540     uint64_t cluster_offset;
 541
 542     qemu_co_mutex_lock(&s->lock);
 543     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
 544     qemu_co_mutex_unlock(&s->lock);
 545     if (ret < 0) {
 546         return ret;
 547     }
 548     index_in_cluster = offset & (s->cluster_size - 1);
 549     n = s->cluster_size - index_in_cluster;
 550     if (n > bytes) {
 551         n = bytes;
 552     }
 553     *pnum = n;
 554     if (!cluster_offset) {
 555         return 0;
 556     }
 557     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
 558         return BDRV_BLOCK_DATA;
 559     }
 560     *map = cluster_offset | index_in_cluster;
 561     *file = bs->file->bs;
 562     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 563 }
 564
 565 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 566                              const uint8_t *buf, int buf_size)
 567 {
 568     z_stream strm1, *strm = &strm1;
 569     int ret, out_len;
 570
 571     memset(strm, 0, sizeof(*strm));
 572
 573     strm->next_in = (uint8_t *)buf;
 574     strm->avail_in = buf_size;
 575     strm->next_out = out_buf;
 576     strm->avail_out = out_buf_size;
 577
 578     ret = inflateInit2(strm, -12);
 579     if (ret != Z_OK)
 580         return -1;
 581     ret = inflate(strm, Z_FINISH);
 582     out_len = strm->next_out - out_buf;
 583     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 584         out_len != out_buf_size) {
 585         inflateEnd(strm);
 586         return -1;
 587     }
 588     inflateEnd(strm);
 589     return 0;
 590 }
 591
 592 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 593 {
 594     BDRVQcowState *s = bs->opaque;
 595     int ret, csize;
 596     uint64_t coffset;
 597
 598     coffset = cluster_offset & s->cluster_offset_mask;
 599     if (s->cluster_cache_offset != coffset) {
 600         csize = cluster_offset >> (63 - s->cluster_bits);
 601         csize &= (s->cluster_size - 1);
 602         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
 603         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
 604         if (ret != csize)
 605             return -1;
 606         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 607                               s->cluster_data, csize) < 0) {
 608             return -1;
 609         }
 610         s->cluster_cache_offset = coffset;
 611     }
 612     return 0;
 613 }
 614
 615 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
 616                          int nb_sectors, QEMUIOVector *qiov)
 617 {
 618     BDRVQcowState *s = bs->opaque;
 619     int index_in_cluster;
 620     int ret = 0, n;
 621     uint64_t cluster_offset;
 622     struct iovec hd_iov;
 623     QEMUIOVector hd_qiov;
 624     uint8_t *buf;
 625     void *orig_buf;
 626
 627     if (qiov->niov > 1) {
 628         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
 629         if (buf == NULL) {
 630             return -ENOMEM;
 631         }
 632     } else {
 633         orig_buf = NULL;
 634         buf = (uint8_t *)qiov->iov->iov_base;
 635     }
 636
 637     qemu_co_mutex_lock(&s->lock);
 638
 639     while (nb_sectors != 0) {
 640         /* prepare next request */
 641         ret = get_cluster_offset(bs, sector_num << 9,
 642                                  0, 0, 0, 0, &cluster_offset);
 643         if (ret < 0) {
 644             break;
 645         }
 646         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 647         n = s->cluster_sectors - index_in_cluster;
 648         if (n > nb_sectors) {
 649             n = nb_sectors;
 650         }
 651
 652         if (!cluster_offset) {
 653             if (bs->backing) {
 654                 /* read from the base image */
 655                 hd_iov.iov_base = (void *)buf;
 656                 hd_iov.iov_len = n * 512;
 657                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 658                 qemu_co_mutex_unlock(&s->lock);
 659                 /* qcow2 emits this on bs->file instead of bs->backing */
 660                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
 661                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
 662                 qemu_co_mutex_lock(&s->lock);
 663                 if (ret < 0) {
 664                     break;
 665                 }
 666             } else {
 667                 /* Note: in this case, no need to wait */
 668                 memset(buf, 0, 512 * n);
 669             }
 670         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 671             /* add AIO support for compressed blocks ? */
 672             if (decompress_cluster(bs, cluster_offset) < 0) {
 673                 ret = -EIO;
 674                 break;
 675             }
 676             memcpy(buf,
 677                    s->cluster_cache + index_in_cluster * 512, 512 * n);
 678         } else {
 679             if ((cluster_offset & 511) != 0) {
 680                 ret = -EIO;
 681                 break;
 682             }
 683             hd_iov.iov_base = (void *)buf;
 684             hd_iov.iov_len = n * 512;
 685             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 686             qemu_co_mutex_unlock(&s->lock);
 687             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
 688             ret = bdrv_co_readv(bs->file,
 689                                 (cluster_offset >> 9) + index_in_cluster,
 690                                 n, &hd_qiov);
 691             qemu_co_mutex_lock(&s->lock);
 692             if (ret < 0) {
 693                 break;
 694             }
 695             if (bs->encrypted) {
 696                 assert(s->crypto);
 697                 if (qcrypto_block_decrypt(s->crypto,
 698                                           sector_num * BDRV_SECTOR_SIZE, buf,
 699                                           n * BDRV_SECTOR_SIZE, NULL) < 0) {
 700                     ret = -EIO;
 701                     break;
 702                 }
 703             }
 704         }
 705         ret = 0;
 706
 707         nb_sectors -= n;
 708         sector_num += n;
 709         buf += n * 512;
 710     }
 711
 712     qemu_co_mutex_unlock(&s->lock);
 713
 714     if (qiov->niov > 1) {
 715         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
 716         qemu_vfree(orig_buf);
 717     }
 718
 719     return ret;
 720 }
 721
 722 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
 723                           int nb_sectors, QEMUIOVector *qiov)
 724 {
 725     BDRVQcowState *s = bs->opaque;
 726     int index_in_cluster;
 727     uint64_t cluster_offset;
 728     int ret = 0, n;
 729     struct iovec hd_iov;
 730     QEMUIOVector hd_qiov;
 731     uint8_t *buf;
 732     void *orig_buf;
 733
 734     s->cluster_cache_offset = -1; /* disable compressed cache */
 735
 736     /* We must always copy the iov when encrypting, so we
 737      * don't modify the original data buffer during encryption */
 738     if (bs->encrypted || qiov->niov > 1) {
 739         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
 740         if (buf == NULL) {
 741             return -ENOMEM;
 742         }
 743         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
 744     } else {
 745         orig_buf = NULL;
 746         buf = (uint8_t *)qiov->iov->iov_base;
 747     }
 748
 749     qemu_co_mutex_lock(&s->lock);
 750
 751     while (nb_sectors != 0) {
 752
 753         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 754         n = s->cluster_sectors - index_in_cluster;
 755         if (n > nb_sectors) {
 756             n = nb_sectors;
 757         }
 758         ret = get_cluster_offset(bs, sector_num << 9, 1, 0,
 759                                  index_in_cluster,
 760                                  index_in_cluster + n, &cluster_offset);
 761         if (ret < 0) {
 762             break;
 763         }
 764         if (!cluster_offset || (cluster_offset & 511) != 0) {
 765             ret = -EIO;
 766             break;
 767         }
 768         if (bs->encrypted) {
 769             assert(s->crypto);
 770             if (qcrypto_block_encrypt(s->crypto, sector_num * BDRV_SECTOR_SIZE,
 771                                       buf, n * BDRV_SECTOR_SIZE, NULL) < 0) {
 772                 ret = -EIO;
 773                 break;
 774             }
 775         }
 776
 777         hd_iov.iov_base = (void *)buf;
 778         hd_iov.iov_len = n * 512;
 779         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 780         qemu_co_mutex_unlock(&s->lock);
 781         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
 782         ret = bdrv_co_writev(bs->file,
 783                              (cluster_offset >> 9) + index_in_cluster,
 784                              n, &hd_qiov);
 785         qemu_co_mutex_lock(&s->lock);
 786         if (ret < 0) {
 787             break;
 788         }
 789         ret = 0;
 790
 791         nb_sectors -= n;
 792         sector_num += n;
 793         buf += n * 512;
 794     }
 795     qemu_co_mutex_unlock(&s->lock);
 796
 797     qemu_vfree(orig_buf);
 798
 799     return ret;
 800 }
 801
 802 static void qcow_close(BlockDriverState *bs)
 803 {
 804     BDRVQcowState *s = bs->opaque;
 805
 806     qcrypto_block_free(s->crypto);
 807     s->crypto = NULL;
 808     g_free(s->l1_table);
 809     qemu_vfree(s->l2_cache);
 810     g_free(s->cluster_cache);
 811     g_free(s->cluster_data);
 812
 813     migrate_del_blocker(s->migration_blocker);
 814     error_free(s->migration_blocker);
 815 }
 816
 817 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts,
 818                                        Error **errp)
 819 {
 820     BlockdevCreateOptionsQcow *qcow_opts;
 821     int header_size, backing_filename_len, l1_size, shift, i;
 822     QCowHeader header;
 823     uint8_t *tmp;
 824     int64_t total_size = 0;
 825     int ret;
 826     BlockDriverState *bs;
 827     BlockBackend *qcow_blk;
 828     QCryptoBlock *crypto = NULL;
 829
 830     assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
 831     qcow_opts = &opts->u.qcow;
 832
 833     /* Sanity checks */
 834     total_size = qcow_opts->size;
 835     if (total_size == 0) {
 836         error_setg(errp, "Image size is too small, cannot be zero length");
 837         return -EINVAL;
 838     }
 839
 840     if (qcow_opts->has_encrypt &&
 841         qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
 842     {
 843         error_setg(errp, "Unsupported encryption format");
 844         return -EINVAL;
 845     }
 846
 847     /* Create BlockBackend to write to the image */
 848     bs = bdrv_open_blockdev_ref(qcow_opts->file, errp);
 849     if (bs == NULL) {
 850         return -EIO;
 851     }
 852
 853     qcow_blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
 854     ret = blk_insert_bs(qcow_blk, bs, errp);
 855     if (ret < 0) {
 856         goto exit;
 857     }
 858     blk_set_allow_write_beyond_eof(qcow_blk, true);
 859
 860     /* Create image format */
 861     ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
 862     if (ret < 0) {
 863         goto exit;
 864     }
 865
 866     memset(&header, 0, sizeof(header));
 867     header.magic = cpu_to_be32(QCOW_MAGIC);
 868     header.version = cpu_to_be32(QCOW_VERSION);
 869     header.size = cpu_to_be64(total_size);
 870     header_size = sizeof(header);
 871     backing_filename_len = 0;
 872     if (qcow_opts->has_backing_file) {
 873         if (strcmp(qcow_opts->backing_file, "fat:")) {
 874             header.backing_file_offset = cpu_to_be64(header_size);
 875             backing_filename_len = strlen(qcow_opts->backing_file);
 876             header.backing_file_size = cpu_to_be32(backing_filename_len);
 877             header_size += backing_filename_len;
 878         } else {
 879             /* special backing file for vvfat */
 880             qcow_opts->has_backing_file = false;
 881         }
 882         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 883                                     unmodified sectors */
 884         header.l2_bits = 12; /* 32 KB L2 tables */
 885     } else {
 886         header.cluster_bits = 12; /* 4 KB clusters */
 887         header.l2_bits = 9; /* 4 KB L2 tables */
 888     }
 889     header_size = (header_size + 7) & ~7;
 890     shift = header.cluster_bits + header.l2_bits;
 891     l1_size = (total_size + (1LL << shift) - 1) >> shift;
 892
 893     header.l1_table_offset = cpu_to_be64(header_size);
 894
 895     if (qcow_opts->has_encrypt) {
 896         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 897
 898         crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
 899                                       NULL, NULL, NULL, errp);
 900         if (!crypto) {
 901             ret = -EINVAL;
 902             goto exit;
 903         }
 904     } else {
 905         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 906     }
 907
 908     /* write all the data */
 909     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
 910     if (ret != sizeof(header)) {
 911         goto exit;
 912     }
 913
 914     if (qcow_opts->has_backing_file) {
 915         ret = blk_pwrite(qcow_blk, sizeof(header),
 916                          qcow_opts->backing_file, backing_filename_len, 0);
 917         if (ret != backing_filename_len) {
 918             goto exit;
 919         }
 920     }
 921
 922     tmp = g_malloc0(BDRV_SECTOR_SIZE);
 923     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
 924          i++) {
 925         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
 926                          tmp, BDRV_SECTOR_SIZE, 0);
 927         if (ret != BDRV_SECTOR_SIZE) {
 928             g_free(tmp);
 929             goto exit;
 930         }
 931     }
 932
 933     g_free(tmp);
 934     ret = 0;
 935 exit:
 936     blk_unref(qcow_blk);
 937     qcrypto_block_free(crypto);
 938     return ret;
 939 }
 940
 941 static int coroutine_fn qcow_co_create_opts(const char *filename,
 942                                             QemuOpts *opts, Error **errp)
 943 {
 944     BlockdevCreateOptions *create_options = NULL;
 945     BlockDriverState *bs = NULL;
 946     QDict *qdict = NULL;
 947     QObject *qobj;
 948     Visitor *v;
 949     const char *val;
 950     Error *local_err = NULL;
 951     int ret;
 952
 953     static const QDictRenames opt_renames[] = {
 954         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
 955         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
 956         { NULL, NULL },
 957     };
 958
 959     /* Parse options and convert legacy syntax */
 960     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
 961
 962     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
 963     if (val && !strcmp(val, "on")) {
 964         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
 965     } else if (val && !strcmp(val, "off")) {
 966         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
 967     }
 968
 969     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
 970     if (val && !strcmp(val, "aes")) {
 971         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
 972     }
 973
 974     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
 975         ret = -EINVAL;
 976         goto fail;
 977     }
 978
 979     /* Create and open the file (protocol layer) */
 980     ret = bdrv_create_file(filename, opts, &local_err);
 981     if (ret < 0) {
 982         error_propagate(errp, local_err);
 983         goto fail;
 984     }
 985
 986     bs = bdrv_open(filename, NULL, NULL,
 987                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
 988     if (bs == NULL) {
 989         ret = -EIO;
 990         goto fail;
 991     }
 992
 993     /* Now get the QAPI type BlockdevCreateOptions */
 994     qdict_put_str(qdict, "driver", "qcow");
 995     qdict_put_str(qdict, "file", bs->node_name);
 996
 997     qobj = qdict_crumple(qdict, errp);
 998     qobject_unref(qdict);
 999     qdict = qobject_to(QDict, qobj);
1000     if (qdict == NULL) {
1001         ret = -EINVAL;
1002         goto fail;
1003     }
1004
1005     v = qobject_input_visitor_new_keyval(QOBJECT(qdict));
1006     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
1007     visit_free(v);
1008
1009     if (local_err) {
1010         error_propagate(errp, local_err);
1011         ret = -EINVAL;
1012         goto fail;
1013     }
1014
1015     /* Silently round up size */
1016     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1017     create_options->u.qcow.size =
1018         ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1019
1020     /* Create the qcow image (format layer) */
1021     ret = qcow_co_create(create_options, errp);
1022     if (ret < 0) {
1023         goto fail;
1024     }
1025
1026     ret = 0;
1027 fail:
1028     qobject_unref(qdict);
1029     bdrv_unref(bs);
1030     qapi_free_BlockdevCreateOptions(create_options);
1031     return ret;
1032 }
1033
1034 static int qcow_make_empty(BlockDriverState *bs)
1035 {
1036     BDRVQcowState *s = bs->opaque;
1037     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1038     int ret;
1039
1040     memset(s->l1_table, 0, l1_length);
1041     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
1042             l1_length) < 0)
1043         return -1;
1044     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
1045                         PREALLOC_MODE_OFF, NULL);
1046     if (ret < 0)
1047         return ret;
1048
1049     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1050     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1051     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1052
1053     return 0;
1054 }
1055
1056 /* XXX: put compressed sectors first, then all the cluster aligned
1057    tables to avoid losing bytes in alignment */
1058 static coroutine_fn int
1059 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1060                            uint64_t bytes, QEMUIOVector *qiov)
1061 {
1062     BDRVQcowState *s = bs->opaque;
1063     QEMUIOVector hd_qiov;
1064     struct iovec iov;
1065     z_stream strm;
1066     int ret, out_len;
1067     uint8_t *buf, *out_buf;
1068     uint64_t cluster_offset;
1069
1070     buf = qemu_blockalign(bs, s->cluster_size);
1071     if (bytes != s->cluster_size) {
1072         if (bytes > s->cluster_size ||
1073             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1074         {
1075             qemu_vfree(buf);
1076             return -EINVAL;
1077         }
1078         /* Zero-pad last write if image size is not cluster aligned */
1079         memset(buf + bytes, 0, s->cluster_size - bytes);
1080     }
1081     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1082
1083     out_buf = g_malloc(s->cluster_size);
1084
1085     /* best compression, small window, no zlib header */
1086     memset(&strm, 0, sizeof(strm));
1087     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1088                        Z_DEFLATED, -12,
1089                        9, Z_DEFAULT_STRATEGY);
1090     if (ret != 0) {
1091         ret = -EINVAL;
1092         goto fail;
1093     }
1094
1095     strm.avail_in = s->cluster_size;
1096     strm.next_in = (uint8_t *)buf;
1097     strm.avail_out = s->cluster_size;
1098     strm.next_out = out_buf;
1099
1100     ret = deflate(&strm, Z_FINISH);
1101     if (ret != Z_STREAM_END && ret != Z_OK) {
1102         deflateEnd(&strm);
1103         ret = -EINVAL;
1104         goto fail;
1105     }
1106     out_len = strm.next_out - out_buf;
1107
1108     deflateEnd(&strm);
1109
1110     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1111         /* could not compress: write normal cluster */
1112         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
1113                              bytes >> BDRV_SECTOR_BITS, qiov);
1114         if (ret < 0) {
1115             goto fail;
1116         }
1117         goto success;
1118     }
1119     qemu_co_mutex_lock(&s->lock);
1120     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1121     qemu_co_mutex_unlock(&s->lock);
1122     if (ret < 0) {
1123         goto fail;
1124     }
1125     if (cluster_offset == 0) {
1126         ret = -EIO;
1127         goto fail;
1128     }
1129     cluster_offset &= s->cluster_offset_mask;
1130
1131     iov = (struct iovec) {
1132         .iov_base   = out_buf,
1133         .iov_len    = out_len,
1134     };
1135     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1136     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1137     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1138     if (ret < 0) {
1139         goto fail;
1140     }
1141 success:
1142     ret = 0;
1143 fail:
1144     qemu_vfree(buf);
1145     g_free(out_buf);
1146     return ret;
1147 }
1148
1149 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1150 {
1151     BDRVQcowState *s = bs->opaque;
1152     bdi->cluster_size = s->cluster_size;
1153     return 0;
1154 }
1155
1156 static QemuOptsList qcow_create_opts = {
1157     .name = "qcow-create-opts",
1158     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1159     .desc = {
1160         {
1161             .name = BLOCK_OPT_SIZE,
1162             .type = QEMU_OPT_SIZE,
1163             .help = "Virtual disk size"
1164         },
1165         {
1166             .name = BLOCK_OPT_BACKING_FILE,
1167             .type = QEMU_OPT_STRING,
1168             .help = "File name of a base image"
1169         },
1170         {
1171             .name = BLOCK_OPT_ENCRYPT,
1172             .type = QEMU_OPT_BOOL,
1173             .help = "Encrypt the image with format 'aes'. (Deprecated "
1174                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1175         },
1176         {
1177             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1178             .type = QEMU_OPT_STRING,
1179             .help = "Encrypt the image, format choices: 'aes'",
1180         },
1181         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1182         { /* end of list */ }
1183     }
1184 };
1185
1186 static BlockDriver bdrv_qcow = {
1187     .format_name        = "qcow",
1188     .instance_size      = sizeof(BDRVQcowState),
1189     .bdrv_probe         = qcow_probe,
1190     .bdrv_open          = qcow_open,
1191     .bdrv_close         = qcow_close,
1192     .bdrv_child_perm        = bdrv_format_default_perms,
1193     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1194     .bdrv_co_create         = qcow_co_create,
1195     .bdrv_co_create_opts    = qcow_co_create_opts,
1196     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1197     .supports_backing       = true,
1198
1199     .bdrv_co_readv          = qcow_co_readv,
1200     .bdrv_co_writev         = qcow_co_writev,
1201     .bdrv_co_block_status   = qcow_co_block_status,
1202
1203     .bdrv_make_empty        = qcow_make_empty,
1204     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1205     .bdrv_get_info          = qcow_get_info,
1206
1207     .create_opts            = &qcow_create_opts,
1208 };
1209
1210 static void bdrv_qcow_init(void)
1211 {
1212     bdrv_register(&bdrv_qcow);
1213 }
1214
1215 block_init(bdrv_qcow_init);