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 #include "qemu/osdep.h"
  25 #include "qapi/error.h"
  26 #include "qemu-common.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/bswap.h"
  32 #include <zlib.h>
  33 #include "qapi/qmp/qerror.h"
  34 #include "crypto/cipher.h"
  35 #include "migration/blocker.h"
  36
  37 /**************************************************************/
  38 /* QEMU COW block driver with compression and encryption support */
  39
  40 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  41 #define QCOW_VERSION 1
  42
  43 #define QCOW_CRYPT_NONE 0
  44 #define QCOW_CRYPT_AES  1
  45
  46 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
  47
  48 typedef struct QCowHeader {
  49     uint32_t magic;
  50     uint32_t version;
  51     uint64_t backing_file_offset;
  52     uint32_t backing_file_size;
  53     uint32_t mtime;
  54     uint64_t size; /* in bytes */
  55     uint8_t cluster_bits;
  56     uint8_t l2_bits;
  57     uint16_t padding;
  58     uint32_t crypt_method;
  59     uint64_t l1_table_offset;
  60 } QEMU_PACKED QCowHeader;
  61
  62 #define L2_CACHE_SIZE 16
  63
  64 typedef struct BDRVQcowState {
  65     int cluster_bits;
  66     int cluster_size;
  67     int cluster_sectors;
  68     int l2_bits;
  69     int l2_size;
  70     unsigned int l1_size;
  71     uint64_t cluster_offset_mask;
  72     uint64_t l1_table_offset;
  73     uint64_t *l1_table;
  74     uint64_t *l2_cache;
  75     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  76     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  77     uint8_t *cluster_cache;
  78     uint8_t *cluster_data;
  79     uint64_t cluster_cache_offset;
  80     QCryptoCipher *cipher; /* NULL if no key yet */
  81     uint32_t crypt_method_header;
  82     CoMutex lock;
  83     Error *migration_blocker;
  84 } BDRVQcowState;
  85
  86 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
  87
  88 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  89 {
  90     const QCowHeader *cow_header = (const void *)buf;
  91
  92     if (buf_size >= sizeof(QCowHeader) &&
  93         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  94         be32_to_cpu(cow_header->version) == QCOW_VERSION)
  95         return 100;
  96     else
  97         return 0;
  98 }
  99
 100 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
 101                      Error **errp)
 102 {
 103     BDRVQcowState *s = bs->opaque;
 104     unsigned int len, i, shift;
 105     int ret;
 106     QCowHeader header;
 107     Error *local_err = NULL;
 108
 109     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
 110                                false, errp);
 111     if (!bs->file) {
 112         return -EINVAL;
 113     }
 114
 115     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
 116     if (ret < 0) {
 117         goto fail;
 118     }
 119     be32_to_cpus(&header.magic);
 120     be32_to_cpus(&header.version);
 121     be64_to_cpus(&header.backing_file_offset);
 122     be32_to_cpus(&header.backing_file_size);
 123     be32_to_cpus(&header.mtime);
 124     be64_to_cpus(&header.size);
 125     be32_to_cpus(&header.crypt_method);
 126     be64_to_cpus(&header.l1_table_offset);
 127
 128     if (header.magic != QCOW_MAGIC) {
 129         error_setg(errp, "Image not in qcow format");
 130         ret = -EINVAL;
 131         goto fail;
 132     }
 133     if (header.version != QCOW_VERSION) {
 134         error_setg(errp, "Unsupported qcow version %" PRIu32, header.version);
 135         ret = -ENOTSUP;
 136         goto fail;
 137     }
 138
 139     if (header.size <= 1) {
 140         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
 141         ret = -EINVAL;
 142         goto fail;
 143     }
 144     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
 145         error_setg(errp, "Cluster size must be between 512 and 64k");
 146         ret = -EINVAL;
 147         goto fail;
 148     }
 149
 150     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
 151      * so bytes = num_entries << 3. */
 152     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
 153         error_setg(errp, "L2 table size must be between 512 and 64k");
 154         ret = -EINVAL;
 155         goto fail;
 156     }
 157
 158     if (header.crypt_method > QCOW_CRYPT_AES) {
 159         error_setg(errp, "invalid encryption method in qcow header");
 160         ret = -EINVAL;
 161         goto fail;
 162     }
 163     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128,
 164                                  QCRYPTO_CIPHER_MODE_CBC)) {
 165         error_setg(errp, "AES cipher not available");
 166         ret = -EINVAL;
 167         goto fail;
 168     }
 169     s->crypt_method_header = header.crypt_method;
 170     if (s->crypt_method_header) {
 171         if (bdrv_uses_whitelist() &&
 172             s->crypt_method_header == QCOW_CRYPT_AES) {
 173             error_setg(errp,
 174                        "Use of AES-CBC encrypted qcow images is no longer "
 175                        "supported in system emulators");
 176             error_append_hint(errp,
 177                               "You can use 'qemu-img convert' to convert your "
 178                               "image to an alternative supported format, such "
 179                               "as unencrypted qcow, or raw with the LUKS "
 180                               "format instead.\n");
 181             ret = -ENOSYS;
 182             goto fail;
 183         }
 184
 185         bs->encrypted = true;
 186     }
 187     s->cluster_bits = header.cluster_bits;
 188     s->cluster_size = 1 << s->cluster_bits;
 189     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 190     s->l2_bits = header.l2_bits;
 191     s->l2_size = 1 << s->l2_bits;
 192     bs->total_sectors = header.size / 512;
 193     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 194
 195     /* read the level 1 table */
 196     shift = s->cluster_bits + s->l2_bits;
 197     if (header.size > UINT64_MAX - (1LL << shift)) {
 198         error_setg(errp, "Image too large");
 199         ret = -EINVAL;
 200         goto fail;
 201     } else {
 202         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
 203         if (l1_size > INT_MAX / sizeof(uint64_t)) {
 204             error_setg(errp, "Image too large");
 205             ret = -EINVAL;
 206             goto fail;
 207         }
 208         s->l1_size = l1_size;
 209     }
 210
 211     s->l1_table_offset = header.l1_table_offset;
 212     s->l1_table = g_try_new(uint64_t, s->l1_size);
 213     if (s->l1_table == NULL) {
 214         error_setg(errp, "Could not allocate memory for L1 table");
 215         ret = -ENOMEM;
 216         goto fail;
 217     }
 218
 219     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
 220                s->l1_size * sizeof(uint64_t));
 221     if (ret < 0) {
 222         goto fail;
 223     }
 224
 225     for(i = 0;i < s->l1_size; i++) {
 226         be64_to_cpus(&s->l1_table[i]);
 227     }
 228
 229     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
 230     s->l2_cache =
 231         qemu_try_blockalign(bs->file->bs,
 232                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 233     if (s->l2_cache == NULL) {
 234         error_setg(errp, "Could not allocate L2 table cache");
 235         ret = -ENOMEM;
 236         goto fail;
 237     }
 238     s->cluster_cache = g_malloc(s->cluster_size);
 239     s->cluster_data = g_malloc(s->cluster_size);
 240     s->cluster_cache_offset = -1;
 241
 242     /* read the backing file name */
 243     if (header.backing_file_offset != 0) {
 244         len = header.backing_file_size;
 245         if (len > 1023 || len >= sizeof(bs->backing_file)) {
 246             error_setg(errp, "Backing file name too long");
 247             ret = -EINVAL;
 248             goto fail;
 249         }
 250         ret = bdrv_pread(bs->file, header.backing_file_offset,
 251                    bs->backing_file, len);
 252         if (ret < 0) {
 253             goto fail;
 254         }
 255         bs->backing_file[len] = '\0';
 256     }
 257
 258     /* Disable migration when qcow images are used */
 259     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
 260                "does not support live migration",
 261                bdrv_get_device_or_node_name(bs));
 262     ret = migrate_add_blocker(s->migration_blocker, &local_err);
 263     if (local_err) {
 264         error_propagate(errp, local_err);
 265         error_free(s->migration_blocker);
 266         goto fail;
 267     }
 268
 269     qemu_co_mutex_init(&s->lock);
 270     return 0;
 271
 272  fail:
 273     g_free(s->l1_table);
 274     qemu_vfree(s->l2_cache);
 275     g_free(s->cluster_cache);
 276     g_free(s->cluster_data);
 277     return ret;
 278 }
 279
 280
 281 /* We have nothing to do for QCOW reopen, stubs just return
 282  * success */
 283 static int qcow_reopen_prepare(BDRVReopenState *state,
 284                                BlockReopenQueue *queue, Error **errp)
 285 {
 286     return 0;
 287 }
 288
 289 static int qcow_set_key(BlockDriverState *bs, const char *key)
 290 {
 291     BDRVQcowState *s = bs->opaque;
 292     uint8_t keybuf[16];
 293     int len, i;
 294     Error *err;
 295
 296     memset(keybuf, 0, 16);
 297     len = strlen(key);
 298     if (len > 16)
 299         len = 16;
 300     /* XXX: we could compress the chars to 7 bits to increase
 301        entropy */
 302     for(i = 0;i < len;i++) {
 303         keybuf[i] = key[i];
 304     }
 305     assert(bs->encrypted);
 306
 307     qcrypto_cipher_free(s->cipher);
 308     s->cipher = qcrypto_cipher_new(
 309         QCRYPTO_CIPHER_ALG_AES_128,
 310         QCRYPTO_CIPHER_MODE_CBC,
 311         keybuf, G_N_ELEMENTS(keybuf),
 312         &err);
 313
 314     if (!s->cipher) {
 315         /* XXX would be nice if errors in this method could
 316          * be properly propagate to the caller. Would need
 317          * the bdrv_set_key() API signature to be fixed. */
 318         error_free(err);
 319         return -1;
 320     }
 321     return 0;
 322 }
 323
 324 /* The crypt function is compatible with the linux cryptoloop
 325    algorithm for < 4 GB images. */
 326 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 327                            uint8_t *buf, int nb_sectors, bool enc,
 328                            Error **errp)
 329 {
 330     union {
 331         uint64_t ll[2];
 332         uint8_t b[16];
 333     } ivec;
 334     int i;
 335     int ret;
 336
 337     for(i = 0; i < nb_sectors; i++) {
 338         ivec.ll[0] = cpu_to_le64(sector_num);
 339         ivec.ll[1] = 0;
 340         if (qcrypto_cipher_setiv(s->cipher,
 341                                  ivec.b, G_N_ELEMENTS(ivec.b),
 342                                  errp) < 0) {
 343             return -1;
 344         }
 345         if (enc) {
 346             ret = qcrypto_cipher_encrypt(s->cipher,
 347                                          buf, buf,
 348                                          512,
 349                                          errp);
 350         } else {
 351             ret = qcrypto_cipher_decrypt(s->cipher,
 352                                          buf, buf,
 353                                          512,
 354                                          errp);
 355         }
 356         if (ret < 0) {
 357             return -1;
 358         }
 359         sector_num++;
 360         buf += 512;
 361     }
 362     return 0;
 363 }
 364
 365 /* 'allocate' is:
 366  *
 367  * 0 to not allocate.
 368  *
 369  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 370  * 'n_end')
 371  *
 372  * 2 to allocate a compressed cluster of size
 373  * 'compressed_size'. 'compressed_size' must be > 0 and <
 374  * cluster_size
 375  *
 376  * return 0 if not allocated.
 377  */
 378 static uint64_t get_cluster_offset(BlockDriverState *bs,
 379                                    uint64_t offset, int allocate,
 380                                    int compressed_size,
 381                                    int n_start, int n_end)
 382 {
 383     BDRVQcowState *s = bs->opaque;
 384     int min_index, i, j, l1_index, l2_index;
 385     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
 386     uint32_t min_count;
 387     int new_l2_table;
 388
 389     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 390     l2_offset = s->l1_table[l1_index];
 391     new_l2_table = 0;
 392     if (!l2_offset) {
 393         if (!allocate)
 394             return 0;
 395         /* allocate a new l2 entry */
 396         l2_offset = bdrv_getlength(bs->file->bs);
 397         /* round to cluster size */
 398         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
 399         /* update the L1 entry */
 400         s->l1_table[l1_index] = l2_offset;
 401         tmp = cpu_to_be64(l2_offset);
 402         if (bdrv_pwrite_sync(bs->file,
 403                 s->l1_table_offset + l1_index * sizeof(tmp),
 404                 &tmp, sizeof(tmp)) < 0)
 405             return 0;
 406         new_l2_table = 1;
 407     }
 408     for(i = 0; i < L2_CACHE_SIZE; i++) {
 409         if (l2_offset == s->l2_cache_offsets[i]) {
 410             /* increment the hit count */
 411             if (++s->l2_cache_counts[i] == 0xffffffff) {
 412                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 413                     s->l2_cache_counts[j] >>= 1;
 414                 }
 415             }
 416             l2_table = s->l2_cache + (i << s->l2_bits);
 417             goto found;
 418         }
 419     }
 420     /* not found: load a new entry in the least used one */
 421     min_index = 0;
 422     min_count = 0xffffffff;
 423     for(i = 0; i < L2_CACHE_SIZE; i++) {
 424         if (s->l2_cache_counts[i] < min_count) {
 425             min_count = s->l2_cache_counts[i];
 426             min_index = i;
 427         }
 428     }
 429     l2_table = s->l2_cache + (min_index << s->l2_bits);
 430     if (new_l2_table) {
 431         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 432         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
 433                 s->l2_size * sizeof(uint64_t)) < 0)
 434             return 0;
 435     } else {
 436         if (bdrv_pread(bs->file, l2_offset, l2_table,
 437                        s->l2_size * sizeof(uint64_t)) !=
 438             s->l2_size * sizeof(uint64_t))
 439             return 0;
 440     }
 441     s->l2_cache_offsets[min_index] = l2_offset;
 442     s->l2_cache_counts[min_index] = 1;
 443  found:
 444     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 445     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 446     if (!cluster_offset ||
 447         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 448         if (!allocate)
 449             return 0;
 450         /* allocate a new cluster */
 451         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 452             (n_end - n_start) < s->cluster_sectors) {
 453             /* if the cluster is already compressed, we must
 454                decompress it in the case it is not completely
 455                overwritten */
 456             if (decompress_cluster(bs, cluster_offset) < 0)
 457                 return 0;
 458             cluster_offset = bdrv_getlength(bs->file->bs);
 459             cluster_offset = (cluster_offset + s->cluster_size - 1) &
 460                 ~(s->cluster_size - 1);
 461             /* write the cluster content */
 462             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
 463                             s->cluster_size) !=
 464                 s->cluster_size)
 465                 return -1;
 466         } else {
 467             cluster_offset = bdrv_getlength(bs->file->bs);
 468             if (allocate == 1) {
 469                 /* round to cluster size */
 470                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
 471                     ~(s->cluster_size - 1);
 472                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size, NULL);
 473                 /* if encrypted, we must initialize the cluster
 474                    content which won't be written */
 475                 if (bs->encrypted &&
 476                     (n_end - n_start) < s->cluster_sectors) {
 477                     uint64_t start_sect;
 478                     assert(s->cipher);
 479                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 480                     for(i = 0; i < s->cluster_sectors; i++) {
 481                         if (i < n_start || i >= n_end) {
 482                             Error *err = NULL;
 483                             memset(s->cluster_data, 0x00, 512);
 484                             if (encrypt_sectors(s, start_sect + i,
 485                                                 s->cluster_data, 1,
 486                                                 true, &err) < 0) {
 487                                 error_free(err);
 488                                 errno = EIO;
 489                                 return -1;
 490                             }
 491                             if (bdrv_pwrite(bs->file,
 492                                             cluster_offset + i * 512,
 493                                             s->cluster_data, 512) != 512)
 494                                 return -1;
 495                         }
 496                     }
 497                 }
 498             } else if (allocate == 2) {
 499                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 500                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 501             }
 502         }
 503         /* update L2 table */
 504         tmp = cpu_to_be64(cluster_offset);
 505         l2_table[l2_index] = tmp;
 506         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
 507                 &tmp, sizeof(tmp)) < 0)
 508             return 0;
 509     }
 510     return cluster_offset;
 511 }
 512
 513 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
 514         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
 515 {
 516     BDRVQcowState *s = bs->opaque;
 517     int index_in_cluster, n;
 518     uint64_t cluster_offset;
 519
 520     qemu_co_mutex_lock(&s->lock);
 521     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 522     qemu_co_mutex_unlock(&s->lock);
 523     index_in_cluster = sector_num & (s->cluster_sectors - 1);
 524     n = s->cluster_sectors - index_in_cluster;
 525     if (n > nb_sectors)
 526         n = nb_sectors;
 527     *pnum = n;
 528     if (!cluster_offset) {
 529         return 0;
 530     }
 531     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
 532         return BDRV_BLOCK_DATA;
 533     }
 534     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
 535     *file = bs->file->bs;
 536     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
 537 }
 538
 539 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 540                              const uint8_t *buf, int buf_size)
 541 {
 542     z_stream strm1, *strm = &strm1;
 543     int ret, out_len;
 544
 545     memset(strm, 0, sizeof(*strm));
 546
 547     strm->next_in = (uint8_t *)buf;
 548     strm->avail_in = buf_size;
 549     strm->next_out = out_buf;
 550     strm->avail_out = out_buf_size;
 551
 552     ret = inflateInit2(strm, -12);
 553     if (ret != Z_OK)
 554         return -1;
 555     ret = inflate(strm, Z_FINISH);
 556     out_len = strm->next_out - out_buf;
 557     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 558         out_len != out_buf_size) {
 559         inflateEnd(strm);
 560         return -1;
 561     }
 562     inflateEnd(strm);
 563     return 0;
 564 }
 565
 566 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 567 {
 568     BDRVQcowState *s = bs->opaque;
 569     int ret, csize;
 570     uint64_t coffset;
 571
 572     coffset = cluster_offset & s->cluster_offset_mask;
 573     if (s->cluster_cache_offset != coffset) {
 574         csize = cluster_offset >> (63 - s->cluster_bits);
 575         csize &= (s->cluster_size - 1);
 576         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
 577         if (ret != csize)
 578             return -1;
 579         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 580                               s->cluster_data, csize) < 0) {
 581             return -1;
 582         }
 583         s->cluster_cache_offset = coffset;
 584     }
 585     return 0;
 586 }
 587
 588 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
 589                          int nb_sectors, QEMUIOVector *qiov)
 590 {
 591     BDRVQcowState *s = bs->opaque;
 592     int index_in_cluster;
 593     int ret = 0, n;
 594     uint64_t cluster_offset;
 595     struct iovec hd_iov;
 596     QEMUIOVector hd_qiov;
 597     uint8_t *buf;
 598     void *orig_buf;
 599     Error *err = NULL;
 600
 601     if (qiov->niov > 1) {
 602         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
 603         if (buf == NULL) {
 604             return -ENOMEM;
 605         }
 606     } else {
 607         orig_buf = NULL;
 608         buf = (uint8_t *)qiov->iov->iov_base;
 609     }
 610
 611     qemu_co_mutex_lock(&s->lock);
 612
 613     while (nb_sectors != 0) {
 614         /* prepare next request */
 615         cluster_offset = get_cluster_offset(bs, sector_num << 9,
 616                                                  0, 0, 0, 0);
 617         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 618         n = s->cluster_sectors - index_in_cluster;
 619         if (n > nb_sectors) {
 620             n = nb_sectors;
 621         }
 622
 623         if (!cluster_offset) {
 624             if (bs->backing) {
 625                 /* read from the base image */
 626                 hd_iov.iov_base = (void *)buf;
 627                 hd_iov.iov_len = n * 512;
 628                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 629                 qemu_co_mutex_unlock(&s->lock);
 630                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
 631                 qemu_co_mutex_lock(&s->lock);
 632                 if (ret < 0) {
 633                     goto fail;
 634                 }
 635             } else {
 636                 /* Note: in this case, no need to wait */
 637                 memset(buf, 0, 512 * n);
 638             }
 639         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 640             /* add AIO support for compressed blocks ? */
 641             if (decompress_cluster(bs, cluster_offset) < 0) {
 642                 goto fail;
 643             }
 644             memcpy(buf,
 645                    s->cluster_cache + index_in_cluster * 512, 512 * n);
 646         } else {
 647             if ((cluster_offset & 511) != 0) {
 648                 goto fail;
 649             }
 650             hd_iov.iov_base = (void *)buf;
 651             hd_iov.iov_len = n * 512;
 652             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 653             qemu_co_mutex_unlock(&s->lock);
 654             ret = bdrv_co_readv(bs->file,
 655                                 (cluster_offset >> 9) + index_in_cluster,
 656                                 n, &hd_qiov);
 657             qemu_co_mutex_lock(&s->lock);
 658             if (ret < 0) {
 659                 break;
 660             }
 661             if (bs->encrypted) {
 662                 assert(s->cipher);
 663                 if (encrypt_sectors(s, sector_num, buf,
 664                                     n, false, &err) < 0) {
 665                     goto fail;
 666                 }
 667             }
 668         }
 669         ret = 0;
 670
 671         nb_sectors -= n;
 672         sector_num += n;
 673         buf += n * 512;
 674     }
 675
 676 done:
 677     qemu_co_mutex_unlock(&s->lock);
 678
 679     if (qiov->niov > 1) {
 680         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
 681         qemu_vfree(orig_buf);
 682     }
 683
 684     return ret;
 685
 686 fail:
 687     error_free(err);
 688     ret = -EIO;
 689     goto done;
 690 }
 691
 692 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
 693                           int nb_sectors, QEMUIOVector *qiov)
 694 {
 695     BDRVQcowState *s = bs->opaque;
 696     int index_in_cluster;
 697     uint64_t cluster_offset;
 698     int ret = 0, n;
 699     struct iovec hd_iov;
 700     QEMUIOVector hd_qiov;
 701     uint8_t *buf;
 702     void *orig_buf;
 703
 704     s->cluster_cache_offset = -1; /* disable compressed cache */
 705
 706     /* We must always copy the iov when encrypting, so we
 707      * don't modify the original data buffer during encryption */
 708     if (bs->encrypted || qiov->niov > 1) {
 709         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
 710         if (buf == NULL) {
 711             return -ENOMEM;
 712         }
 713         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
 714     } else {
 715         orig_buf = NULL;
 716         buf = (uint8_t *)qiov->iov->iov_base;
 717     }
 718
 719     qemu_co_mutex_lock(&s->lock);
 720
 721     while (nb_sectors != 0) {
 722
 723         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 724         n = s->cluster_sectors - index_in_cluster;
 725         if (n > nb_sectors) {
 726             n = nb_sectors;
 727         }
 728         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
 729                                             index_in_cluster,
 730                                             index_in_cluster + n);
 731         if (!cluster_offset || (cluster_offset & 511) != 0) {
 732             ret = -EIO;
 733             break;
 734         }
 735         if (bs->encrypted) {
 736             Error *err = NULL;
 737             assert(s->cipher);
 738             if (encrypt_sectors(s, sector_num, buf, n, true, &err) < 0) {
 739                 error_free(err);
 740                 ret = -EIO;
 741                 break;
 742             }
 743         }
 744
 745         hd_iov.iov_base = (void *)buf;
 746         hd_iov.iov_len = n * 512;
 747         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 748         qemu_co_mutex_unlock(&s->lock);
 749         ret = bdrv_co_writev(bs->file,
 750                              (cluster_offset >> 9) + index_in_cluster,
 751                              n, &hd_qiov);
 752         qemu_co_mutex_lock(&s->lock);
 753         if (ret < 0) {
 754             break;
 755         }
 756         ret = 0;
 757
 758         nb_sectors -= n;
 759         sector_num += n;
 760         buf += n * 512;
 761     }
 762     qemu_co_mutex_unlock(&s->lock);
 763
 764     qemu_vfree(orig_buf);
 765
 766     return ret;
 767 }
 768
 769 static void qcow_close(BlockDriverState *bs)
 770 {
 771     BDRVQcowState *s = bs->opaque;
 772
 773     qcrypto_cipher_free(s->cipher);
 774     s->cipher = NULL;
 775     g_free(s->l1_table);
 776     qemu_vfree(s->l2_cache);
 777     g_free(s->cluster_cache);
 778     g_free(s->cluster_data);
 779
 780     migrate_del_blocker(s->migration_blocker);
 781     error_free(s->migration_blocker);
 782 }
 783
 784 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
 785 {
 786     int header_size, backing_filename_len, l1_size, shift, i;
 787     QCowHeader header;
 788     uint8_t *tmp;
 789     int64_t total_size = 0;
 790     char *backing_file = NULL;
 791     Error *local_err = NULL;
 792     int ret;
 793     BlockBackend *qcow_blk;
 794     const char *encryptfmt = NULL;
 795
 796     /* Read out options */
 797     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 798                           BDRV_SECTOR_SIZE);
 799     if (total_size == 0) {
 800         error_setg(errp, "Image size is too small, cannot be zero length");
 801         ret = -EINVAL;
 802         goto cleanup;
 803     }
 804
 805     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
 806     encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
 807     if (encryptfmt) {
 808         if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) {
 809             error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and "
 810                        BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive");
 811             ret = -EINVAL;
 812             goto cleanup;
 813         }
 814     } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
 815         encryptfmt = "aes";
 816     }
 817
 818     ret = bdrv_create_file(filename, opts, &local_err);
 819     if (ret < 0) {
 820         error_propagate(errp, local_err);
 821         goto cleanup;
 822     }
 823
 824     qcow_blk = blk_new_open(filename, NULL, NULL,
 825                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
 826                             &local_err);
 827     if (qcow_blk == NULL) {
 828         error_propagate(errp, local_err);
 829         ret = -EIO;
 830         goto cleanup;
 831     }
 832
 833     blk_set_allow_write_beyond_eof(qcow_blk, true);
 834
 835     ret = blk_truncate(qcow_blk, 0, errp);
 836     if (ret < 0) {
 837         goto exit;
 838     }
 839
 840     memset(&header, 0, sizeof(header));
 841     header.magic = cpu_to_be32(QCOW_MAGIC);
 842     header.version = cpu_to_be32(QCOW_VERSION);
 843     header.size = cpu_to_be64(total_size);
 844     header_size = sizeof(header);
 845     backing_filename_len = 0;
 846     if (backing_file) {
 847         if (strcmp(backing_file, "fat:")) {
 848             header.backing_file_offset = cpu_to_be64(header_size);
 849             backing_filename_len = strlen(backing_file);
 850             header.backing_file_size = cpu_to_be32(backing_filename_len);
 851             header_size += backing_filename_len;
 852         } else {
 853             /* special backing file for vvfat */
 854             g_free(backing_file);
 855             backing_file = NULL;
 856         }
 857         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 858                                     unmodified sectors */
 859         header.l2_bits = 12; /* 32 KB L2 tables */
 860     } else {
 861         header.cluster_bits = 12; /* 4 KB clusters */
 862         header.l2_bits = 9; /* 4 KB L2 tables */
 863     }
 864     header_size = (header_size + 7) & ~7;
 865     shift = header.cluster_bits + header.l2_bits;
 866     l1_size = (total_size + (1LL << shift) - 1) >> shift;
 867
 868     header.l1_table_offset = cpu_to_be64(header_size);
 869     if (encryptfmt) {
 870         if (!g_str_equal(encryptfmt, "aes")) {
 871             error_setg(errp, "Unknown encryption format '%s', expected 'aes'",
 872                        encryptfmt);
 873             ret = -EINVAL;
 874             goto exit;
 875         }
 876         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 877     } else {
 878         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 879     }
 880
 881     /* write all the data */
 882     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
 883     if (ret != sizeof(header)) {
 884         goto exit;
 885     }
 886
 887     if (backing_file) {
 888         ret = blk_pwrite(qcow_blk, sizeof(header),
 889                          backing_file, backing_filename_len, 0);
 890         if (ret != backing_filename_len) {
 891             goto exit;
 892         }
 893     }
 894
 895     tmp = g_malloc0(BDRV_SECTOR_SIZE);
 896     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
 897          i++) {
 898         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
 899                          tmp, BDRV_SECTOR_SIZE, 0);
 900         if (ret != BDRV_SECTOR_SIZE) {
 901             g_free(tmp);
 902             goto exit;
 903         }
 904     }
 905
 906     g_free(tmp);
 907     ret = 0;
 908 exit:
 909     blk_unref(qcow_blk);
 910 cleanup:
 911     g_free(backing_file);
 912     return ret;
 913 }
 914
 915 static int qcow_make_empty(BlockDriverState *bs)
 916 {
 917     BDRVQcowState *s = bs->opaque;
 918     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 919     int ret;
 920
 921     memset(s->l1_table, 0, l1_length);
 922     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
 923             l1_length) < 0)
 924         return -1;
 925     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, NULL);
 926     if (ret < 0)
 927         return ret;
 928
 929     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 930     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
 931     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 932
 933     return 0;
 934 }
 935
 936 /* XXX: put compressed sectors first, then all the cluster aligned
 937    tables to avoid losing bytes in alignment */
 938 static coroutine_fn int
 939 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
 940                            uint64_t bytes, QEMUIOVector *qiov)
 941 {
 942     BDRVQcowState *s = bs->opaque;
 943     QEMUIOVector hd_qiov;
 944     struct iovec iov;
 945     z_stream strm;
 946     int ret, out_len;
 947     uint8_t *buf, *out_buf;
 948     uint64_t cluster_offset;
 949
 950     buf = qemu_blockalign(bs, s->cluster_size);
 951     if (bytes != s->cluster_size) {
 952         if (bytes > s->cluster_size ||
 953             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
 954         {
 955             qemu_vfree(buf);
 956             return -EINVAL;
 957         }
 958         /* Zero-pad last write if image size is not cluster aligned */
 959         memset(buf + bytes, 0, s->cluster_size - bytes);
 960     }
 961     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
 962
 963     out_buf = g_malloc(s->cluster_size);
 964
 965     /* best compression, small window, no zlib header */
 966     memset(&strm, 0, sizeof(strm));
 967     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 968                        Z_DEFLATED, -12,
 969                        9, Z_DEFAULT_STRATEGY);
 970     if (ret != 0) {
 971         ret = -EINVAL;
 972         goto fail;
 973     }
 974
 975     strm.avail_in = s->cluster_size;
 976     strm.next_in = (uint8_t *)buf;
 977     strm.avail_out = s->cluster_size;
 978     strm.next_out = out_buf;
 979
 980     ret = deflate(&strm, Z_FINISH);
 981     if (ret != Z_STREAM_END && ret != Z_OK) {
 982         deflateEnd(&strm);
 983         ret = -EINVAL;
 984         goto fail;
 985     }
 986     out_len = strm.next_out - out_buf;
 987
 988     deflateEnd(&strm);
 989
 990     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 991         /* could not compress: write normal cluster */
 992         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
 993                              bytes >> BDRV_SECTOR_BITS, qiov);
 994         if (ret < 0) {
 995             goto fail;
 996         }
 997         goto success;
 998     }
 999     qemu_co_mutex_lock(&s->lock);
1000     cluster_offset = get_cluster_offset(bs, offset, 2, out_len, 0, 0);
1001     qemu_co_mutex_unlock(&s->lock);
1002     if (cluster_offset == 0) {
1003         ret = -EIO;
1004         goto fail;
1005     }
1006     cluster_offset &= s->cluster_offset_mask;
1007
1008     iov = (struct iovec) {
1009         .iov_base   = out_buf,
1010         .iov_len    = out_len,
1011     };
1012     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1013     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1014     if (ret < 0) {
1015         goto fail;
1016     }
1017 success:
1018     ret = 0;
1019 fail:
1020     qemu_vfree(buf);
1021     g_free(out_buf);
1022     return ret;
1023 }
1024
1025 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1026 {
1027     BDRVQcowState *s = bs->opaque;
1028     bdi->cluster_size = s->cluster_size;
1029     return 0;
1030 }
1031
1032 static QemuOptsList qcow_create_opts = {
1033     .name = "qcow-create-opts",
1034     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1035     .desc = {
1036         {
1037             .name = BLOCK_OPT_SIZE,
1038             .type = QEMU_OPT_SIZE,
1039             .help = "Virtual disk size"
1040         },
1041         {
1042             .name = BLOCK_OPT_BACKING_FILE,
1043             .type = QEMU_OPT_STRING,
1044             .help = "File name of a base image"
1045         },
1046         {
1047             .name = BLOCK_OPT_ENCRYPT,
1048             .type = QEMU_OPT_BOOL,
1049             .help = "Encrypt the image with format 'aes'. (Deprecated "
1050                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1051         },
1052         {
1053             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1054             .type = QEMU_OPT_STRING,
1055             .help = "Encrypt the image, format choices: 'aes'",
1056         },
1057         { /* end of list */ }
1058     }
1059 };
1060
1061 static BlockDriver bdrv_qcow = {
1062     .format_name        = "qcow",
1063     .instance_size      = sizeof(BDRVQcowState),
1064     .bdrv_probe         = qcow_probe,
1065     .bdrv_open          = qcow_open,
1066     .bdrv_close         = qcow_close,
1067     .bdrv_child_perm        = bdrv_format_default_perms,
1068     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1069     .bdrv_create            = qcow_create,
1070     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1071     .supports_backing       = true,
1072
1073     .bdrv_co_readv          = qcow_co_readv,
1074     .bdrv_co_writev         = qcow_co_writev,
1075     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1076
1077     .bdrv_set_key           = qcow_set_key,
1078     .bdrv_make_empty        = qcow_make_empty,
1079     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1080     .bdrv_get_info          = qcow_get_info,
1081
1082     .create_opts            = &qcow_create_opts,
1083 };
1084
1085 static void bdrv_qcow_init(void)
1086 {
1087     bdrv_register(&bdrv_qcow);
1088 }
1089
1090 block_init(bdrv_qcow_init);