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