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