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