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