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-common.h"
  25 #include "block_int.h"
  26 #include "module.h"
  27 #include <zlib.h>
  28 #include "aes.h"
  29
  30 /**************************************************************/
  31 /* QEMU COW block driver with compression and encryption support */
  32
  33 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  34 #define QCOW_VERSION 1
  35
  36 #define QCOW_CRYPT_NONE 0
  37 #define QCOW_CRYPT_AES  1
  38
  39 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
  40
  41 typedef struct QCowHeader {
  42     uint32_t magic;
  43     uint32_t version;
  44     uint64_t backing_file_offset;
  45     uint32_t backing_file_size;
  46     uint32_t mtime;
  47     uint64_t size; /* in bytes */
  48     uint8_t cluster_bits;
  49     uint8_t l2_bits;
  50     uint32_t crypt_method;
  51     uint64_t l1_table_offset;
  52 } QCowHeader;
  53
  54 #define L2_CACHE_SIZE 16
  55
  56 typedef struct BDRVQcowState {
  57     BlockDriverState *hd;
  58     int cluster_bits;
  59     int cluster_size;
  60     int cluster_sectors;
  61     int l2_bits;
  62     int l2_size;
  63     int l1_size;
  64     uint64_t cluster_offset_mask;
  65     uint64_t l1_table_offset;
  66     uint64_t *l1_table;
  67     uint64_t *l2_cache;
  68     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  69     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  70     uint8_t *cluster_cache;
  71     uint8_t *cluster_data;
  72     uint64_t cluster_cache_offset;
  73     uint32_t crypt_method; /* current crypt method, 0 if no key yet */
  74     uint32_t crypt_method_header;
  75     AES_KEY aes_encrypt_key;
  76     AES_KEY aes_decrypt_key;
  77 } BDRVQcowState;
  78
  79 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
  80
  81 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  82 {
  83     const QCowHeader *cow_header = (const void *)buf;
  84
  85     if (buf_size >= sizeof(QCowHeader) &&
  86         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  87         be32_to_cpu(cow_header->version) == QCOW_VERSION)
  88         return 100;
  89     else
  90         return 0;
  91 }
  92
  93 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
  94 {
  95     BDRVQcowState *s = bs->opaque;
  96     int len, i, shift, ret;
  97     QCowHeader header;
  98
  99     ret = bdrv_file_open(&s->hd, filename, flags);
 100     if (ret < 0)
 101         return ret;
 102     if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
 103         goto fail;
 104     be32_to_cpus(&header.magic);
 105     be32_to_cpus(&header.version);
 106     be64_to_cpus(&header.backing_file_offset);
 107     be32_to_cpus(&header.backing_file_size);
 108     be32_to_cpus(&header.mtime);
 109     be64_to_cpus(&header.size);
 110     be32_to_cpus(&header.crypt_method);
 111     be64_to_cpus(&header.l1_table_offset);
 112
 113     if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
 114         goto fail;
 115     if (header.size <= 1 || header.cluster_bits < 9)
 116         goto fail;
 117     if (header.crypt_method > QCOW_CRYPT_AES)
 118         goto fail;
 119     s->crypt_method_header = header.crypt_method;
 120     if (s->crypt_method_header)
 121         bs->encrypted = 1;
 122     s->cluster_bits = header.cluster_bits;
 123     s->cluster_size = 1 << s->cluster_bits;
 124     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 125     s->l2_bits = header.l2_bits;
 126     s->l2_size = 1 << s->l2_bits;
 127     bs->total_sectors = header.size / 512;
 128     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 129
 130     /* read the level 1 table */
 131     shift = s->cluster_bits + s->l2_bits;
 132     s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
 133
 134     s->l1_table_offset = header.l1_table_offset;
 135     s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
 136     if (!s->l1_table)
 137         goto fail;
 138     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
 139         s->l1_size * sizeof(uint64_t))
 140         goto fail;
 141     for(i = 0;i < s->l1_size; i++) {
 142         be64_to_cpus(&s->l1_table[i]);
 143     }
 144     /* alloc L2 cache */
 145     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 146     if (!s->l2_cache)
 147         goto fail;
 148     s->cluster_cache = qemu_malloc(s->cluster_size);
 149     if (!s->cluster_cache)
 150         goto fail;
 151     s->cluster_data = qemu_malloc(s->cluster_size);
 152     if (!s->cluster_data)
 153         goto fail;
 154     s->cluster_cache_offset = -1;
 155
 156     /* read the backing file name */
 157     if (header.backing_file_offset != 0) {
 158         len = header.backing_file_size;
 159         if (len > 1023)
 160             len = 1023;
 161         if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
 162             goto fail;
 163         bs->backing_file[len] = '\0';
 164     }
 165     return 0;
 166
 167  fail:
 168     qemu_free(s->l1_table);
 169     qemu_free(s->l2_cache);
 170     qemu_free(s->cluster_cache);
 171     qemu_free(s->cluster_data);
 172     bdrv_delete(s->hd);
 173     return -1;
 174 }
 175
 176 static int qcow_set_key(BlockDriverState *bs, const char *key)
 177 {
 178     BDRVQcowState *s = bs->opaque;
 179     uint8_t keybuf[16];
 180     int len, i;
 181
 182     memset(keybuf, 0, 16);
 183     len = strlen(key);
 184     if (len > 16)
 185         len = 16;
 186     /* XXX: we could compress the chars to 7 bits to increase
 187        entropy */
 188     for(i = 0;i < len;i++) {
 189         keybuf[i] = key[i];
 190     }
 191     s->crypt_method = s->crypt_method_header;
 192
 193     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
 194         return -1;
 195     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
 196         return -1;
 197 #if 0
 198     /* test */
 199     {
 200         uint8_t in[16];
 201         uint8_t out[16];
 202         uint8_t tmp[16];
 203         for(i=0;i<16;i++)
 204             in[i] = i;
 205         AES_encrypt(in, tmp, &s->aes_encrypt_key);
 206         AES_decrypt(tmp, out, &s->aes_decrypt_key);
 207         for(i = 0; i < 16; i++)
 208             printf(" %02x", tmp[i]);
 209         printf("\n");
 210         for(i = 0; i < 16; i++)
 211             printf(" %02x", out[i]);
 212         printf("\n");
 213     }
 214 #endif
 215     return 0;
 216 }
 217
 218 /* The crypt function is compatible with the linux cryptoloop
 219    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
 220    supported */
 221 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 222                             uint8_t *out_buf, const uint8_t *in_buf,
 223                             int nb_sectors, int enc,
 224                             const AES_KEY *key)
 225 {
 226     union {
 227         uint64_t ll[2];
 228         uint8_t b[16];
 229     } ivec;
 230     int i;
 231
 232     for(i = 0; i < nb_sectors; i++) {
 233         ivec.ll[0] = cpu_to_le64(sector_num);
 234         ivec.ll[1] = 0;
 235         AES_cbc_encrypt(in_buf, out_buf, 512, key,
 236                         ivec.b, enc);
 237         sector_num++;
 238         in_buf += 512;
 239         out_buf += 512;
 240     }
 241 }
 242
 243 /* 'allocate' is:
 244  *
 245  * 0 to not allocate.
 246  *
 247  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 248  * 'n_end')
 249  *
 250  * 2 to allocate a compressed cluster of size
 251  * 'compressed_size'. 'compressed_size' must be > 0 and <
 252  * cluster_size
 253  *
 254  * return 0 if not allocated.
 255  */
 256 static uint64_t get_cluster_offset(BlockDriverState *bs,
 257                                    uint64_t offset, int allocate,
 258                                    int compressed_size,
 259                                    int n_start, int n_end)
 260 {
 261     BDRVQcowState *s = bs->opaque;
 262     int min_index, i, j, l1_index, l2_index;
 263     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
 264     uint32_t min_count;
 265     int new_l2_table;
 266
 267     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 268     l2_offset = s->l1_table[l1_index];
 269     new_l2_table = 0;
 270     if (!l2_offset) {
 271         if (!allocate)
 272             return 0;
 273         /* allocate a new l2 entry */
 274         l2_offset = bdrv_getlength(s->hd);
 275         /* round to cluster size */
 276         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
 277         /* update the L1 entry */
 278         s->l1_table[l1_index] = l2_offset;
 279         tmp = cpu_to_be64(l2_offset);
 280         if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
 281                         &tmp, sizeof(tmp)) != sizeof(tmp))
 282             return 0;
 283         new_l2_table = 1;
 284     }
 285     for(i = 0; i < L2_CACHE_SIZE; i++) {
 286         if (l2_offset == s->l2_cache_offsets[i]) {
 287             /* increment the hit count */
 288             if (++s->l2_cache_counts[i] == 0xffffffff) {
 289                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 290                     s->l2_cache_counts[j] >>= 1;
 291                 }
 292             }
 293             l2_table = s->l2_cache + (i << s->l2_bits);
 294             goto found;
 295         }
 296     }
 297     /* not found: load a new entry in the least used one */
 298     min_index = 0;
 299     min_count = 0xffffffff;
 300     for(i = 0; i < L2_CACHE_SIZE; i++) {
 301         if (s->l2_cache_counts[i] < min_count) {
 302             min_count = s->l2_cache_counts[i];
 303             min_index = i;
 304         }
 305     }
 306     l2_table = s->l2_cache + (min_index << s->l2_bits);
 307     if (new_l2_table) {
 308         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 309         if (bdrv_pwrite(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
 310             s->l2_size * sizeof(uint64_t))
 311             return 0;
 312     } else {
 313         if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
 314             s->l2_size * sizeof(uint64_t))
 315             return 0;
 316     }
 317     s->l2_cache_offsets[min_index] = l2_offset;
 318     s->l2_cache_counts[min_index] = 1;
 319  found:
 320     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 321     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 322     if (!cluster_offset ||
 323         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 324         if (!allocate)
 325             return 0;
 326         /* allocate a new cluster */
 327         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 328             (n_end - n_start) < s->cluster_sectors) {
 329             /* if the cluster is already compressed, we must
 330                decompress it in the case it is not completely
 331                overwritten */
 332             if (decompress_cluster(s, cluster_offset) < 0)
 333                 return 0;
 334             cluster_offset = bdrv_getlength(s->hd);
 335             cluster_offset = (cluster_offset + s->cluster_size - 1) &
 336                 ~(s->cluster_size - 1);
 337             /* write the cluster content */
 338             if (bdrv_pwrite(s->hd, cluster_offset, s->cluster_cache, s->cluster_size) !=
 339                 s->cluster_size)
 340                 return -1;
 341         } else {
 342             cluster_offset = bdrv_getlength(s->hd);
 343             if (allocate == 1) {
 344                 /* round to cluster size */
 345                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
 346                     ~(s->cluster_size - 1);
 347                 bdrv_truncate(s->hd, cluster_offset + s->cluster_size);
 348                 /* if encrypted, we must initialize the cluster
 349                    content which won't be written */
 350                 if (s->crypt_method &&
 351                     (n_end - n_start) < s->cluster_sectors) {
 352                     uint64_t start_sect;
 353                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 354                     memset(s->cluster_data + 512, 0x00, 512);
 355                     for(i = 0; i < s->cluster_sectors; i++) {
 356                         if (i < n_start || i >= n_end) {
 357                             encrypt_sectors(s, start_sect + i,
 358                                             s->cluster_data,
 359                                             s->cluster_data + 512, 1, 1,
 360                                             &s->aes_encrypt_key);
 361                             if (bdrv_pwrite(s->hd, cluster_offset + i * 512,
 362                                             s->cluster_data, 512) != 512)
 363                                 return -1;
 364                         }
 365                     }
 366                 }
 367             } else if (allocate == 2) {
 368                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 369                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 370             }
 371         }
 372         /* update L2 table */
 373         tmp = cpu_to_be64(cluster_offset);
 374         l2_table[l2_index] = tmp;
 375         if (bdrv_pwrite(s->hd,
 376                         l2_offset + l2_index * sizeof(tmp), &tmp, sizeof(tmp)) != sizeof(tmp))
 377             return 0;
 378     }
 379     return cluster_offset;
 380 }
 381
 382 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
 383                              int nb_sectors, int *pnum)
 384 {
 385     BDRVQcowState *s = bs->opaque;
 386     int index_in_cluster, n;
 387     uint64_t cluster_offset;
 388
 389     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 390     index_in_cluster = sector_num & (s->cluster_sectors - 1);
 391     n = s->cluster_sectors - index_in_cluster;
 392     if (n > nb_sectors)
 393         n = nb_sectors;
 394     *pnum = n;
 395     return (cluster_offset != 0);
 396 }
 397
 398 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 399                              const uint8_t *buf, int buf_size)
 400 {
 401     z_stream strm1, *strm = &strm1;
 402     int ret, out_len;
 403
 404     memset(strm, 0, sizeof(*strm));
 405
 406     strm->next_in = (uint8_t *)buf;
 407     strm->avail_in = buf_size;
 408     strm->next_out = out_buf;
 409     strm->avail_out = out_buf_size;
 410
 411     ret = inflateInit2(strm, -12);
 412     if (ret != Z_OK)
 413         return -1;
 414     ret = inflate(strm, Z_FINISH);
 415     out_len = strm->next_out - out_buf;
 416     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 417         out_len != out_buf_size) {
 418         inflateEnd(strm);
 419         return -1;
 420     }
 421     inflateEnd(strm);
 422     return 0;
 423 }
 424
 425 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
 426 {
 427     int ret, csize;
 428     uint64_t coffset;
 429
 430     coffset = cluster_offset & s->cluster_offset_mask;
 431     if (s->cluster_cache_offset != coffset) {
 432         csize = cluster_offset >> (63 - s->cluster_bits);
 433         csize &= (s->cluster_size - 1);
 434         ret = bdrv_pread(s->hd, coffset, s->cluster_data, csize);
 435         if (ret != csize)
 436             return -1;
 437         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 438                               s->cluster_data, csize) < 0) {
 439             return -1;
 440         }
 441         s->cluster_cache_offset = coffset;
 442     }
 443     return 0;
 444 }
 445
 446 #if 0
 447
 448 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
 449                      uint8_t *buf, int nb_sectors)
 450 {
 451     BDRVQcowState *s = bs->opaque;
 452     int ret, index_in_cluster, n;
 453     uint64_t cluster_offset;
 454
 455     while (nb_sectors > 0) {
 456         cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 457         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 458         n = s->cluster_sectors - index_in_cluster;
 459         if (n > nb_sectors)
 460             n = nb_sectors;
 461         if (!cluster_offset) {
 462             if (bs->backing_hd) {
 463                 /* read from the base image */
 464                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
 465                 if (ret < 0)
 466                     return -1;
 467             } else {
 468                 memset(buf, 0, 512 * n);
 469             }
 470         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 471             if (decompress_cluster(s, cluster_offset) < 0)
 472                 return -1;
 473             memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
 474         } else {
 475             ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
 476             if (ret != n * 512)
 477                 return -1;
 478             if (s->crypt_method) {
 479                 encrypt_sectors(s, sector_num, buf, buf, n, 0,
 480                                 &s->aes_decrypt_key);
 481             }
 482         }
 483         nb_sectors -= n;
 484         sector_num += n;
 485         buf += n * 512;
 486     }
 487     return 0;
 488 }
 489 #endif
 490
 491 typedef struct QCowAIOCB {
 492     BlockDriverAIOCB common;
 493     int64_t sector_num;
 494     QEMUIOVector *qiov;
 495     uint8_t *buf;
 496     void *orig_buf;
 497     int nb_sectors;
 498     int n;
 499     uint64_t cluster_offset;
 500     uint8_t *cluster_data;
 501     struct iovec hd_iov;
 502     QEMUIOVector hd_qiov;
 503     BlockDriverAIOCB *hd_aiocb;
 504 } QCowAIOCB;
 505
 506 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
 507 {
 508     QCowAIOCB *acb = (QCowAIOCB *)blockacb;
 509     if (acb->hd_aiocb)
 510         bdrv_aio_cancel(acb->hd_aiocb);
 511     qemu_aio_release(acb);
 512 }
 513
 514 static AIOPool qcow_aio_pool = {
 515     .aiocb_size         = sizeof(QCowAIOCB),
 516     .cancel             = qcow_aio_cancel,
 517 };
 518
 519 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
 520         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 521         BlockDriverCompletionFunc *cb, void *opaque, int is_write)
 522 {
 523     QCowAIOCB *acb;
 524
 525     acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
 526     if (!acb)
 527         return NULL;
 528     acb->hd_aiocb = NULL;
 529     acb->sector_num = sector_num;
 530     acb->qiov = qiov;
 531     if (qiov->niov > 1) {
 532         acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
 533         if (is_write)
 534             qemu_iovec_to_buffer(qiov, acb->buf);
 535     } else {
 536         acb->buf = (uint8_t *)qiov->iov->iov_base;
 537     }
 538     acb->nb_sectors = nb_sectors;
 539     acb->n = 0;
 540     acb->cluster_offset = 0;
 541     return acb;
 542 }
 543
 544 static void qcow_aio_read_cb(void *opaque, int ret)
 545 {
 546     QCowAIOCB *acb = opaque;
 547     BlockDriverState *bs = acb->common.bs;
 548     BDRVQcowState *s = bs->opaque;
 549     int index_in_cluster;
 550
 551     acb->hd_aiocb = NULL;
 552     if (ret < 0)
 553         goto done;
 554
 555  redo:
 556     /* post process the read buffer */
 557     if (!acb->cluster_offset) {
 558         /* nothing to do */
 559     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
 560         /* nothing to do */
 561     } else {
 562         if (s->crypt_method) {
 563             encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
 564                             acb->n, 0,
 565                             &s->aes_decrypt_key);
 566         }
 567     }
 568
 569     acb->nb_sectors -= acb->n;
 570     acb->sector_num += acb->n;
 571     acb->buf += acb->n * 512;
 572
 573     if (acb->nb_sectors == 0) {
 574         /* request completed */
 575         ret = 0;
 576         goto done;
 577     }
 578
 579     /* prepare next AIO request */
 580     acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9,
 581                                              0, 0, 0, 0);
 582     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
 583     acb->n = s->cluster_sectors - index_in_cluster;
 584     if (acb->n > acb->nb_sectors)
 585         acb->n = acb->nb_sectors;
 586
 587     if (!acb->cluster_offset) {
 588         if (bs->backing_hd) {
 589             /* read from the base image */
 590             acb->hd_iov.iov_base = (void *)acb->buf;
 591             acb->hd_iov.iov_len = acb->n * 512;
 592             qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 593             acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
 594                 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
 595             if (acb->hd_aiocb == NULL)
 596                 goto done;
 597         } else {
 598             /* Note: in this case, no need to wait */
 599             memset(acb->buf, 0, 512 * acb->n);
 600             goto redo;
 601         }
 602     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
 603         /* add AIO support for compressed blocks ? */
 604         if (decompress_cluster(s, acb->cluster_offset) < 0)
 605             goto done;
 606         memcpy(acb->buf,
 607                s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
 608         goto redo;
 609     } else {
 610         if ((acb->cluster_offset & 511) != 0) {
 611             ret = -EIO;
 612             goto done;
 613         }
 614         acb->hd_iov.iov_base = (void *)acb->buf;
 615         acb->hd_iov.iov_len = acb->n * 512;
 616         qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 617         acb->hd_aiocb = bdrv_aio_readv(s->hd,
 618                             (acb->cluster_offset >> 9) + index_in_cluster,
 619                             &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
 620         if (acb->hd_aiocb == NULL)
 621             goto done;
 622     }
 623
 624     return;
 625
 626 done:
 627     if (acb->qiov->niov > 1) {
 628         qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
 629         qemu_vfree(acb->orig_buf);
 630     }
 631     acb->common.cb(acb->common.opaque, ret);
 632     qemu_aio_release(acb);
 633 }
 634
 635 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
 636         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 637         BlockDriverCompletionFunc *cb, void *opaque)
 638 {
 639     QCowAIOCB *acb;
 640
 641     acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
 642     if (!acb)
 643         return NULL;
 644
 645     qcow_aio_read_cb(acb, 0);
 646     return &acb->common;
 647 }
 648
 649 static void qcow_aio_write_cb(void *opaque, int ret)
 650 {
 651     QCowAIOCB *acb = opaque;
 652     BlockDriverState *bs = acb->common.bs;
 653     BDRVQcowState *s = bs->opaque;
 654     int index_in_cluster;
 655     uint64_t cluster_offset;
 656     const uint8_t *src_buf;
 657
 658     acb->hd_aiocb = NULL;
 659
 660     if (ret < 0)
 661         goto done;
 662
 663     acb->nb_sectors -= acb->n;
 664     acb->sector_num += acb->n;
 665     acb->buf += acb->n * 512;
 666
 667     if (acb->nb_sectors == 0) {
 668         /* request completed */
 669         ret = 0;
 670         goto done;
 671     }
 672
 673     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
 674     acb->n = s->cluster_sectors - index_in_cluster;
 675     if (acb->n > acb->nb_sectors)
 676         acb->n = acb->nb_sectors;
 677     cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, 1, 0,
 678                                         index_in_cluster,
 679                                         index_in_cluster + acb->n);
 680     if (!cluster_offset || (cluster_offset & 511) != 0) {
 681         ret = -EIO;
 682         goto done;
 683     }
 684     if (s->crypt_method) {
 685         if (!acb->cluster_data) {
 686             acb->cluster_data = qemu_mallocz(s->cluster_size);
 687             if (!acb->cluster_data) {
 688                 ret = -ENOMEM;
 689                 goto done;
 690             }
 691         }
 692         encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
 693                         acb->n, 1, &s->aes_encrypt_key);
 694         src_buf = acb->cluster_data;
 695     } else {
 696         src_buf = acb->buf;
 697     }
 698
 699     acb->hd_iov.iov_base = (void *)src_buf;
 700     acb->hd_iov.iov_len = acb->n * 512;
 701     qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 702     acb->hd_aiocb = bdrv_aio_writev(s->hd,
 703                                     (cluster_offset >> 9) + index_in_cluster,
 704                                     &acb->hd_qiov, acb->n,
 705                                     qcow_aio_write_cb, acb);
 706     if (acb->hd_aiocb == NULL)
 707         goto done;
 708     return;
 709
 710 done:
 711     if (acb->qiov->niov > 1)
 712         qemu_vfree(acb->orig_buf);
 713     acb->common.cb(acb->common.opaque, ret);
 714     qemu_aio_release(acb);
 715 }
 716
 717 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
 718         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 719         BlockDriverCompletionFunc *cb, void *opaque)
 720 {
 721     BDRVQcowState *s = bs->opaque;
 722     QCowAIOCB *acb;
 723
 724     s->cluster_cache_offset = -1; /* disable compressed cache */
 725
 726     acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
 727     if (!acb)
 728         return NULL;
 729
 730
 731     qcow_aio_write_cb(acb, 0);
 732     return &acb->common;
 733 }
 734
 735 static void qcow_close(BlockDriverState *bs)
 736 {
 737     BDRVQcowState *s = bs->opaque;
 738     qemu_free(s->l1_table);
 739     qemu_free(s->l2_cache);
 740     qemu_free(s->cluster_cache);
 741     qemu_free(s->cluster_data);
 742     bdrv_delete(s->hd);
 743 }
 744
 745 static int qcow_create(const char *filename, QEMUOptionParameter *options)
 746 {
 747     int fd, header_size, backing_filename_len, l1_size, i, shift;
 748     QCowHeader header;
 749     uint64_t tmp;
 750     int64_t total_size = 0;
 751     const char *backing_file = NULL;
 752     int flags = 0;
 753
 754     /* Read out options */
 755     while (options && options->name) {
 756         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 757             total_size = options->value.n / 512;
 758         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 759             backing_file = options->value.s;
 760         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
 761             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
 762         }
 763         options++;
 764     }
 765
 766     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
 767     if (fd < 0)
 768         return -1;
 769     memset(&header, 0, sizeof(header));
 770     header.magic = cpu_to_be32(QCOW_MAGIC);
 771     header.version = cpu_to_be32(QCOW_VERSION);
 772     header.size = cpu_to_be64(total_size * 512);
 773     header_size = sizeof(header);
 774     backing_filename_len = 0;
 775     if (backing_file) {
 776         if (strcmp(backing_file, "fat:")) {
 777             header.backing_file_offset = cpu_to_be64(header_size);
 778             backing_filename_len = strlen(backing_file);
 779             header.backing_file_size = cpu_to_be32(backing_filename_len);
 780             header_size += backing_filename_len;
 781         } else {
 782             /* special backing file for vvfat */
 783             backing_file = NULL;
 784         }
 785         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 786                                     unmodifyed sectors */
 787         header.l2_bits = 12; /* 32 KB L2 tables */
 788     } else {
 789         header.cluster_bits = 12; /* 4 KB clusters */
 790         header.l2_bits = 9; /* 4 KB L2 tables */
 791     }
 792     header_size = (header_size + 7) & ~7;
 793     shift = header.cluster_bits + header.l2_bits;
 794     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
 795
 796     header.l1_table_offset = cpu_to_be64(header_size);
 797     if (flags & BLOCK_FLAG_ENCRYPT) {
 798         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 799     } else {
 800         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 801     }
 802
 803     /* write all the data */
 804     write(fd, &header, sizeof(header));
 805     if (backing_file) {
 806         write(fd, backing_file, backing_filename_len);
 807     }
 808     lseek(fd, header_size, SEEK_SET);
 809     tmp = 0;
 810     for(i = 0;i < l1_size; i++) {
 811         write(fd, &tmp, sizeof(tmp));
 812     }
 813     close(fd);
 814     return 0;
 815 }
 816
 817 static int qcow_make_empty(BlockDriverState *bs)
 818 {
 819     BDRVQcowState *s = bs->opaque;
 820     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 821     int ret;
 822
 823     memset(s->l1_table, 0, l1_length);
 824     if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
 825         return -1;
 826     ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
 827     if (ret < 0)
 828         return ret;
 829
 830     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 831     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
 832     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 833
 834     return 0;
 835 }
 836
 837 /* XXX: put compressed sectors first, then all the cluster aligned
 838    tables to avoid losing bytes in alignment */
 839 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
 840                                  const uint8_t *buf, int nb_sectors)
 841 {
 842     BDRVQcowState *s = bs->opaque;
 843     z_stream strm;
 844     int ret, out_len;
 845     uint8_t *out_buf;
 846     uint64_t cluster_offset;
 847
 848     if (nb_sectors != s->cluster_sectors)
 849         return -EINVAL;
 850
 851     out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 852     if (!out_buf)
 853         return -1;
 854
 855     /* best compression, small window, no zlib header */
 856     memset(&strm, 0, sizeof(strm));
 857     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 858                        Z_DEFLATED, -12,
 859                        9, Z_DEFAULT_STRATEGY);
 860     if (ret != 0) {
 861         qemu_free(out_buf);
 862         return -1;
 863     }
 864
 865     strm.avail_in = s->cluster_size;
 866     strm.next_in = (uint8_t *)buf;
 867     strm.avail_out = s->cluster_size;
 868     strm.next_out = out_buf;
 869
 870     ret = deflate(&strm, Z_FINISH);
 871     if (ret != Z_STREAM_END && ret != Z_OK) {
 872         qemu_free(out_buf);
 873         deflateEnd(&strm);
 874         return -1;
 875     }
 876     out_len = strm.next_out - out_buf;
 877
 878     deflateEnd(&strm);
 879
 880     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 881         /* could not compress: write normal cluster */
 882         bdrv_write(bs, sector_num, buf, s->cluster_sectors);
 883     } else {
 884         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
 885                                             out_len, 0, 0);
 886         cluster_offset &= s->cluster_offset_mask;
 887         if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
 888             qemu_free(out_buf);
 889             return -1;
 890         }
 891     }
 892
 893     qemu_free(out_buf);
 894     return 0;
 895 }
 896
 897 static void qcow_flush(BlockDriverState *bs)
 898 {
 899     BDRVQcowState *s = bs->opaque;
 900     bdrv_flush(s->hd);
 901 }
 902
 903 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 904 {
 905     BDRVQcowState *s = bs->opaque;
 906     bdi->cluster_size = s->cluster_size;
 907     return 0;
 908 }
 909
 910
 911 static QEMUOptionParameter qcow_create_options[] = {
 912     {
 913         .name = BLOCK_OPT_SIZE,
 914         .type = OPT_SIZE,
 915         .help = "Virtual disk size"
 916     },
 917     {
 918         .name = BLOCK_OPT_BACKING_FILE,
 919         .type = OPT_STRING,
 920         .help = "File name of a base image"
 921     },
 922     {
 923         .name = BLOCK_OPT_ENCRYPT,
 924         .type = OPT_FLAG,
 925         .help = "Encrypt the image"
 926     },
 927     { NULL }
 928 };
 929
 930 static BlockDriver bdrv_qcow = {
 931     .format_name        = "qcow",
 932     .instance_size      = sizeof(BDRVQcowState),
 933     .bdrv_probe         = qcow_probe,
 934     .bdrv_open          = qcow_open,
 935     .bdrv_close         = qcow_close,
 936     .bdrv_create        = qcow_create,
 937     .bdrv_flush         = qcow_flush,
 938     .bdrv_is_allocated  = qcow_is_allocated,
 939     .bdrv_set_key       = qcow_set_key,
 940     .bdrv_make_empty    = qcow_make_empty,
 941     .bdrv_aio_readv     = qcow_aio_readv,
 942     .bdrv_aio_writev    = qcow_aio_writev,
 943     .bdrv_write_compressed = qcow_write_compressed,
 944     .bdrv_get_info      = qcow_get_info,
 945
 946     .create_options = qcow_create_options,
 947 };
 948
 949 static void bdrv_qcow_init(void)
 950 {
 951     bdrv_register(&bdrv_qcow);
 952 }
 953
 954 block_init(bdrv_qcow_init);