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(BlockDriverState *bs, 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, int flags)
  94 {
  95     BDRVQcowState *s = bs->opaque;
  96     int len, i, shift;
  97     QCowHeader header;
  98
  99     if (bdrv_pread(bs->file, 0, &header, sizeof(header)) != sizeof(header))
 100         goto fail;
 101     be32_to_cpus(&header.magic);
 102     be32_to_cpus(&header.version);
 103     be64_to_cpus(&header.backing_file_offset);
 104     be32_to_cpus(&header.backing_file_size);
 105     be32_to_cpus(&header.mtime);
 106     be64_to_cpus(&header.size);
 107     be32_to_cpus(&header.crypt_method);
 108     be64_to_cpus(&header.l1_table_offset);
 109
 110     if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
 111         goto fail;
 112     if (header.size <= 1 || header.cluster_bits < 9)
 113         goto fail;
 114     if (header.crypt_method > QCOW_CRYPT_AES)
 115         goto fail;
 116     s->crypt_method_header = header.crypt_method;
 117     if (s->crypt_method_header)
 118         bs->encrypted = 1;
 119     s->cluster_bits = header.cluster_bits;
 120     s->cluster_size = 1 << s->cluster_bits;
 121     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 122     s->l2_bits = header.l2_bits;
 123     s->l2_size = 1 << s->l2_bits;
 124     bs->total_sectors = header.size / 512;
 125     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 126
 127     /* read the level 1 table */
 128     shift = s->cluster_bits + s->l2_bits;
 129     s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
 130
 131     s->l1_table_offset = header.l1_table_offset;
 132     s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
 133     if (!s->l1_table)
 134         goto fail;
 135     if (bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
 136         s->l1_size * sizeof(uint64_t))
 137         goto fail;
 138     for(i = 0;i < s->l1_size; i++) {
 139         be64_to_cpus(&s->l1_table[i]);
 140     }
 141     /* alloc L2 cache */
 142     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 143     if (!s->l2_cache)
 144         goto fail;
 145     s->cluster_cache = qemu_malloc(s->cluster_size);
 146     if (!s->cluster_cache)
 147         goto fail;
 148     s->cluster_data = qemu_malloc(s->cluster_size);
 149     if (!s->cluster_data)
 150         goto fail;
 151     s->cluster_cache_offset = -1;
 152
 153     /* read the backing file name */
 154     if (header.backing_file_offset != 0) {
 155         len = header.backing_file_size;
 156         if (len > 1023)
 157             len = 1023;
 158         if (bdrv_pread(bs->file, header.backing_file_offset, bs->backing_file, len) != len)
 159             goto fail;
 160         bs->backing_file[len] = '\0';
 161     }
 162     return 0;
 163
 164  fail:
 165     qemu_free(s->l1_table);
 166     qemu_free(s->l2_cache);
 167     qemu_free(s->cluster_cache);
 168     qemu_free(s->cluster_data);
 169     return -1;
 170 }
 171
 172 static int qcow_set_key(BlockDriverState *bs, const char *key)
 173 {
 174     BDRVQcowState *s = bs->opaque;
 175     uint8_t keybuf[16];
 176     int len, i;
 177
 178     memset(keybuf, 0, 16);
 179     len = strlen(key);
 180     if (len > 16)
 181         len = 16;
 182     /* XXX: we could compress the chars to 7 bits to increase
 183        entropy */
 184     for(i = 0;i < len;i++) {
 185         keybuf[i] = key[i];
 186     }
 187     s->crypt_method = s->crypt_method_header;
 188
 189     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
 190         return -1;
 191     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
 192         return -1;
 193 #if 0
 194     /* test */
 195     {
 196         uint8_t in[16];
 197         uint8_t out[16];
 198         uint8_t tmp[16];
 199         for(i=0;i<16;i++)
 200             in[i] = i;
 201         AES_encrypt(in, tmp, &s->aes_encrypt_key);
 202         AES_decrypt(tmp, out, &s->aes_decrypt_key);
 203         for(i = 0; i < 16; i++)
 204             printf(" %02x", tmp[i]);
 205         printf("\n");
 206         for(i = 0; i < 16; i++)
 207             printf(" %02x", out[i]);
 208         printf("\n");
 209     }
 210 #endif
 211     return 0;
 212 }
 213
 214 /* The crypt function is compatible with the linux cryptoloop
 215    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
 216    supported */
 217 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 218                             uint8_t *out_buf, const uint8_t *in_buf,
 219                             int nb_sectors, int enc,
 220                             const AES_KEY *key)
 221 {
 222     union {
 223         uint64_t ll[2];
 224         uint8_t b[16];
 225     } ivec;
 226     int i;
 227
 228     for(i = 0; i < nb_sectors; i++) {
 229         ivec.ll[0] = cpu_to_le64(sector_num);
 230         ivec.ll[1] = 0;
 231         AES_cbc_encrypt(in_buf, out_buf, 512, key,
 232                         ivec.b, enc);
 233         sector_num++;
 234         in_buf += 512;
 235         out_buf += 512;
 236     }
 237 }
 238
 239 /* 'allocate' is:
 240  *
 241  * 0 to not allocate.
 242  *
 243  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 244  * 'n_end')
 245  *
 246  * 2 to allocate a compressed cluster of size
 247  * 'compressed_size'. 'compressed_size' must be > 0 and <
 248  * cluster_size
 249  *
 250  * return 0 if not allocated.
 251  */
 252 static uint64_t get_cluster_offset(BlockDriverState *bs,
 253                                    uint64_t offset, int allocate,
 254                                    int compressed_size,
 255                                    int n_start, int n_end)
 256 {
 257     BDRVQcowState *s = bs->opaque;
 258     int min_index, i, j, l1_index, l2_index;
 259     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
 260     uint32_t min_count;
 261     int new_l2_table;
 262
 263     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 264     l2_offset = s->l1_table[l1_index];
 265     new_l2_table = 0;
 266     if (!l2_offset) {
 267         if (!allocate)
 268             return 0;
 269         /* allocate a new l2 entry */
 270         l2_offset = bdrv_getlength(bs->file);
 271         /* round to cluster size */
 272         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
 273         /* update the L1 entry */
 274         s->l1_table[l1_index] = l2_offset;
 275         tmp = cpu_to_be64(l2_offset);
 276         if (bdrv_pwrite_sync(bs->file,
 277                 s->l1_table_offset + l1_index * sizeof(tmp),
 278                 &tmp, sizeof(tmp)) < 0)
 279             return 0;
 280         new_l2_table = 1;
 281     }
 282     for(i = 0; i < L2_CACHE_SIZE; i++) {
 283         if (l2_offset == s->l2_cache_offsets[i]) {
 284             /* increment the hit count */
 285             if (++s->l2_cache_counts[i] == 0xffffffff) {
 286                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 287                     s->l2_cache_counts[j] >>= 1;
 288                 }
 289             }
 290             l2_table = s->l2_cache + (i << s->l2_bits);
 291             goto found;
 292         }
 293     }
 294     /* not found: load a new entry in the least used one */
 295     min_index = 0;
 296     min_count = 0xffffffff;
 297     for(i = 0; i < L2_CACHE_SIZE; i++) {
 298         if (s->l2_cache_counts[i] < min_count) {
 299             min_count = s->l2_cache_counts[i];
 300             min_index = i;
 301         }
 302     }
 303     l2_table = s->l2_cache + (min_index << s->l2_bits);
 304     if (new_l2_table) {
 305         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 306         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
 307                 s->l2_size * sizeof(uint64_t)) < 0)
 308             return 0;
 309     } else {
 310         if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
 311             s->l2_size * sizeof(uint64_t))
 312             return 0;
 313     }
 314     s->l2_cache_offsets[min_index] = l2_offset;
 315     s->l2_cache_counts[min_index] = 1;
 316  found:
 317     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 318     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 319     if (!cluster_offset ||
 320         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 321         if (!allocate)
 322             return 0;
 323         /* allocate a new cluster */
 324         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 325             (n_end - n_start) < s->cluster_sectors) {
 326             /* if the cluster is already compressed, we must
 327                decompress it in the case it is not completely
 328                overwritten */
 329             if (decompress_cluster(bs, cluster_offset) < 0)
 330                 return 0;
 331             cluster_offset = bdrv_getlength(bs->file);
 332             cluster_offset = (cluster_offset + s->cluster_size - 1) &
 333                 ~(s->cluster_size - 1);
 334             /* write the cluster content */
 335             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
 336                 s->cluster_size)
 337                 return -1;
 338         } else {
 339             cluster_offset = bdrv_getlength(bs->file);
 340             if (allocate == 1) {
 341                 /* round to cluster size */
 342                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
 343                     ~(s->cluster_size - 1);
 344                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
 345                 /* if encrypted, we must initialize the cluster
 346                    content which won't be written */
 347                 if (s->crypt_method &&
 348                     (n_end - n_start) < s->cluster_sectors) {
 349                     uint64_t start_sect;
 350                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 351                     memset(s->cluster_data + 512, 0x00, 512);
 352                     for(i = 0; i < s->cluster_sectors; i++) {
 353                         if (i < n_start || i >= n_end) {
 354                             encrypt_sectors(s, start_sect + i,
 355                                             s->cluster_data,
 356                                             s->cluster_data + 512, 1, 1,
 357                                             &s->aes_encrypt_key);
 358                             if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
 359                                             s->cluster_data, 512) != 512)
 360                                 return -1;
 361                         }
 362                     }
 363                 }
 364             } else if (allocate == 2) {
 365                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 366                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 367             }
 368         }
 369         /* update L2 table */
 370         tmp = cpu_to_be64(cluster_offset);
 371         l2_table[l2_index] = tmp;
 372         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
 373                 &tmp, sizeof(tmp)) < 0)
 374             return 0;
 375     }
 376     return cluster_offset;
 377 }
 378
 379 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
 380                              int nb_sectors, int *pnum)
 381 {
 382     BDRVQcowState *s = bs->opaque;
 383     int index_in_cluster, n;
 384     uint64_t cluster_offset;
 385
 386     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 387     index_in_cluster = sector_num & (s->cluster_sectors - 1);
 388     n = s->cluster_sectors - index_in_cluster;
 389     if (n > nb_sectors)
 390         n = nb_sectors;
 391     *pnum = n;
 392     return (cluster_offset != 0);
 393 }
 394
 395 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 396                              const uint8_t *buf, int buf_size)
 397 {
 398     z_stream strm1, *strm = &strm1;
 399     int ret, out_len;
 400
 401     memset(strm, 0, sizeof(*strm));
 402
 403     strm->next_in = (uint8_t *)buf;
 404     strm->avail_in = buf_size;
 405     strm->next_out = out_buf;
 406     strm->avail_out = out_buf_size;
 407
 408     ret = inflateInit2(strm, -12);
 409     if (ret != Z_OK)
 410         return -1;
 411     ret = inflate(strm, Z_FINISH);
 412     out_len = strm->next_out - out_buf;
 413     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 414         out_len != out_buf_size) {
 415         inflateEnd(strm);
 416         return -1;
 417     }
 418     inflateEnd(strm);
 419     return 0;
 420 }
 421
 422 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 423 {
 424     BDRVQcowState *s = bs->opaque;
 425     int ret, csize;
 426     uint64_t coffset;
 427
 428     coffset = cluster_offset & s->cluster_offset_mask;
 429     if (s->cluster_cache_offset != coffset) {
 430         csize = cluster_offset >> (63 - s->cluster_bits);
 431         csize &= (s->cluster_size - 1);
 432         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
 433         if (ret != csize)
 434             return -1;
 435         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 436                               s->cluster_data, csize) < 0) {
 437             return -1;
 438         }
 439         s->cluster_cache_offset = coffset;
 440     }
 441     return 0;
 442 }
 443
 444 #if 0
 445
 446 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
 447                      uint8_t *buf, int nb_sectors)
 448 {
 449     BDRVQcowState *s = bs->opaque;
 450     int ret, index_in_cluster, n;
 451     uint64_t cluster_offset;
 452
 453     while (nb_sectors > 0) {
 454         cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 455         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 456         n = s->cluster_sectors - index_in_cluster;
 457         if (n > nb_sectors)
 458             n = nb_sectors;
 459         if (!cluster_offset) {
 460             if (bs->backing_hd) {
 461                 /* read from the base image */
 462                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
 463                 if (ret < 0)
 464                     return -1;
 465             } else {
 466                 memset(buf, 0, 512 * n);
 467             }
 468         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 469             if (decompress_cluster(bs, cluster_offset) < 0)
 470                 return -1;
 471             memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
 472         } else {
 473             ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512);
 474             if (ret != n * 512)
 475                 return -1;
 476             if (s->crypt_method) {
 477                 encrypt_sectors(s, sector_num, buf, buf, n, 0,
 478                                 &s->aes_decrypt_key);
 479             }
 480         }
 481         nb_sectors -= n;
 482         sector_num += n;
 483         buf += n * 512;
 484     }
 485     return 0;
 486 }
 487 #endif
 488
 489 typedef struct QCowAIOCB {
 490     BlockDriverAIOCB common;
 491     int64_t sector_num;
 492     QEMUIOVector *qiov;
 493     uint8_t *buf;
 494     void *orig_buf;
 495     int nb_sectors;
 496     int n;
 497     uint64_t cluster_offset;
 498     uint8_t *cluster_data;
 499     struct iovec hd_iov;
 500     QEMUIOVector hd_qiov;
 501     BlockDriverAIOCB *hd_aiocb;
 502 } QCowAIOCB;
 503
 504 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
 505 {
 506     QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common);
 507     if (acb->hd_aiocb)
 508         bdrv_aio_cancel(acb->hd_aiocb);
 509     qemu_aio_release(acb);
 510 }
 511
 512 static AIOPool qcow_aio_pool = {
 513     .aiocb_size         = sizeof(QCowAIOCB),
 514     .cancel             = qcow_aio_cancel,
 515 };
 516
 517 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
 518         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 519         BlockDriverCompletionFunc *cb, void *opaque, int is_write)
 520 {
 521     QCowAIOCB *acb;
 522
 523     acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
 524     if (!acb)
 525         return NULL;
 526     acb->hd_aiocb = NULL;
 527     acb->sector_num = sector_num;
 528     acb->qiov = qiov;
 529     if (qiov->niov > 1) {
 530         acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
 531         if (is_write)
 532             qemu_iovec_to_buffer(qiov, acb->buf);
 533     } else {
 534         acb->buf = (uint8_t *)qiov->iov->iov_base;
 535     }
 536     acb->nb_sectors = nb_sectors;
 537     acb->n = 0;
 538     acb->cluster_offset = 0;
 539     return acb;
 540 }
 541
 542 static void qcow_aio_read_cb(void *opaque, int ret)
 543 {
 544     QCowAIOCB *acb = opaque;
 545     BlockDriverState *bs = acb->common.bs;
 546     BDRVQcowState *s = bs->opaque;
 547     int index_in_cluster;
 548
 549     acb->hd_aiocb = NULL;
 550     if (ret < 0)
 551         goto done;
 552
 553  redo:
 554     /* post process the read buffer */
 555     if (!acb->cluster_offset) {
 556         /* nothing to do */
 557     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
 558         /* nothing to do */
 559     } else {
 560         if (s->crypt_method) {
 561             encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
 562                             acb->n, 0,
 563                             &s->aes_decrypt_key);
 564         }
 565     }
 566
 567     acb->nb_sectors -= acb->n;
 568     acb->sector_num += acb->n;
 569     acb->buf += acb->n * 512;
 570
 571     if (acb->nb_sectors == 0) {
 572         /* request completed */
 573         ret = 0;
 574         goto done;
 575     }
 576
 577     /* prepare next AIO request */
 578     acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9,
 579                                              0, 0, 0, 0);
 580     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
 581     acb->n = s->cluster_sectors - index_in_cluster;
 582     if (acb->n > acb->nb_sectors)
 583         acb->n = acb->nb_sectors;
 584
 585     if (!acb->cluster_offset) {
 586         if (bs->backing_hd) {
 587             /* read from the base image */
 588             acb->hd_iov.iov_base = (void *)acb->buf;
 589             acb->hd_iov.iov_len = acb->n * 512;
 590             qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 591             acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
 592                 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
 593             if (acb->hd_aiocb == NULL)
 594                 goto done;
 595         } else {
 596             /* Note: in this case, no need to wait */
 597             memset(acb->buf, 0, 512 * acb->n);
 598             goto redo;
 599         }
 600     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
 601         /* add AIO support for compressed blocks ? */
 602         if (decompress_cluster(bs, acb->cluster_offset) < 0)
 603             goto done;
 604         memcpy(acb->buf,
 605                s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
 606         goto redo;
 607     } else {
 608         if ((acb->cluster_offset & 511) != 0) {
 609             ret = -EIO;
 610             goto done;
 611         }
 612         acb->hd_iov.iov_base = (void *)acb->buf;
 613         acb->hd_iov.iov_len = acb->n * 512;
 614         qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 615         acb->hd_aiocb = bdrv_aio_readv(bs->file,
 616                             (acb->cluster_offset >> 9) + index_in_cluster,
 617                             &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
 618         if (acb->hd_aiocb == NULL)
 619             goto done;
 620     }
 621
 622     return;
 623
 624 done:
 625     if (acb->qiov->niov > 1) {
 626         qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
 627         qemu_vfree(acb->orig_buf);
 628     }
 629     acb->common.cb(acb->common.opaque, ret);
 630     qemu_aio_release(acb);
 631 }
 632
 633 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
 634         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 635         BlockDriverCompletionFunc *cb, void *opaque)
 636 {
 637     QCowAIOCB *acb;
 638
 639     acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
 640     if (!acb)
 641         return NULL;
 642
 643     qcow_aio_read_cb(acb, 0);
 644     return &acb->common;
 645 }
 646
 647 static void qcow_aio_write_cb(void *opaque, int ret)
 648 {
 649     QCowAIOCB *acb = opaque;
 650     BlockDriverState *bs = acb->common.bs;
 651     BDRVQcowState *s = bs->opaque;
 652     int index_in_cluster;
 653     uint64_t cluster_offset;
 654     const uint8_t *src_buf;
 655
 656     acb->hd_aiocb = NULL;
 657
 658     if (ret < 0)
 659         goto done;
 660
 661     acb->nb_sectors -= acb->n;
 662     acb->sector_num += acb->n;
 663     acb->buf += acb->n * 512;
 664
 665     if (acb->nb_sectors == 0) {
 666         /* request completed */
 667         ret = 0;
 668         goto done;
 669     }
 670
 671     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
 672     acb->n = s->cluster_sectors - index_in_cluster;
 673     if (acb->n > acb->nb_sectors)
 674         acb->n = acb->nb_sectors;
 675     cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, 1, 0,
 676                                         index_in_cluster,
 677                                         index_in_cluster + acb->n);
 678     if (!cluster_offset || (cluster_offset & 511) != 0) {
 679         ret = -EIO;
 680         goto done;
 681     }
 682     if (s->crypt_method) {
 683         if (!acb->cluster_data) {
 684             acb->cluster_data = qemu_mallocz(s->cluster_size);
 685             if (!acb->cluster_data) {
 686                 ret = -ENOMEM;
 687                 goto done;
 688             }
 689         }
 690         encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
 691                         acb->n, 1, &s->aes_encrypt_key);
 692         src_buf = acb->cluster_data;
 693     } else {
 694         src_buf = acb->buf;
 695     }
 696
 697     acb->hd_iov.iov_base = (void *)src_buf;
 698     acb->hd_iov.iov_len = acb->n * 512;
 699     qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 700     acb->hd_aiocb = bdrv_aio_writev(bs->file,
 701                                     (cluster_offset >> 9) + index_in_cluster,
 702                                     &acb->hd_qiov, acb->n,
 703                                     qcow_aio_write_cb, acb);
 704     if (acb->hd_aiocb == NULL)
 705         goto done;
 706     return;
 707
 708 done:
 709     if (acb->qiov->niov > 1)
 710         qemu_vfree(acb->orig_buf);
 711     acb->common.cb(acb->common.opaque, ret);
 712     qemu_aio_release(acb);
 713 }
 714
 715 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
 716         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 717         BlockDriverCompletionFunc *cb, void *opaque)
 718 {
 719     BDRVQcowState *s = bs->opaque;
 720     QCowAIOCB *acb;
 721
 722     s->cluster_cache_offset = -1; /* disable compressed cache */
 723
 724     acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
 725     if (!acb)
 726         return NULL;
 727
 728
 729     qcow_aio_write_cb(acb, 0);
 730     return &acb->common;
 731 }
 732
 733 static void qcow_close(BlockDriverState *bs)
 734 {
 735     BDRVQcowState *s = bs->opaque;
 736     qemu_free(s->l1_table);
 737     qemu_free(s->l2_cache);
 738     qemu_free(s->cluster_cache);
 739     qemu_free(s->cluster_data);
 740 }
 741
 742 static int qcow_create(const char *filename, QEMUOptionParameter *options)
 743 {
 744     int fd, header_size, backing_filename_len, l1_size, i, shift;
 745     QCowHeader header;
 746     uint64_t tmp;
 747     int64_t total_size = 0;
 748     const char *backing_file = NULL;
 749     int flags = 0;
 750     int ret;
 751
 752     /* Read out options */
 753     while (options && options->name) {
 754         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 755             total_size = options->value.n / 512;
 756         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 757             backing_file = options->value.s;
 758         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
 759             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
 760         }
 761         options++;
 762     }
 763
 764     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
 765     if (fd < 0)
 766         return -errno;
 767     memset(&header, 0, sizeof(header));
 768     header.magic = cpu_to_be32(QCOW_MAGIC);
 769     header.version = cpu_to_be32(QCOW_VERSION);
 770     header.size = cpu_to_be64(total_size * 512);
 771     header_size = sizeof(header);
 772     backing_filename_len = 0;
 773     if (backing_file) {
 774         if (strcmp(backing_file, "fat:")) {
 775             header.backing_file_offset = cpu_to_be64(header_size);
 776             backing_filename_len = strlen(backing_file);
 777             header.backing_file_size = cpu_to_be32(backing_filename_len);
 778             header_size += backing_filename_len;
 779         } else {
 780             /* special backing file for vvfat */
 781             backing_file = NULL;
 782         }
 783         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 784                                     unmodifyed sectors */
 785         header.l2_bits = 12; /* 32 KB L2 tables */
 786     } else {
 787         header.cluster_bits = 12; /* 4 KB clusters */
 788         header.l2_bits = 9; /* 4 KB L2 tables */
 789     }
 790     header_size = (header_size + 7) & ~7;
 791     shift = header.cluster_bits + header.l2_bits;
 792     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
 793
 794     header.l1_table_offset = cpu_to_be64(header_size);
 795     if (flags & BLOCK_FLAG_ENCRYPT) {
 796         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 797     } else {
 798         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 799     }
 800
 801     /* write all the data */
 802     ret = qemu_write_full(fd, &header, sizeof(header));
 803     if (ret != sizeof(header)) {
 804         ret = -errno;
 805         goto exit;
 806     }
 807
 808     if (backing_file) {
 809         ret = qemu_write_full(fd, backing_file, backing_filename_len);
 810         if (ret != backing_filename_len) {
 811             ret = -errno;
 812             goto exit;
 813         }
 814
 815     }
 816     lseek(fd, header_size, SEEK_SET);
 817     tmp = 0;
 818     for(i = 0;i < l1_size; i++) {
 819         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
 820         if (ret != sizeof(tmp)) {
 821             ret = -errno;
 822             goto exit;
 823         }
 824     }
 825
 826     ret = 0;
 827 exit:
 828     close(fd);
 829     return ret;
 830 }
 831
 832 static int qcow_make_empty(BlockDriverState *bs)
 833 {
 834     BDRVQcowState *s = bs->opaque;
 835     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 836     int ret;
 837
 838     memset(s->l1_table, 0, l1_length);
 839     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
 840             l1_length) < 0)
 841         return -1;
 842     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
 843     if (ret < 0)
 844         return ret;
 845
 846     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 847     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
 848     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 849
 850     return 0;
 851 }
 852
 853 /* XXX: put compressed sectors first, then all the cluster aligned
 854    tables to avoid losing bytes in alignment */
 855 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
 856                                  const uint8_t *buf, int nb_sectors)
 857 {
 858     BDRVQcowState *s = bs->opaque;
 859     z_stream strm;
 860     int ret, out_len;
 861     uint8_t *out_buf;
 862     uint64_t cluster_offset;
 863
 864     if (nb_sectors != s->cluster_sectors)
 865         return -EINVAL;
 866
 867     out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 868     if (!out_buf)
 869         return -1;
 870
 871     /* best compression, small window, no zlib header */
 872     memset(&strm, 0, sizeof(strm));
 873     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 874                        Z_DEFLATED, -12,
 875                        9, Z_DEFAULT_STRATEGY);
 876     if (ret != 0) {
 877         qemu_free(out_buf);
 878         return -1;
 879     }
 880
 881     strm.avail_in = s->cluster_size;
 882     strm.next_in = (uint8_t *)buf;
 883     strm.avail_out = s->cluster_size;
 884     strm.next_out = out_buf;
 885
 886     ret = deflate(&strm, Z_FINISH);
 887     if (ret != Z_STREAM_END && ret != Z_OK) {
 888         qemu_free(out_buf);
 889         deflateEnd(&strm);
 890         return -1;
 891     }
 892     out_len = strm.next_out - out_buf;
 893
 894     deflateEnd(&strm);
 895
 896     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 897         /* could not compress: write normal cluster */
 898         bdrv_write(bs, sector_num, buf, s->cluster_sectors);
 899     } else {
 900         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
 901                                             out_len, 0, 0);
 902         cluster_offset &= s->cluster_offset_mask;
 903         if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
 904             qemu_free(out_buf);
 905             return -1;
 906         }
 907     }
 908
 909     qemu_free(out_buf);
 910     return 0;
 911 }
 912
 913 static void qcow_flush(BlockDriverState *bs)
 914 {
 915     bdrv_flush(bs->file);
 916 }
 917
 918 static BlockDriverAIOCB *qcow_aio_flush(BlockDriverState *bs,
 919         BlockDriverCompletionFunc *cb, void *opaque)
 920 {
 921     return bdrv_aio_flush(bs->file, cb, opaque);
 922 }
 923
 924 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 925 {
 926     BDRVQcowState *s = bs->opaque;
 927     bdi->cluster_size = s->cluster_size;
 928     return 0;
 929 }
 930
 931
 932 static QEMUOptionParameter qcow_create_options[] = {
 933     {
 934         .name = BLOCK_OPT_SIZE,
 935         .type = OPT_SIZE,
 936         .help = "Virtual disk size"
 937     },
 938     {
 939         .name = BLOCK_OPT_BACKING_FILE,
 940         .type = OPT_STRING,
 941         .help = "File name of a base image"
 942     },
 943     {
 944         .name = BLOCK_OPT_ENCRYPT,
 945         .type = OPT_FLAG,
 946         .help = "Encrypt the image"
 947     },
 948     { NULL }
 949 };
 950
 951 static BlockDriver bdrv_qcow = {
 952     .format_name        = "qcow",
 953     .instance_size      = sizeof(BDRVQcowState),
 954     .bdrv_probe         = qcow_probe,
 955     .bdrv_open          = qcow_open,
 956     .bdrv_close         = qcow_close,
 957     .bdrv_create        = qcow_create,
 958     .bdrv_flush         = qcow_flush,
 959     .bdrv_is_allocated  = qcow_is_allocated,
 960     .bdrv_set_key       = qcow_set_key,
 961     .bdrv_make_empty    = qcow_make_empty,
 962     .bdrv_aio_readv     = qcow_aio_readv,
 963     .bdrv_aio_writev    = qcow_aio_writev,
 964     .bdrv_aio_flush     = qcow_aio_flush,
 965     .bdrv_write_compressed = qcow_write_compressed,
 966     .bdrv_get_info      = qcow_get_info,
 967
 968     .create_options = qcow_create_options,
 969 };
 970
 971 static void bdrv_qcow_init(void)
 972 {
 973     bdrv_register(&bdrv_qcow);
 974 }
 975
 976 block_init(bdrv_qcow_init);