block/qcow.c

   1 /*
   2  * Block driver for the QCOW format
   3  *
   4  * Copyright (c) 2004-2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "qemu/osdep.h"
  26 #include "qapi/error.h"
  27 #include "qemu/error-report.h"
  28 #include "block/block_int.h"
  29 #include "block/qdict.h"
  30 #include "sysemu/block-backend.h"
  31 #include "qemu/module.h"
  32 #include "qemu/option.h"
  33 #include "qemu/bswap.h"
  34 #include "qemu/cutils.h"
  35 #include "qemu/memalign.h"
  36 #include <zlib.h>
  37 #include "qapi/qmp/qdict.h"
  38 #include "qapi/qmp/qstring.h"
  39 #include "qapi/qobject-input-visitor.h"
  40 #include "qapi/qapi-visit-block-core.h"
  41 #include "crypto/block.h"
  42 #include "migration/blocker.h"
  43 #include "crypto.h"
  44
  45 /**************************************************************/
  46 /* QEMU COW block driver with compression and encryption support */
  47
  48 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  49 #define QCOW_VERSION 1
  50
  51 #define QCOW_CRYPT_NONE 0
  52 #define QCOW_CRYPT_AES  1
  53
  54 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
  55
  56 typedef struct QCowHeader {
  57     uint32_t magic;
  58     uint32_t version;
  59     uint64_t backing_file_offset;
  60     uint32_t backing_file_size;
  61     uint32_t mtime;
  62     uint64_t size; /* in bytes */
  63     uint8_t cluster_bits;
  64     uint8_t l2_bits;
  65     uint16_t padding;
  66     uint32_t crypt_method;
  67     uint64_t l1_table_offset;
  68 } QEMU_PACKED QCowHeader;
  69
  70 #define L2_CACHE_SIZE 16
  71
  72 typedef struct BDRVQcowState {
  73     int cluster_bits;
  74     int cluster_size;
  75     int l2_bits;
  76     int l2_size;
  77     unsigned int l1_size;
  78     uint64_t cluster_offset_mask;
  79     uint64_t l1_table_offset;
  80     uint64_t *l1_table;
  81     uint64_t *l2_cache;
  82     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  83     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  84     uint8_t *cluster_cache;
  85     uint8_t *cluster_data;
  86     uint64_t cluster_cache_offset;
  87     QCryptoBlock *crypto; /* Disk encryption format driver */
  88     uint32_t crypt_method_header;
  89     CoMutex lock;
  90     Error *migration_blocker;
  91 } BDRVQcowState;
  92
  93 static QemuOptsList qcow_create_opts;
  94
  95 static int coroutine_fn GRAPH_RDLOCK
  96 decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
  97
  98 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  99 {
 100     const QCowHeader *cow_header = (const void *)buf;
 101
 102     if (buf_size >= sizeof(QCowHeader) &&
 103         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
 104         be32_to_cpu(cow_header->version) == QCOW_VERSION)
 105         return 100;
 106     else
 107         return 0;
 108 }
 109
 110 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
 111                      Error **errp)
 112 {
 113     BDRVQcowState *s = bs->opaque;
 114     unsigned int len, i, shift;
 115     int ret;
 116     QCowHeader header;
 117     QCryptoBlockOpenOptions *crypto_opts = NULL;
 118     unsigned int cflags = 0;
 119     QDict *encryptopts = NULL;
 120     const char *encryptfmt;
 121
 122     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
 123     encryptfmt = qdict_get_try_str(encryptopts, "format");
 124
 125     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
 126     if (ret < 0) {
 127         goto fail_unlocked;
 128     }
 129
 130     bdrv_graph_rdlock_main_loop();
 131
 132     ret = bdrv_pread(bs->file, 0, sizeof(header), &header, 0);
 133     if (ret < 0) {
 134         goto fail;
 135     }
 136     header.magic = be32_to_cpu(header.magic);
 137     header.version = be32_to_cpu(header.version);
 138     header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
 139     header.backing_file_size = be32_to_cpu(header.backing_file_size);
 140     header.mtime = be32_to_cpu(header.mtime);
 141     header.size = be64_to_cpu(header.size);
 142     header.crypt_method = be32_to_cpu(header.crypt_method);
 143     header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
 144
 145     if (header.magic != QCOW_MAGIC) {
 146         error_setg(errp, "Image not in qcow format");
 147         ret = -EINVAL;
 148         goto fail;
 149     }
 150     if (header.version != QCOW_VERSION) {
 151         error_setg(errp, "qcow (v%d) does not support qcow version %" PRIu32,
 152                    QCOW_VERSION, header.version);
 153         if (header.version == 2 || header.version == 3) {
 154             error_append_hint(errp, "Try the 'qcow2' driver instead.\n");
 155         }
 156
 157         ret = -ENOTSUP;
 158         goto fail;
 159     }
 160
 161     if (header.size <= 1) {
 162         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
 163         ret = -EINVAL;
 164         goto fail;
 165     }
 166     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
 167         error_setg(errp, "Cluster size must be between 512 and 64k");
 168         ret = -EINVAL;
 169         goto fail;
 170     }
 171
 172     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
 173      * so bytes = num_entries << 3. */
 174     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
 175         error_setg(errp, "L2 table size must be between 512 and 64k");
 176         ret = -EINVAL;
 177         goto fail;
 178     }
 179
 180     s->crypt_method_header = header.crypt_method;
 181     if (s->crypt_method_header) {
 182         if (bdrv_uses_whitelist() &&
 183             s->crypt_method_header == QCOW_CRYPT_AES) {
 184             error_setg(errp,
 185                        "Use of AES-CBC encrypted qcow images is no longer "
 186                        "supported in system emulators");
 187             error_append_hint(errp,
 188                               "You can use 'qemu-img convert' to convert your "
 189                               "image to an alternative supported format, such "
 190                               "as unencrypted qcow, or raw with the LUKS "
 191                               "format instead.\n");
 192             ret = -ENOSYS;
 193             goto fail;
 194         }
 195         if (s->crypt_method_header == QCOW_CRYPT_AES) {
 196             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
 197                 error_setg(errp,
 198                            "Header reported 'aes' encryption format but "
 199                            "options specify '%s'", encryptfmt);
 200                 ret = -EINVAL;
 201                 goto fail;
 202             }
 203             qdict_put_str(encryptopts, "format", "qcow");
 204             crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
 205             if (!crypto_opts) {
 206                 ret = -EINVAL;
 207                 goto fail;
 208             }
 209
 210             if (flags & BDRV_O_NO_IO) {
 211                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
 212             }
 213             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
 214                                            NULL, NULL, cflags, 1, errp);
 215             if (!s->crypto) {
 216                 ret = -EINVAL;
 217                 goto fail;
 218             }
 219         } else {
 220             error_setg(errp, "invalid encryption method in qcow header");
 221             ret = -EINVAL;
 222             goto fail;
 223         }
 224         bs->encrypted = true;
 225     } else {
 226         if (encryptfmt) {
 227             error_setg(errp, "No encryption in image header, but options "
 228                        "specified format '%s'", encryptfmt);
 229             ret = -EINVAL;
 230             goto fail;
 231         }
 232     }
 233     s->cluster_bits = header.cluster_bits;
 234     s->cluster_size = 1 << s->cluster_bits;
 235     s->l2_bits = header.l2_bits;
 236     s->l2_size = 1 << s->l2_bits;
 237     bs->total_sectors = header.size / 512;
 238     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 239
 240     /* read the level 1 table */
 241     shift = s->cluster_bits + s->l2_bits;
 242     if (header.size > UINT64_MAX - (1LL << shift)) {
 243         error_setg(errp, "Image too large");
 244         ret = -EINVAL;
 245         goto fail;
 246     } else {
 247         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
 248         if (l1_size > INT_MAX / sizeof(uint64_t)) {
 249             error_setg(errp, "Image too large");
 250             ret = -EINVAL;
 251             goto fail;
 252         }
 253         s->l1_size = l1_size;
 254     }
 255
 256     s->l1_table_offset = header.l1_table_offset;
 257     s->l1_table = g_try_new(uint64_t, s->l1_size);
 258     if (s->l1_table == NULL) {
 259         error_setg(errp, "Could not allocate memory for L1 table");
 260         ret = -ENOMEM;
 261         goto fail;
 262     }
 263
 264     ret = bdrv_pread(bs->file, s->l1_table_offset,
 265                      s->l1_size * sizeof(uint64_t), s->l1_table, 0);
 266     if (ret < 0) {
 267         goto fail;
 268     }
 269
 270     for(i = 0;i < s->l1_size; i++) {
 271         s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
 272     }
 273
 274     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
 275     s->l2_cache =
 276         qemu_try_blockalign(bs->file->bs,
 277                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 278     if (s->l2_cache == NULL) {
 279         error_setg(errp, "Could not allocate L2 table cache");
 280         ret = -ENOMEM;
 281         goto fail;
 282     }
 283     s->cluster_cache = g_malloc(s->cluster_size);
 284     s->cluster_data = g_malloc(s->cluster_size);
 285     s->cluster_cache_offset = -1;
 286
 287     /* read the backing file name */
 288     if (header.backing_file_offset != 0) {
 289         len = header.backing_file_size;
 290         if (len > 1023 || len >= sizeof(bs->backing_file)) {
 291             error_setg(errp, "Backing file name too long");
 292             ret = -EINVAL;
 293             goto fail;
 294         }
 295         ret = bdrv_pread(bs->file, header.backing_file_offset, len,
 296                          bs->auto_backing_file, 0);
 297         if (ret < 0) {
 298             goto fail;
 299         }
 300         bs->auto_backing_file[len] = '\0';
 301         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
 302                 bs->auto_backing_file);
 303     }
 304
 305     /* Disable migration when qcow images are used */
 306     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
 307                "does not support live migration",
 308                bdrv_get_device_or_node_name(bs));
 309
 310     ret = migrate_add_blocker_normal(&s->migration_blocker, errp);
 311     if (ret < 0) {
 312         goto fail;
 313     }
 314
 315     qobject_unref(encryptopts);
 316     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
 317     qemu_co_mutex_init(&s->lock);
 318     bdrv_graph_rdunlock_main_loop();
 319     return 0;
 320
 321 fail:
 322     bdrv_graph_rdunlock_main_loop();
 323 fail_unlocked:
 324     g_free(s->l1_table);
 325     qemu_vfree(s->l2_cache);
 326     g_free(s->cluster_cache);
 327     g_free(s->cluster_data);
 328     qcrypto_block_free(s->crypto);
 329     qobject_unref(encryptopts);
 330     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
 331     return ret;
 332 }
 333
 334
 335 /* We have nothing to do for QCOW reopen, stubs just return
 336  * success */
 337 static int qcow_reopen_prepare(BDRVReopenState *state,
 338                                BlockReopenQueue *queue, Error **errp)
 339 {
 340     return 0;
 341 }
 342
 343
 344 /* 'allocate' is:
 345  *
 346  * 0 to not allocate.
 347  *
 348  * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
 349  * to 'n_end' within the cluster)
 350  *
 351  * 2 to allocate a compressed cluster of size
 352  * 'compressed_size'. 'compressed_size' must be > 0 and <
 353  * cluster_size
 354  *
 355  * return 0 if not allocated, 1 if *result is assigned, and negative
 356  * errno on failure.
 357  */
 358 static int coroutine_fn GRAPH_RDLOCK
 359 get_cluster_offset(BlockDriverState *bs, uint64_t offset, int allocate,
 360                    int compressed_size, int n_start, int n_end,
 361                    uint64_t *result)
 362 {
 363     BDRVQcowState *s = bs->opaque;
 364     int min_index, i, j, l1_index, l2_index, ret;
 365     int64_t l2_offset;
 366     uint64_t *l2_table, cluster_offset, tmp;
 367     uint32_t min_count;
 368     int new_l2_table;
 369
 370     *result = 0;
 371     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 372     l2_offset = s->l1_table[l1_index];
 373     new_l2_table = 0;
 374     if (!l2_offset) {
 375         if (!allocate)
 376             return 0;
 377         /* allocate a new l2 entry */
 378         l2_offset = bdrv_co_getlength(bs->file->bs);
 379         if (l2_offset < 0) {
 380             return l2_offset;
 381         }
 382         /* round to cluster size */
 383         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
 384         /* update the L1 entry */
 385         s->l1_table[l1_index] = l2_offset;
 386         tmp = cpu_to_be64(l2_offset);
 387         BLKDBG_CO_EVENT(bs->file, BLKDBG_L1_UPDATE);
 388         ret = bdrv_co_pwrite_sync(bs->file,
 389                                   s->l1_table_offset + l1_index * sizeof(tmp),
 390                                   sizeof(tmp), &tmp, 0);
 391         if (ret < 0) {
 392             return ret;
 393         }
 394         new_l2_table = 1;
 395     }
 396     for(i = 0; i < L2_CACHE_SIZE; i++) {
 397         if (l2_offset == s->l2_cache_offsets[i]) {
 398             /* increment the hit count */
 399             if (++s->l2_cache_counts[i] == 0xffffffff) {
 400                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 401                     s->l2_cache_counts[j] >>= 1;
 402                 }
 403             }
 404             l2_table = s->l2_cache + (i << s->l2_bits);
 405             goto found;
 406         }
 407     }
 408     /* not found: load a new entry in the least used one */
 409     min_index = 0;
 410     min_count = 0xffffffff;
 411     for(i = 0; i < L2_CACHE_SIZE; i++) {
 412         if (s->l2_cache_counts[i] < min_count) {
 413             min_count = s->l2_cache_counts[i];
 414             min_index = i;
 415         }
 416     }
 417     l2_table = s->l2_cache + (min_index << s->l2_bits);
 418     BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_LOAD);
 419     if (new_l2_table) {
 420         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 421         ret = bdrv_co_pwrite_sync(bs->file, l2_offset,
 422                                   s->l2_size * sizeof(uint64_t), l2_table, 0);
 423         if (ret < 0) {
 424             return ret;
 425         }
 426     } else {
 427         ret = bdrv_co_pread(bs->file, l2_offset,
 428                             s->l2_size * sizeof(uint64_t), l2_table, 0);
 429         if (ret < 0) {
 430             return ret;
 431         }
 432     }
 433     s->l2_cache_offsets[min_index] = l2_offset;
 434     s->l2_cache_counts[min_index] = 1;
 435  found:
 436     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 437     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 438     if (!cluster_offset ||
 439         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 440         if (!allocate)
 441             return 0;
 442         BLKDBG_CO_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
 443         assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
 444         /* allocate a new cluster */
 445         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 446             (n_end - n_start) < s->cluster_size) {
 447             /* if the cluster is already compressed, we must
 448                decompress it in the case it is not completely
 449                overwritten */
 450             if (decompress_cluster(bs, cluster_offset) < 0) {
 451                 return -EIO;
 452             }
 453             cluster_offset = bdrv_co_getlength(bs->file->bs);
 454             if ((int64_t) cluster_offset < 0) {
 455                 return cluster_offset;
 456             }
 457             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
 458             /* write the cluster content */
 459             BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
 460             ret = bdrv_co_pwrite(bs->file, cluster_offset, s->cluster_size,
 461                                  s->cluster_cache, 0);
 462             if (ret < 0) {
 463                 return ret;
 464             }
 465         } else {
 466             cluster_offset = bdrv_co_getlength(bs->file->bs);
 467             if ((int64_t) cluster_offset < 0) {
 468                 return cluster_offset;
 469             }
 470             if (allocate == 1) {
 471                 /* round to cluster size */
 472                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
 473                 if (cluster_offset + s->cluster_size > INT64_MAX) {
 474                     return -E2BIG;
 475                 }
 476                 ret = bdrv_co_truncate(bs->file,
 477                                        cluster_offset + s->cluster_size,
 478                                        false, PREALLOC_MODE_OFF, 0, NULL);
 479                 if (ret < 0) {
 480                     return ret;
 481                 }
 482                 /* if encrypted, we must initialize the cluster
 483                    content which won't be written */
 484                 if (bs->encrypted &&
 485                     (n_end - n_start) < s->cluster_size) {
 486                     uint64_t start_offset;
 487                     assert(s->crypto);
 488                     start_offset = offset & ~(s->cluster_size - 1);
 489                     for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
 490                         if (i < n_start || i >= n_end) {
 491                             memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
 492                             if (qcrypto_block_encrypt(s->crypto,
 493                                                       start_offset + i,
 494                                                       s->cluster_data,
 495                                                       BDRV_SECTOR_SIZE,
 496                                                       NULL) < 0) {
 497                                 return -EIO;
 498                             }
 499                             BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
 500                             ret = bdrv_co_pwrite(bs->file, cluster_offset + i,
 501                                                  BDRV_SECTOR_SIZE,
 502                                                  s->cluster_data, 0);
 503                             if (ret < 0) {
 504                                 return ret;
 505                             }
 506                         }
 507                     }
 508                 }
 509             } else if (allocate == 2) {
 510                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 511                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 512             }
 513         }
 514         /* update L2 table */
 515         tmp = cpu_to_be64(cluster_offset);
 516         l2_table[l2_index] = tmp;
 517         if (allocate == 2) {
 518             BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
 519         } else {
 520             BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE);
 521         }
 522         ret = bdrv_co_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
 523                                   sizeof(tmp), &tmp, 0);
 524         if (ret < 0) {
 525             return ret;
 526         }
 527     }
 528     *result = cluster_offset;
 529     return 1;
 530 }
 531
 532 static int coroutine_fn GRAPH_RDLOCK
 533 qcow_co_block_status(BlockDriverState *bs, bool want_zero,
 534                      int64_t offset, int64_t bytes, int64_t *pnum,
 535                      int64_t *map, BlockDriverState **file)
 536 {
 537     BDRVQcowState *s = bs->opaque;
 538     int index_in_cluster, ret;
 539     int64_t n;
 540     uint64_t cluster_offset;
 541
 542     qemu_co_mutex_lock(&s->lock);
 543     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
 544     qemu_co_mutex_unlock(&s->lock);
 545     if (ret < 0) {
 546         return ret;
 547     }
 548     index_in_cluster = offset & (s->cluster_size - 1);
 549     n = s->cluster_size - index_in_cluster;
 550     if (n > bytes) {
 551         n = bytes;
 552     }
 553     *pnum = n;
 554     if (!cluster_offset) {
 555         return 0;
 556     }
 557     if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 558         return BDRV_BLOCK_DATA | BDRV_BLOCK_COMPRESSED;
 559     }
 560     if (s->crypto) {
 561         return BDRV_BLOCK_DATA;
 562     }
 563     *map = cluster_offset | index_in_cluster;
 564     *file = bs->file->bs;
 565     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 566 }
 567
 568 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 569                              const uint8_t *buf, int buf_size)
 570 {
 571     z_stream strm1, *strm = &strm1;
 572     int ret, out_len;
 573
 574     memset(strm, 0, sizeof(*strm));
 575
 576     strm->next_in = (uint8_t *)buf;
 577     strm->avail_in = buf_size;
 578     strm->next_out = out_buf;
 579     strm->avail_out = out_buf_size;
 580
 581     ret = inflateInit2(strm, -12);
 582     if (ret != Z_OK)
 583         return -1;
 584     ret = inflate(strm, Z_FINISH);
 585     out_len = strm->next_out - out_buf;
 586     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 587         out_len != out_buf_size) {
 588         inflateEnd(strm);
 589         return -1;
 590     }
 591     inflateEnd(strm);
 592     return 0;
 593 }
 594
 595 static int coroutine_fn GRAPH_RDLOCK
 596 decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 597 {
 598     BDRVQcowState *s = bs->opaque;
 599     int ret, csize;
 600     uint64_t coffset;
 601
 602     coffset = cluster_offset & s->cluster_offset_mask;
 603     if (s->cluster_cache_offset != coffset) {
 604         csize = cluster_offset >> (63 - s->cluster_bits);
 605         csize &= (s->cluster_size - 1);
 606         BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
 607         ret = bdrv_co_pread(bs->file, coffset, csize, s->cluster_data, 0);
 608         if (ret < 0)
 609             return -1;
 610         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 611                               s->cluster_data, csize) < 0) {
 612             return -1;
 613         }
 614         s->cluster_cache_offset = coffset;
 615     }
 616     return 0;
 617 }
 618
 619 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
 620 {
 621     /* At least encrypted images require 512-byte alignment. Apply the
 622      * limit universally, rather than just on encrypted images, as
 623      * it's easier to let the block layer handle rounding than to
 624      * audit this code further. */
 625     bs->bl.request_alignment = BDRV_SECTOR_SIZE;
 626 }
 627
 628 static int coroutine_fn GRAPH_RDLOCK
 629 qcow_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
 630                QEMUIOVector *qiov, BdrvRequestFlags flags)
 631 {
 632     BDRVQcowState *s = bs->opaque;
 633     int offset_in_cluster;
 634     int ret = 0, n;
 635     uint64_t cluster_offset;
 636     uint8_t *buf;
 637     void *orig_buf;
 638
 639     if (qiov->niov > 1) {
 640         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
 641         if (buf == NULL) {
 642             return -ENOMEM;
 643         }
 644     } else {
 645         orig_buf = NULL;
 646         buf = (uint8_t *)qiov->iov->iov_base;
 647     }
 648
 649     qemu_co_mutex_lock(&s->lock);
 650
 651     while (bytes != 0) {
 652         /* prepare next request */
 653         ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
 654         if (ret < 0) {
 655             break;
 656         }
 657         offset_in_cluster = offset & (s->cluster_size - 1);
 658         n = s->cluster_size - offset_in_cluster;
 659         if (n > bytes) {
 660             n = bytes;
 661         }
 662
 663         if (!cluster_offset) {
 664             if (bs->backing) {
 665                 /* read from the base image */
 666                 qemu_co_mutex_unlock(&s->lock);
 667                 /* qcow2 emits this on bs->file instead of bs->backing */
 668                 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
 669                 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0);
 670                 qemu_co_mutex_lock(&s->lock);
 671                 if (ret < 0) {
 672                     break;
 673                 }
 674             } else {
 675                 /* Note: in this case, no need to wait */
 676                 memset(buf, 0, n);
 677             }
 678         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 679             /* add AIO support for compressed blocks ? */
 680             if (decompress_cluster(bs, cluster_offset) < 0) {
 681                 ret = -EIO;
 682                 break;
 683             }
 684             memcpy(buf, s->cluster_cache + offset_in_cluster, n);
 685         } else {
 686             if ((cluster_offset & 511) != 0) {
 687                 ret = -EIO;
 688                 break;
 689             }
 690             qemu_co_mutex_unlock(&s->lock);
 691             BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_AIO);
 692             ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster,
 693                                 n, buf, 0);
 694             qemu_co_mutex_lock(&s->lock);
 695             if (ret < 0) {
 696                 break;
 697             }
 698             if (bs->encrypted) {
 699                 assert(s->crypto);
 700                 if (qcrypto_block_decrypt(s->crypto,
 701                                           offset, buf, n, NULL) < 0) {
 702                     ret = -EIO;
 703                     break;
 704                 }
 705             }
 706         }
 707         ret = 0;
 708
 709         bytes -= n;
 710         offset += n;
 711         buf += n;
 712     }
 713
 714     qemu_co_mutex_unlock(&s->lock);
 715
 716     if (qiov->niov > 1) {
 717         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
 718         qemu_vfree(orig_buf);
 719     }
 720
 721     return ret;
 722 }
 723
 724 static int coroutine_fn GRAPH_RDLOCK
 725 qcow_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
 726                 QEMUIOVector *qiov, BdrvRequestFlags flags)
 727 {
 728     BDRVQcowState *s = bs->opaque;
 729     int offset_in_cluster;
 730     uint64_t cluster_offset;
 731     int ret = 0, n;
 732     uint8_t *buf;
 733     void *orig_buf;
 734
 735     s->cluster_cache_offset = -1; /* disable compressed cache */
 736
 737     /* We must always copy the iov when encrypting, so we
 738      * don't modify the original data buffer during encryption */
 739     if (bs->encrypted || qiov->niov > 1) {
 740         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
 741         if (buf == NULL) {
 742             return -ENOMEM;
 743         }
 744         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
 745     } else {
 746         orig_buf = NULL;
 747         buf = (uint8_t *)qiov->iov->iov_base;
 748     }
 749
 750     qemu_co_mutex_lock(&s->lock);
 751
 752     while (bytes != 0) {
 753         offset_in_cluster = offset & (s->cluster_size - 1);
 754         n = s->cluster_size - offset_in_cluster;
 755         if (n > bytes) {
 756             n = bytes;
 757         }
 758         ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
 759                                  offset_in_cluster + n, &cluster_offset);
 760         if (ret < 0) {
 761             break;
 762         }
 763         if (!cluster_offset || (cluster_offset & 511) != 0) {
 764             ret = -EIO;
 765             break;
 766         }
 767         if (bs->encrypted) {
 768             assert(s->crypto);
 769             if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
 770                 ret = -EIO;
 771                 break;
 772             }
 773         }
 774
 775         qemu_co_mutex_unlock(&s->lock);
 776         BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
 777         ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster,
 778                              n, buf, 0);
 779         qemu_co_mutex_lock(&s->lock);
 780         if (ret < 0) {
 781             break;
 782         }
 783         ret = 0;
 784
 785         bytes -= n;
 786         offset += n;
 787         buf += n;
 788     }
 789     qemu_co_mutex_unlock(&s->lock);
 790
 791     qemu_vfree(orig_buf);
 792
 793     return ret;
 794 }
 795
 796 static void qcow_close(BlockDriverState *bs)
 797 {
 798     BDRVQcowState *s = bs->opaque;
 799
 800     qcrypto_block_free(s->crypto);
 801     s->crypto = NULL;
 802     g_free(s->l1_table);
 803     qemu_vfree(s->l2_cache);
 804     g_free(s->cluster_cache);
 805     g_free(s->cluster_data);
 806
 807     migrate_del_blocker(&s->migration_blocker);
 808 }
 809
 810 static int coroutine_fn GRAPH_UNLOCKED
 811 qcow_co_create(BlockdevCreateOptions *opts, Error **errp)
 812 {
 813     BlockdevCreateOptionsQcow *qcow_opts;
 814     int header_size, backing_filename_len, l1_size, shift, i;
 815     QCowHeader header;
 816     uint8_t *tmp;
 817     int64_t total_size = 0;
 818     int ret;
 819     BlockDriverState *bs;
 820     BlockBackend *qcow_blk;
 821     QCryptoBlock *crypto = NULL;
 822
 823     assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
 824     qcow_opts = &opts->u.qcow;
 825
 826     /* Sanity checks */
 827     total_size = qcow_opts->size;
 828     if (total_size == 0) {
 829         error_setg(errp, "Image size is too small, cannot be zero length");
 830         return -EINVAL;
 831     }
 832
 833     if (qcow_opts->encrypt &&
 834         qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
 835     {
 836         error_setg(errp, "Unsupported encryption format");
 837         return -EINVAL;
 838     }
 839
 840     /* Create BlockBackend to write to the image */
 841     bs = bdrv_co_open_blockdev_ref(qcow_opts->file, errp);
 842     if (bs == NULL) {
 843         return -EIO;
 844     }
 845
 846     qcow_blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE,
 847                                   BLK_PERM_ALL, errp);
 848     if (!qcow_blk) {
 849         ret = -EPERM;
 850         goto exit;
 851     }
 852     blk_set_allow_write_beyond_eof(qcow_blk, true);
 853
 854     /* Create image format */
 855     memset(&header, 0, sizeof(header));
 856     header.magic = cpu_to_be32(QCOW_MAGIC);
 857     header.version = cpu_to_be32(QCOW_VERSION);
 858     header.size = cpu_to_be64(total_size);
 859     header_size = sizeof(header);
 860     backing_filename_len = 0;
 861     if (qcow_opts->backing_file) {
 862         if (strcmp(qcow_opts->backing_file, "fat:")) {
 863             header.backing_file_offset = cpu_to_be64(header_size);
 864             backing_filename_len = strlen(qcow_opts->backing_file);
 865             header.backing_file_size = cpu_to_be32(backing_filename_len);
 866             header_size += backing_filename_len;
 867         } else {
 868             /* special backing file for vvfat */
 869             qcow_opts->backing_file = NULL;
 870         }
 871         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 872                                     unmodified sectors */
 873         header.l2_bits = 12; /* 32 KB L2 tables */
 874     } else {
 875         header.cluster_bits = 12; /* 4 KB clusters */
 876         header.l2_bits = 9; /* 4 KB L2 tables */
 877     }
 878     header_size = (header_size + 7) & ~7;
 879     shift = header.cluster_bits + header.l2_bits;
 880     l1_size = (total_size + (1LL << shift) - 1) >> shift;
 881
 882     header.l1_table_offset = cpu_to_be64(header_size);
 883
 884     if (qcow_opts->encrypt) {
 885         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 886
 887         crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
 888                                       NULL, NULL, NULL, 0, errp);
 889         if (!crypto) {
 890             ret = -EINVAL;
 891             goto exit;
 892         }
 893     } else {
 894         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 895     }
 896
 897     /* write all the data */
 898     ret = blk_co_pwrite(qcow_blk, 0, sizeof(header), &header, 0);
 899     if (ret < 0) {
 900         goto exit;
 901     }
 902
 903     if (qcow_opts->backing_file) {
 904         ret = blk_co_pwrite(qcow_blk, sizeof(header), backing_filename_len,
 905                             qcow_opts->backing_file, 0);
 906         if (ret < 0) {
 907             goto exit;
 908         }
 909     }
 910
 911     tmp = g_malloc0(BDRV_SECTOR_SIZE);
 912     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
 913          i++) {
 914         ret = blk_co_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
 915                             BDRV_SECTOR_SIZE, tmp, 0);
 916         if (ret < 0) {
 917             g_free(tmp);
 918             goto exit;
 919         }
 920     }
 921
 922     g_free(tmp);
 923     ret = 0;
 924 exit:
 925     blk_co_unref(qcow_blk);
 926     bdrv_co_unref(bs);
 927     qcrypto_block_free(crypto);
 928     return ret;
 929 }
 930
 931 static int coroutine_fn GRAPH_UNLOCKED
 932 qcow_co_create_opts(BlockDriver *drv, const char *filename,
 933                     QemuOpts *opts, Error **errp)
 934 {
 935     BlockdevCreateOptions *create_options = NULL;
 936     BlockDriverState *bs = NULL;
 937     QDict *qdict = NULL;
 938     Visitor *v;
 939     const char *val;
 940     int ret;
 941     char *backing_fmt;
 942
 943     static const QDictRenames opt_renames[] = {
 944         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
 945         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
 946         { NULL, NULL },
 947     };
 948
 949     /*
 950      * We can't actually store a backing format, but can check that
 951      * the user's request made sense.
 952      */
 953     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
 954     if (backing_fmt && !bdrv_find_format(backing_fmt)) {
 955         error_setg(errp, "unrecognized backing format '%s'", backing_fmt);
 956         ret = -EINVAL;
 957         goto fail;
 958     }
 959
 960     /* Parse options and convert legacy syntax */
 961     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
 962
 963     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
 964     if (val && !strcmp(val, "on")) {
 965         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
 966     } else if (val && !strcmp(val, "off")) {
 967         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
 968     }
 969
 970     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
 971     if (val && !strcmp(val, "aes")) {
 972         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
 973     }
 974
 975     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
 976         ret = -EINVAL;
 977         goto fail;
 978     }
 979
 980     /* Create and open the file (protocol layer) */
 981     ret = bdrv_co_create_file(filename, opts, errp);
 982     if (ret < 0) {
 983         goto fail;
 984     }
 985
 986     bs = bdrv_co_open(filename, NULL, NULL,
 987                       BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
 988     if (bs == NULL) {
 989         ret = -EIO;
 990         goto fail;
 991     }
 992
 993     /* Now get the QAPI type BlockdevCreateOptions */
 994     qdict_put_str(qdict, "driver", "qcow");
 995     qdict_put_str(qdict, "file", bs->node_name);
 996
 997     v = qobject_input_visitor_new_flat_confused(qdict, errp);
 998     if (!v) {
 999         ret = -EINVAL;
1000         goto fail;
1001     }
1002
1003     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
1004     visit_free(v);
1005     if (!create_options) {
1006         ret = -EINVAL;
1007         goto fail;
1008     }
1009
1010     /* Silently round up size */
1011     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1012     create_options->u.qcow.size =
1013         ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1014
1015     /* Create the qcow image (format layer) */
1016     ret = qcow_co_create(create_options, errp);
1017     if (ret < 0) {
1018         goto fail;
1019     }
1020
1021     ret = 0;
1022 fail:
1023     g_free(backing_fmt);
1024     qobject_unref(qdict);
1025     bdrv_co_unref(bs);
1026     qapi_free_BlockdevCreateOptions(create_options);
1027     return ret;
1028 }
1029
1030 static int GRAPH_RDLOCK qcow_make_empty(BlockDriverState *bs)
1031 {
1032     BDRVQcowState *s = bs->opaque;
1033     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1034     int ret;
1035
1036     memset(s->l1_table, 0, l1_length);
1037     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, l1_length, s->l1_table,
1038                          0) < 0)
1039         return -1;
1040     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, false,
1041                         PREALLOC_MODE_OFF, 0, NULL);
1042     if (ret < 0)
1043         return ret;
1044
1045     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1046     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1047     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1048
1049     return 0;
1050 }
1051
1052 /* XXX: put compressed sectors first, then all the cluster aligned
1053    tables to avoid losing bytes in alignment */
1054 static int coroutine_fn GRAPH_RDLOCK
1055 qcow_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
1056                            QEMUIOVector *qiov)
1057 {
1058     BDRVQcowState *s = bs->opaque;
1059     z_stream strm;
1060     int ret, out_len;
1061     uint8_t *buf, *out_buf;
1062     uint64_t cluster_offset;
1063
1064     buf = qemu_blockalign(bs, s->cluster_size);
1065     if (bytes != s->cluster_size) {
1066         if (bytes > s->cluster_size ||
1067             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1068         {
1069             qemu_vfree(buf);
1070             return -EINVAL;
1071         }
1072         /* Zero-pad last write if image size is not cluster aligned */
1073         memset(buf + bytes, 0, s->cluster_size - bytes);
1074     }
1075     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1076
1077     out_buf = g_malloc(s->cluster_size);
1078
1079     /* best compression, small window, no zlib header */
1080     memset(&strm, 0, sizeof(strm));
1081     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1082                        Z_DEFLATED, -12,
1083                        9, Z_DEFAULT_STRATEGY);
1084     if (ret != 0) {
1085         ret = -EINVAL;
1086         goto fail;
1087     }
1088
1089     strm.avail_in = s->cluster_size;
1090     strm.next_in = (uint8_t *)buf;
1091     strm.avail_out = s->cluster_size;
1092     strm.next_out = out_buf;
1093
1094     ret = deflate(&strm, Z_FINISH);
1095     if (ret != Z_STREAM_END && ret != Z_OK) {
1096         deflateEnd(&strm);
1097         ret = -EINVAL;
1098         goto fail;
1099     }
1100     out_len = strm.next_out - out_buf;
1101
1102     deflateEnd(&strm);
1103
1104     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1105         /* could not compress: write normal cluster */
1106         ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1107         if (ret < 0) {
1108             goto fail;
1109         }
1110         goto success;
1111     }
1112     qemu_co_mutex_lock(&s->lock);
1113     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1114     qemu_co_mutex_unlock(&s->lock);
1115     if (ret < 0) {
1116         goto fail;
1117     }
1118     if (cluster_offset == 0) {
1119         ret = -EIO;
1120         goto fail;
1121     }
1122     cluster_offset &= s->cluster_offset_mask;
1123
1124     BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1125     ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0);
1126     if (ret < 0) {
1127         goto fail;
1128     }
1129 success:
1130     ret = 0;
1131 fail:
1132     qemu_vfree(buf);
1133     g_free(out_buf);
1134     return ret;
1135 }
1136
1137 static int coroutine_fn
1138 qcow_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1139 {
1140     BDRVQcowState *s = bs->opaque;
1141     bdi->cluster_size = s->cluster_size;
1142     return 0;
1143 }
1144
1145 static QemuOptsList qcow_create_opts = {
1146     .name = "qcow-create-opts",
1147     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1148     .desc = {
1149         {
1150             .name = BLOCK_OPT_SIZE,
1151             .type = QEMU_OPT_SIZE,
1152             .help = "Virtual disk size"
1153         },
1154         {
1155             .name = BLOCK_OPT_BACKING_FILE,
1156             .type = QEMU_OPT_STRING,
1157             .help = "File name of a base image"
1158         },
1159         {
1160             .name = BLOCK_OPT_BACKING_FMT,
1161             .type = QEMU_OPT_STRING,
1162             .help = "Format of the backing image",
1163         },
1164         {
1165             .name = BLOCK_OPT_ENCRYPT,
1166             .type = QEMU_OPT_BOOL,
1167             .help = "Encrypt the image with format 'aes'. (Deprecated "
1168                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1169         },
1170         {
1171             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1172             .type = QEMU_OPT_STRING,
1173             .help = "Encrypt the image, format choices: 'aes'",
1174         },
1175         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1176         { /* end of list */ }
1177     }
1178 };
1179
1180 static const char *const qcow_strong_runtime_opts[] = {
1181     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
1182
1183     NULL
1184 };
1185
1186 static BlockDriver bdrv_qcow = {
1187     .format_name        = "qcow",
1188     .instance_size      = sizeof(BDRVQcowState),
1189     .bdrv_probe         = qcow_probe,
1190     .bdrv_open          = qcow_open,
1191     .bdrv_close         = qcow_close,
1192     .bdrv_child_perm        = bdrv_default_perms,
1193     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1194     .bdrv_co_create         = qcow_co_create,
1195     .bdrv_co_create_opts    = qcow_co_create_opts,
1196     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1197     .is_format              = true,
1198     .supports_backing       = true,
1199     .bdrv_refresh_limits    = qcow_refresh_limits,
1200
1201     .bdrv_co_preadv         = qcow_co_preadv,
1202     .bdrv_co_pwritev        = qcow_co_pwritev,
1203     .bdrv_co_block_status   = qcow_co_block_status,
1204
1205     .bdrv_make_empty        = qcow_make_empty,
1206     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1207     .bdrv_co_get_info       = qcow_co_get_info,
1208
1209     .create_opts            = &qcow_create_opts,
1210     .strong_runtime_opts    = qcow_strong_runtime_opts,
1211 };
1212
1213 static void bdrv_qcow_init(void)
1214 {
1215     bdrv_register(&bdrv_qcow);
1216 }
1217
1218 block_init(bdrv_qcow_init);