block/qcow2-cluster.c

   1 /*
   2  * Block driver for the QCOW version 2 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 <zlib.h>
  26
  27 #include "qemu-common.h"
  28 #include "block_int.h"
  29 #include "block/qcow2.h"
  30
  31 int qcow2_grow_l1_table(BlockDriverState *bs, int min_size)
  32 {
  33     BDRVQcowState *s = bs->opaque;
  34     int new_l1_size, new_l1_size2, ret, i;
  35     uint64_t *new_l1_table;
  36     int64_t new_l1_table_offset;
  37     uint8_t data[12];
  38
  39     new_l1_size = s->l1_size;
  40     if (min_size <= new_l1_size)
  41         return 0;
  42     if (new_l1_size == 0) {
  43         new_l1_size = 1;
  44     }
  45     while (min_size > new_l1_size) {
  46         new_l1_size = (new_l1_size * 3 + 1) / 2;
  47     }
  48 #ifdef DEBUG_ALLOC2
  49     printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
  50 #endif
  51
  52     new_l1_size2 = sizeof(uint64_t) * new_l1_size;
  53     new_l1_table = qemu_mallocz(align_offset(new_l1_size2, 512));
  54     memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
  55
  56     /* write new table (align to cluster) */
  57     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
  58     new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
  59     if (new_l1_table_offset < 0) {
  60         qemu_free(new_l1_table);
  61         return new_l1_table_offset;
  62     }
  63
  64     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
  65     for(i = 0; i < s->l1_size; i++)
  66         new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
  67     ret = bdrv_pwrite(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2);
  68     if (ret != new_l1_size2)
  69         goto fail;
  70     for(i = 0; i < s->l1_size; i++)
  71         new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
  72
  73     /* set new table */
  74     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
  75     cpu_to_be32w((uint32_t*)data, new_l1_size);
  76     cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
  77     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
  78     if (ret != sizeof(data)) {
  79         goto fail;
  80     }
  81     qemu_free(s->l1_table);
  82     qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
  83     s->l1_table_offset = new_l1_table_offset;
  84     s->l1_table = new_l1_table;
  85     s->l1_size = new_l1_size;
  86     return 0;
  87  fail:
  88     qemu_free(new_l1_table);
  89     qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
  90     return ret < 0 ? ret : -EIO;
  91 }
  92
  93 void qcow2_l2_cache_reset(BlockDriverState *bs)
  94 {
  95     BDRVQcowState *s = bs->opaque;
  96
  97     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
  98     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
  99     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 100 }
 101
 102 static inline int l2_cache_new_entry(BlockDriverState *bs)
 103 {
 104     BDRVQcowState *s = bs->opaque;
 105     uint32_t min_count;
 106     int min_index, i;
 107
 108     /* find a new entry in the least used one */
 109     min_index = 0;
 110     min_count = 0xffffffff;
 111     for(i = 0; i < L2_CACHE_SIZE; i++) {
 112         if (s->l2_cache_counts[i] < min_count) {
 113             min_count = s->l2_cache_counts[i];
 114             min_index = i;
 115         }
 116     }
 117     return min_index;
 118 }
 119
 120 /*
 121  * seek_l2_table
 122  *
 123  * seek l2_offset in the l2_cache table
 124  * if not found, return NULL,
 125  * if found,
 126  *   increments the l2 cache hit count of the entry,
 127  *   if counter overflow, divide by two all counters
 128  *   return the pointer to the l2 cache entry
 129  *
 130  */
 131
 132 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
 133 {
 134     int i, j;
 135
 136     for(i = 0; i < L2_CACHE_SIZE; i++) {
 137         if (l2_offset == s->l2_cache_offsets[i]) {
 138             /* increment the hit count */
 139             if (++s->l2_cache_counts[i] == 0xffffffff) {
 140                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 141                     s->l2_cache_counts[j] >>= 1;
 142                 }
 143             }
 144             return s->l2_cache + (i << s->l2_bits);
 145         }
 146     }
 147     return NULL;
 148 }
 149
 150 /*
 151  * l2_load
 152  *
 153  * Loads a L2 table into memory. If the table is in the cache, the cache
 154  * is used; otherwise the L2 table is loaded from the image file.
 155  *
 156  * Returns a pointer to the L2 table on success, or NULL if the read from
 157  * the image file failed.
 158  */
 159
 160 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
 161 {
 162     BDRVQcowState *s = bs->opaque;
 163     int min_index;
 164     uint64_t *l2_table;
 165
 166     /* seek if the table for the given offset is in the cache */
 167
 168     l2_table = seek_l2_table(s, l2_offset);
 169     if (l2_table != NULL)
 170         return l2_table;
 171
 172     /* not found: load a new entry in the least used one */
 173
 174     min_index = l2_cache_new_entry(bs);
 175     l2_table = s->l2_cache + (min_index << s->l2_bits);
 176
 177     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
 178     if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
 179         s->l2_size * sizeof(uint64_t))
 180         return NULL;
 181     s->l2_cache_offsets[min_index] = l2_offset;
 182     s->l2_cache_counts[min_index] = 1;
 183
 184     return l2_table;
 185 }
 186
 187 /*
 188  * Writes one sector of the L1 table to the disk (can't update single entries
 189  * and we really don't want bdrv_pread to perform a read-modify-write)
 190  */
 191 #define L1_ENTRIES_PER_SECTOR (512 / 8)
 192 static int write_l1_entry(BlockDriverState *bs, int l1_index)
 193 {
 194     BDRVQcowState *s = bs->opaque;
 195     uint64_t buf[L1_ENTRIES_PER_SECTOR];
 196     int l1_start_index;
 197     int i, ret;
 198
 199     l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
 200     for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
 201         buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
 202     }
 203
 204     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
 205     ret = bdrv_pwrite(bs->file, s->l1_table_offset + 8 * l1_start_index,
 206         buf, sizeof(buf));
 207     if (ret < 0) {
 208         return ret;
 209     }
 210
 211     return 0;
 212 }
 213
 214 /*
 215  * l2_allocate
 216  *
 217  * Allocate a new l2 entry in the file. If l1_index points to an already
 218  * used entry in the L2 table (i.e. we are doing a copy on write for the L2
 219  * table) copy the contents of the old L2 table into the newly allocated one.
 220  * Otherwise the new table is initialized with zeros.
 221  *
 222  */
 223
 224 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
 225 {
 226     BDRVQcowState *s = bs->opaque;
 227     int min_index;
 228     uint64_t old_l2_offset;
 229     uint64_t *l2_table;
 230     int64_t l2_offset;
 231     int ret;
 232
 233     old_l2_offset = s->l1_table[l1_index];
 234
 235     /* allocate a new l2 entry */
 236
 237     l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
 238     if (l2_offset < 0) {
 239         return l2_offset;
 240     }
 241
 242     /* update the L1 entry */
 243
 244     s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
 245     ret = write_l1_entry(bs, l1_index);
 246     if (ret < 0) {
 247         return ret;
 248     }
 249
 250     /* allocate a new entry in the l2 cache */
 251
 252     min_index = l2_cache_new_entry(bs);
 253     l2_table = s->l2_cache + (min_index << s->l2_bits);
 254
 255     if (old_l2_offset == 0) {
 256         /* if there was no old l2 table, clear the new table */
 257         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 258     } else {
 259         /* if there was an old l2 table, read it from the disk */
 260         BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
 261         ret = bdrv_pread(bs->file, old_l2_offset, l2_table,
 262             s->l2_size * sizeof(uint64_t));
 263         if (ret < 0) {
 264             return ret;
 265         }
 266     }
 267     /* write the l2 table to the file */
 268     BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
 269     ret = bdrv_pwrite(bs->file, l2_offset, l2_table,
 270         s->l2_size * sizeof(uint64_t));
 271     if (ret < 0) {
 272         return ret;
 273     }
 274
 275     /* update the l2 cache entry */
 276
 277     s->l2_cache_offsets[min_index] = l2_offset;
 278     s->l2_cache_counts[min_index] = 1;
 279
 280     *table = l2_table;
 281     return 0;
 282 }
 283
 284 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
 285         uint64_t *l2_table, uint64_t start, uint64_t mask)
 286 {
 287     int i;
 288     uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
 289
 290     if (!offset)
 291         return 0;
 292
 293     for (i = start; i < start + nb_clusters; i++)
 294         if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
 295             break;
 296
 297         return (i - start);
 298 }
 299
 300 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
 301 {
 302     int i = 0;
 303
 304     while(nb_clusters-- && l2_table[i] == 0)
 305         i++;
 306
 307     return i;
 308 }
 309
 310 /* The crypt function is compatible with the linux cryptoloop
 311    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
 312    supported */
 313 void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 314                            uint8_t *out_buf, const uint8_t *in_buf,
 315                            int nb_sectors, int enc,
 316                            const AES_KEY *key)
 317 {
 318     union {
 319         uint64_t ll[2];
 320         uint8_t b[16];
 321     } ivec;
 322     int i;
 323
 324     for(i = 0; i < nb_sectors; i++) {
 325         ivec.ll[0] = cpu_to_le64(sector_num);
 326         ivec.ll[1] = 0;
 327         AES_cbc_encrypt(in_buf, out_buf, 512, key,
 328                         ivec.b, enc);
 329         sector_num++;
 330         in_buf += 512;
 331         out_buf += 512;
 332     }
 333 }
 334
 335
 336 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
 337                      uint8_t *buf, int nb_sectors)
 338 {
 339     BDRVQcowState *s = bs->opaque;
 340     int ret, index_in_cluster, n, n1;
 341     uint64_t cluster_offset;
 342
 343     while (nb_sectors > 0) {
 344         n = nb_sectors;
 345         cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, &n);
 346         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 347         if (!cluster_offset) {
 348             if (bs->backing_hd) {
 349                 /* read from the base image */
 350                 n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n);
 351                 if (n1 > 0) {
 352                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING);
 353                     ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
 354                     if (ret < 0)
 355                         return -1;
 356                 }
 357             } else {
 358                 memset(buf, 0, 512 * n);
 359             }
 360         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 361             if (qcow2_decompress_cluster(bs, cluster_offset) < 0)
 362                 return -1;
 363             memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
 364         } else {
 365             BLKDBG_EVENT(bs->file, BLKDBG_READ);
 366             ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512);
 367             if (ret != n * 512)
 368                 return -1;
 369             if (s->crypt_method) {
 370                 qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
 371                                 &s->aes_decrypt_key);
 372             }
 373         }
 374         nb_sectors -= n;
 375         sector_num += n;
 376         buf += n * 512;
 377     }
 378     return 0;
 379 }
 380
 381 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
 382                         uint64_t cluster_offset, int n_start, int n_end)
 383 {
 384     BDRVQcowState *s = bs->opaque;
 385     int n, ret;
 386
 387     n = n_end - n_start;
 388     if (n <= 0)
 389         return 0;
 390     BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
 391     ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
 392     if (ret < 0)
 393         return ret;
 394     if (s->crypt_method) {
 395         qcow2_encrypt_sectors(s, start_sect + n_start,
 396                         s->cluster_data,
 397                         s->cluster_data, n, 1,
 398                         &s->aes_encrypt_key);
 399     }
 400     BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
 401     ret = bdrv_write(bs->file, (cluster_offset >> 9) + n_start,
 402                      s->cluster_data, n);
 403     if (ret < 0)
 404         return ret;
 405     return 0;
 406 }
 407
 408
 409 /*
 410  * get_cluster_offset
 411  *
 412  * For a given offset of the disk image, return cluster offset in
 413  * qcow2 file.
 414  *
 415  * on entry, *num is the number of contiguous clusters we'd like to
 416  * access following offset.
 417  *
 418  * on exit, *num is the number of contiguous clusters we can read.
 419  *
 420  * Return 1, if the offset is found
 421  * Return 0, otherwise.
 422  *
 423  */
 424
 425 uint64_t qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
 426     int *num)
 427 {
 428     BDRVQcowState *s = bs->opaque;
 429     unsigned int l1_index, l2_index;
 430     uint64_t l2_offset, *l2_table, cluster_offset;
 431     int l1_bits, c;
 432     unsigned int index_in_cluster, nb_clusters;
 433     uint64_t nb_available, nb_needed;
 434
 435     index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
 436     nb_needed = *num + index_in_cluster;
 437
 438     l1_bits = s->l2_bits + s->cluster_bits;
 439
 440     /* compute how many bytes there are between the offset and
 441      * the end of the l1 entry
 442      */
 443
 444     nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
 445
 446     /* compute the number of available sectors */
 447
 448     nb_available = (nb_available >> 9) + index_in_cluster;
 449
 450     if (nb_needed > nb_available) {
 451         nb_needed = nb_available;
 452     }
 453
 454     cluster_offset = 0;
 455
 456     /* seek the the l2 offset in the l1 table */
 457
 458     l1_index = offset >> l1_bits;
 459     if (l1_index >= s->l1_size)
 460         goto out;
 461
 462     l2_offset = s->l1_table[l1_index];
 463
 464     /* seek the l2 table of the given l2 offset */
 465
 466     if (!l2_offset)
 467         goto out;
 468
 469     /* load the l2 table in memory */
 470
 471     l2_offset &= ~QCOW_OFLAG_COPIED;
 472     l2_table = l2_load(bs, l2_offset);
 473     if (l2_table == NULL)
 474         return 0;
 475
 476     /* find the cluster offset for the given disk offset */
 477
 478     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 479     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 480     nb_clusters = size_to_clusters(s, nb_needed << 9);
 481
 482     if (!cluster_offset) {
 483         /* how many empty clusters ? */
 484         c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
 485     } else {
 486         /* how many allocated clusters ? */
 487         c = count_contiguous_clusters(nb_clusters, s->cluster_size,
 488                 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
 489     }
 490
 491    nb_available = (c * s->cluster_sectors);
 492 out:
 493     if (nb_available > nb_needed)
 494         nb_available = nb_needed;
 495
 496     *num = nb_available - index_in_cluster;
 497
 498     return cluster_offset & ~QCOW_OFLAG_COPIED;
 499 }
 500
 501 /*
 502  * get_cluster_table
 503  *
 504  * for a given disk offset, load (and allocate if needed)
 505  * the l2 table.
 506  *
 507  * the l2 table offset in the qcow2 file and the cluster index
 508  * in the l2 table are given to the caller.
 509  *
 510  * Returns 0 on success, -errno in failure case
 511  */
 512 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
 513                              uint64_t **new_l2_table,
 514                              uint64_t *new_l2_offset,
 515                              int *new_l2_index)
 516 {
 517     BDRVQcowState *s = bs->opaque;
 518     unsigned int l1_index, l2_index;
 519     uint64_t l2_offset;
 520     uint64_t *l2_table = NULL;
 521     int ret;
 522
 523     /* seek the the l2 offset in the l1 table */
 524
 525     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 526     if (l1_index >= s->l1_size) {
 527         ret = qcow2_grow_l1_table(bs, l1_index + 1);
 528         if (ret < 0) {
 529             return ret;
 530         }
 531     }
 532     l2_offset = s->l1_table[l1_index];
 533
 534     /* seek the l2 table of the given l2 offset */
 535
 536     if (l2_offset & QCOW_OFLAG_COPIED) {
 537         /* load the l2 table in memory */
 538         l2_offset &= ~QCOW_OFLAG_COPIED;
 539         l2_table = l2_load(bs, l2_offset);
 540         if (l2_table == NULL) {
 541             return -EIO;
 542         }
 543     } else {
 544         if (l2_offset)
 545             qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
 546         ret = l2_allocate(bs, l1_index, &l2_table);
 547         if (ret < 0) {
 548             return ret;
 549         }
 550         l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
 551     }
 552
 553     /* find the cluster offset for the given disk offset */
 554
 555     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 556
 557     *new_l2_table = l2_table;
 558     *new_l2_offset = l2_offset;
 559     *new_l2_index = l2_index;
 560
 561     return 0;
 562 }
 563
 564 /*
 565  * alloc_compressed_cluster_offset
 566  *
 567  * For a given offset of the disk image, return cluster offset in
 568  * qcow2 file.
 569  *
 570  * If the offset is not found, allocate a new compressed cluster.
 571  *
 572  * Return the cluster offset if successful,
 573  * Return 0, otherwise.
 574  *
 575  */
 576
 577 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
 578                                                uint64_t offset,
 579                                                int compressed_size)
 580 {
 581     BDRVQcowState *s = bs->opaque;
 582     int l2_index, ret;
 583     uint64_t l2_offset, *l2_table;
 584     int64_t cluster_offset;
 585     int nb_csectors;
 586
 587     ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
 588     if (ret < 0) {
 589         return 0;
 590     }
 591
 592     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 593     if (cluster_offset & QCOW_OFLAG_COPIED)
 594         return cluster_offset & ~QCOW_OFLAG_COPIED;
 595
 596     if (cluster_offset)
 597         qcow2_free_any_clusters(bs, cluster_offset, 1);
 598
 599     cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
 600     if (cluster_offset < 0) {
 601         return 0;
 602     }
 603
 604     nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
 605                   (cluster_offset >> 9);
 606
 607     cluster_offset |= QCOW_OFLAG_COMPRESSED |
 608                       ((uint64_t)nb_csectors << s->csize_shift);
 609
 610     /* update L2 table */
 611
 612     /* compressed clusters never have the copied flag */
 613
 614     BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
 615     l2_table[l2_index] = cpu_to_be64(cluster_offset);
 616     if (bdrv_pwrite(bs->file,
 617                     l2_offset + l2_index * sizeof(uint64_t),
 618                     l2_table + l2_index,
 619                     sizeof(uint64_t)) != sizeof(uint64_t))
 620         return 0;
 621
 622     return cluster_offset;
 623 }
 624
 625 /*
 626  * Write L2 table updates to disk, writing whole sectors to avoid a
 627  * read-modify-write in bdrv_pwrite
 628  */
 629 #define L2_ENTRIES_PER_SECTOR (512 / 8)
 630 static int write_l2_entries(BlockDriverState *bs, uint64_t *l2_table,
 631     uint64_t l2_offset, int l2_index, int num)
 632 {
 633     int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1);
 634     int start_offset = (8 * l2_index) & ~511;
 635     int end_offset = (8 * (l2_index + num) + 511) & ~511;
 636     size_t len = end_offset - start_offset;
 637     int ret;
 638
 639     BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
 640     ret = bdrv_pwrite(bs->file, l2_offset + start_offset,
 641         &l2_table[l2_start_index], len);
 642     if (ret < 0) {
 643         return ret;
 644     }
 645
 646     return 0;
 647 }
 648
 649 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
 650 {
 651     BDRVQcowState *s = bs->opaque;
 652     int i, j = 0, l2_index, ret;
 653     uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
 654     uint64_t cluster_offset = m->cluster_offset;
 655
 656     if (m->nb_clusters == 0)
 657         return 0;
 658
 659     old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
 660
 661     /* copy content of unmodified sectors */
 662     start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
 663     if (m->n_start) {
 664         ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
 665         if (ret < 0)
 666             goto err;
 667     }
 668
 669     if (m->nb_available & (s->cluster_sectors - 1)) {
 670         uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
 671         ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
 672                 m->nb_available - end, s->cluster_sectors);
 673         if (ret < 0)
 674             goto err;
 675     }
 676
 677     /* update L2 table */
 678     ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index);
 679     if (ret < 0) {
 680         goto err;
 681     }
 682
 683     for (i = 0; i < m->nb_clusters; i++) {
 684         /* if two concurrent writes happen to the same unallocated cluster
 685          * each write allocates separate cluster and writes data concurrently.
 686          * The first one to complete updates l2 table with pointer to its
 687          * cluster the second one has to do RMW (which is done above by
 688          * copy_sectors()), update l2 table with its cluster pointer and free
 689          * old cluster. This is what this loop does */
 690         if(l2_table[l2_index + i] != 0)
 691             old_cluster[j++] = l2_table[l2_index + i];
 692
 693         l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
 694                     (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
 695      }
 696
 697     ret = write_l2_entries(bs, l2_table, l2_offset, l2_index, m->nb_clusters);
 698     if (ret < 0) {
 699         goto err;
 700     }
 701
 702     for (i = 0; i < j; i++)
 703         qcow2_free_any_clusters(bs,
 704             be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
 705
 706     ret = 0;
 707 err:
 708     qemu_free(old_cluster);
 709     return ret;
 710  }
 711
 712 /*
 713  * alloc_cluster_offset
 714  *
 715  * For a given offset of the disk image, return cluster offset in qcow2 file.
 716  * If the offset is not found, allocate a new cluster.
 717  *
 718  * If the cluster was already allocated, m->nb_clusters is set to 0,
 719  * m->depends_on is set to NULL and the other fields in m are meaningless.
 720  *
 721  * If the cluster is newly allocated, m->nb_clusters is set to the number of
 722  * contiguous clusters that have been allocated. This may be 0 if the request
 723  * conflict with another write request in flight; in this case, m->depends_on
 724  * is set and the remaining fields of m are meaningless.
 725  *
 726  * If m->nb_clusters is non-zero, the other fields of m are valid and contain
 727  * information about the first allocated cluster.
 728  *
 729  * Return 0 on success and -errno in error cases
 730  */
 731 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
 732     int n_start, int n_end, int *num, QCowL2Meta *m)
 733 {
 734     BDRVQcowState *s = bs->opaque;
 735     int l2_index, ret;
 736     uint64_t l2_offset, *l2_table;
 737     int64_t cluster_offset;
 738     unsigned int nb_clusters, i = 0;
 739     QCowL2Meta *old_alloc;
 740
 741     ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
 742     if (ret < 0) {
 743         return ret;
 744     }
 745
 746     nb_clusters = size_to_clusters(s, n_end << 9);
 747
 748     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
 749
 750     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 751
 752     /* We keep all QCOW_OFLAG_COPIED clusters */
 753
 754     if (cluster_offset & QCOW_OFLAG_COPIED) {
 755         nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
 756                 &l2_table[l2_index], 0, 0);
 757
 758         cluster_offset &= ~QCOW_OFLAG_COPIED;
 759         m->nb_clusters = 0;
 760         m->depends_on = NULL;
 761
 762         goto out;
 763     }
 764
 765     /* for the moment, multiple compressed clusters are not managed */
 766
 767     if (cluster_offset & QCOW_OFLAG_COMPRESSED)
 768         nb_clusters = 1;
 769
 770     /* how many available clusters ? */
 771
 772     while (i < nb_clusters) {
 773         i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
 774                 &l2_table[l2_index], i, 0);
 775         if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) {
 776             break;
 777         }
 778
 779         i += count_contiguous_free_clusters(nb_clusters - i,
 780                 &l2_table[l2_index + i]);
 781         if (i >= nb_clusters) {
 782             break;
 783         }
 784
 785         cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
 786
 787         if ((cluster_offset & QCOW_OFLAG_COPIED) ||
 788                 (cluster_offset & QCOW_OFLAG_COMPRESSED))
 789             break;
 790     }
 791     assert(i <= nb_clusters);
 792     nb_clusters = i;
 793
 794     /*
 795      * Check if there already is an AIO write request in flight which allocates
 796      * the same cluster. In this case we need to wait until the previous
 797      * request has completed and updated the L2 table accordingly.
 798      */
 799     QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
 800
 801         uint64_t end_offset = offset + nb_clusters * s->cluster_size;
 802         uint64_t old_offset = old_alloc->offset;
 803         uint64_t old_end_offset = old_alloc->offset +
 804             old_alloc->nb_clusters * s->cluster_size;
 805
 806         if (end_offset < old_offset || offset > old_end_offset) {
 807             /* No intersection */
 808         } else {
 809             if (offset < old_offset) {
 810                 /* Stop at the start of a running allocation */
 811                 nb_clusters = (old_offset - offset) >> s->cluster_bits;
 812             } else {
 813                 nb_clusters = 0;
 814             }
 815
 816             if (nb_clusters == 0) {
 817                 /* Set dependency and wait for a callback */
 818                 m->depends_on = old_alloc;
 819                 m->nb_clusters = 0;
 820                 *num = 0;
 821                 return 0;
 822             }
 823         }
 824     }
 825
 826     if (!nb_clusters) {
 827         abort();
 828     }
 829
 830     QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
 831
 832     /* allocate a new cluster */
 833
 834     cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
 835     if (cluster_offset < 0) {
 836         QLIST_REMOVE(m, next_in_flight);
 837         return cluster_offset;
 838     }
 839
 840     /* save info needed for meta data update */
 841     m->offset = offset;
 842     m->n_start = n_start;
 843     m->nb_clusters = nb_clusters;
 844
 845 out:
 846     m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
 847     m->cluster_offset = cluster_offset;
 848
 849     *num = m->nb_available - n_start;
 850
 851     return 0;
 852 }
 853
 854 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 855                              const uint8_t *buf, int buf_size)
 856 {
 857     z_stream strm1, *strm = &strm1;
 858     int ret, out_len;
 859
 860     memset(strm, 0, sizeof(*strm));
 861
 862     strm->next_in = (uint8_t *)buf;
 863     strm->avail_in = buf_size;
 864     strm->next_out = out_buf;
 865     strm->avail_out = out_buf_size;
 866
 867     ret = inflateInit2(strm, -12);
 868     if (ret != Z_OK)
 869         return -1;
 870     ret = inflate(strm, Z_FINISH);
 871     out_len = strm->next_out - out_buf;
 872     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 873         out_len != out_buf_size) {
 874         inflateEnd(strm);
 875         return -1;
 876     }
 877     inflateEnd(strm);
 878     return 0;
 879 }
 880
 881 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 882 {
 883     BDRVQcowState *s = bs->opaque;
 884     int ret, csize, nb_csectors, sector_offset;
 885     uint64_t coffset;
 886
 887     coffset = cluster_offset & s->cluster_offset_mask;
 888     if (s->cluster_cache_offset != coffset) {
 889         nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
 890         sector_offset = coffset & 511;
 891         csize = nb_csectors * 512 - sector_offset;
 892         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
 893         ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
 894         if (ret < 0) {
 895             return -1;
 896         }
 897         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 898                               s->cluster_data + sector_offset, csize) < 0) {
 899             return -1;
 900         }
 901         s->cluster_cache_offset = coffset;
 902     }
 903     return 0;
 904 }