Home | History | Annotate | Download | only in block
      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/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 = g_malloc0(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         g_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_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2);
     68     if (ret < 0)
     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_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
     78     if (ret < 0) {
     79         goto fail;
     80     }
     81     g_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     g_free(new_l1_table);
     89     qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
     90     return ret;
     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 int l2_load(BlockDriverState *bs, uint64_t l2_offset,
    161     uint64_t **l2_table)
    162 {
    163     BDRVQcowState *s = bs->opaque;
    164     int min_index;
    165     int ret;
    166 
    167     /* seek if the table for the given offset is in the cache */
    168 
    169     *l2_table = seek_l2_table(s, l2_offset);
    170     if (*l2_table != NULL) {
    171         return 0;
    172     }
    173 
    174     /* not found: load a new entry in the least used one */
    175 
    176     min_index = l2_cache_new_entry(bs);
    177     *l2_table = s->l2_cache + (min_index << s->l2_bits);
    178 
    179     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
    180     ret = bdrv_pread(bs->file, l2_offset, *l2_table,
    181         s->l2_size * sizeof(uint64_t));
    182     if (ret < 0) {
    183         return ret;
    184     }
    185 
    186     s->l2_cache_offsets[min_index] = l2_offset;
    187     s->l2_cache_counts[min_index] = 1;
    188 
    189     return 0;
    190 }
    191 
    192 /*
    193  * Writes one sector of the L1 table to the disk (can't update single entries
    194  * and we really don't want bdrv_pread to perform a read-modify-write)
    195  */
    196 #define L1_ENTRIES_PER_SECTOR (512 / 8)
    197 static int write_l1_entry(BlockDriverState *bs, int l1_index)
    198 {
    199     BDRVQcowState *s = bs->opaque;
    200     uint64_t buf[L1_ENTRIES_PER_SECTOR];
    201     int l1_start_index;
    202     int i, ret;
    203 
    204     l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
    205     for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
    206         buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
    207     }
    208 
    209     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
    210     ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index,
    211         buf, sizeof(buf));
    212     if (ret < 0) {
    213         return ret;
    214     }
    215 
    216     return 0;
    217 }
    218 
    219 /*
    220  * l2_allocate
    221  *
    222  * Allocate a new l2 entry in the file. If l1_index points to an already
    223  * used entry in the L2 table (i.e. we are doing a copy on write for the L2
    224  * table) copy the contents of the old L2 table into the newly allocated one.
    225  * Otherwise the new table is initialized with zeros.
    226  *
    227  */
    228 
    229 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
    230 {
    231     BDRVQcowState *s = bs->opaque;
    232     int min_index;
    233     uint64_t old_l2_offset;
    234     uint64_t *l2_table;
    235     int64_t l2_offset;
    236     int ret;
    237 
    238     old_l2_offset = s->l1_table[l1_index];
    239 
    240     /* allocate a new l2 entry */
    241 
    242     l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
    243     if (l2_offset < 0) {
    244         return l2_offset;
    245     }
    246 
    247     /* allocate a new entry in the l2 cache */
    248 
    249     min_index = l2_cache_new_entry(bs);
    250     l2_table = s->l2_cache + (min_index << s->l2_bits);
    251 
    252     if (old_l2_offset == 0) {
    253         /* if there was no old l2 table, clear the new table */
    254         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
    255     } else {
    256         /* if there was an old l2 table, read it from the disk */
    257         BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
    258         ret = bdrv_pread(bs->file, old_l2_offset, l2_table,
    259             s->l2_size * sizeof(uint64_t));
    260         if (ret < 0) {
    261             goto fail;
    262         }
    263     }
    264     /* write the l2 table to the file */
    265     BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
    266     ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
    267         s->l2_size * sizeof(uint64_t));
    268     if (ret < 0) {
    269         goto fail;
    270     }
    271 
    272     /* update the L1 entry */
    273     s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
    274     ret = write_l1_entry(bs, l1_index);
    275     if (ret < 0) {
    276         goto fail;
    277     }
    278 
    279     /* update the l2 cache entry */
    280 
    281     s->l2_cache_offsets[min_index] = l2_offset;
    282     s->l2_cache_counts[min_index] = 1;
    283 
    284     *table = l2_table;
    285     return 0;
    286 
    287 fail:
    288     s->l1_table[l1_index] = old_l2_offset;
    289     qcow2_l2_cache_reset(bs);
    290     return ret;
    291 }
    292 
    293 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
    294         uint64_t *l2_table, uint64_t start, uint64_t mask)
    295 {
    296     int i;
    297     uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
    298 
    299     if (!offset)
    300         return 0;
    301 
    302     for (i = start; i < start + nb_clusters; i++)
    303         if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
    304             break;
    305 
    306 	return (i - start);
    307 }
    308 
    309 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
    310 {
    311     int i = 0;
    312 
    313     while(nb_clusters-- && l2_table[i] == 0)
    314         i++;
    315 
    316     return i;
    317 }
    318 
    319 /* The crypt function is compatible with the linux cryptoloop
    320    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
    321    supported */
    322 void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
    323                            uint8_t *out_buf, const uint8_t *in_buf,
    324                            int nb_sectors, int enc,
    325                            const AES_KEY *key)
    326 {
    327     union {
    328         uint64_t ll[2];
    329         uint8_t b[16];
    330     } ivec;
    331     int i;
    332 
    333     for(i = 0; i < nb_sectors; i++) {
    334         ivec.ll[0] = cpu_to_le64(sector_num);
    335         ivec.ll[1] = 0;
    336         AES_cbc_encrypt(in_buf, out_buf, 512, key,
    337                         ivec.b, enc);
    338         sector_num++;
    339         in_buf += 512;
    340         out_buf += 512;
    341     }
    342 }
    343 
    344 
    345 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
    346                      uint8_t *buf, int nb_sectors)
    347 {
    348     BDRVQcowState *s = bs->opaque;
    349     int ret, index_in_cluster, n, n1;
    350     uint64_t cluster_offset;
    351 
    352     while (nb_sectors > 0) {
    353         n = nb_sectors;
    354 
    355         ret = qcow2_get_cluster_offset(bs, sector_num << 9, &n,
    356             &cluster_offset);
    357         if (ret < 0) {
    358             return ret;
    359         }
    360 
    361         index_in_cluster = sector_num & (s->cluster_sectors - 1);
    362         if (!cluster_offset) {
    363             if (bs->backing_hd) {
    364                 /* read from the base image */
    365                 n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n);
    366                 if (n1 > 0) {
    367                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING);
    368                     ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
    369                     if (ret < 0)
    370                         return -1;
    371                 }
    372             } else {
    373                 memset(buf, 0, 512 * n);
    374             }
    375         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
    376             if (qcow2_decompress_cluster(bs, cluster_offset) < 0)
    377                 return -1;
    378             memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
    379         } else {
    380             BLKDBG_EVENT(bs->file, BLKDBG_READ);
    381             ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512);
    382             if (ret != n * 512)
    383                 return -1;
    384             if (s->crypt_method) {
    385                 qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
    386                                 &s->aes_decrypt_key);
    387             }
    388         }
    389         nb_sectors -= n;
    390         sector_num += n;
    391         buf += n * 512;
    392     }
    393     return 0;
    394 }
    395 
    396 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
    397                         uint64_t cluster_offset, int n_start, int n_end)
    398 {
    399     BDRVQcowState *s = bs->opaque;
    400     int n, ret;
    401 
    402     n = n_end - n_start;
    403     if (n <= 0)
    404         return 0;
    405     BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
    406     ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
    407     if (ret < 0)
    408         return ret;
    409     if (s->crypt_method) {
    410         qcow2_encrypt_sectors(s, start_sect + n_start,
    411                         s->cluster_data,
    412                         s->cluster_data, n, 1,
    413                         &s->aes_encrypt_key);
    414     }
    415     BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
    416     ret = bdrv_write_sync(bs->file, (cluster_offset >> 9) + n_start,
    417         s->cluster_data, n);
    418     if (ret < 0)
    419         return ret;
    420     return 0;
    421 }
    422 
    423 
    424 /*
    425  * get_cluster_offset
    426  *
    427  * For a given offset of the disk image, find the cluster offset in
    428  * qcow2 file. The offset is stored in *cluster_offset.
    429  *
    430  * on entry, *num is the number of contiguous clusters we'd like to
    431  * access following offset.
    432  *
    433  * on exit, *num is the number of contiguous clusters we can read.
    434  *
    435  * Return 0, if the offset is found
    436  * Return -errno, otherwise.
    437  *
    438  */
    439 
    440 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
    441     int *num, uint64_t *cluster_offset)
    442 {
    443     BDRVQcowState *s = bs->opaque;
    444     unsigned int l1_index, l2_index;
    445     uint64_t l2_offset, *l2_table;
    446     int l1_bits, c;
    447     unsigned int index_in_cluster, nb_clusters;
    448     uint64_t nb_available, nb_needed;
    449     int ret;
    450 
    451     index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
    452     nb_needed = *num + index_in_cluster;
    453 
    454     l1_bits = s->l2_bits + s->cluster_bits;
    455 
    456     /* compute how many bytes there are between the offset and
    457      * the end of the l1 entry
    458      */
    459 
    460     nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
    461 
    462     /* compute the number of available sectors */
    463 
    464     nb_available = (nb_available >> 9) + index_in_cluster;
    465 
    466     if (nb_needed > nb_available) {
    467         nb_needed = nb_available;
    468     }
    469 
    470     *cluster_offset = 0;
    471 
    472     /* seek the the l2 offset in the l1 table */
    473 
    474     l1_index = offset >> l1_bits;
    475     if (l1_index >= s->l1_size)
    476         goto out;
    477 
    478     l2_offset = s->l1_table[l1_index];
    479 
    480     /* seek the l2 table of the given l2 offset */
    481 
    482     if (!l2_offset)
    483         goto out;
    484 
    485     /* load the l2 table in memory */
    486 
    487     l2_offset &= ~QCOW_OFLAG_COPIED;
    488     ret = l2_load(bs, l2_offset, &l2_table);
    489     if (ret < 0) {
    490         return ret;
    491     }
    492 
    493     /* find the cluster offset for the given disk offset */
    494 
    495     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
    496     *cluster_offset = be64_to_cpu(l2_table[l2_index]);
    497     nb_clusters = size_to_clusters(s, nb_needed << 9);
    498 
    499     if (!*cluster_offset) {
    500         /* how many empty clusters ? */
    501         c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
    502     } else {
    503         /* how many allocated clusters ? */
    504         c = count_contiguous_clusters(nb_clusters, s->cluster_size,
    505                 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
    506     }
    507 
    508    nb_available = (c * s->cluster_sectors);
    509 out:
    510     if (nb_available > nb_needed)
    511         nb_available = nb_needed;
    512 
    513     *num = nb_available - index_in_cluster;
    514 
    515     *cluster_offset &=~QCOW_OFLAG_COPIED;
    516     return 0;
    517 }
    518 
    519 /*
    520  * get_cluster_table
    521  *
    522  * for a given disk offset, load (and allocate if needed)
    523  * the l2 table.
    524  *
    525  * the l2 table offset in the qcow2 file and the cluster index
    526  * in the l2 table are given to the caller.
    527  *
    528  * Returns 0 on success, -errno in failure case
    529  */
    530 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
    531                              uint64_t **new_l2_table,
    532                              uint64_t *new_l2_offset,
    533                              int *new_l2_index)
    534 {
    535     BDRVQcowState *s = bs->opaque;
    536     unsigned int l1_index, l2_index;
    537     uint64_t l2_offset;
    538     uint64_t *l2_table = NULL;
    539     int ret;
    540 
    541     /* seek the the l2 offset in the l1 table */
    542 
    543     l1_index = offset >> (s->l2_bits + s->cluster_bits);
    544     if (l1_index >= s->l1_size) {
    545         ret = qcow2_grow_l1_table(bs, l1_index + 1);
    546         if (ret < 0) {
    547             return ret;
    548         }
    549     }
    550     l2_offset = s->l1_table[l1_index];
    551 
    552     /* seek the l2 table of the given l2 offset */
    553 
    554     if (l2_offset & QCOW_OFLAG_COPIED) {
    555         /* load the l2 table in memory */
    556         l2_offset &= ~QCOW_OFLAG_COPIED;
    557         ret = l2_load(bs, l2_offset, &l2_table);
    558         if (ret < 0) {
    559             return ret;
    560         }
    561     } else {
    562         if (l2_offset)
    563             qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
    564         ret = l2_allocate(bs, l1_index, &l2_table);
    565         if (ret < 0) {
    566             return ret;
    567         }
    568         l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
    569     }
    570 
    571     /* find the cluster offset for the given disk offset */
    572 
    573     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
    574 
    575     *new_l2_table = l2_table;
    576     *new_l2_offset = l2_offset;
    577     *new_l2_index = l2_index;
    578 
    579     return 0;
    580 }
    581 
    582 /*
    583  * alloc_compressed_cluster_offset
    584  *
    585  * For a given offset of the disk image, return cluster offset in
    586  * qcow2 file.
    587  *
    588  * If the offset is not found, allocate a new compressed cluster.
    589  *
    590  * Return the cluster offset if successful,
    591  * Return 0, otherwise.
    592  *
    593  */
    594 
    595 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
    596                                                uint64_t offset,
    597                                                int compressed_size)
    598 {
    599     BDRVQcowState *s = bs->opaque;
    600     int l2_index, ret;
    601     uint64_t l2_offset, *l2_table;
    602     int64_t cluster_offset;
    603     int nb_csectors;
    604 
    605     ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
    606     if (ret < 0) {
    607         return 0;
    608     }
    609 
    610     cluster_offset = be64_to_cpu(l2_table[l2_index]);
    611     if (cluster_offset & QCOW_OFLAG_COPIED)
    612         return cluster_offset & ~QCOW_OFLAG_COPIED;
    613 
    614     if (cluster_offset)
    615         qcow2_free_any_clusters(bs, cluster_offset, 1);
    616 
    617     cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
    618     if (cluster_offset < 0) {
    619         return 0;
    620     }
    621 
    622     nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
    623                   (cluster_offset >> 9);
    624 
    625     cluster_offset |= QCOW_OFLAG_COMPRESSED |
    626                       ((uint64_t)nb_csectors << s->csize_shift);
    627 
    628     /* update L2 table */
    629 
    630     /* compressed clusters never have the copied flag */
    631 
    632     BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
    633     l2_table[l2_index] = cpu_to_be64(cluster_offset);
    634     if (bdrv_pwrite_sync(bs->file,
    635                     l2_offset + l2_index * sizeof(uint64_t),
    636                     l2_table + l2_index,
    637                     sizeof(uint64_t)) < 0)
    638         return 0;
    639 
    640     return cluster_offset;
    641 }
    642 
    643 /*
    644  * Write L2 table updates to disk, writing whole sectors to avoid a
    645  * read-modify-write in bdrv_pwrite
    646  */
    647 #define L2_ENTRIES_PER_SECTOR (512 / 8)
    648 static int write_l2_entries(BlockDriverState *bs, uint64_t *l2_table,
    649     uint64_t l2_offset, int l2_index, int num)
    650 {
    651     int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1);
    652     int start_offset = (8 * l2_index) & ~511;
    653     int end_offset = (8 * (l2_index + num) + 511) & ~511;
    654     size_t len = end_offset - start_offset;
    655     int ret;
    656 
    657     BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
    658     ret = bdrv_pwrite_sync(bs->file, l2_offset + start_offset,
    659         &l2_table[l2_start_index], len);
    660     if (ret < 0) {
    661         return ret;
    662     }
    663 
    664     return 0;
    665 }
    666 
    667 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
    668 {
    669     BDRVQcowState *s = bs->opaque;
    670     int i, j = 0, l2_index, ret;
    671     uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
    672     uint64_t cluster_offset = m->cluster_offset;
    673 
    674     if (m->nb_clusters == 0)
    675         return 0;
    676 
    677     old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t));
    678 
    679     /* copy content of unmodified sectors */
    680     start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
    681     if (m->n_start) {
    682         ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
    683         if (ret < 0)
    684             goto err;
    685     }
    686 
    687     if (m->nb_available & (s->cluster_sectors - 1)) {
    688         uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
    689         ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
    690                 m->nb_available - end, s->cluster_sectors);
    691         if (ret < 0)
    692             goto err;
    693     }
    694 
    695     /* update L2 table */
    696     ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index);
    697     if (ret < 0) {
    698         goto err;
    699     }
    700 
    701     for (i = 0; i < m->nb_clusters; i++) {
    702         /* if two concurrent writes happen to the same unallocated cluster
    703 	 * each write allocates separate cluster and writes data concurrently.
    704 	 * The first one to complete updates l2 table with pointer to its
    705 	 * cluster the second one has to do RMW (which is done above by
    706 	 * copy_sectors()), update l2 table with its cluster pointer and free
    707 	 * old cluster. This is what this loop does */
    708         if(l2_table[l2_index + i] != 0)
    709             old_cluster[j++] = l2_table[l2_index + i];
    710 
    711         l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
    712                     (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
    713      }
    714 
    715     ret = write_l2_entries(bs, l2_table, l2_offset, l2_index, m->nb_clusters);
    716     if (ret < 0) {
    717         qcow2_l2_cache_reset(bs);
    718         goto err;
    719     }
    720 
    721     for (i = 0; i < j; i++)
    722         qcow2_free_any_clusters(bs,
    723             be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
    724 
    725     ret = 0;
    726 err:
    727     g_free(old_cluster);
    728     return ret;
    729  }
    730 
    731 /*
    732  * alloc_cluster_offset
    733  *
    734  * For a given offset of the disk image, return cluster offset in qcow2 file.
    735  * If the offset is not found, allocate a new cluster.
    736  *
    737  * If the cluster was already allocated, m->nb_clusters is set to 0,
    738  * m->depends_on is set to NULL and the other fields in m are meaningless.
    739  *
    740  * If the cluster is newly allocated, m->nb_clusters is set to the number of
    741  * contiguous clusters that have been allocated. This may be 0 if the request
    742  * conflict with another write request in flight; in this case, m->depends_on
    743  * is set and the remaining fields of m are meaningless.
    744  *
    745  * If m->nb_clusters is non-zero, the other fields of m are valid and contain
    746  * information about the first allocated cluster.
    747  *
    748  * Return 0 on success and -errno in error cases
    749  */
    750 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
    751     int n_start, int n_end, int *num, QCowL2Meta *m)
    752 {
    753     BDRVQcowState *s = bs->opaque;
    754     int l2_index, ret;
    755     uint64_t l2_offset, *l2_table;
    756     int64_t cluster_offset;
    757     unsigned int nb_clusters, i = 0;
    758     QCowL2Meta *old_alloc;
    759 
    760     ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
    761     if (ret < 0) {
    762         return ret;
    763     }
    764 
    765     nb_clusters = size_to_clusters(s, n_end << 9);
    766 
    767     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
    768 
    769     cluster_offset = be64_to_cpu(l2_table[l2_index]);
    770 
    771     /* We keep all QCOW_OFLAG_COPIED clusters */
    772 
    773     if (cluster_offset & QCOW_OFLAG_COPIED) {
    774         nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
    775                 &l2_table[l2_index], 0, 0);
    776 
    777         cluster_offset &= ~QCOW_OFLAG_COPIED;
    778         m->nb_clusters = 0;
    779         m->depends_on = NULL;
    780 
    781         goto out;
    782     }
    783 
    784     /* for the moment, multiple compressed clusters are not managed */
    785 
    786     if (cluster_offset & QCOW_OFLAG_COMPRESSED)
    787         nb_clusters = 1;
    788 
    789     /* how many available clusters ? */
    790 
    791     while (i < nb_clusters) {
    792         i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
    793                 &l2_table[l2_index], i, 0);
    794         if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) {
    795             break;
    796         }
    797 
    798         i += count_contiguous_free_clusters(nb_clusters - i,
    799                 &l2_table[l2_index + i]);
    800         if (i >= nb_clusters) {
    801             break;
    802         }
    803 
    804         cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
    805 
    806         if ((cluster_offset & QCOW_OFLAG_COPIED) ||
    807                 (cluster_offset & QCOW_OFLAG_COMPRESSED))
    808             break;
    809     }
    810     assert(i <= nb_clusters);
    811     nb_clusters = i;
    812 
    813     /*
    814      * Check if there already is an AIO write request in flight which allocates
    815      * the same cluster. In this case we need to wait until the previous
    816      * request has completed and updated the L2 table accordingly.
    817      */
    818     QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
    819 
    820         uint64_t end_offset = offset + nb_clusters * s->cluster_size;
    821         uint64_t old_offset = old_alloc->offset;
    822         uint64_t old_end_offset = old_alloc->offset +
    823             old_alloc->nb_clusters * s->cluster_size;
    824 
    825         if (end_offset < old_offset || offset > old_end_offset) {
    826             /* No intersection */
    827         } else {
    828             if (offset < old_offset) {
    829                 /* Stop at the start of a running allocation */
    830                 nb_clusters = (old_offset - offset) >> s->cluster_bits;
    831             } else {
    832                 nb_clusters = 0;
    833             }
    834 
    835             if (nb_clusters == 0) {
    836                 /* Set dependency and wait for a callback */
    837                 m->depends_on = old_alloc;
    838                 m->nb_clusters = 0;
    839                 *num = 0;
    840                 return 0;
    841             }
    842         }
    843     }
    844 
    845     if (!nb_clusters) {
    846         abort();
    847     }
    848 
    849     QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
    850 
    851     /* allocate a new cluster */
    852 
    853     cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
    854     if (cluster_offset < 0) {
    855         QLIST_REMOVE(m, next_in_flight);
    856         return cluster_offset;
    857     }
    858 
    859     /* save info needed for meta data update */
    860     m->offset = offset;
    861     m->n_start = n_start;
    862     m->nb_clusters = nb_clusters;
    863 
    864 out:
    865     m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
    866     m->cluster_offset = cluster_offset;
    867 
    868     *num = m->nb_available - n_start;
    869 
    870     return 0;
    871 }
    872 
    873 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
    874                              const uint8_t *buf, int buf_size)
    875 {
    876     z_stream strm1, *strm = &strm1;
    877     int ret, out_len;
    878 
    879     memset(strm, 0, sizeof(*strm));
    880 
    881     strm->next_in = (uint8_t *)buf;
    882     strm->avail_in = buf_size;
    883     strm->next_out = out_buf;
    884     strm->avail_out = out_buf_size;
    885 
    886     ret = inflateInit2(strm, -12);
    887     if (ret != Z_OK)
    888         return -1;
    889     ret = inflate(strm, Z_FINISH);
    890     out_len = strm->next_out - out_buf;
    891     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
    892         out_len != out_buf_size) {
    893         inflateEnd(strm);
    894         return -1;
    895     }
    896     inflateEnd(strm);
    897     return 0;
    898 }
    899 
    900 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
    901 {
    902     BDRVQcowState *s = bs->opaque;
    903     int ret, csize, nb_csectors, sector_offset;
    904     uint64_t coffset;
    905 
    906     coffset = cluster_offset & s->cluster_offset_mask;
    907     if (s->cluster_cache_offset != coffset) {
    908         nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
    909         sector_offset = coffset & 511;
    910         csize = nb_csectors * 512 - sector_offset;
    911         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
    912         ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
    913         if (ret < 0) {
    914             return -1;
    915         }
    916         if (decompress_buffer(s->cluster_cache, s->cluster_size,
    917                               s->cluster_data + sector_offset, csize) < 0) {
    918             return -1;
    919         }
    920         s->cluster_cache_offset = coffset;
    921     }
    922     return 0;
    923 }
    924