]> Git Repo - qemu.git/blobdiff - block/qcow2-cluster.c
qcow2: Handle failure for potentially large allocations
[qemu.git] / block / qcow2-cluster.c
index cca76d4fcdd12a1722f21af3ef60297c501eca4c..e7c5f486cd569d98713eb757ae6bd9aaf1c565b5 100644 (file)
@@ -35,12 +35,20 @@ int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
     BDRVQcowState *s = bs->opaque;
     int new_l1_size2, ret, i;
     uint64_t *new_l1_table;
+    int64_t old_l1_table_offset, old_l1_size;
     int64_t new_l1_table_offset, new_l1_size;
     uint8_t data[12];
 
     if (min_size <= s->l1_size)
         return 0;
 
+    /* Do a sanity check on min_size before trying to calculate new_l1_size
+     * (this prevents overflows during the while loop for the calculation of
+     * new_l1_size) */
+    if (min_size > INT_MAX / sizeof(uint64_t)) {
+        return -EFBIG;
+    }
+
     if (exact_size) {
         new_l1_size = min_size;
     } else {
@@ -54,7 +62,7 @@ int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
         }
     }
 
-    if (new_l1_size > INT_MAX) {
+    if (new_l1_size > INT_MAX / sizeof(uint64_t)) {
         return -EFBIG;
     }
 
@@ -64,14 +72,20 @@ int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
 #endif
 
     new_l1_size2 = sizeof(uint64_t) * new_l1_size;
-    new_l1_table = g_malloc0(align_offset(new_l1_size2, 512));
+    new_l1_table = qemu_try_blockalign(bs->file,
+                                       align_offset(new_l1_size2, 512));
+    if (new_l1_table == NULL) {
+        return -ENOMEM;
+    }
+    memset(new_l1_table, 0, align_offset(new_l1_size2, 512));
+
     memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
 
     /* write new table (align to cluster) */
     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
     new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
     if (new_l1_table_offset < 0) {
-        g_free(new_l1_table);
+        qemu_vfree(new_l1_table);
         return new_l1_table_offset;
     }
 
@@ -80,6 +94,14 @@ int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
         goto fail;
     }
 
+    /* the L1 position has not yet been updated, so these clusters must
+     * indeed be completely free */
+    ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset,
+                                        new_l1_size2);
+    if (ret < 0) {
+        goto fail;
+    }
+
     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
     for(i = 0; i < s->l1_size; i++)
         new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
@@ -92,20 +114,22 @@ int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
     /* set new table */
     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
     cpu_to_be32w((uint32_t*)data, new_l1_size);
-    cpu_to_be64wu((uint64_t*)(data + 4), new_l1_table_offset);
+    stq_be_p(data + 4, new_l1_table_offset);
     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
     if (ret < 0) {
         goto fail;
     }
-    g_free(s->l1_table);
-    qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t),
-                        QCOW2_DISCARD_OTHER);
+    qemu_vfree(s->l1_table);
+    old_l1_table_offset = s->l1_table_offset;
     s->l1_table_offset = new_l1_table_offset;
     s->l1_table = new_l1_table;
+    old_l1_size = s->l1_size;
     s->l1_size = new_l1_size;
+    qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t),
+                        QCOW2_DISCARD_OTHER);
     return 0;
  fail:
-    g_free(new_l1_table);
+    qemu_vfree(new_l1_table);
     qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2,
                         QCOW2_DISCARD_OTHER);
     return ret;
@@ -137,7 +161,7 @@ static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
  * and we really don't want bdrv_pread to perform a read-modify-write)
  */
 #define L1_ENTRIES_PER_SECTOR (512 / 8)
-static int write_l1_entry(BlockDriverState *bs, int l1_index)
+int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
 {
     BDRVQcowState *s = bs->opaque;
     uint64_t buf[L1_ENTRIES_PER_SECTOR];
@@ -149,6 +173,12 @@ static int write_l1_entry(BlockDriverState *bs, int l1_index)
         buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
     }
 
+    ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
+            s->l1_table_offset + 8 * l1_start_index, sizeof(buf));
+    if (ret < 0) {
+        return ret;
+    }
+
     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
     ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index,
         buf, sizeof(buf));
@@ -173,7 +203,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
 {
     BDRVQcowState *s = bs->opaque;
     uint64_t old_l2_offset;
-    uint64_t *l2_table;
+    uint64_t *l2_table = NULL;
     int64_t l2_offset;
     int ret;
 
@@ -185,7 +215,8 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
 
     l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
     if (l2_offset < 0) {
-        return l2_offset;
+        ret = l2_offset;
+        goto fail;
     }
 
     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
@@ -198,7 +229,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
     trace_qcow2_l2_allocate_get_empty(bs, l1_index);
     ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
     if (ret < 0) {
-        return ret;
+        goto fail;
     }
 
     l2_table = *table;
@@ -239,7 +270,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
     /* update the L1 entry */
     trace_qcow2_l2_allocate_write_l1(bs, l1_index);
     s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
-    ret = write_l1_entry(bs, l1_index);
+    ret = qcow2_write_l1_entry(bs, l1_index);
     if (ret < 0) {
         goto fail;
     }
@@ -250,8 +281,14 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
 
 fail:
     trace_qcow2_l2_allocate_done(bs, l1_index, ret);
-    qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
+    if (l2_table != NULL) {
+        qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
+    }
     s->l1_table[l1_index] = old_l2_offset;
+    if (l2_offset > 0) {
+        qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
+                            QCOW2_DISCARD_ALWAYS);
+    }
     return ret;
 }
 
@@ -263,23 +300,26 @@ fail:
  * cluster which may require a different handling)
  */
 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
-        uint64_t *l2_table, uint64_t start, uint64_t stop_flags)
+        uint64_t *l2_table, uint64_t stop_flags)
 {
     int i;
-    uint64_t mask = stop_flags | L2E_OFFSET_MASK;
-    uint64_t offset = be64_to_cpu(l2_table[0]) & mask;
+    uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
+    uint64_t first_entry = be64_to_cpu(l2_table[0]);
+    uint64_t offset = first_entry & mask;
 
     if (!offset)
         return 0;
 
-    for (i = start; i < start + nb_clusters; i++) {
+    assert(qcow2_get_cluster_type(first_entry) != QCOW2_CLUSTER_COMPRESSED);
+
+    for (i = 0; i < nb_clusters; i++) {
         uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask;
         if (offset + (uint64_t) i * cluster_size != l2_entry) {
             break;
         }
     }
 
-       return (i - start);
+       return i;
 }
 
 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
@@ -332,27 +372,26 @@ static int coroutine_fn copy_sectors(BlockDriverState *bs,
     struct iovec iov;
     int n, ret;
 
-    /*
-     * If this is the last cluster and it is only partially used, we must only
-     * copy until the end of the image, or bdrv_check_request will fail for the
-     * bdrv_read/write calls below.
-     */
-    if (start_sect + n_end > bs->total_sectors) {
-        n_end = bs->total_sectors - start_sect;
-    }
-
     n = n_end - n_start;
     if (n <= 0) {
         return 0;
     }
 
     iov.iov_len = n * BDRV_SECTOR_SIZE;
-    iov.iov_base = qemu_blockalign(bs, iov.iov_len);
+    iov.iov_base = qemu_try_blockalign(bs, iov.iov_len);
+    if (iov.iov_base == NULL) {
+        return -ENOMEM;
+    }
 
     qemu_iovec_init_external(&qiov, &iov, 1);
 
     BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
 
+    if (!bs->drv) {
+        ret = -ENOMEDIUM;
+        goto out;
+    }
+
     /* Call .bdrv_co_readv() directly instead of using the public block-layer
      * interface.  This avoids double I/O throttling and request tracking,
      * which can lead to deadlock when block layer copy-on-read is enabled.
@@ -368,6 +407,12 @@ static int coroutine_fn copy_sectors(BlockDriverState *bs,
                         &s->aes_encrypt_key);
     }
 
+    ret = qcow2_pre_write_overlap_check(bs, 0,
+            cluster_offset + n_start * BDRV_SECTOR_SIZE, n * BDRV_SECTOR_SIZE);
+    if (ret < 0) {
+        goto out;
+    }
+
     BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
     ret = bdrv_co_writev(bs->file, (cluster_offset >> 9) + n_start, n, &qiov);
     if (ret < 0) {
@@ -463,11 +508,11 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
         break;
     case QCOW2_CLUSTER_ZERO:
         if (s->qcow_version < 3) {
+            qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
             return -EIO;
         }
         c = count_contiguous_clusters(nb_clusters, s->cluster_size,
-                &l2_table[l2_index], 0,
-                QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO);
+                &l2_table[l2_index], QCOW_OFLAG_ZERO);
         *cluster_offset = 0;
         break;
     case QCOW2_CLUSTER_UNALLOCATED:
@@ -478,8 +523,7 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
     case QCOW2_CLUSTER_NORMAL:
         /* how many allocated clusters ? */
         c = count_contiguous_clusters(nb_clusters, s->cluster_size,
-                &l2_table[l2_index], 0,
-                QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO);
+                &l2_table[l2_index], QCOW_OFLAG_ZERO);
         *cluster_offset &= L2E_OFFSET_MASK;
         break;
     default:
@@ -667,7 +711,11 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
     trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
     assert(m->nb_clusters > 0);
 
-    old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t));
+    old_cluster = g_try_malloc(m->nb_clusters * sizeof(uint64_t));
+    if (old_cluster == NULL) {
+        ret = -ENOMEM;
+        goto err;
+    }
 
     /* copy content of unmodified sectors */
     ret = perform_cow(bs, m, &m->cow_start);
@@ -695,6 +743,7 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
     }
     qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
 
+    assert(l2_index + m->nb_clusters <= s->l2_size);
     for (i = 0; i < m->nb_clusters; i++) {
         /* if two concurrent writes happen to the same unallocated cluster
         * each write allocates separate cluster and writes data concurrently.
@@ -908,7 +957,7 @@ static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
         /* We keep all QCOW_OFLAG_COPIED clusters */
         keep_clusters =
             count_contiguous_clusters(nb_clusters, s->cluster_size,
-                                      &l2_table[l2_index], 0,
+                                      &l2_table[l2_index],
                                       QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
         assert(keep_clusters <= nb_clusters);
 
@@ -1150,7 +1199,7 @@ fail:
  * Return 0 on success and -errno in error cases
  */
 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
-    int n_start, int n_end, int *num, uint64_t *host_offset, QCowL2Meta **m)
+    int *num, uint64_t *host_offset, QCowL2Meta **m)
 {
     BDRVQcowState *s = bs->opaque;
     uint64_t start, remaining;
@@ -1158,15 +1207,13 @@ int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
     uint64_t cur_bytes;
     int ret;
 
-    trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset,
-                                      n_start, n_end);
+    trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *num);
 
-    assert(n_start * BDRV_SECTOR_SIZE == offset_into_cluster(s, offset));
-    offset = start_of_cluster(s, offset);
+    assert((offset & ~BDRV_SECTOR_MASK) == 0);
 
 again:
-    start = offset + (n_start << BDRV_SECTOR_BITS);
-    remaining = (n_end - n_start) << BDRV_SECTOR_BITS;
+    start = offset;
+    remaining = *num << BDRV_SECTOR_BITS;
     cluster_offset = 0;
     *host_offset = 0;
     cur_bytes = 0;
@@ -1252,7 +1299,7 @@ again:
         }
     }
 
-    *num = (n_end - n_start) - (remaining >> BDRV_SECTOR_BITS);
+    *num -= remaining >> BDRV_SECTOR_BITS;
     assert(*num > 0);
     assert(*host_offset != 0);
 
@@ -1317,7 +1364,7 @@ int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
  * clusters.
  */
 static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
-    unsigned int nb_clusters)
+    unsigned int nb_clusters, enum qcow2_discard_type type)
 {
     BDRVQcowState *s = bs->opaque;
     uint64_t *l2_table;
@@ -1334,19 +1381,47 @@ static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
 
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t old_offset;
+        uint64_t old_l2_entry;
 
-        old_offset = be64_to_cpu(l2_table[l2_index + i]);
-        if ((old_offset & L2E_OFFSET_MASK) == 0) {
-            continue;
+        old_l2_entry = be64_to_cpu(l2_table[l2_index + i]);
+
+        /*
+         * Make sure that a discarded area reads back as zeroes for v3 images
+         * (we cannot do it for v2 without actually writing a zero-filled
+         * buffer). We can skip the operation if the cluster is already marked
+         * as zero, or if it's unallocated and we don't have a backing file.
+         *
+         * TODO We might want to use bdrv_get_block_status(bs) here, but we're
+         * holding s->lock, so that doesn't work today.
+         */
+        switch (qcow2_get_cluster_type(old_l2_entry)) {
+            case QCOW2_CLUSTER_UNALLOCATED:
+                if (!bs->backing_hd) {
+                    continue;
+                }
+                break;
+
+            case QCOW2_CLUSTER_ZERO:
+                continue;
+
+            case QCOW2_CLUSTER_NORMAL:
+            case QCOW2_CLUSTER_COMPRESSED:
+                break;
+
+            default:
+                abort();
         }
 
         /* First remove L2 entries */
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
-        l2_table[l2_index + i] = cpu_to_be64(0);
+        if (s->qcow_version >= 3) {
+            l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
+        } else {
+            l2_table[l2_index + i] = cpu_to_be64(0);
+        }
 
         /* Then decrease the refcount */
-        qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
+        qcow2_free_any_clusters(bs, old_l2_entry, 1, type);
     }
 
     ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
@@ -1358,7 +1433,7 @@ static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
 }
 
 int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
-    int nb_sectors)
+    int nb_sectors, enum qcow2_discard_type type)
 {
     BDRVQcowState *s = bs->opaque;
     uint64_t end_offset;
@@ -1369,7 +1444,7 @@ int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
 
     /* Round start up and end down */
     offset = align_offset(offset, s->cluster_size);
-    end_offset &= ~(s->cluster_size - 1);
+    end_offset = start_of_cluster(s, end_offset);
 
     if (offset > end_offset) {
         return 0;
@@ -1381,7 +1456,7 @@ int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
 
     /* Each L2 table is handled by its own loop iteration */
     while (nb_clusters > 0) {
-        ret = discard_single_l2(bs, offset, nb_clusters);
+        ret = discard_single_l2(bs, offset, nb_clusters, type);
         if (ret < 0) {
             goto fail;
         }
@@ -1476,3 +1551,262 @@ fail:
 
     return ret;
 }
+
+/*
+ * Expands all zero clusters in a specific L1 table (or deallocates them, for
+ * non-backed non-pre-allocated zero clusters).
+ *
+ * expanded_clusters is a bitmap where every bit corresponds to one cluster in
+ * the image file; a bit gets set if the corresponding cluster has been used for
+ * zero expansion (i.e., has been filled with zeroes and is referenced from an
+ * L2 table). nb_clusters contains the total cluster count of the image file,
+ * i.e., the number of bits in expanded_clusters.
+ */
+static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
+                                      int l1_size, uint8_t **expanded_clusters,
+                                      uint64_t *nb_clusters)
+{
+    BDRVQcowState *s = bs->opaque;
+    bool is_active_l1 = (l1_table == s->l1_table);
+    uint64_t *l2_table = NULL;
+    int ret;
+    int i, j;
+
+    if (!is_active_l1) {
+        /* inactive L2 tables require a buffer to be stored in when loading
+         * them from disk */
+        l2_table = qemu_try_blockalign(bs->file, s->cluster_size);
+        if (l2_table == NULL) {
+            return -ENOMEM;
+        }
+    }
+
+    for (i = 0; i < l1_size; i++) {
+        uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK;
+        bool l2_dirty = false;
+
+        if (!l2_offset) {
+            /* unallocated */
+            continue;
+        }
+
+        if (is_active_l1) {
+            /* get active L2 tables from cache */
+            ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
+                    (void **)&l2_table);
+        } else {
+            /* load inactive L2 tables from disk */
+            ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE,
+                    (void *)l2_table, s->cluster_sectors);
+        }
+        if (ret < 0) {
+            goto fail;
+        }
+
+        for (j = 0; j < s->l2_size; j++) {
+            uint64_t l2_entry = be64_to_cpu(l2_table[j]);
+            int64_t offset = l2_entry & L2E_OFFSET_MASK, cluster_index;
+            int cluster_type = qcow2_get_cluster_type(l2_entry);
+            bool preallocated = offset != 0;
+
+            if (cluster_type == QCOW2_CLUSTER_NORMAL) {
+                cluster_index = offset >> s->cluster_bits;
+                assert((cluster_index >= 0) && (cluster_index < *nb_clusters));
+                if ((*expanded_clusters)[cluster_index / 8] &
+                    (1 << (cluster_index % 8))) {
+                    /* Probably a shared L2 table; this cluster was a zero
+                     * cluster which has been expanded, its refcount
+                     * therefore most likely requires an update. */
+                    ret = qcow2_update_cluster_refcount(bs, cluster_index, 1,
+                                                        QCOW2_DISCARD_NEVER);
+                    if (ret < 0) {
+                        goto fail;
+                    }
+                    /* Since we just increased the refcount, the COPIED flag may
+                     * no longer be set. */
+                    l2_table[j] = cpu_to_be64(l2_entry & ~QCOW_OFLAG_COPIED);
+                    l2_dirty = true;
+                }
+                continue;
+            }
+            else if (qcow2_get_cluster_type(l2_entry) != QCOW2_CLUSTER_ZERO) {
+                continue;
+            }
+
+            if (!preallocated) {
+                if (!bs->backing_hd) {
+                    /* not backed; therefore we can simply deallocate the
+                     * cluster */
+                    l2_table[j] = 0;
+                    l2_dirty = true;
+                    continue;
+                }
+
+                offset = qcow2_alloc_clusters(bs, s->cluster_size);
+                if (offset < 0) {
+                    ret = offset;
+                    goto fail;
+                }
+            }
+
+            ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
+            if (ret < 0) {
+                if (!preallocated) {
+                    qcow2_free_clusters(bs, offset, s->cluster_size,
+                                        QCOW2_DISCARD_ALWAYS);
+                }
+                goto fail;
+            }
+
+            ret = bdrv_write_zeroes(bs->file, offset / BDRV_SECTOR_SIZE,
+                                    s->cluster_sectors, 0);
+            if (ret < 0) {
+                if (!preallocated) {
+                    qcow2_free_clusters(bs, offset, s->cluster_size,
+                                        QCOW2_DISCARD_ALWAYS);
+                }
+                goto fail;
+            }
+
+            l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED);
+            l2_dirty = true;
+
+            cluster_index = offset >> s->cluster_bits;
+
+            if (cluster_index >= *nb_clusters) {
+                uint64_t old_bitmap_size = (*nb_clusters + 7) / 8;
+                uint64_t new_bitmap_size;
+                /* The offset may lie beyond the old end of the underlying image
+                 * file for growable files only */
+                assert(bs->file->growable);
+                *nb_clusters = size_to_clusters(s, bs->file->total_sectors *
+                                                BDRV_SECTOR_SIZE);
+                new_bitmap_size = (*nb_clusters + 7) / 8;
+                *expanded_clusters = g_realloc(*expanded_clusters,
+                                               new_bitmap_size);
+                /* clear the newly allocated space */
+                memset(&(*expanded_clusters)[old_bitmap_size], 0,
+                       new_bitmap_size - old_bitmap_size);
+            }
+
+            assert((cluster_index >= 0) && (cluster_index < *nb_clusters));
+            (*expanded_clusters)[cluster_index / 8] |= 1 << (cluster_index % 8);
+        }
+
+        if (is_active_l1) {
+            if (l2_dirty) {
+                qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
+                qcow2_cache_depends_on_flush(s->l2_table_cache);
+            }
+            ret = qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
+            if (ret < 0) {
+                l2_table = NULL;
+                goto fail;
+            }
+        } else {
+            if (l2_dirty) {
+                ret = qcow2_pre_write_overlap_check(bs,
+                        QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset,
+                        s->cluster_size);
+                if (ret < 0) {
+                    goto fail;
+                }
+
+                ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE,
+                        (void *)l2_table, s->cluster_sectors);
+                if (ret < 0) {
+                    goto fail;
+                }
+            }
+        }
+    }
+
+    ret = 0;
+
+fail:
+    if (l2_table) {
+        if (!is_active_l1) {
+            qemu_vfree(l2_table);
+        } else {
+            if (ret < 0) {
+                qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
+            } else {
+                ret = qcow2_cache_put(bs, s->l2_table_cache,
+                        (void **)&l2_table);
+            }
+        }
+    }
+    return ret;
+}
+
+/*
+ * For backed images, expands all zero clusters on the image. For non-backed
+ * images, deallocates all non-pre-allocated zero clusters (and claims the
+ * allocation for pre-allocated ones). This is important for downgrading to a
+ * qcow2 version which doesn't yet support metadata zero clusters.
+ */
+int qcow2_expand_zero_clusters(BlockDriverState *bs)
+{
+    BDRVQcowState *s = bs->opaque;
+    uint64_t *l1_table = NULL;
+    uint64_t nb_clusters;
+    uint8_t *expanded_clusters;
+    int ret;
+    int i, j;
+
+    nb_clusters = size_to_clusters(s, bs->file->total_sectors *
+                                   BDRV_SECTOR_SIZE);
+    expanded_clusters = g_try_malloc0((nb_clusters + 7) / 8);
+    if (expanded_clusters == NULL) {
+        ret = -ENOMEM;
+        goto fail;
+    }
+
+    ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size,
+                                     &expanded_clusters, &nb_clusters);
+    if (ret < 0) {
+        goto fail;
+    }
+
+    /* Inactive L1 tables may point to active L2 tables - therefore it is
+     * necessary to flush the L2 table cache before trying to access the L2
+     * tables pointed to by inactive L1 entries (else we might try to expand
+     * zero clusters that have already been expanded); furthermore, it is also
+     * necessary to empty the L2 table cache, since it may contain tables which
+     * are now going to be modified directly on disk, bypassing the cache.
+     * qcow2_cache_empty() does both for us. */
+    ret = qcow2_cache_empty(bs, s->l2_table_cache);
+    if (ret < 0) {
+        goto fail;
+    }
+
+    for (i = 0; i < s->nb_snapshots; i++) {
+        int l1_sectors = (s->snapshots[i].l1_size * sizeof(uint64_t) +
+                BDRV_SECTOR_SIZE - 1) / BDRV_SECTOR_SIZE;
+
+        l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE);
+
+        ret = bdrv_read(bs->file, s->snapshots[i].l1_table_offset /
+                BDRV_SECTOR_SIZE, (void *)l1_table, l1_sectors);
+        if (ret < 0) {
+            goto fail;
+        }
+
+        for (j = 0; j < s->snapshots[i].l1_size; j++) {
+            be64_to_cpus(&l1_table[j]);
+        }
+
+        ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size,
+                                         &expanded_clusters, &nb_clusters);
+        if (ret < 0) {
+            goto fail;
+        }
+    }
+
+    ret = 0;
+
+fail:
+    g_free(expanded_clusters);
+    g_free(l1_table);
+    return ret;
+}
This page took 0.043226 seconds and 4 git commands to generate.