On 2014-11-11 at 09:37, Max Reitz wrote:
On 2014-11-10 at 23:49, Eric Blake wrote:
On 11/10/2014 06:45 AM, Max Reitz wrote:
Add a helper function which correctly calculates the byte size of a
refcount array for any refcount order, and use that function.
Signed-off-by: Max Reitz <mre...@redhat.com>
---
block/qcow2-refcount.c | 39 ++++++++++++++++++++++++++++-----------
1 file changed, 28 insertions(+), 11 deletions(-)
diff --git a/block/qcow2-refcount.c b/block/qcow2-refcount.c
index 16652da..cfb4807 100644
--- a/block/qcow2-refcount.c
+++ b/block/qcow2-refcount.c
@@ -1132,6 +1132,20 @@ fail:
/* refcount checking functions */
+static size_t refcount_array_byte_size(BDRVQcowState *s,
uint64_t entries)
+{
+ if (s->refcount_order < 3) {
+ /* sub-byte width */
+ int shift = 3 - s->refcount_order;
+ return (entries + (1 << shift) - 1) >> shift;
+ } else if (s->refcount_order == 3) {
+ /* byte width */
+ return entries;
+ } else {
+ /* multiple bytes wide */
+ return entries << (s->refcount_order - 3);
+ }
A comment proving why this can't overflow might be nice (if I analyzed
correctly, entries will be computed by file size / clusters, and in the
worst case, the smallest cluster and largest refcount_order results in
'(size >> 9) << (6 - 3)' which is still safe).
Yes, will do.
@@ -1161,12 +1175,13 @@ static int inc_refcounts(BlockDriverState *bs,
k = cluster_offset >> s->cluster_bits;
if (k >= *refcount_table_size) {
int64_t old_refcount_table_size = *refcount_table_size;
+ size_t new_byte_size;
void *new_refcount_table;
*refcount_table_size = k + 1;
- new_refcount_table = g_try_realloc(*refcount_table,
- *refcount_table_size *
- s->refcount_bits / 8);
+ new_byte_size = refcount_array_byte_size(s,
*refcount_table_size);
+
+ new_refcount_table = g_try_realloc(*refcount_table,
new_byte_size);
Yay - this addresses one of my possible overflow comments on 6/21.
I wonder if the series would have less churn if you rearranged this
patch to come before 6/21.
Why not, I'll add an __attribute__((used)) to it (which should be fine
for the duration of a single patch).
I'm not sure why I thought that might be necessary. Of course it isn't.
Max
