diff mbox

[v4,2/4] qcow2: fix offset overflow in qcow2_alloc_clusters_at()

Message ID bfc390e142739e08943fad1014109869f4f9a7e8.1390445921.git.hutao@cn.fujitsu.com
State New
Headers show

Commit Message

Hu Tao Jan. 23, 2014, 3:04 a.m. UTC
When cluster size is big enough it can lead offset overflow
in qcow2_alloc_clusters_at(). This patch fixes it.

The allocation each time is stopped at L2 table boundary
(see handle_alloc()), so the possible maximum bytes could be

  2^(cluster_bits - 3 + cluster_bits)

so int is safe for cluster_bits<=17, unsafe otherwise.

Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Hu Tao <hutao@cn.fujitsu.com>
---
 block/qcow2-refcount.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

Comments

Benoît Canet Jan. 23, 2014, 5:13 p.m. UTC | #1
Le Thursday 23 Jan 2014 à 11:04:06 (+0800), Hu Tao a écrit :
> When cluster size is big enough it can lead offset overflow

Maybe "it can lead to an offset overflow in"
> in qcow2_alloc_clusters_at(). This patch fixes it.
> 
> The allocation each time is stopped at L2 table boundary
"The allocation is stopped each time at"

> (see handle_alloc()), so the possible maximum bytes could be
> 
>   2^(cluster_bits - 3 + cluster_bits)

So if understand cluster_bits - 3 is used to compute the number of entry by L2
and the additional cluster_bits is to take into account each clusters referenced
by the L2 entries.
It makes sense.

> 
> so int is safe for cluster_bits<=17, unsafe otherwise.
> 
> Reviewed-by: Max Reitz <mreitz@redhat.com>
> Signed-off-by: Hu Tao <hutao@cn.fujitsu.com>
> ---
>  block/qcow2-refcount.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/block/qcow2-refcount.c b/block/qcow2-refcount.c
> index c974abe..8712d8b 100644
> --- a/block/qcow2-refcount.c
> +++ b/block/qcow2-refcount.c
> @@ -676,7 +676,13 @@ int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
>      BDRVQcowState *s = bs->opaque;
>      uint64_t cluster_index;
>      uint64_t old_free_cluster_index;
> -    int i, refcount, ret;
> +    uint64_t i;
> +    int refcount, ret;
> +


> +    assert(nb_clusters >= 0);
> +    if (nb_clusters == 0) {
> +        return 0;
> +    }
^
Adding a a line on the commit message about this assertion and chunk of code
would be helpful.

Best regards

Benoît

>  
>      /* Check how many clusters there are free */
>      cluster_index = offset >> s->cluster_bits;
> -- 
> 1.8.5.2.229.g4448466
> 
>
Hu Tao Jan. 24, 2014, 10:01 a.m. UTC | #2
On Thu, Jan 23, 2014 at 06:13:48PM +0100, Benoît Canet wrote:
> Le Thursday 23 Jan 2014 à 11:04:06 (+0800), Hu Tao a écrit :
> > When cluster size is big enough it can lead offset overflow
> 
> Maybe "it can lead to an offset overflow in"
> > in qcow2_alloc_clusters_at(). This patch fixes it.

Sure.

> > 
> > The allocation each time is stopped at L2 table boundary
> "The allocation is stopped each time at"

Sure.

> 
> > (see handle_alloc()), so the possible maximum bytes could be
> > 
> >   2^(cluster_bits - 3 + cluster_bits)
> 
> So if understand cluster_bits - 3 is used to compute the number of entry by L2
> and the additional cluster_bits is to take into account each clusters referenced
> by the L2 entries.

Exactly. This is clearer than just one calculation. Do you mind I
put the sentence in commit message?

> It makes sense.
> 
> > 
> > so int is safe for cluster_bits<=17, unsafe otherwise.
> > 
> > Reviewed-by: Max Reitz <mreitz@redhat.com>
> > Signed-off-by: Hu Tao <hutao@cn.fujitsu.com>
> > ---
> >  block/qcow2-refcount.c | 8 +++++++-
> >  1 file changed, 7 insertions(+), 1 deletion(-)
> > 
> > diff --git a/block/qcow2-refcount.c b/block/qcow2-refcount.c
> > index c974abe..8712d8b 100644
> > --- a/block/qcow2-refcount.c
> > +++ b/block/qcow2-refcount.c
> > @@ -676,7 +676,13 @@ int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
> >      BDRVQcowState *s = bs->opaque;
> >      uint64_t cluster_index;
> >      uint64_t old_free_cluster_index;
> > -    int i, refcount, ret;
> > +    uint64_t i;
> > +    int refcount, ret;
> > +
> 
> 
> > +    assert(nb_clusters >= 0);
> > +    if (nb_clusters == 0) {
> > +        return 0;
> > +    }
> ^
> Adding a a line on the commit message about this assertion and chunk of code
> would be helpful.

How about

  Add an assert to guard the comparation between signed and unsigned

?

> 
> Best regards
> 
> Benoît
> 
> >  
> >      /* Check how many clusters there are free */
> >      cluster_index = offset >> s->cluster_bits;
> > -- 
> > 1.8.5.2.229.g4448466
> > 
> >
Benoît Canet Jan. 24, 2014, 3:22 p.m. UTC | #3
Le Friday 24 Jan 2014 à 18:01:20 (+0800), Hu Tao a écrit :
> On Thu, Jan 23, 2014 at 06:13:48PM +0100, Benoît Canet wrote:
> > Le Thursday 23 Jan 2014 à 11:04:06 (+0800), Hu Tao a écrit :
> > > When cluster size is big enough it can lead offset overflow
> > 
> > Maybe "it can lead to an offset overflow in"
> > > in qcow2_alloc_clusters_at(). This patch fixes it.
> 
> Sure.
> 
> > > 
> > > The allocation each time is stopped at L2 table boundary
> > "The allocation is stopped each time at"
> 
> Sure.
> 
> > 
> > > (see handle_alloc()), so the possible maximum bytes could be
> > > 
> > >   2^(cluster_bits - 3 + cluster_bits)
> > 
> > So if understand cluster_bits - 3 is used to compute the number of entry by L2
> > and the additional cluster_bits is to take into account each clusters referenced
> > by the L2 entries.
> 
> Exactly. This is clearer than just one calculation. Do you mind I
> put the sentence in commit message?
yes It would be fine for future reference.

> 
> > It makes sense.
> > 
> > > 
> > > so int is safe for cluster_bits<=17, unsafe otherwise.
> > > 
> > > Reviewed-by: Max Reitz <mreitz@redhat.com>
> > > Signed-off-by: Hu Tao <hutao@cn.fujitsu.com>
> > > ---
> > >  block/qcow2-refcount.c | 8 +++++++-
> > >  1 file changed, 7 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/block/qcow2-refcount.c b/block/qcow2-refcount.c
> > > index c974abe..8712d8b 100644
> > > --- a/block/qcow2-refcount.c
> > > +++ b/block/qcow2-refcount.c
> > > @@ -676,7 +676,13 @@ int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
> > >      BDRVQcowState *s = bs->opaque;
> > >      uint64_t cluster_index;
> > >      uint64_t old_free_cluster_index;
> > > -    int i, refcount, ret;
> > > +    uint64_t i;
> > > +    int refcount, ret;
> > > +
> > 
> > 
> > > +    assert(nb_clusters >= 0);
> > > +    if (nb_clusters == 0) {
> > > +        return 0;
> > > +    }
> > ^
> > Adding a a line on the commit message about this assertion and chunk of code
> > would be helpful.
> 
> How about
> 
>   Add an assert to guard the comparation between signed and unsigned
yes
> 
> ?
> 
> > 
> > Best regards
> > 
> > Benoît
> > 
> > >  
> > >      /* Check how many clusters there are free */
> > >      cluster_index = offset >> s->cluster_bits;
> > > -- 
> > > 1.8.5.2.229.g4448466
> > > 
> > > 
>
diff mbox

Patch

diff --git a/block/qcow2-refcount.c b/block/qcow2-refcount.c
index c974abe..8712d8b 100644
--- a/block/qcow2-refcount.c
+++ b/block/qcow2-refcount.c
@@ -676,7 +676,13 @@  int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
     BDRVQcowState *s = bs->opaque;
     uint64_t cluster_index;
     uint64_t old_free_cluster_index;
-    int i, refcount, ret;
+    uint64_t i;
+    int refcount, ret;
+
+    assert(nb_clusters >= 0);
+    if (nb_clusters == 0) {
+        return 0;
+    }
 
     /* Check how many clusters there are free */
     cluster_index = offset >> s->cluster_bits;