[Spice-devel] [spice-server PATCH 2/8] image_encoders: check shared_dict before accessing it
Uri Lublin
uril at redhat.com
Wed Oct 26 15:18:53 UTC 2016
On 10/17/2016 01:29 PM, Frediano Ziglio wrote:
>>
>> In both image_encoders_restore_glz_dictionary() and
>> image_encoders_get_glz_dictionary() shared-dict may
>> be NULL if size is too large, and the server gets
>> size from the network.
>>
>> Both functions end up calling glz_enc_dictionary_create()
>> that calls glz_dictionary_window_create() where size is
>> checked.
>>
>> Found by coverity.
>>
>> Signed-off-by: Uri Lublin <uril at redhat.com>
>> ---
>> server/image-encoders.c | 6 ++++--
>> 1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/server/image-encoders.c b/server/image-encoders.c
>> index 39aca6c..9dfabd6 100644
>> --- a/server/image-encoders.c
>> +++ b/server/image-encoders.c
>> @@ -746,12 +746,13 @@ gboolean
>> image_encoders_get_glz_dictionary(ImageEncoders *enc,
>> shared_dict->refs++;
>> } else {
>> shared_dict = create_glz_dictionary(enc, client, id, window_size);
>> + spice_return_val_if_fail(shared_dict != NULL, FALSE);
>> glz_dictionary_list = g_list_prepend(glz_dictionary_list,
>> shared_dict);
>> }
>>
>> pthread_mutex_unlock(&glz_dictionary_list_lock);
>> enc->glz_dict = shared_dict;
>> - return shared_dict != NULL;
>> + return TRUE;
>> }
>>
>> static GlzSharedDictionary *restore_glz_dictionary(ImageEncoders *enc,
>> @@ -782,12 +783,13 @@ gboolean
>> image_encoders_restore_glz_dictionary(ImageEncoders *enc,
>> shared_dict->refs++;
>> } else {
>> shared_dict = restore_glz_dictionary(enc, client, id, restore_data);
>> + spice_return_val_if_fail(shared_dict != NULL, FALSE);
>> glz_dictionary_list = g_list_prepend(glz_dictionary_list,
>> shared_dict);
>> }
>>
>> pthread_mutex_unlock(&glz_dictionary_list_lock);
>> enc->glz_dict = shared_dict;
>> - return shared_dict != NULL;
>> + return TRUE;
>> }
>>
>> gboolean image_encoders_glz_create(ImageEncoders *enc, uint8_t id)
>
> Note that you are creating dead locks.
Yes, your are right.
I'll send a replacement patch that fixes that.
> Beside that is not clear what the change could cause to the
> upper layer.
Upper layer should already handle a case where these functions
return NULL.
> I think the actual logic is supposing that dictionary creation (like
> memory allocation) is not failing.
Since it depends on a value that comes from the client that
assumption should not be made. Alternatively we
can specifically check the number early
Thanks,
Uri.
More information about the Spice-devel
mailing list