Re: [Qemu-devel] [PATCH] qapi: change QmpInputVisitor to QSLIST

[Markus Armbruster]

Eric Blake <address@hidden> writes:

> On 07/06/2016 06:43 AM, Paolo Bonzini wrote:
>> This saves a lot of memory compared to a statically-sized array.
>> 
>> Signed-off-by: Paolo Bonzini <address@hidden>
>> ---
>>  qapi/qmp-input-visitor.c | 53 
>> ++++++++++++++++++++++++------------------------
>>  1 file changed, 26 insertions(+), 27 deletions(-)
>> 
>
>> @@ -99,17 +100,10 @@ static const QListEntry *qmp_input_push(QmpInputVisitor 
>> *qiv, QObject *obj,
>>                                          Error **errp)
>>  {
>>      GHashTable *h;
>> -    StackObject *tos = &qiv->stack[qiv->nb_stack];
>> +    StackObject *tos = g_new0(StackObject, 1);
>>  
>>      assert(obj);
>> -    if (qiv->nb_stack >= QIV_STACK_SIZE) {
>
> You should also delete QIV_STACK_SIZE as it is now unused.

Actually, you should either prove that untrusted input still cannot make
us allocated unbounded amounts of memory, or bring the limit right back.

>> @@ -127,9 +121,7 @@ static const QListEntry *qmp_input_push(QmpInputVisitor 
>> *qiv, QObject *obj,
>>  static void qmp_input_check_struct(Visitor *v, Error **errp)
>>  {
>>      QmpInputVisitor *qiv = to_qiv(v);
>> -    StackObject *tos = &qiv->stack[qiv->nb_stack - 1];
>> -
>> -    assert(qiv->nb_stack > 0);
>> +    StackObject *tos = QSLIST_FIRST(&qiv->stack);
>
> Does QSLIST_FIRST() properly crash if the list is empty, or do we need

It returns null.

> to add an assert(tos) to replace the assertion on nb_stack being non-zero?

We do need to.

> Otherwise looking reasonable; looking forward to v2.