Le 06/12/2018 à 17:40, Laurent Vivier a écrit :
> Le 06/12/2018 à 16:18, Eric Blake a écrit :
>> POSIX states that the value of endptr is unspecified if strtol()
>> fails with EINVAL due to an invalid base argument. Since none of
>> the callers to check_strtox_error() initialized endptr, we could
>> end up propagating uninitialized data back to a caller on error.
>> However, passing an out-of-range base is already a sign of poor
>> programming, so let's just assert that base is in range, at which
>> point check_strtox_error() can be tightened to assert that it is
>> receiving an initialized ep that points somewhere within the
>> caller's original string, regardless of whether strto*() succeeded
>> or failed with ERANGE.
>>
>> Reported-by: Vladimir Sementsov-Ogievskiy <vsement...@virtuozzo.com>
>> Signed-off-by: Eric Blake <ebl...@redhat.com>
>> ---
>>
>> Also tested that this does not negatively impact David's pending
>> additions of qemu_strtod{,_finite}(). Thus:
>> Based-on: <20181121164421.20780-1-da...@redhat.com>
>>
>> util/cutils.c | 8 ++++++++
>> 1 file changed, 8 insertions(+)
>>
>> diff --git a/util/cutils.c b/util/cutils.c
>> index 91930d1bbeb..e098debdc0c 100644
>> --- a/util/cutils.c
>> +++ b/util/cutils.c
>> @@ -278,6 +278,7 @@ int qemu_strtosz_metric(const char *nptr, const char
>> **end, uint64_t *result)
>> static int check_strtox_error(const char *nptr, char *ep,
>> const char **endptr, int libc_errno)
>> {
>> + assert(ep >= nptr);
>> if (endptr) {
>> *endptr = ep;
>> }
>> @@ -325,6 +326,7 @@ int qemu_strtoi(const char *nptr, const char **endptr,
>> int base,
>> char *ep;
>> long long lresult;
>>
>> + assert((unsigned) base <= 36 && base != 1);
>
> If you want to play with type, I think you can do:
>
> assert((unsigned)(base - 2) <= 34)
oops, no, '0' is a valid case. Forgive this...
Laurent
