David Howells <dhowe...@redhat.com> writes: > Mitsuo Hayasaka <mitsuo.hayasaka...@hitachi.com> wrote: > >> When we read the sysctl parameter, they are always treated >> as signed integer, and are casted into unsigned long type >> in the current kernel. If we set a value equivalent to >> (the maximum value in signed integer + 1) > > Wouldn't it be better to return EINVAL or EDOM?
Yes we should definitely fail the write in the case where we write an unsigned value and we can not fit that value in an integer. There will still remain the bug of reading the integer where (-val == val) && val < 0. In that case we do want to an (unsigned int) before storing it in an unsinged long. The decription of the patch is confusing. The problem is not the cast to unsigned long, the problem is the implicit cast to signed long which happens before the cast to unsigned long. I think this is a case where C's casting rules get it wrong. If I have an explicit cast why add an implicit cast to do sign extension. Sigh. Mitsuo since you are looking at this do you think you could fix the write side of the problem as well, and check to make cetain the unsigned value we write will fit in an interger. Eric -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/