Re: LTP: list of failures on 32bit and compat mode

[Arnd Bergmann]

On Thu, Apr 6, 2023, at 14:48, Petr Vorel wrote:
>> On Thu, Apr 6, 2023, at 12:56, Petr Vorel wrote:

>> Ah, I see. This must be the padding code then, not the time64
>> conversion:
>
>> +int recvmmsg(int fd, struct mmsghdr *msgvec, unsigned int vlen, unsigned int flags, struct timespec *timeout)
>> +{
>> +#if LONG_MAX > INT_MAX
>> +       struct mmsghdr *mh = msgvec;
>> +       unsigned int i;
>> +       for (i = vlen; i; i--, mh++)
>> +               mh->msg_hdr.__pad1 = mh->msg_hdr.__pad2 = 0;
>> +#endif
>
> I suppose this is a suggestion for fix in LTP. I'd expect is should go into
> testcases/kernel/syscalls/sendmmsg/sendmmsg_var.h into static inline int
> sys_recvmmsg(...)
>
> But that at least on glibc 64bit compilation does not see __pad1 member:
>
> ../sendmmsg/sendmmsg_var.h: In function ‘sys_recvmmsg’:
> ../sendmmsg/sendmmsg_var.h:47:28: error: ‘struct msghdr’ has no member 
> named ‘__pad1’
>    47 |                 mh->msg_hdr.__pad1 = mh->msg_hdr.__pad2 = 0;
>       |                            ^

Sorry, I should have been clearer, the snippet I cited is
from the musl sources, and the __pad access is what causes the
segfault. The fix is to catch the fault on ltp, same as for the
time64 conversions.

      Arnd