[PATCH rtems-libbsd 1/7] rtemsbsd: Catch timeout overflows

[Chris Johns]

On 28/7/21 7:38 am, Gedare Bloom wrote:
> On Tue, Jul 27, 2021 at 2:59 AM <chrisj at rtems.org> wrote:
>>
>> From: Chris Johns <chrisj at rtems.org>
>>
>> Update #4475
> This change could probably use its own ticket.

Without the change the RPC client fails to connect in a reliable way because the
connection may time out for no real reason. A connection that is stable appears
fragile. I used this ticket because the change was needed for NFS to work.

https://github.com/freebsd/freebsd-src/blob/main/sys/rpc/clnt_dg.c#L388

>> ---
>>  rtemsbsd/rtems/rtems-kernel-timesupport.c | 8 +++++++-
>>  1 file changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/rtemsbsd/rtems/rtems-kernel-timesupport.c b/rtemsbsd/rtems/rtems-kernel-timesupport.c
>> index ef14d1fa..5d290d66 100644
>> --- a/rtemsbsd/rtems/rtems-kernel-timesupport.c
>> +++ b/rtemsbsd/rtems/rtems-kernel-timesupport.c
>> @@ -35,6 +35,7 @@
>>
>>  #include <machine/rtems-bsd-kernel-space.h>
>>
>> +#include <limits.h>
>>  #include <sys/time.h>
>>
>>  #include <rtems/score/timespec.h>
>> @@ -46,9 +47,14 @@ int
>>  tvtohz(struct timeval *tv)
>>  {
>>    struct timespec ts;
>> +  uint32_t ticks;
>>
>>    ts.tv_sec = tv->tv_sec;
>>    ts.tv_nsec = tv->tv_usec * 1000;
>>
>> -  return (int) _Timespec_To_ticks( &ts );
>> +  ticks = _Timespec_To_ticks( &ts );
>> +  if (ticks > INT_MAX)
>> +    ticks = INT_MAX;
>> +
> This changes the behavior to saturating in the overflow case, which is
> at least well-defined, but is it the best thing to do?  (I have no
> idea.)

I do not have a better suggestion? The user is providing a timeval that exceeds
the size of the return value so providing the max value at least the gets you
closer to the actual value than any other value?

Chris