crypt and POSIX

[Sebastian Huber]

On 13/08/2020 21:52, Chris Johns wrote:

> On 13/8/20 11:47 pm, Joel Sherrill wrote:
>> On Thu, Aug 13, 2020 at 7:52 AM Sebastian Huber
>> <sebastian.huber at embedded-brains.de <mailto:sebastian.huber at embedded-brains.de>>
>> wrote:
>>
>>      On 13/08/2020 14:49, Joel Sherrill wrote:
>>
>>      > Why was crypt() removed? It is in the FreeBSD version at the bottom of
>>      > the file.
>>      >
>>      > https://github.com/freebsd/freebsd/blob/master/lib/libcrypt/crypt.c
>>      It is not thread-safe.
>>
>> But it is standards compliant.
> Could it be added and just return NULL or even generate an internal error? The
> lack of crypt stopped me from using and it made me consider the issues. It is
> easy to forget these things when adding passwords to the console or telnet
> server and really hard to see an issue.

If you prefer a run time error to a link time error you can add 
something like this:

char *
crypt(const char *passwd, const char *salt)
{
     errno = ENOSYS;
     return NULL;
}

The ready to use shell function to do a login check is:

extern bool rtems_shell_login_check(
   const char *user,
   const char *passphrase
);

>
>> And this would seem to violate the rules of source transparency in the
>> way the code was removed. If you wanted to disable it, it should have
>> been wrapped in ifndef __rtems__.
> I am confused, the code in cpukit/libcrypt is not from FreeBSD?
The file "cpukit/libcrypt/crypt.c" is not from FreeBSD, so there is no 
point of having #ifndef __rtems__ stuff in it. Other files in the 
libcrypt are from FreeBSD.