Re: __LINUX_ERRNO_EXTENSIONS__ and RTEMS

[Jeff Johnston <jjohnstn at redhat dot com>]

On 16/10/09 12:58 PM, Joel Sherrill wrote:
> Hi,
>
> Ralf Corsepius didn't like my patch to turn on
> __LINUX_ERRNO_EXTENSIONS__ for RTEMS. So in fairness,
> I did some analysis from an RTEMS perspective
> and here are the results.
> RTEMS includes a port of the FreeBSD TCP/IP stack
> and a few other BSD network services. I grep'ed
> the RTEMS source for every errno marked as a
> Linux extension by newlib. Here is the result:
>
> ======== EBADR
> cpukit/libnetworking/lib/ftpfs.c
> ========= EBADRQC
> cpukit/libnetworking/lib/ftpfs.c
> ========= EPFNOSUPPORT
> cpukit/librpc/src/rpc/bindresvport.c
> cpukit/librpc/src/rpc/clnt_generic.c
> ========= EHOSTDOWN
> cpukit/libnetworking/netinet/ip_input.c
> cpukit/libnetworking/netinet/tcp_subr.c
> cpukit/libnetworking/nfs/bootp_subr.c
> cpukit/libnetworking/net/if_ethersubr.c
> ========= ETOOMANYREFS
> cpukit/libnetworking/netinet/ip_output.c
> ========= EPROCLIM
> cpukit/httpd/sock.c
>
> What are the above files from?
> + libnetworking is the TCP/IP stack and some services.
> + libnetworking/ftpfs is RTEMS specific but tests for an error
> returned by the stack.
> + librpc is Sun RPC/XDR.
> + httpd is the GoAhead webserver.
>
> Are there any relevant standards?
>
> + EHOSTDOWN is from
> http://www.opengroup.org/onlinepubs/009695399/functions/xsh_chap02_10.html
> so it is probably really not Linux specific anyway.
>
> + Every errno in this list appears in the Open Group test suite.
>
> http://tetworks.opengroup.org/tet//sample_binaries/3.3/tet3.3-bin-linux2-dist.cpio.Z
>
> I was unable to uncompress this so don't know what that means.
> I didn't see any obvious references to them at opengroup.org in
> a standards page in my search.
>
> It would be interesting to see if other errnos marked
> Linux extensions are also used by *BSD but the set above
> are the problems for RTEMS. We cannot build RTEMS with
> newlib cvs without those defined.
>
> What is the best solution?

I believe it was RTEMS originally that had the problem with all the 
errnos being exposed so I am totally ok with you creating a new flag: 
__OPEN_GROUP_ERRNO_EXTENSIONS__ and using this flag or'd with the 
__LINUX_ERRNO_EXTENSIONS__ flag for the errnos above.

Another alternative would be to override sys/errno.h for RTEMS.

-- Jeff J.