GNAT.Sockets Sockaddr-structures do not match with C-structures
Joel Sherrill
joel.sherrill at OARcorp.com
Tue Oct 28 14:52:32 UTC 2008
Petri Rokka wrote:
> Hi!
>
> I had some problems with sending an UDP packet via GNAT.Sockets.
> The Socket_Send procedure kept raising exception with message like this:
> "Error [125]: Address family not supported by protocol family"
>
> There was also a problem with binding an IP address to the local socket.
>
> Environment that was used:
> - RTEMS 4.9
> - GCC 4.3.2 (with RTEMS patches)
> - NXP LPC2468 ARM7 (little-endian)
> - Self made ethernet driver
>
> To find the cause of the problem I added some debug messages to
> the GNAT.Sockets files and to the RTEMS networking code. I discovered
> that the protocol family information did not transfer intact to the
> C-socket API. In fact the C-sendto function got always the value
> zero instead of two(AF_INET).
>
> I did some more tracing and noticed that the C-structures and Ada-
> structures differed with the size of protocol family variable and that
> there was also an extra length variable in the C-structures. C-structures
> had protocol family variable of size 8bits while the GNAT.Sockets had size
> of 16bits. The extra length variable had size of 8bits maybe to compensate
> with smaller protocol family variable.
>
> I modified the Ada-structures to match with the C-structures and
> there were no problems anymore.
>
> Could it be possible that this bug does not affect big-endian hosts?
>
>
I don't know where this could be a problem but this file
is generated offline and then the output placed
in the GCC source tree. If the one I generated on the PowerPC
is not right for all RTEMS targets, you will have to help
me get it right in the tree. Here is the procedure:
+ check out the RTEMS CVS module ada-examples
+ Verify the files gen-soccon.c and gsocket.h match
those in the 4.3.2 source you have.
+ Run the program gen-soccon on your ARM target.
+ See if the output is right. If not, figure out where
the two files above are broken. It is possible they
are broken for all targets or just the ARM.
The end result of this is that we will have to file a GCC PR
and get it into GCC.
+ Could be a simple update of gen-soccon.c, gsocket.h and
the g-soccon-rtems.ads for 4.3.2.
+ It could also require a different g-soccon-rtems.ads for
arm from other CPUs. That would require adding a file
and fixing a Makefile.in.
I think on the SVN trunk, this file is generated on the fly
by the build system using a different mechanism. So we
will have to look at that as a separate issue. I can build
that for ARM and send you the file for review once we
have fixed 4.3.2.
--joel
> Ada-structures (from gcc/ada/g-socthi.ads) with modifications
> -------------------------------------------------------------
> type Sockaddr is record
> Sa_Len : C.unsigned_char; -- added
> Sa_Family : C.unsigned_char; -- changed from C.unsigned_short
>
> Sa_Data : C.char_array (1 .. 14);
> end record;
> pragma Convention (C, Sockaddr);
>
> type Sockaddr_In is record
> Sin_Len : C.unsigned_char := 0; -- added
> Sin_Family : C.unsigned_char := Constants.AF_INET; -- changed from C.unsigned_short
>
> Sin_Port : C.unsigned_short := 0;
> Sin_Addr : In_Addr := Inaddr_Any;
> Sin_Zero : C.char_array (1 .. 8) := (others => C.char'Val (0));
> end record;
> pragma Convention (C, Sockaddr_In);
>
>
> C-structures for comparison:
> -----------------------------
> cpukit/libnetworking/rtems/bsdnet/_types.h:typedef uint8_t __sa_family_t;
>
> cpukit/libnetworking/sys/socket.h-struct sockaddr {
> cpukit/libnetworking/sys/socket.h- unsigned char sa_len; /* total length */
> cpukit/libnetworking/sys/socket.h: sa_family_t sa_family; /* address family */
> cpukit/libnetworking/sys/socket.h- char sa_data[14]; /* actually longer; address value */
> cpukit/libnetworking/sys/socket.h-};
>
> cpukit/libnetworking/netinet/in.h-/* Socket address, internet style. */
> cpukit/libnetworking/netinet/in.h-struct sockaddr_in {
> cpukit/libnetworking/netinet/in.h- uint8_t sin_len;
> cpukit/libnetworking/netinet/in.h: sa_family_t sin_family;
> cpukit/libnetworking/netinet/in.h- in_port_t sin_port;
> cpukit/libnetworking/netinet/in.h- struct in_addr sin_addr;
> cpukit/libnetworking/netinet/in.h- char sin_zero[8];
> cpukit/libnetworking/netinet/in.h-};
>
>
> Regards
> Petri Rokka
>
> _______________________________________________
> rtems-users mailing list
> rtems-users at rtems.com
> http://rtems.rtems.org/mailman/listinfo/rtems-users
>
--
Joel Sherrill, Ph.D. Director of Research & Development
joel.sherrill at OARcorp.com On-Line Applications Research
Ask me about RTEMS: a free RTOS Huntsville AL 35805
Support Available (256) 722-9985
More information about the users
mailing list