[parisc-linux] definition of EWOULDBLOCK in /usr/include/asm/errno.h

[parisc-linux] definition of EWOULDBLOCK in /usr/include/asm/errno.h

[Jochen Friedrich]

Hi,

because of the unusual definition of EWOULDBLOCK, zebra currently doesn't
work (zebra checks for EWOULDBLOCK in zebra/rt_netlink.c and will loop if
it gets EAGAIN instead and both are not the same).

I'm not sure what's the correct fix for this, change errno.h to match the
definition of the other archs or change zebra to check for EAGAIN as
well...

Cheers,
--jochen

Re: [parisc-linux] definition of EWOULDBLOCK in /usr/include/asm/errno.h

[Randolph Chung]

> because of the unusual definition of EWOULDBLOCK, zebra currently doesn't
> work (zebra checks for EWOULDBLOCK in zebra/rt_netlink.c and will loop if
> it gets EAGAIN instead and both are not the same).
> 
> I'm not sure what's the correct fix for this, change errno.h to match the
> definition of the other archs or change zebra to check for EAGAIN as
> well...

I assume you are refering to this bit of code:

      status = recvmsg (nl->sock, &msg, 0);

      if (status < 0)
        {
          if (errno == EINTR)
            continue;
          if (errno == EWOULDBLOCK)
            break;
          zlog (NULL, LOG_ERR, "%s recvmsg overrun", nl->name);
          continue;
        }

On some SysV systems EAGAIN != EWOULDBLOCK. I think we inherited the
errno definitions from HPUX...

glibc's documentation says:

Portability Note: In many older Unix systems, this condition was indicated by EWOULDBLOCK, which was a distinct error code different from EAGAIN. To make your program portable, you should check for both codes and treat them the same. 

sounds like good advice to me.. :-)

it does say, however, that for glibc EWOULDBLOCK and EAGAIN should have
the same value though, so i guess ours is wrong... i wonder if we'll
break things if we changed it.

randolph
--  
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

[parisc-linux] HPUX guru needed -> definition of EWOULDBLOCK in /usr/include/asm/errno.h

[Carlos O'Donell]

> 
> On some SysV systems EAGAIN != EWOULDBLOCK. I think we inherited the
> errno definitions from HPUX...
> 
> glibc's documentation says:
> 
> Portability Note: In many older Unix systems, this condition was indicated by EWOULDBLOCK, which was a distinct error code different from EAGAIN. To make your program portable, you should check for both codes and treat them the same. 
> 
> sounds like good advice to me.. :-)
> 
> it does say, however, that for glibc EWOULDBLOCK and EAGAIN should have
> the same value though, so i guess ours is wrong... i wonder if we'll
> break things if we changed it.
> 
> randolph

I agree with randolph, you should change the code to check for EAGAIN aswell.

On the other hand...

We have to do both of the following in order to maintain backwards compatibility.
I'm all for ignoring that and just doing 'b' :}

a. Emulate EWOULDBLOCK == EAGAIN in assembly during syscall errors.
	- Added code to glibc :(

See:
glibc-2.2.5/sysdeps/unix/i386/sysdep.S
glibc-2.2.5/sysdeps/unix/sparc/sysdep.S
glibc-2.2.5/sysdeps/unix/x86_64/sysdep.S
glibc-2.2.5/sysdeps/unix/arm/sysdep.S
...

b. Just change the kernel header to alias EWOULDBLOCK as EAGAIN.

--- linux/include/asm-parisc/errno.h	1999-12-24 12:05:04.000000000 -0500
+++ linux/include/asm-parisc/errno.h.new	2002-09-10 12:09:52.000000000 -0400
@@ -134,7 +134,7 @@
 
 #define	EALREADY	244	/* Operation already in progress */
 #define	EINPROGRESS	245	/* Operation now in progress */
-#define	EWOULDBLOCK	246	/* Operation would block (Linux returns EAGAIN) */
+#define	EWOULDBLOCK	EAGAIN	/* Operation would block (Linux returns EAGAIN) */
 #define	ENOTEMPTY	247	/* Directory not empty */
 #define	ENAMETOOLONG	248	/* File name too long */
 #define	ELOOP		249	/* Too many symbolic links encountered */

This seems best since I don't really see any code in ioctl32.c that returns
EWOULDBLOCK :) We just confuse the applicaiton by providing a different value.
Does this have anything to do with HPUX compatibility? Would an HPUX application
expect different semantic meansings from these two values?

c.

Re: [parisc-linux] definition of EWOULDBLOCK in /usr/include/asm/errno.h

[Jochen Friedrich]

Hi Randolph,

> Portability Note: In many older Unix systems, this condition was indicated by EWOULDBLOCK, which was a distinct error code different from EAGAIN. To make your program portable, you should check for both codes and treat them the same.
>
> sounds like good advice to me.. :-)

It's fixed this way in current zebra CVS. However, there might be other
applications which might be hit by this, as well.

Cheers,
--jochen