linux binary blues

[Dan Nelson]

In the last episode (May 14), Bogdan TARU said:
> I have a linux binary that runs well under a jail on a FreeBSD 4.8,
> but fails to run on a jaili on 4.7 (obviously, both have linux
> compatibility installed).
> 
> As far as I know (don't have the sources of this linux binary), it
> tries to get a lock on a file which is called engine.pid. That failes
> on 4.7.
> 
>  The ktraces on the systems look like:

BTW - it didn't matter in this case, but make sure you use the
linux_kdump program (in the ports tree) when dumping traces from Linux
executables.  Some syscalls have different numbers under Linux and
FreeBSD.
 
>  Good (jail on 4.8):
> 
>    582 engine   NAMI  "/home/test/testbot/var/run/engine.pid"
>    582 engine   RET   open 3
>    582 engine   CALL  semget(0x3,0x6,0xbfbfe820)
>    582 engine   RET   semget 0
>    582 engine   CALL  getpid
>    582 engine   RET   getpid 582/0x246
>    582 engine   CALL  select(0x3,0)
>    582 engine   RET   select 0
>    582 engine   CALL  write(0x3,0x864c28c,0x4)
>    582 engine   GIO   fd 3 wrote 4 bytes
> 
>  Failed (jail on 4.7):
> 
>  10995 engine   NAMI  "/usr/home/ohsandy/sandy/var/run/engine.pid"
>  10995 engine   RET   open 3
>  10995 engine   CALL  semget(0x3,0x6,0xbfbfe800)
>  10995 engine   RET   semget -1 errno -22 Unknown error: -22(

Do you have sysv semaphores enabled on this box? Put this in your
kernel config file and rebuild:

options         SYSVSEM

>  I don't know, though, why the semget(2) has different values for the
> third argument (0xbfbfe820 != 0xbfbfe800), if the binary is the same and
> the linux libraries (as far as I can tell) are the same (tried with the
> 4.7 linux_base package, and as well with the 4.8 linux_base package
> installed on the 4.7 system).

The third argument is a pointer value, so it's probably going to be
different every time.
 
-- 
	Dan Nelson
	dnelson at allantgroup.com