[OpenAFS] 1.4.0 on Solaris 10 x86/amd64

chas williams - CONTRACTOR chas@cmf.nrl.navy.mil
Wed, 09 Nov 2005 09:08:10 -0500


In message <20051108220005.GA24280@ccali22.in2p3.fr>,Loic Tortay writes:
>time 91.000000, pid 335: Lookup adp 0x8d9e4000 name in2p3.fr fid (11:536871133.2.5412), code=0
>time 91.065536, pid 0: Tue Nov  8 22:27:55 2005
>time 91.065536, pid 335: RPC GetVolumeByName for ~M~V~Mrw ( at 0x88b62779)
>time -1912.-901544, pid 0: Tue Nov  8 22:27:55 2005
>time -1912.-901544, pid 335: Analyze RPC op -1 conn 0x888ee398 code 0x58c07 user 0x0
>time -185.-971996, pid 0: Tue Nov  8 22:27:55 2005
>time -185.-971996, pid 335: Returning code 19 from 19 
>time -1828.-08768, pid 0: Tue Nov  8 22:27:55 2005

something is wrong with the timestamps on 64-bit machines but that is
likely besides the point.  you are getting ENODEV and that would make
sense if afs is really trying to GetVolumeByName "INSERT LINE NOISE".

watching the afs transaction with tcpdump (and a big enough snap 
length) would tell you if afs is really requesting this or the
trace facility is broken on amd64.  but since this same code
works on 32-bit machines, its likely some pointer is getting truncated.
on another side note, in CM_TRACE_GETVOL %x might need to be %p.
that would help debug this problem somewhat.

anyway, can you watch with tcpdump and see what afs is requesting?