[854] in NetBSD-Development
Some form of cache corruption
daemon@ATHENA.MIT.EDU (Greg Hudson)
Thu Jul 6 02:36:22 1995
Date: Thu, 6 Jul 1995 02:37:44 -0400
From: Greg Hudson <ghudson@MIT.EDU>
To: netbsd-afs@MIT.EDU
I've had an odd problem on both limekiller and glacier tonight. The
problem seems to have occurred when project.gnu.i386_nbsd1 was
released and a new version of gnu ls propagated into
/mit/gnu/arch/i386_nbsd1/bin. The version from the read-only volume
is dumping core, while the version from the read-write volume works.
However, "sum" and "diff" think the two are the same:
glacier% echo $gnu
/afs/.athena/project/gnu
glacier% sum $gnu/bin/gls /mit/gnu/bin/gls
5668 36 /afs/.athena/project/gnu/bin/gls
5668 36 /mit/gnu/bin/gls
glacier% $gnu/bin/gls
CVS distrib games lib sbin usr.bin
Makefile domestic gnu libexec share usr.sbin
bin etc include regress sys
glacier% /mit/gnu/bin/gls
Bus error
Here is the ktrace output from running /mit/gnu/bin/gls:
23262 ktrace RET ktrace 0
23262 ktrace CALL execve(0xf7bfd65f,0xf7bfd5d4,0xf7bfd5dc)
23262 ktrace NAMI "/mit/gnu/bin/gls"
23262 gls RET execve 0
23262 gls PSIG SIGBUS SIG_DFL
This problem did not happen when I logged into granola and tested it
after I encountered the problem on limekiller, suggesting that the
problem is conditional on ls having been in the AFS cache prior to the
volume release.
"fs checkb", "fs checkv", and various "fs flush" and "fs flushvol
commands on limekiller had no effect on the problem.
