[4088] in Athena Bugs
Re: Circular mountpoint strikes again
daemon@ATHENA.MIT.EDU (marc@ATHENA.MIT.EDU)
Tue Jan 30 15:26:32 1990
From: marc@ATHENA.MIT.EDU
To: garner@ATHENA.MIT.EDU
Cc: bugs@ATHENA.MIT.EDU
In-Reply-To: Your message of Tue, 30 Jan 90 14:20:50 -0500.
Reply-To: marc@MIT.EDU
Date: Tue, 30 Jan 90 15:26:02 EST
The bug itself is kind of long and complex. How about this:
kernel bug:
AFS (including 3.0beta) on all architectures has a bug in the
kernel. Basically, when a mountpoint is referenced for the first
time, the pointer to the parent (referenced by ..) is set to the
current parent. If you have a mountpoint existing in it's own path,
this can result in a directory being its own parent. Unfortunately,
the kernel structures aren't complex enough to make the solution
simple. Lose lose. Why is this bad? See the next two pieces of the
error:
libc bug:
Nowhere. I was confused.
csh bug:
When a new shell begins to execute, it attempts to execute
getwd(). With the circular parents, it fails, and prints the almost
useless message "getwd: can't open .." and the completely bogus error
message "(Did you su in a fascist directory?)". This is only
completely annoying. The more significant bug (which I can't
replicate, unfortunately) was that for some reason (I think) the shell
seemed to be trying to find out the current dir by stating ., then
./.., then ./../.., etc. It lost big time inside the kernel, as kill
-9 wouldn't even kill it.
That better robin? :-)
Marc
