Subject: Re: Per-user temp storage
To: Bill Rees <firstname.lastname@example.org>
From: Michael Graff <email@example.com>
Date: 02/24/1997 00:27:44
Bill Rees <firstname.lastname@example.org> writes:
> This seems awfully heavy handed for resolving something that needs to be
> fixed in the kernel in the first place. Aside from a server side inode
> depletion problem, the access times on tmp would soar as users collided
> during path lookup.
Why should file creation race conditions be handled in the kernel?
I also don't see how inodes can be depelted -- you would only have one
more than normal per user, and two more for the system's use.
> What exactly are the race conditions that are problematic?
There are quite a few... Mostly ones where a process is braindead and
copies /etc/master.passwd to /tmp/ptmp.12345, edits it, and then removes
it. The problem is that some programs (as FreeBSD found out recently I
think) don't do this sanely.
There are other less severe ones too -- elm races, etc.