apr-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Greg Stein <gst...@lyra.org>
Subject Re: Why lock pagf?
Date Tue, 01 May 2001 05:14:04 GMT
On Mon, Apr 30, 2001 at 08:19:06PM -0500, William A. Rowe, Jr. wrote:
> Greg (and dbm folks),
>   did you have anything specific in mind when you choose pagf rather
> than dirf?

pagf was opened first, so I added the locking stuff on it (before we
bothered to open dirf).

> I'm asking, since my cache-refresh patch to fstat the 
> mtime to determine cache validity aught to correspond to the fstat
> we already need for the dirf.  To wrap these all against the same
> file handle would be a good thing.

Nope. pagf can change without a change in dirf. Any mtime-based system you
use will need to use pagf.

>   Any reason not to lock against dirf instead?

Not really, you could reorder the two files in prep(). Open one, lock it (if
necessary), then open the other. But see above re: fstat.

>   The upshot - realized if we will cache stuff across processes for
> mod_auth_digest, and if users will update passwords in dbm, we don't
> see changing 'trusting' the cache.

The patch below won't work properly.


More on this series in the other thread.


Greg Stein, http://www.lyra.org/

View raw message