[Date Prev][Date Next]
Re: Logic problem in the LDBM backend
At 01:11 PM 12/17/98 -0700, Daniel Carroll wrote:
> I think I've found a login problem in the LDBM backend of slapd
> (Relating to the caching of entries).
> The problem stems from the fact that DNs are not stored in canonical
> from in the database.
> The result is that when the DELETE command is issued, cache entries
> are not always correctly deleted (out of the cache) -- which has the
> potential to cause further problems (mainly at the application level)
> when the deleted entries still appear in search results.
> (I have an example at the end of this document to demonstrate the problem)
> As I see it there are a number of possible ways to fix this:
> 1) Always store the DNs in canonical form in the database
> 2) Always canonicalize the DN when loading it from the database
> 3) In the cache_entrydn_cmp() routine, canonicalize the cache entry
> before doing the comparison.
This (#3) is likely the best option. It preserved the DN form of the
> 4) Modify the ravl_delete() routine. (This is where the problem
> manifests itself).
> 5) Never normalize/de-normalize a DN once it's in the cache.
> (Not doing this has created the problem I'm trying to fix)
I believe should not be changed... just the cmp function should be fixed.
(this suggestion is just off the cuff... we should discuss ramifications)
> 6) (HACK) Re-write the cache_delete_entry_internal() routine.
> I think this routine should be written slightly differently
> anyway, to make it slightly more robust. (However, my re-write
> does obscures the bug, and doesn't fix the current memory leak,
> but it _does_ do a better cleanup job than the current code.)
Be sure to post a copy of your "HACK" to this list for review proposes.
> The basic problem stems from attempting to compare a normalized
> DN to a non-normalized DN. E.g.:
> Compare: cn=Sample, ou=Some org, o=Main org, c=Whatever
> and: CN=SAMPLE,OU=SOME ORG,O=MAIN ORG,C=WHATEVER
> Removing the spaces causes the strcasecmp() routine to
> fail, both for sorting (the main problem), and for comparisons.
> Does anyone have suggestions on which way I (or someone else) should
> pursue this?
> - Dan (Daniel Carroll - email@example.com)
>P.S. I'm submitting the patch for cache_delete_entry_internal() in
> a separate message to openldap-devel.