RE: Re: on database logging
"Tina Bird" <[email protected]>
| Newsgroups | gmane.comp.sysutils.loganalysis |
|---|---|
| Message-ID | <02ca01c771bd$4af81b40$1701a8c0@lindesfarne> |
> I would prefer if application programmers would map the users at the > application layer of a web application to a database user. I would > prefer if the RBAC groups implemented in that application were > manifested in the database as well. Finally I would prefer if > the groups > of logical operations that the groups of users were mapped > back into the > database also. > > With phpBB, all application users connect to the backend database as > user phpbbuser (or whatever I happen to configure). That "phpbbuser" > will have the superset of database access privileges that admins, > moderators, validated users, unvalidated users, suspended > users require. > > This application cannot do not fine-grained access control at the > database layer. The control objectives I see in ISO 17799 and other > standards assume that it can. And HIPAA. Don't forget HIPAA. I have been blithely making statements along the lines of "medical applications need to be able to record every write, change, read or delete to an electronic medical record at a per-user level" for years, until it finally sunk in (a year or so ago) that the vast majority of DB applications -- including those with Web front ends -- did generic authentication "to the database." No hope of individual accountability. This discussion brings me back to one of my persistent chants, when it comes to logging -- the end users (in this case, the sys admins and auditors who must demonstrate compliance to laws) have *got* to provide the technical requirements to the developers for more detailed auditing; systems have to be architected and maintained to support the additional strain created by the logging. Does anyone have any numbers on the additional CPU load (or performance hit) created by this level of auditing in a high-load DB environment? And then, of course, we can worry about how much worse the performance gets if we insist that user crendentials and authorization tokens are suitably encrypted... cheers -- tbird p.s. Wynn, I am still laughing about your last posting: "IT systems often do not fit into simple little boxes and that is part of the problem. We build cloverleafs at dirt-road intersections to avoid teaching people about the 4-way stop." What a *perfect* description.