Or, actually, maybe I don't see it: I can see how master/service.c loads the service name into config_ident. And I can see how config_getoverflowstring checks for that as a prefix.
But I still can't see any calls to read an admins value that actually use config_getoverflowstring. Everything I look at seems to be calling through either a) global_authisa(..., IMAPOPT_ADMINS) directly -- which uses config_getstring, not config_getoverflowstring; or through b) mysasl_proxy_policy, which in turn just calls global_authisa anyway. I suppose the thing to do here is test it from the outside rather than tracing it from the inside. Further into the rabbit hole... But just quickly, on a hunch, a bit of history digging finds where we started using global_authisa: https://cgit.cyrus.foundation/cyrus-imapd/commit/?id=8d6cf338 -- which is a very big commit, but, interestingly, contains: > - imapd_userisadmin = authisa(imapd_authstate, "imap", "admins"); > + imapd_userisadmin = global_authisa(imapd_authstate, IMAPOPT_ADMINS); Uh oh? Okay now I'm actually going to try wrangling Cassandane into testing whether an "imap_admins" setting does anything, because at the moment it's looking a lot like this stopped existing ~2.2. On Fri, Jun 17, 2016, at 09:58 AM, ellie timoney via Cyrus-devel wrote: > Oh, right, yeah, I see it now. Thanks > > On Wed, Jun 15, 2016, at 10:48 PM, Ken Murchison wrote: > > In config_getoverflowstring(): > > > > /* First lookup <ident>_key, to see if we have a service-specific > > * override */ > > > > if (config_ident) { > > if (snprintf(buf,sizeof(buf),"%s_%s",config_ident,key) == -1) > > fatal("key too long in config_getoverflowstring", > > EC_TEMPFAIL); > > > > lcase(buf); > > ret = hash_lookup(buf, &confighash); > > } > > > > > > On 06/14/2016 09:35 PM, ellie timoney via Cyrus-devel wrote: > > > Bits of our documentation refer to these various service-specific admins > > > settings, and googling a bit I see a bunch of people using them in the > > > imapd.confs they post when asking for help with stuff. > > > > > > But I went looking for where/how they're implemented (to see what was > > > needed to support the same sort of thing for backups) and-- this > > > functionality doesn't seem to exist anywhere in our source? I was > > > expecting to find something building a config key from a service name > > > plus "admins" and then using config_getoverflowstring to check for it > > > (like how the channel-specific config keys for sync_client work) but... > > > nothing. > > > > > > Is this a thing that used to exist, and we've removed it at some point > > > for some reason? Is it a thing that was planned and documented, but > > > never implemented? Am I missing something? > > > > -- > > Kenneth Murchison > > Principal Systems Software Engineer > > Carnegie Mellon University > >
