On Thu, Jun 06, 2013 at 08:45:34PM +0200, Sumit Bose wrote:
On Thu, Jun 06, 2013 at 08:06:22PM +0200, Jakub Hrozek wrote:
> On Thu, Jun 06, 2013 at 01:58:18PM +0200, Sumit Bose wrote:
> > On Wed, Jun 05, 2013 at 10:06:28PM +0200, Jakub Hrozek wrote:
> > > On Wed, Jun 05, 2013 at 11:32:39AM +0200, Jakub Hrozek wrote:
> > > > I pushed the code into my gc branch and I'm attaching an updated
tarball
> > > > again. Sorry for the confusion, I only developed and tested on F19.
> > >
> > > Sumit found out that the code crashed after performing a SRV lookup if
> > > subdomain user was requested. The attached patches implement SRV lookups
> > > for GC servers properly and protect against cases where no GC servers
> > > are found at all during service discovery, but later subdomain user is
> > > requested.
> > >
> > > I've pushed the code into my fedorapeople branch and I'm attaching
a
> > > tarball with the latest patches as well.
> >
> > Thank you, I see no coredump anymore, but still the GC cannot be found
> > with SRV lookups. But I hve to check my DNS setup as well, so this
> > might not be a real issue and if, it can be fixed after the beta.
> >
> > Here are my review comments:
> >
> > [PATCH 01/15] Do not obfuscate calls with booleans
> > ACK. I only have one comment. I'm not sure what would be more
gdb-friendly,
> > using defined like in you patch or using real functions for
> > *_primary_servers_init and _backup_servers_init?
> >
>
> OK, that's a fair comment, I used inline functions in this iteration.
> Lukas reminded me that inline functions might not be usable in gdb with
> high debug levels, but I think we can live with that.
>
> > [PATCH 02/15] LDAP: sdap_id_ctx might contain several connections
> > ACK
> >
> > [PATCH 03/15] LDAP: Refactor account info handler into a tevent request
> > ACK
> >
> > [PATCH 04/15] LDAP: Pass in a connection to ID functions
> > ACK
> >
> > [PATCH 05/15] LDAP: new SDAP domain structure
> >
> > +int
> > +sdap_domain_destructor(TALLOC_CTX *ctx)
> > +{
> > + struct sdap_domain *dom =
> > + talloc_get_type(ctx, struct sdap_domain);
> > + DLIST_REMOVE(*(dom->head), dom);
> > + return 0;
> > +}
> >
> > destructors use void * as argument.
>
> Fixed.
>
> >
> > +void
> > +sdap_domain_remove(struct sss_domain_info *subdom)
> > +{
> > + struct sss_domain_info *diter;
> > +
> > + if (subdom->parent == NULL) return;
> > +
> > + for (diter = subdom->parent->subdomains;
> > + diter;
> > + diter = get_next_domain(diter, false)) {
> > + if (diter == subdom) break;
> > + }
> > +
> > + talloc_free(diter);
> > +}
> >
> > I think we should be careful to delete sss_domain_info objects, because they
> > might be still used by some requests.
>
> That's true, the current patch just removes the domain from the linked
> list. I think that removing a domain is such a rare event that for the
> time being we can live with the leak. Maybe there could be a ticket to
> remove the domain after some period of time or provide a more elaborate
> solution, maybe some refcount based on number of requests.
>
> >
> > [PATCH 06/15] LDAP: return sdap search return code to ID
> > ACK
> >
> > [PATCH 07/15] Move domain_to_basedn outside IPA subtree
> > ACK
> >
> > [PATCH 08/15] New utility function sss_get_domain_name
> > ACK
> >
> > PATCH 09/15] LDAP: split a function to create search bases
> > ACK
> >
> > [PATCH 10/15] LDAP: store FQDNs for trusted users and groups
> > ACK
> >
> > [PATCH 11/15] Split generating primary GID for ID mapped users into a separate
function
> > ACK
> >
> > [PATCH 12/15] LDAP: Do not store separate GID for subdomain users
> > ACK
> >
> > PATCH 13/15] AD: Add additional service to support Global Catalog lookups
> > ACK
> >
> > [PATCH 14/15] AD ID lookups - choose GC or LDAP as appropriate
> >
> > + switch (ar->entry_type & BE_REQ_TYPE_MASK) {
> > + case BE_REQ_USER: /* user */
> > + case BE_REQ_GROUP: /* group */
> > + case BE_REQ_INITGROUPS: /* init groups for user */
> > + if (ad_ctx->gc_ctx && IS_SUBDOMAIN(dom)) {
> > + clist[i] = ad_ctx->gc_ctx;
> > + i++;
> > + } else {
> > + clist[i] = ad_ctx->ldap_ctx;
> > + }
> > + break;
> > +
> > + default:
> > + clist[0] = ad_ctx->ldap_ctx;
> > + break;
> > + }
> >
> > I guess BE_REQ_BY_SECID and BE_REQ_USER_AND_GROUP should be handled like
BE_REQ_USER?
> >
>
> Ah, true. The BE_REQ_BY_SECID and BE_REQ_USER_AND_GROUP were developed
> in parallel with these patches, so I probably forgot to include them
> here.
>
> > [PATCH 15/15] AD: Store trusted AD domains as subdomains
> >
> > typo in the commit message 'newe' -> 'newer'
>
> Fixed.
>
> Thank you for the review. I rebased the patches on top of your PAC
> patches (there was one change in pacsrv where I used the new function
> sss_get_domain_name) and pushed the patches to my fedorapeople branch
> called "gc".
>
> A tarball is also provided for convenience.
ACK, please do not forget to open a ticket about the leaking domain.
bye,
Sumit
Thank you for the review, all patches were pushed to master.
I didn't push your additional patch to amend the service discovery to
use the forest value instead of the domain. Can you please open a ticket
to fix lookups when enrolled with a different server than the forest
root? I think that given you already had a partial patch you know the
details better.