On Wed, Jan 28, 2015 at 08:18:04PM +0100, Sumit Bose wrote:
> On Wed, Jan 28, 2015 at 04:21:53PM +0100, Sumit Bose wrote:
> > On Wed, Jan 28, 2015 at 03:27:44PM +0100, Jakub Hrozek wrote:
> > > On Wed, Jan 28, 2015 at 03:05:00PM +0100, Sumit Bose wrote:
> > > > Hi,
> > > > 
> > > > another issue found by Steeve during testing. To reproduce this you need
> > > > a universal group with members from different domains. Then either look
> > > > up the group by SID e.g. with
> > > > 
> > > > python -c "import pysss_nss_idmap; print 
> > > > pysss_nss_idmap.getnamebysid('S-1-5-21-3456664713-2053453454-4165325232-1234')"
> > > > 
> > > > and then with getent group groupname.
> > > > 
> > > > Or use IPA views, override the group name in the 'default trust view'
> > > > on the IPA server and look up the group by the overridden name. In both
> > > > case the group should not already be in the cache. Only members from the
> > > > domain of the group should be show without the patch.
> > > > 
> > > > bye,
> > > > Sumit
> > > 
> > > The patch is correct, but I'm worried about the implications. What kind
> > > of requests by SID does the server receive? Do we also resolve requests
> > > for users by SID? In that case, we might be surprised that some POSIX
> > > attributes are not available in GC..
> > 
> > good point, I'm working on a group only alternative.
> 
> Please find attached a slightly extended version. It tries to switch to
> the LDAP connection in the common LDAP code. We already do similar
> things during group requests in the common LDAP code so I hope this is
> ok. This switch only happens during user_and_group request where it is
> not clear if the match will be a user or a group, e.g. a search by SID.
> I have to admit that there is a side effect because the AD provider uses
> GC lookups with LDAP fallback for these types of requests and will be
> affected by this change as well. But since the more detailed user
> information will be available via LDAP this might even be a benefit.
> 
> bye,
> Sumit

[...]

> -    if (state->sdap_ret == ENOENT) {
> +    if (state->sdap_ret == ENOENT && state->noexist_delete == true) {

I don't think this change is correct, the noexist_delete check should be
in its own nested if, otherwise looking up a SID that matches neither user
nor group fails with EIO:
Breakpoint 3, get_user_and_group_users_done (subreq=0x18beca0) at 
src/providers/ldap/ldap_id.c:1799
1799        struct tevent_req *req = tevent_req_callback_data(subreq,
(gdb) n
1801        struct get_user_and_group_state *state = tevent_req_data(req,
(gdb) 
1805        ret = users_get_recv(subreq, &state->dp_error, &state->sdap_ret);
(gdb) 
1806        talloc_zfree(subreq);
(gdb) 
1808        if (ret != EOK) {
(gdb) 
1812        if (state->sdap_ret == ENOENT && state->noexist_delete == true) {
(gdb) 
1822        } else if (state->sdap_ret != EOK) {
(gdb) 
1823            tevent_req_error(req, EIO);
(gdb) 
1824            return;
(gdb) p state->sdap_ret
$1 = 2
(gdb) p state->noexist_delete 
$2 = false
(gdb) quit
_______________________________________________
sssd-devel mailing list
sssd-devel@lists.fedorahosted.org
https://lists.fedorahosted.org/mailman/listinfo/sssd-devel

Reply via email to