[SSSD] [PATCH] LDAP: retain external members

Sumit Bose sbose at redhat.com
Fri Dec 12 15:08:34 UTC 2014


On Fri, Dec 12, 2014 at 01:16:29PM +0100, Pavel Reichl wrote:
> 
> On 12/11/2014 06:47 PM, Sumit Bose wrote:
> >On Thu, Dec 11, 2014 at 06:07:07PM +0100, Pavel Reichl wrote:
> >>On 12/10/2014 11:28 PM, Lukas Slebodnik wrote:
> >>>On (10/12/14 21:32), Sumit Bose wrote:
> >>>>On Wed, Dec 10, 2014 at 05:43:37PM +0100, Lukas Slebodnik wrote:
> >>>>>On (10/12/14 11:26), Sumit Bose wrote:
> >>>>>>On Wed, Dec 10, 2014 at 10:44:32AM +0100, Pavel Reichl wrote:
> >>>>>>>On 12/09/2014 01:36 PM, Sumit Bose wrote:
> >>>>>>>>On Tue, Nov 25, 2014 at 03:42:14PM +0100, Pavel Reichl wrote:
> >>>>>>>>>Hello,
> >>>>>>>>>
> >>>>>>>>>please see attached patch for https://fedorahosted.org/sssd/ticket/2492
> >>>>>>>>>
> >>>>>>>>>Thanks!
> >>>>>>>>Hi Pavel,
> >>>>>>>>
> >>>>>>>>thank you for the patch, it works well in my tests and I didn't see any
> >>>>>>>>regressions in IPA setup with and without trsut to AD, so ACK.
> >>>>>>>>
> >>>>>>>>I would just like to ask you to add a comment to
> >>>>>>>>
> >>>>>>>>>@@ -842,6 +913,23 @@ static int sdap_save_grpmem(TALLOC_CTX *memctx,
> >>>>>>>>>              goto fail;
> >>>>>>>>>          }
> >>>>>>>>>      }
> >>>>>>>>>+    if (opts->schema_type == SDAP_SCHEMA_IPA_V1) {
> >>>>>>>>>+        ret = sysdb_attrs_get_string(attrs, SYSDB_SID_STR, &group_sid);
> >>>>>>>>>+        if (ret != EOK) {
> >>>>>>>>>+            DEBUG(SSSDBG_TRACE_FUNC, "Failed to get group sid\n");
> >>>>>>>>>+            group_sid = NULL;
> >>>>>>>>>+        }
> >>>>>>>>>+
> >>>>>>>>>+        if (group_sid != NULL) {
> >>>>>>>>>+            ret = retain_extern_members(memctx, dom, group_name, group_sid,
> >>>>>>>>>+                                        &userdns, &nuserdns);
> >>>>>>>>>+            if (ret != EOK) {
> >>>>>>>>>+                DEBUG(SSSDBG_MINOR_FAILURE,
> >>>>>>>>>+                      "retain_extern_members failed: %d:[%s].\n",
> >>>>>>>>>+                      ret, sss_strerror(ret));
> >>>>>>>>>+            }
> >>>>>>>>>+        }
> >>>>>>>>>+    }
> >>>>>>>>which explains that this is a temporary solution until the IPA provider
> >>>>>>>>can resolve external group membership. I have created
> >>>>>>>>https://fedorahosted.org/sssd/ticket/2522 for this. Feel free to
> >>>>>>>>explicitly add the ticket URL into the comment.
> >>>>>>>>
> >>>>>>>>bye,
> >>>>>>>>Sumit
> >>>>>>>Thanks for review.
> >>>>>>>Please see updated patch.
> >>>>>>>
> >>>>>>Thank you, ACK.
> >>>>>NACK, sssd crashed with this patch.
> >>>>>How to reproduce:
> >>>>>sssd with ipa server mode:
> >>>>>
> >>>>>id admin at rdustv1911.test  //ipa admin
> >>>>>id aduser1 at ipaad2012r2.test // aduser
> >>>>>id aduser2 at ipaad2012r2.test
> >>>>>sss_cache -E
> >>>>>id aduser1 at ipaad2012r2.test // aduser
> >>>>good catch, thank you. Looks like I always only tested with one external
> >>>>member. Does
> >>>>
> >>>>diff --git a/src/providers/ldap/sdap_async_groups.c
> >>>>b/src/providers/ldap/sdap_async_groups.c
> >>>>index 43744ac..4bc2aa7 100644
> >>>>--- a/src/providers/ldap/sdap_async_groups.c
> >>>>+++ b/src/providers/ldap/sdap_async_groups.c
> >>>>@@ -861,7 +861,7 @@ retain_extern_members(TALLOC_CTX *mem_ctx,
> >>>>
> >>>>             (*_nuserdns)++;
> >>>>             *_userdns = talloc_realloc(mem_ctx, *_userdns, char*, *_nuserdns);
> >>>>-            *_userdns[(*_nuserdns)-1] = talloc_steal(mem_ctx, dns[i]);
> >>>>+            (*_userdns)[(*_nuserdns)-1] = talloc_steal(mem_ctx, dns[i]);
> >>>>         }
> >>>>     }
> >>>>
> >>>>
> >>>>fix the issue for you?
> >>>Yet, it fixed.
> >>>I could see the mistake in code and I was looking at the code at least fo 10
> >>>minutes. Someone should write 100 times I will learn C operator precedence
> >>>http://en.cppreference.com/w/c/language/operator_precedence
> >>>
> >>>>Maybe this should be better written with local variables to avoid this
> >>>>kind of issues?
> >>>+1
> >>>
> >>>>Additionally there is a NULL check missing after talloc_realloc().
> >>>+1
> >>>
> >>>LS
> >>>_______________________________________________
> >>>sssd-devel mailing list
> >>>sssd-devel at lists.fedorahosted.org
> >>>https://lists.fedorahosted.org/mailman/listinfo/sssd-devel
> >>I'm really sorry about the crash.
> >>Please see if the new patch is crash free.
> >>
> >>Thanks!
> >Hi Pavel,
> >
> >thank you for the new version, I haven't had a close look yet, but I hav
> >a nitpick about using the memory contexts ...
> >
> >...
> >
> >With the if (userdns == NULL), there is a chance to leak memory. I would
> >suggest to following:
> >
> >>+
> >>+    for (i=0; i < n; i++) {
> >>+        ret = are_sids_from_same_dom(group_sid, sids[i], &same_domain);
> >>+        if (ret == EOK && !same_domain) {
> >>+            DEBUG(SSSDBG_TRACE_ALL, "extern member: %s\n", dns[i]);
> >>+            nuserdns++;
> >>+            userdns = talloc_realloc(mem_ctx, userdns, const char*, nuserdns);
> >                userdns = talloc_realloc(tmp_ctx, userdns, const char*, nuserdns);
> >>+            if (userdns == NULL) {
> >>+                return ENOMEM;
> >>+            }
> >>+            userdns[nuserdns-1] = talloc_steal(mem_ctx, dns[i]);
> >                userdns[nuserdns-1] = talloc_steal(userdns, dns[i]);
> >>+        }
> >>+    }
> >>+    *_nuserdns = nuserdns;
> >        *_nuserdns = talloc_steal(mem_ctx, nuserdns);
> >>+    *_userdns = discard_const(userdns);
> >>+    ret = EOK;
> >>+
> >talloc_steal() is not recursive, but with the scheme above the items of
> >userdns are children of userdns talloc-wise and stealing userdns on
> >mem_ctx does not change it and a talloc_free(userdns) will free userdns
> >and all its elements.
> >
> >bye,
> >Sumit
> Thank you for patience.  Updated patch attached.

Thank you for the new version, it is working as expected and I didn't
see a crash neither in IPA clients nor on a server.

I think all of Lukas' comments are addressed as well. I just have a
minor comment but imo it does not has to be fixed.

bye,
Sumit

> +
> +        if (sidstr != NULL) {
> +            sids[n] = talloc_steal(sids, sidstr);

Currently talloc_steal() only returns NULL if the 2nd argument is NULL
and since we know that sidstr != NULL the following check will never be
true. But since it is quite cheap it can be left here.

> +            if (sids[n] == NULL) {
> +                ret = ENOMEM;
> +                goto done;
> +            }
> +
> +            dns[n] = talloc_steal(dns,
> +                                  ldb_dn_get_linearized(members[i]->dn));

In this case the check is valid because ldb_dn_get_linearized() can
return NULL.

> +
> +            if (dns[n] == NULL) {
> +                ret = ENOMEM;
> +                goto done;
> +            }
> +            n++;
> +        }
> +    }
> +
> +    if (n == 0) {
> +        ret = ENOENT;
> +        goto done;
> +    }
> +
> +    *_n = n;
> +    *_sids = talloc_steal(mem_ctx, sids);
> +    *_dns = talloc_steal(mem_ctx, dns);
> +
> +    ret = EOK;
> +
> +done:
> +    if (ret == ENOENT) {
> +        DEBUG(SSSDBG_TRACE_FUNC, "No such entry\n");
> +    } else if (ret) {
> +        DEBUG(SSSDBG_OP_FAILURE, "Error: %d (%s)\n", ret, strerror(ret));
> +    }
> +    talloc_free(tmp_ctx);
> +    return ret;
> +}



More information about the sssd-devel mailing list