Issue 6417 - If an entry RDN is identical to the suffix, then Entryrd…

[tbordaz]

…n gets broken during a reindex

Bug description:
During a reindex, the entryrdn index is built at the end from
each entry in the suffix.
If one entry has a RDN that is identical to the suffix DN,
then entryrdn_lookup_dn may erroneously return the suffix DN
as the DN of the entry.

Fix description:
When the lookup entry has no parent (because index is under
work) the loop lookup the entry using the RDN.
If this RDN matches the suffix DN, then it exits from the loop
with the suffix DN.
Before exiting it checks that the original lookup entryID
is equal to suffix entryID. If it does not match
the function fails and then the DN from the entry will be
built from id2enty

fixes: #6417

Reviewed by:

[progier389]

Nice finding and nice test case but the test case fails on lmdb 😉
IMHO entryrdn_lookup_dn is not even called in lmdb case ...

I wonder if the bdb fix could not be more simple. Something like:

         if (maybesuffix) {
+            if (id  >= 1) { 
+                rc = -1;
+                goto bail;
+            }

[droideck]

And yep, looks like LMDB wasn't able to reindex... The ds2index output in the LMDB test:

2
  ID: 2; RDN: "ou=my_org"; NRDN: "ou=my_org"
C1
  ID: 2; RDN: "ou=my_org"; NRDN: "ou=my_org"
P2
  ID: 1; RDN: "dc=dup_rdn"; NRDN: "dc=dup_rdn"
dc=dup_rdn
  ID: 1; RDN: "dc=dup_rdn"; NRDN: "dc=dup_rdn"

[progier389]

Looks like last commit generated a serious regression on lmdb
BTW: seems that issue #6386 also induced a regression ...

[tbordaz]

Absolutely... it fixes my original testcase but broke everything else ;)
I am investigating a better fix

[progier389]

LGTM, This time all the tests are PASS except the backup one which is fixed by PR #6423

[droideck]

Looks good! Ack