Re: (ITS#8443) segfault in lmdb following mdb_env_set_mapsize

[hyc@symas.com]

gdochev@yahoo.com wrote:
> Full_Name: George Dochev
> Version:
> OS: Windows/OSX/Linux
> URL: ftp://ftp.openldap.org/incoming/
> Submission from: (NULL) (2601:582:4500:e134:1d59:965c:aecc:5697)
>
>
> I updated from 0.9.15 to 0.9.18 and started getting intermittent segfaults in
> one of my unit tests. In my test when I issue mdb_txn_commit and get back
> MDB_MAP_FULL I abort the transaction,

You've misused the API. The txn has already been released by txn_commit, you 
cannot call abort on it at that point. This was already clearly spelled out in 
the docs for mdb_txn_commit.

Closing this ITS.

  expand the database through a call to
> mdb_env_set_mapsize and attempt a retry. In the retry I issue mdb_txn_begin
> again which is where I crash when it calls mdb_txn_renew0(MDB_txn *txn) with txn
> pointing to garbage.
>
> Some additional info which may or may not be relevant:
>
> The db size is 1MB aligned and there are no active transactions prior to the
> call to mdb_env_set_mapsize other than the one that I aborted as a result of the
>   MDB_MAP_FULL.
> The environment was initially opened with  MDB_NOSYNC | MDB_WRITEMAP |
> MDB_NOMETASYNC | MDB_NOTLS.
>
> The crash is intermittent but reproducible in all my configurations.
>
> Regards,
> George
>
>


-- 
   -- Howard Chu
   CTO, Symas Corp.           http://www.symas.com
   Director, Highland Sun     http://highlandsun.com/hyc/
   Chief Architect, OpenLDAP  http://www.openldap.org/project/