Details
-
Type:
Bug
-
Status: Closed
-
Priority:
Major
-
Resolution: Won't Fix
-
Affects Version/s: 10.0.1
-
Fix Version/s: 10.0.4
-
Component/s: None
-
Labels:None
Description
(Context: I'm working with Arjen Lentz on opengraph )
If a plugin passes a NULL tablename to init_tmp_table_share() this causes a segfault.
If I assume that the core mysqld should be robust, this should be trapped (although not necessarily able to prevent a related segfault elsewhere in the storage plugin)
I have attached a patch that deals with this, by checking the arguments that are otherwise dereferenced by strlen() before it is used and either ASSERTs in debug build or returning with a warning log.
Gliffy Diagrams
Attachments
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
Not really. We don't add checks everywhere against improper usage.
Not even in the client-server API. Think of it, even glibc does not — try to call strcmp with a NULL pointer.
And in this particular case if the plugin invokes init_tmp_table_share without a table name, than something is seriously wrong with it. And failing gracefully here won't help, the plugin will probably just crash later or produce some weird results.