rbtree problems in libpthread

[Matthias Drochner <M.Drochner%fz-juelich.de@localhost>]

I've observed that some applications sometimes get an ESRCH error
on pthread_join(). Added diagnostics but everything looks OK -
just the RB_FIND() in libpthread:pthread__find() returns that error
while the thread should be there.

This could be a bug in the rbtree implementation, or some
misuse or race condition in libpthread, or some memory
corruption.

Now I've added some fallback code to pthread__find() to
look up the thread in pthread__allqueue if it wasn't found
in pthread__alltree, and this did always succeed.

I can't imagine a memory corruption so subtle that it
damages the rbtree structures in a way that doesn't cause
a crash, and still leaves the queue structures intact, and I
also didn't find a misuse in libpthread, so I suspect
the rbtree implementation.

Has anyone seen that kind of problem? Is the rbtree code
well tested or used in other subsystems?

(btw, I think the use of rbtree in libpthread is not completely
clean because the pointer subtraction in pthread__cmp() can
overflow but this doesn't seem to be the problem here)

best regards
Matthias




-------------------------------------------------------------------
-------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich

Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzende des Aufsichtsrats: MinDir'in Baerbel Brumme-Bothe
Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender),
Dr. Ulrich Krafft (stellv. Vorsitzender), Prof. Dr. Harald Bolt,
Dr. Sebastian M. Schmidt
-------------------------------------------------------------------
-------------------------------------------------------------------