svn commit: r291301 - head/sys/netinet

[Gleb Smirnoff]

  Fabien,

On Wed, Nov 25, 2015 at 02:45:44PM +0000, Fabien Thomas wrote:
F> Author: fabient
F> Date: Wed Nov 25 14:45:43 2015
F> New Revision: 291301
F> URL: https://svnweb.freebsd.org/changeset/base/291301
F> 
F> Log:
F>   The r241129 description was wrong that the scenario is possible
F>   only for read locks on pcbs. The same race can happen with write
F>   lock semantics as well.
F>   
F>   The race scenario:
F>   
F>   - Two threads (1 and 2) locate pcb with writer semantics (INPLOOKUP_WLOCKPCB)
F>    and do in_pcbref() on it.
F>   - 1 and 2 both drop the inp hash lock.
F>   - Another thread (3) grabs the inp hash lock. Then it runs in_pcbfree(),
F>    which wlocks the pcb. They must happen faster than 1 or 2 come INP_WLOCK()!
F>   - 1 and 2 congest in INP_WLOCK().
F>   - 3 does in_pcbremlists(), drops hash lock, and runs in_pcbrele_wlocked(),
F>    which doesn't free the pcb due to two references on it.
F>    Then it unlocks the pcb.
F>   - 1 (or 2) gets wlock on the pcb, runs in_pcbrele_wlocked(), which doesn't
F>    report inp as freed, due to 2 (or 1) still helding extra reference on it.
F>    The thread tries to do smth with a disconnected pcb and crashes.
F>   
F>   Submitted by:	emeric.poupon at stormshield.eu
F>   Reviewed by:	gleb@

  Note that gleb at FreeBSD.org and glebius at FreeBSD.org are two different persons.

-- 
Totus tuus, Glebius.