Fix: futex wait: handle spurious futex wakeups
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Thu, 23 Jun 2022 20:15:51 +0000 (16:15 -0400)
committerJérémie Galarneau <jeremie.galarneau@efficios.com>
Mon, 22 Aug 2022 18:01:41 +0000 (14:01 -0400)
commit57bbde26b025e9012a3089331d370896a5952c0d
treed585498c270350950ddf1a5ba417856fa12e0e4d
parent4e15f915216d4956a31011eb10c26398f81adafa
Fix: futex wait: handle spurious futex wakeups

Observed issue
==============

The futex futex_nto1_wait() implements a futex wait/wakeup scheme
identical to the liburcu workqueue code, which has an issue with
spurious wakeups.

A spurious wakeup on futex_nto1_wait can cause futex_nto1_wait to return
with a futex state of -1, which is unexpected.

futex_nto1_wait is used by the relayd live dispatcher thread, by the
relayd main dispatcher thread, as well as by the sessiond dispatcher
thread.

Given that following a futex_nto1_wait returning due to a spurious
wakeup futex_nto1_prepare will set the futex value to -1, things go
back to normal for the following futex_nto1_wait calls.

Therefore, the only impact of this issue is to spuriously use slightly
more CPU time than strictly required.

The effect is even shorter-lasting that in the liburcu counterparts
because futex_nto1_prepare explicitly sets the futex state to -1 rather
than use an atomic decrement, which immediately sets to state back to
a consistent state.

Cause
=====

From futex(5):

       FUTEX_WAIT
              Returns 0 if the caller was woken up.  Note that a  wake-up  can
              also  be caused by common futex usage patterns in unrelated code
              that happened to have previously used the  futex  word's  memory
              location  (e.g., typical futex-based implementations of Pthreads
              mutexes can cause this under some conditions).  Therefore, call‐
              ers should always conservatively assume that a return value of 0
              can mean a spurious wake-up, and  use  the  futex  word's  value
              (i.e.,  the user-space synchronization scheme) to decide whether
              to continue to block or not.

Solution
========

We therefore need to validate whether the value differs from -1 in
user-space after the call to FUTEX_WAIT returns 0.

Known drawbacks
===============

None.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8a1b6aaf77b6a2f941fd4f89b61bed71cf17906b
src/common/futex.c
This page took 0.026246 seconds and 4 git commands to generate.