In the Linux kernel, the following vulnerability has been resolved:
rcu/nocb: Fix missed RCU barrier on deoffloading
Currently, running rcutorture test with torturetype=rcu fwdprogress=8 nbarriercbs=8 nocbsnthreads=8 nocbstoggle=100 onoffinterval=60 testboost=2, will trigger the following warning:
WARNING: CPU: 19 PID: 100 at kernel/rcu/tree_nocb.h:1061 rcu_nocb_rdp_deoffload+0x292/0x2a0
RIP: 0010:rcu_nocb_rdp_deoffload+0x292/0x2a0
Call Trace:
<TASK>
? __warn+0x7e/0x120
? rcu_nocb_rdp_deoffload+0x292/0x2a0
? report_bug+0x18e/0x1a0
? handle_bug+0x3d/0x70
? exc_invalid_op+0x18/0x70
? asm_exc_invalid_op+0x1a/0x20
? rcu_nocb_rdp_deoffload+0x292/0x2a0
rcu_nocb_cpu_deoffload+0x70/0xa0
rcu_nocb_toggle+0x136/0x1c0
? __pfx_rcu_nocb_toggle+0x10/0x10
kthread+0xd1/0x100
? __pfx_kthread+0x10/0x10
ret_from_fork+0x2f/0x50
? __pfx_kthread+0x10/0x10
ret_from_fork_asm+0x1a/0x30
</TASK>
CPU0 CPU2 CPU3 //rcunocbtoggle //nocbcbwait //rcutorture
// deoffload CPU1 // process CPU1's rdp rcubarrier() rcusegcblistentrain() rcusegcblistaddlen(1); // len == 2 // enqueue barrier // callback to CPU1's // rdp->cblist rcudobatch() // invoke CPU1's rdp->cblist // callback rcubarriercallback() rcubarrier() mutexlock(&rcustate.barriermutex); // still see len == 2 // enqueue barrier callback // to CPU1's rdp->cblist rcusegcblistentrain() rcusegcblistaddlen(1); // len == 3 // decrement len rcusegcblistaddlen(-2); kthread_parkme()
// CPU1's rdp->cblist len == 1 // Warn because there is // still a pending barrier // trigger warning WARNONONCE(rcusegcblistncbs(&rdp->cblist)); cpusread_unlock();
// wait CPU1 to comes online and
// invoke barrier callback on
// CPU1 rdp's->cblist
wait_for_completion(&rcu_state.barrier_completion);
// deoffload CPU4 cpusreadlock() rcubarrier() mutexlock(&rcustate.barriermutex); // block on barriermutex // wait rcubarrier() on // CPU3 to unlock barriermutex // but CPU3 unlock barriermutex // need to wait CPU1 comes online // when CPU1 going online will block on cpuswritelock
The above scenario will not only trigger a WARNONONCE(), but also trigger a deadlock.
Thanks to nocb locking, a second racing rcubarrier() on an offline CPU will either observe the decremented callback counter down to 0 and spare the callback enqueue, or rcuo will observe the new callback and keep rdp->nocbcb_sleep to false.
Therefore check rdp->nocbcbsleep before parking to make sure no further rcu_barrier() is waiting on the rdp.