Re: [syzbot] KASAN: use-after-free Read in xfs_qm_dqfree_one

[Paul E. McKenney]

On Fri, Dec 09, 2022 at 11:46:05AM +0800, Hillf Danton wrote:
> On 6 Dec 2022 07:32:11 -0800 "Paul E. McKenney" <paulmck@xxxxxxxxxx>
> > On Tue, Dec 06, 2022 at 12:06:10PM +0100, Dmitry Vyukov wrote:
> > > On Tue, 6 Dec 2022 at 04:34, Dave Chinner <david@xxxxxxxxxxxxx> wrote:
> > > >
> > > > On Mon, Dec 05, 2022 at 07:12:15PM -0800, syzbot wrote:
> > > > > Hello,
> > > > >
> > > > > syzbot has tested the proposed patch but the reproducer is still triggering an issue:
> > > > > INFO: rcu detected stall in corrupted
> > > > >
> > > > > rcu: INFO: rcu_preempt detected expedited stalls on CPUs/tasks: { P4122 } 2641 jiffies s: 2877 root: 0x0/T
> > > > > rcu: blocking rcu_node structures (internal RCU debug):
> > > >
> > > > I'm pretty sure this has nothing to do with the reproducer - the
> > > > console log here:
> > > >
> > > > > Tested on:
> > > > >
> > > > > commit:         bce93322 proc: proc_skip_spaces() shouldn't think it i..
> > > > > git tree:       https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> > > > > console output: https://syzkaller.appspot.com/x/log.txt?x=1566216b880000
> > > >
> > > > indicates that syzbot is screwing around with bluetooth, HCI,
> > > > netdevsim, bridging, bonding, etc.
> > > >
> > > > There's no evidence that it actually ran the reproducer for the bug
> > > > reported in this thread - there's no record of a single XFS
> > > > filesystem being mounted in the log....
> > > >
> > > > It look slike someone else also tried a private patch to fix this
> > > > problem (which was obviously broken) and it failed with exactly the
> > > > same RCU warnings. That was run from the same commit id as the
> > > > original reproducer, so this looks like either syzbot is broken or
> > > > there's some other completely unrelated problem that syzbot is
> > > > tripping over here.
> > > >
> > > > Over to the syzbot people to debug the syzbot failure....
> > > 
> > > Hi Dave,
> > > 
> > > It's not uncommon for a single program to trigger multiple bugs.
> > > That's what happens here. The rcu stall issue is reproducible with
> > > this test program.
> > > In such cases you can either submit more test requests, or test manually.
> > > 
> > > I think there is an RCU expedited stall detection.
> > > For some reason CONFIG_RCU_EXP_CPU_STALL_TIMEOUT is limited to 21
> > > seconds, and that's not enough for reliable flake-free stress testing.
> > > We bump other timeouts to 100+ seconds.
> > > +RCU maintainers, do you mind removing the overly restrictive limit on
> > > CONFIG_RCU_EXP_CPU_STALL_TIMEOUT?
> > > Or you think there is something to fix in the kernel to not stall? I
> > > see the test writes to
> > > /proc/sys/vm/drop_caches, maybe there is some issue in that code.
> > 
> > Like this?
> > 
> > If so, I don't see why not.  And in that case, may I please have
> > your Tested-by or similar?
> > 
> > At the same time, I am sure that there are things in the kernel that
> > should be adjusted to avoid stalls, but I recognize that different
> > developers in different situations will have different issues that they
> > choose to focus on.  ;-)
> > 
> > 							Thanx, Paul
> > 
> > ------------------------------------------------------------------------
> > 
> > diff --git a/kernel/rcu/Kconfig.debug b/kernel/rcu/Kconfig.debug
> > index 49da904df6aa6..2984de629f749 100644
> > --- a/kernel/rcu/Kconfig.debug
> > +++ b/kernel/rcu/Kconfig.debug
> > @@ -82,7 +82,7 @@ config RCU_CPU_STALL_TIMEOUT
> >  config RCU_EXP_CPU_STALL_TIMEOUT
> >  	int "Expedited RCU CPU stall timeout in milliseconds"
> >  	depends on RCU_STALL_COMMON
> > -	range 0 21000
> > +	range 0 300000
> >  	default 0
> >  	help
> >  	  If a given expedited RCU grace period extends more than the
> >
> 	// Limit check must be consistent with the Kconfig limits for
> 	// CONFIG_RCU_EXP_CPU_STALL_TIMEOUT, so check the allowed range.
> 	// The minimum clamped value is "2UL", because at least one full
> 	// tick has to be guaranteed.
> 	till_stall_check = clamp(msecs_to_jiffies(cpu_stall_timeout), 2UL, 21UL * HZ); 
> 
> But with 21UL left behind intact?

Good catch, will fix, thank you!

							Thanx, Paul