Re: [PATCH] rcu: fix a race in hlist_nulls_for_each_entry_rcu macro

From: Roman Gushchin
Date: Tue May 21 2013 - 07:55:20 EST

On 21.05.2013 14:40, David Laight wrote:
Some network functions (udp4_lib_lookup2(), for instance) use the
hlist_nulls_for_each_entry_rcu macro in a way that assumes restarting
of a loop. In this case, it is strictly necessary to reread the head->first
value from the memory before each scan.
Without additional hints, gcc caches this value in a register. In this case,
if a cached node is moved to another chain during the scan, we can loop
forever getting wrong nulls values and restarting the loop uninterruptedly.

Hmmm.... if either inet_ehashfn() or next_pseudo_random32() is
called gcc must reread it anyway.
I'm surprised gcc is generating separate code for all the conditional
loop endings. So why is it caching head->first.
The 'list empty' might be short-circuited - but that would only
be relevant after a rescan.
I suspect something else is going on.

What do you mean?

I'd also have thought that this code needs to scan the entire
hash list. If things are moved under its feet this won't happen.
If it can end up on a different list (because a node got moved)
it is also possible for a later node to move it back.
In that case it would end up on the correct list

Things are always moved to the head of the list, so, it's not a problem.

-#define hlist_nulls_first_rcu(head) \
- (*((struct hlist_nulls_node __rcu __force **)&(head)->first))
+#define hlist_nulls_first_rcu(head) \
+ (*((struct hlist_nulls_node __rcu __force **) \
+ &((volatile typeof(*head) *)head)->first))

I'd have thought it would be better to change hlist_nulls_first_rcu().

It's exactly what I suggest. May be I miss something? Please, clarify.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at