xfrm: policy: Read seqcount outside of rcu-read side in xfrm_policy_lookup_bytype
authorVarad Gautam <varad.gautam@suse.com>
Fri, 28 May 2021 16:04:06 +0000 (18:04 +0200)
committerSteffen Klassert <steffen.klassert@secunet.com>
Tue, 1 Jun 2021 05:48:46 +0000 (07:48 +0200)
xfrm_policy_lookup_bytype loops on seqcount mutex xfrm_policy_hash_generation
within an RCU read side critical section. Although ill advised, this is fine if
the loop is bounded.

xfrm_policy_hash_generation wraps mutex hash_resize_mutex, which is used to
serialize writers (xfrm_hash_resize, xfrm_hash_rebuild). This is fine too.

On PREEMPT_RT=y, the read_seqcount_begin call within xfrm_policy_lookup_bytype
emits a mutex lock/unlock for hash_resize_mutex. Mutex locking is fine, since
RCU read side critical sections are allowed to sleep with PREEMPT_RT.

xfrm_hash_resize can, however, block on synchronize_rcu while holding
hash_resize_mutex.

This leads to the following situation on PREEMPT_RT, where the writer is
blocked on RCU grace period expiry, while the reader is blocked on a lock held
by the writer:

Thead 1 (xfrm_hash_resize) Thread 2 (xfrm_policy_lookup_bytype)

rcu_read_lock();
mutex_lock(&hash_resize_mutex);
read_seqcount_begin(&xfrm_policy_hash_generation);
mutex_lock(&hash_resize_mutex); // block
xfrm_bydst_resize();
synchronize_rcu(); // block
<RCU stalls in xfrm_policy_lookup_bytype>

Move the read_seqcount_begin call outside of the RCU read side critical section,
and do an rcu_read_unlock/retry if we got stale data within the critical section.

On non-PREEMPT_RT, this shortens the time spent within RCU read side critical
section in case the seqcount needs a retry, and avoids unbounded looping.

Fixes: 77cc278f7b20 ("xfrm: policy: Use sequence counters with associated lock")
Signed-off-by: Varad Gautam <varad.gautam@suse.com>
Cc: linux-rt-users <linux-rt-users@vger.kernel.org>
Cc: netdev@vger.kernel.org
Cc: stable@vger.kernel.org # v4.9
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Florian Westphal <fw@strlen.de>
Cc: "Ahmed S. Darwish" <a.darwish@linutronix.de>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Acked-by: Ahmed S. Darwish <a.darwish@linutronix.de>
net/xfrm/xfrm_policy.c

index b74f28c..8c56e3e 100644 (file)
@@ -2092,12 +2092,15 @@ static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
        if (unlikely(!daddr || !saddr))
                return NULL;
 
-       rcu_read_lock();
  retry:
-       do {
-               sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
-               chain = policy_hash_direct(net, daddr, saddr, family, dir);
-       } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
+       sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
+       rcu_read_lock();
+
+       chain = policy_hash_direct(net, daddr, saddr, family, dir);
+       if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence)) {
+               rcu_read_unlock();
+               goto retry;
+       }
 
        ret = NULL;
        hlist_for_each_entry_rcu(pol, chain, bydst) {
@@ -2128,11 +2131,15 @@ static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
        }
 
 skip_inexact:
-       if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
+       if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence)) {
+               rcu_read_unlock();
                goto retry;
+       }
 
-       if (ret && !xfrm_pol_hold_rcu(ret))
+       if (ret && !xfrm_pol_hold_rcu(ret)) {
+               rcu_read_unlock();
                goto retry;
+       }
 fail:
        rcu_read_unlock();