diff options
| author | Eric Dumazet <edumazet@google.com> | 2012-07-30 01:20:37 +0200 |
|---|---|---|
| committer | David S. Miller <davem@davemloft.net> | 2012-07-30 23:53:22 +0200 |
| commit | 404e0a8b6a55d5e1cd138c6deb1bca9abdf75d8c (patch) | |
| tree | 38e9748d38c415cc97b973fecb9279cd43f76393 /net/core | |
| parent | net: TCP early demux cleanup (diff) | |
| download | linux-404e0a8b6a55d5e1cd138c6deb1bca9abdf75d8c.tar.xz linux-404e0a8b6a55d5e1cd138c6deb1bca9abdf75d8c.zip | |
net: ipv4: fix RCU races on dst refcounts
commit c6cffba4ffa2 (ipv4: Fix input route performance regression.)
added various fatal races with dst refcounts.
crashes happen on tcp workloads if routes are added/deleted at the same
time.
The dst_free() calls from free_fib_info_rcu() are clearly racy.
We need instead regular dst refcounting (dst_release()) and make
sure dst_release() is aware of RCU grace periods :
Add DST_RCU_FREE flag so that dst_release() respects an RCU grace period
before dst destruction for cached dst
Introduce a new inet_sk_rx_dst_set() helper, using atomic_inc_not_zero()
to make sure we dont increase a zero refcount (On a dst currently
waiting an rcu grace period before destruction)
rt_cache_route() must take a reference on the new cached route, and
release it if was not able to install it.
With this patch, my machines survive various benchmarks.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net/core')
| -rw-r--r-- | net/core/dst.c | 26 |
1 files changed, 21 insertions, 5 deletions
diff --git a/net/core/dst.c b/net/core/dst.c index 069d51d29414..d9e33ebe170f 100644 --- a/net/core/dst.c +++ b/net/core/dst.c @@ -258,6 +258,15 @@ again: } EXPORT_SYMBOL(dst_destroy); +static void dst_rcu_destroy(struct rcu_head *head) +{ + struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head); + + dst = dst_destroy(dst); + if (dst) + __dst_free(dst); +} + void dst_release(struct dst_entry *dst) { if (dst) { @@ -265,10 +274,14 @@ void dst_release(struct dst_entry *dst) newrefcnt = atomic_dec_return(&dst->__refcnt); WARN_ON(newrefcnt < 0); - if (unlikely(dst->flags & DST_NOCACHE) && !newrefcnt) { - dst = dst_destroy(dst); - if (dst) - __dst_free(dst); + if (unlikely(dst->flags & (DST_NOCACHE | DST_RCU_FREE)) && !newrefcnt) { + if (dst->flags & DST_RCU_FREE) { + call_rcu_bh(&dst->rcu_head, dst_rcu_destroy); + } else { + dst = dst_destroy(dst); + if (dst) + __dst_free(dst); + } } } } @@ -320,11 +333,14 @@ EXPORT_SYMBOL(__dst_destroy_metrics_generic); */ void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) { + bool hold; + WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); /* If dst not in cache, we must take a reference, because * dst_release() will destroy dst as soon as its refcount becomes zero */ - if (unlikely(dst->flags & DST_NOCACHE)) { + hold = (dst->flags & (DST_NOCACHE | DST_RCU_FREE)) == DST_NOCACHE; + if (unlikely(hold)) { dst_hold(dst); skb_dst_set(skb, dst); } else { |