Re: [PATCH net] net: openvswitch: fix data race in ovs_vport_get_upcall_stats
flat view
From: Ilya Maximets <i.maximets@ovn.org>
Date: 2026-01-19 21:05:21
Also in:
lkml
On 1/19/26 7:13 PM, David Yang wrote:
In ovs_vport_get_upcall_stats(), some statistics protected by
u64_stats_sync, are read and accumulated in ignorance of possible
u64_stats_fetch_retry() events. These statistics are already accumulated
by u64_stats_inc(). Fix this by reading them into temporary variables
first.
Fixes: 1933ea365aa7 ("net: openvswitch: Add support to count upcall packets")
Signed-off-by: David Yang <mmyangfl@gmail.com>
---
net/openvswitch/vport.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)Hi, David. The change makes sense to me, thanks!
quoted hunk ↗ jump to hunk
diff --git a/net/openvswitch/vport.c b/net/openvswitch/vport.c index 6bbbc16ab778..bc46d661b527 100644 --- a/net/openvswitch/vport.c +++ b/net/openvswitch/vport.c@@ -319,13 +319,17 @@ int ovs_vport_get_upcall_stats(struct vport *vport, struct sk_buff *skb) for_each_possible_cpu(i) { const struct vport_upcall_stats_percpu *stats; unsigned int start; + __u64 n_success; + __u64 n_fail;
In addition to Eric's suggestion on using plain u64, you may also put these two together on the same line, right above the definition for 'start', to keep the reverse x-mass tree ordering.
quoted hunk ↗ jump to hunk
stats = per_cpu_ptr(vport->upcall_stats, i); do { start = u64_stats_fetch_begin(&stats->syncp); - tx_success += u64_stats_read(&stats->n_success); - tx_fail += u64_stats_read(&stats->n_fail); + n_success = u64_stats_read(&stats->n_success); + n_fail = u64_stats_read(&stats->n_fail); } while (u64_stats_fetch_retry(&stats->syncp, start)); + tx_success += n_success; + tx_fail += n_fail; } nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_UPCALL_STATS);
Best regards, Ilya Maximets.