On 2/25/26 6:05 PM, Matteo Perin via dev wrote:
> ovs_router_rule_add() is annotated with OVS_REQUIRES(mutex) but its
> external caller rule_handle_msg() in route-table.c does not hold the
> mutex. This could lead to data races on the rules pvector.
>
> Fix this by changing the annotation to OVS_EXCLUDED(mutex) and
> acquiring the mutex inside ovs_router_rule_add() around the call to
> the internal ovs_router_rule_add__() function.
>
> Fixes: e2a241560080 ("ovs-router: Introduce ovs/route/rule/{add, del}
> commands.")
The Fixes tag for this issue should be the same as in the previous patches,
as the issue was introduced when the call was added to the route-table.c
that adds a rule from the netlink message. All the calls from the unixctl
commands have proper locking.
> Signed-off-by: Matteo Perin <[email protected]>
> ---
> lib/ovs-router.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/lib/ovs-router.c b/lib/ovs-router.c
> index d0cf0f6a9..75d9badfa 100644
> --- a/lib/ovs-router.c
> +++ b/lib/ovs-router.c
> @@ -1338,11 +1338,13 @@ void
> ovs_router_rule_add(uint32_t prio, bool invert, bool user, uint8_t src_len,
> const struct in6_addr *from, uint32_t lookup_table,
> bool ipv4)
> - OVS_REQUIRES(mutex)
> + OVS_EXCLUDED(mutex)
> {
> if (use_system_routing_table) {
> + ovs_mutex_lock(&mutex);
> ovs_router_rule_add__(prio, invert, user, src_len, from,
> lookup_table, ipv4);
> + ovs_mutex_unlock(&mutex);
> }
> }
>
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
