Should REPLACE with an explicit entry list that yields 0 entries return -EINVAL or should it be treated the same as omitting TCA_GATE_ENTRY_LIST and keeping the old schedule?

thanks,
Paul





On Wednesday, January 21st, 2026 at 3:49 PM, Victor Nogueira [off-list ref] wrote:

On 21/01/2026 16:44, Victor Nogueira wrote:

quoted

On 21/01/2026 10:20, Paul Moses wrote:

quoted

Reject empty schedules (num_entries == 0) so next_entry is always
valid and
RCU readers/timer logic never walk an empty list. taprio enforces the
same
constraint on schedules (sch_taprio.c, commit 09dbdf28f9f9fa).

Fixes: a51c328df310 ("net: qos: introduce a gate control flow action")
Signed-off-by: Paul Moses p@1g4.org
Cc: stable@vger.kernel.org
---
net/sched/act_gate.c | 6 ++++++
1 file changed, 6 insertions(+)

diff --git a/net/sched/act_gate.c b/net/sched/act_gate.c
index 48ff378bb051a..e4134b9a4a314 100644
--- a/net/sched/act_gate.c
+++ b/net/sched/act_gate.c

@@ -509,6 +509,12 @@ static int tcf_gate_init(struct net *net, struct

nlattr *nla,
cycletime_ext = nla_get_u64(tb[TCA_GATE_CYCLE_TIME_EXT]);
p->tcfg_cycletime_ext = cycletime_ext;
+ if (p->num_entries == 0) {
+ NL_SET_ERR_MSG(extack, "The entry list is empty");
+ err = -EINVAL;
+ goto release_mem;
+ }

It would be simpler to check this in parse_gate_list.
That way you could return -EINVAL there directly
in case 0 entries were passed.

On second thought, I believe it would be better
to check whether parse_gate_list's return is 0
and the op is a create. Something like:

err = parse_gate_list(tb[TCA_GATE_ENTRY_LIST], p, extack);
...
if (!err && ret == ACT_P_CREATED) {
NL_SET_ERR_MSG(extack, "The entry list is empty");
err = -EINVAL;
goto release_mem;
}

so that you don't need to add new arguments to
parse_gate_list.

cheers,
Victor