This driver currently doesn't support any flower control flags.
Implement check for control flags, such as can be set through
`tc flower ... ip_flags frag`.
Since qede_parse_flow_attr() are called by both qede_add_tc_flower_fltr()
and qede_flow_spec_to_rule(), as the latter doesn't having access to
extack, then flow_rule_*_control_flags() can't be used in this driver.
This patch therefore re-implements flow_rule_match_has_control_flags(),
but with error messaging via DP_NOTICE, instead of NL_SET_ERR_MSG_FMT_MOD.
So in case any control flags are masked, we call DP_NOTICE() and
return -EOPNOTSUPP.
Only compile-tested.
Signed-off-by: Asbjørn Sloth Tønnesen <[email protected]>
---
This is AFAICT the last driver which didn't validate these flags.
$ git grep FLOW_DISSECTOR_KEY_CONTROL drivers/
$ git grep 'flow_rule_.*_control_flags' drivers/
drivers/net/ethernet/qlogic/qede/qede_filter.c | 13 +++++++++++++
1 file changed, 13 insertions(+)
diff --git a/drivers/net/ethernet/qlogic/qede/qede_filter.c b/drivers/net/ethernet/qlogic/qede/qede_filter.c
index a5ac21a0ee33..40f72e700d8e 100644
--- a/drivers/net/ethernet/qlogic/qede/qede_filter.c
+++ b/drivers/net/ethernet/qlogic/qede/qede_filter.c
@@ -1843,6 +1843,19 @@ qede_parse_flow_attr(struct qede_dev *edev, __be16 proto,
return -EPROTONOSUPPORT;
}
+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
+ struct flow_match_control match;
+
+ flow_rule_match_control(rule, &match);
+
+ if (match.mask->flags) {
+ DP_NOTICE(edev,
+ "Unsupported match on control.flags %#x",
+ match.mask->flags);
+ return -EOPNOTSUPP;
+ }
+ }
+
if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
struct flow_match_basic match;
--
2.43.0
Wed, Apr 24, 2024 at 03:42:48PM CEST, [email protected] wrote:
>This driver currently doesn't support any flower control flags.
>
>Implement check for control flags, such as can be set through
>`tc flower ... ip_flags frag`.
>
>Since qede_parse_flow_attr() are called by both qede_add_tc_flower_fltr()
>and qede_flow_spec_to_rule(), as the latter doesn't having access to
>extack, then flow_rule_*_control_flags() can't be used in this driver.
Why? You can pass null.
>
>This patch therefore re-implements flow_rule_match_has_control_flags(),
>but with error messaging via DP_NOTICE, instead of NL_SET_ERR_MSG_FMT_MOD.
>
>So in case any control flags are masked, we call DP_NOTICE() and
>return -EOPNOTSUPP.
>
>Only compile-tested.
>
>Signed-off-by: Asbj?rn Sloth T?nnesen <[email protected]>
>---
>
>This is AFAICT the last driver which didn't validate these flags.
>
>$ git grep FLOW_DISSECTOR_KEY_CONTROL drivers/
>$ git grep 'flow_rule_.*_control_flags' drivers/
>
> drivers/net/ethernet/qlogic/qede/qede_filter.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
>diff --git a/drivers/net/ethernet/qlogic/qede/qede_filter.c b/drivers/net/ethernet/qlogic/qede/qede_filter.c
>index a5ac21a0ee33..40f72e700d8e 100644
>--- a/drivers/net/ethernet/qlogic/qede/qede_filter.c
>+++ b/drivers/net/ethernet/qlogic/qede/qede_filter.c
>@@ -1843,6 +1843,19 @@ qede_parse_flow_attr(struct qede_dev *edev, __be16 proto,
> return -EPROTONOSUPPORT;
> }
>
>+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
>+ struct flow_match_control match;
>+
>+ flow_rule_match_control(rule, &match);
>+
>+ if (match.mask->flags) {
>+ DP_NOTICE(edev,
>+ "Unsupported match on control.flags %#x",
>+ match.mask->flags);
>+ return -EOPNOTSUPP;
>+ }
>+ }
>+
> if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
> struct flow_match_basic match;
>
>--
>2.43.0
>
>
Hi Jiri,
On 4/24/24 2:52 PM, Jiri Pirko wrote:
> Wed, Apr 24, 2024 at 03:42:48PM CEST, [email protected] wrote:
>> This driver currently doesn't support any flower control flags.
>>
>> Implement check for control flags, such as can be set through
>> `tc flower ... ip_flags frag`.
>>
>> Since qede_parse_flow_attr() are called by both qede_add_tc_flower_fltr()
>> and qede_flow_spec_to_rule(), as the latter doesn't having access to
>> extack, then flow_rule_*_control_flags() can't be used in this driver.
>
> Why? You can pass null.
Ah, I see. I hadn't traced that option down through the defines,
I incorrectly assumed that NL_SET_ERR_MSG* didn't allow NULL.
Currently thinking about doing v2 in this style:
if (flow_rule_match_has_control_flags(rule, extack)) {
if (!extack)
DP_NOTICE(edev, "Unsupported match on control.flags");
return -EOPNOTSUPP;
}
pw-bot: changes-requested
--
Best regards
Asbjørn Sloth Tønnesen
Network Engineer
Fiberby - AS42541
Wed, Apr 24, 2024 at 06:43:14PM CEST, [email protected] wrote:
>Hi Jiri,
>
>On 4/24/24 2:52 PM, Jiri Pirko wrote:
>> Wed, Apr 24, 2024 at 03:42:48PM CEST, [email protected] wrote:
>> > This driver currently doesn't support any flower control flags.
>> >
>> > Implement check for control flags, such as can be set through
>> > `tc flower ... ip_flags frag`.
>> >
>> > Since qede_parse_flow_attr() are called by both qede_add_tc_flower_fltr()
>> > and qede_flow_spec_to_rule(), as the latter doesn't having access to
>> > extack, then flow_rule_*_control_flags() can't be used in this driver.
>>
>> Why? You can pass null.
>
>Ah, I see. I hadn't traced that option down through the defines,
>I incorrectly assumed that NL_SET_ERR_MSG* didn't allow NULL.
>
>Currently thinking about doing v2 in this style:
>
>if (flow_rule_match_has_control_flags(rule, extack)) {
> if (!extack)
> DP_NOTICE(edev, "Unsupported match on control.flags");
> return -EOPNOTSUPP;
>}
Looks ok.
>
>pw-bot: changes-requested
>
>--
>Best regards
>Asbj?rn Sloth T?nnesen
>Network Engineer
>Fiberby - AS42541