lkml.org 
[lkml]   [2022]   [Mar]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH net-next v3 1/2] net: flow_offload: add tc police action parameters
On Tue, Mar 15, 2022 at 09:13:58PM +0200, Vladimir Oltean wrote:
> Hello Jianbo,
>
> On Thu, Feb 24, 2022 at 10:29:07AM +0000, Jianbo Liu wrote:
> > The current police offload action entry is missing exceed/notexceed
> > actions and parameters that can be configured by tc police action.
> > Add the missing parameters as a pre-step for offloading police actions
> > to hardware.
> >
> > Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
> > Signed-off-by: Roi Dayan <roid@nvidia.com>
> > Reviewed-by: Ido Schimmel <idosch@nvidia.com>
> > ---
> > include/net/flow_offload.h | 9 +++++++
> > include/net/tc_act/tc_police.h | 30 ++++++++++++++++++++++
> > net/sched/act_police.c | 46 ++++++++++++++++++++++++++++++++++
> > 3 files changed, 85 insertions(+)
> >
> > diff --git a/include/net/flow_offload.h b/include/net/flow_offload.h
> > index 5b8c54eb7a6b..74f44d44abe3 100644
> > --- a/include/net/flow_offload.h
> > +++ b/include/net/flow_offload.h
> > @@ -148,6 +148,8 @@ enum flow_action_id {
> > FLOW_ACTION_MPLS_MANGLE,
> > FLOW_ACTION_GATE,
> > FLOW_ACTION_PPPOE_PUSH,
> > + FLOW_ACTION_JUMP,
> > + FLOW_ACTION_PIPE,
> > NUM_FLOW_ACTIONS,
> > };
> >
> > @@ -235,9 +237,16 @@ struct flow_action_entry {
> > struct { /* FLOW_ACTION_POLICE */
> > u32 burst;
> > u64 rate_bytes_ps;
> > + u64 peakrate_bytes_ps;
> > + u32 avrate;
> > + u16 overhead;
> > u64 burst_pkt;
> > u64 rate_pkt_ps;
> > u32 mtu;
> > + struct {
> > + enum flow_action_id act_id;
> > + u32 extval;
> > + } exceed, notexceed;
> > } police;
> > struct { /* FLOW_ACTION_CT */
> > int action;
> > diff --git a/include/net/tc_act/tc_police.h b/include/net/tc_act/tc_police.h
> > index 72649512dcdd..283bde711a42 100644
> > --- a/include/net/tc_act/tc_police.h
> > +++ b/include/net/tc_act/tc_police.h
> > @@ -159,4 +159,34 @@ static inline u32 tcf_police_tcfp_mtu(const struct tc_action *act)
> > return params->tcfp_mtu;
> > }
> >
> > +static inline u64 tcf_police_peakrate_bytes_ps(const struct tc_action *act)
> > +{
> > + struct tcf_police *police = to_police(act);
> > + struct tcf_police_params *params;
> > +
> > + params = rcu_dereference_protected(police->params,
> > + lockdep_is_held(&police->tcf_lock));
> > + return params->peak.rate_bytes_ps;
> > +}
> > +
> > +static inline u32 tcf_police_tcfp_ewma_rate(const struct tc_action *act)
> > +{
> > + struct tcf_police *police = to_police(act);
> > + struct tcf_police_params *params;
> > +
> > + params = rcu_dereference_protected(police->params,
> > + lockdep_is_held(&police->tcf_lock));
> > + return params->tcfp_ewma_rate;
> > +}
> > +
> > +static inline u16 tcf_police_rate_overhead(const struct tc_action *act)
> > +{
> > + struct tcf_police *police = to_police(act);
> > + struct tcf_police_params *params;
> > +
> > + params = rcu_dereference_protected(police->params,
> > + lockdep_is_held(&police->tcf_lock));
> > + return params->rate.overhead;
> > +}
> > +
> > #endif /* __NET_TC_POLICE_H */
> > diff --git a/net/sched/act_police.c b/net/sched/act_police.c
> > index 0923aa2b8f8a..a2275eef6877 100644
> > --- a/net/sched/act_police.c
> > +++ b/net/sched/act_police.c
> > @@ -405,20 +405,66 @@ static int tcf_police_search(struct net *net, struct tc_action **a, u32 index)
> > return tcf_idr_search(tn, a, index);
> > }
> >
> > +static int tcf_police_act_to_flow_act(int tc_act, u32 *extval)
> > +{
> > + int act_id = -EOPNOTSUPP;
> > +
> > + if (!TC_ACT_EXT_OPCODE(tc_act)) {
> > + if (tc_act == TC_ACT_OK)
> > + act_id = FLOW_ACTION_ACCEPT;
> > + else if (tc_act == TC_ACT_SHOT)
> > + act_id = FLOW_ACTION_DROP;
> > + else if (tc_act == TC_ACT_PIPE)
> > + act_id = FLOW_ACTION_PIPE;
> > + } else if (TC_ACT_EXT_CMP(tc_act, TC_ACT_GOTO_CHAIN)) {
> > + act_id = FLOW_ACTION_GOTO;
> > + *extval = tc_act & TC_ACT_EXT_VAL_MASK;
> > + } else if (TC_ACT_EXT_CMP(tc_act, TC_ACT_JUMP)) {
> > + act_id = FLOW_ACTION_JUMP;
> > + *extval = tc_act & TC_ACT_EXT_VAL_MASK;
> > + }
> > +
> > + return act_id;
> > +}
> > +
> > static int tcf_police_offload_act_setup(struct tc_action *act, void *entry_data,
> > u32 *index_inc, bool bind)
> > {
> > if (bind) {
> > struct flow_action_entry *entry = entry_data;
> > + struct tcf_police *police = to_police(act);
> > + struct tcf_police_params *p;
> > + int act_id;
> > +
> > + p = rcu_dereference_protected(police->params,
> > + lockdep_is_held(&police->tcf_lock));
> >
> > entry->id = FLOW_ACTION_POLICE;
> > entry->police.burst = tcf_police_burst(act);
> > entry->police.rate_bytes_ps =
> > tcf_police_rate_bytes_ps(act);
> > + entry->police.peakrate_bytes_ps = tcf_police_peakrate_bytes_ps(act);
> > + entry->police.avrate = tcf_police_tcfp_ewma_rate(act);
> > + entry->police.overhead = tcf_police_rate_overhead(act);
> > entry->police.burst_pkt = tcf_police_burst_pkt(act);
> > entry->police.rate_pkt_ps =
> > tcf_police_rate_pkt_ps(act);
> > entry->police.mtu = tcf_police_tcfp_mtu(act);
> > +
> > + act_id = tcf_police_act_to_flow_act(police->tcf_action,
> > + &entry->police.exceed.extval);
>
> I don't know why just now, but I observed an apparent regression here
> with these commands:
>
> root@debian:~# tc qdisc add dev swp3 clsact
> root@debian:~# tc filter add dev swp3 ingress protocol ip flower skip_sw ip_proto icmp action police rate 100Mbit burst 10000
> [ 45.767900] tcf_police_act_to_flow_act: 434: tc_act 1
> [ 45.773100] tcf_police_offload_act_setup: 475, act_id -95
> Error: cls_flower: Failed to setup flow action.
> We have an error talking to the kernel, -1
>
> The reason why I'm not sure is because I don't know if this should have
> worked as intended or not. I am remarking just now in "man tc-police"
> that the default conform-exceed action is "reclassify".
>
> So if I specify "conform-exceed drop", things are as expected, but with
> the default (implicitly "conform-exceed reclassify") things fail with
> -EOPNOTSUPP because tcf_police_act_to_flow_act() doesn't handle a
> police->tcf_action of TC_ACT_RECLASSIFY.
>
> Should it?

Even if tcf_police_act_to_flow_act() handled "reclassify", the
configuration would have been rejected later on by the relevant device
driver since they all support "drop" for exceed action and nothing else.

I don't know why iproute2 defaults to "reclassify", but the
configuration in the example does something different in the SW and HW
data paths. One ugly suggestion to keep this case working it to have
tcf_police_act_to_flow_act() default to "drop" and emit a warning via
extack so that user space is at least aware of this misconfiguration.

>
> > + if (act_id < 0)
> > + return act_id;
> > +
> > + entry->police.exceed.act_id = act_id;
> > +
> > + act_id = tcf_police_act_to_flow_act(p->tcfp_result,
> > + &entry->police.notexceed.extval);
> > + if (act_id < 0)
> > + return act_id;
> > +
> > + entry->police.notexceed.act_id = act_id;
> > +
> > *index_inc = 1;
> > } else {
> > struct flow_offload_action *fl_action = entry_data;
> > --
> > 2.26.2
> >

\
 
 \ /
  Last update: 2022-03-17 14:23    [W:4.763 / U:0.024 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site