Message-ID: <871qfw6w8d.fsf@nvidia.com>
Date: Mon, 21 Aug 2023 22:12:51 +0300
From: Vlad Buslov <vladbu@...dia.com>
To: Victor Nogueira <victor@...atatu.com>
CC: <jhs@...atatu.com>, <xiyou.wangcong@...il.com>, <jiri@...nulli.us>,
	<davem@...emloft.net>, <edumazet@...gle.com>, <kuba@...nel.org>,
	<pabeni@...hat.com>, <netdev@...r.kernel.org>, <mleitner@...hat.com>,
	<horms@...nel.org>, <pctammela@...atatu.com>, <kernel@...atatu.com>
Subject: Re: [PATCH net-next v2 1/3] net/sched: Introduce tc block netdev
 tracking infra


On Sat 19 Aug 2023 at 13:35, Victor Nogueira <victor@...atatu.com> wrote:
> The tc block is a collection of netdevs/ports which allow qdiscs to share
> filter block instances (as opposed to the traditional tc filter per port).
> Example:
> $ tc qdisc add dev ens7 ingress block 22
> $ tc qdisc add dev ens8 ingress block 22
>
> Now we can add a filter using the block index:
> $ tc filter add block 22 protocol ip pref 25 \
>   flower dst_ip 192.168.0.0/16 action drop
>
> Up to this point, the block is unaware of its ports. This patch fixes that
> and makes the tc block ports available to the datapath as well as control
> path on offloading.
>
> Suggested-by: Jiri Pirko <jiri@...dia.com>
> Co-developed-by: Jamal Hadi Salim <jhs@...atatu.com>
> Signed-off-by: Jamal Hadi Salim <jhs@...atatu.com>
> Co-developed-by: Pedro Tammela <pctammela@...atatu.com>
> Signed-off-by: Pedro Tammela <pctammela@...atatu.com>
> Signed-off-by: Victor Nogueira <victor@...atatu.com>
> ---
>  include/net/sch_generic.h |  4 ++
>  net/sched/cls_api.c       |  1 +
>  net/sched/sch_api.c       | 79 +++++++++++++++++++++++++++++++++++++--
>  net/sched/sch_generic.c   | 34 ++++++++++++++++-
>  4 files changed, 112 insertions(+), 6 deletions(-)
>
> diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
> index e92f73bb3198..824a0ecb5afc 100644
> --- a/include/net/sch_generic.h
> +++ b/include/net/sch_generic.h
> @@ -19,6 +19,7 @@
>  #include <net/gen_stats.h>
>  #include <net/rtnetlink.h>
>  #include <net/flow_offload.h>
> +#include <linux/xarray.h>
>  
>  struct Qdisc_ops;
>  struct qdisc_walker;
> @@ -126,6 +127,8 @@ struct Qdisc {
>  
>  	struct rcu_head		rcu;
>  	netdevice_tracker	dev_tracker;
> +	netdevice_tracker	in_block_tracker;
> +	netdevice_tracker	eg_block_tracker;
>  	/* private data */
>  	long privdata[] ____cacheline_aligned;
>  };
> @@ -458,6 +461,7 @@ struct tcf_chain {
>  };
>  
>  struct tcf_block {
> +	struct xarray ports; /* datapath accessible */
>  	/* Lock protects tcf_block and lifetime-management data of chains
>  	 * attached to the block (refcnt, action_refcnt, explicitly_created).
>  	 */
> diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c
> index a193cc7b3241..a976792ef02f 100644
> --- a/net/sched/cls_api.c
> +++ b/net/sched/cls_api.c
> @@ -1003,6 +1003,7 @@ static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
>  	refcount_set(&block->refcnt, 1);
>  	block->net = net;
>  	block->index = block_index;
> +	xa_init(&block->ports);

Missing dual call to xa_destroy() for this.

>  
>  	/* Don't store q pointer for blocks which are shared */
>  	if (!tcf_block_shared(block))
> diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
> index aa6b1fe65151..6c0c220cdb21 100644
> --- a/net/sched/sch_api.c
> +++ b/net/sched/sch_api.c
> @@ -1180,6 +1180,71 @@ static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
>  	return 0;
>  }
>  
> +static void qdisc_block_undo_set(struct Qdisc *sch, struct nlattr **tca)
> +{
> +	if (tca[TCA_INGRESS_BLOCK])
> +		sch->ops->ingress_block_set(sch, 0);
> +
> +	if (tca[TCA_EGRESS_BLOCK])
> +		sch->ops->egress_block_set(sch, 0);
> +}
> +
> +static int qdisc_block_add_dev(struct Qdisc *sch, struct net_device *dev,
> +			       struct nlattr **tca,
> +			       struct netlink_ext_ack *extack)
> +{
> +	const struct Qdisc_class_ops *cl_ops = sch->ops->cl_ops;
> +	struct tcf_block *in_block = NULL;
> +	struct tcf_block *eg_block = NULL;
> +	unsigned long cl = 0;
> +	int err;
> +
> +	if (tca[TCA_INGRESS_BLOCK]) {
> +		/* works for both ingress and clsact */
> +		cl = TC_H_MIN_INGRESS;
> +		in_block = cl_ops->tcf_block(sch, cl, NULL);
> +		if (!in_block) {
> +			NL_SET_ERR_MSG(extack, "Shared ingress block missing");
> +			return -EINVAL;
> +		}
> +
> +		err = xa_insert(&in_block->ports, dev->ifindex, dev, GFP_KERNEL);
> +		if (err) {
> +			NL_SET_ERR_MSG(extack, "ingress block dev insert failed");
> +			return err;
> +		}
> +
> +		netdev_hold(dev, &sch->in_block_tracker, GFP_KERNEL);
> +	}
> +
> +	if (tca[TCA_EGRESS_BLOCK]) {
> +		cl = TC_H_MIN_EGRESS;
> +		eg_block = cl_ops->tcf_block(sch, cl, NULL);
> +		if (!eg_block) {
> +			NL_SET_ERR_MSG(extack, "Shared egress block missing");
> +			err = -EINVAL;
> +			goto err_out;
> +		}
> +
> +		err = xa_insert(&eg_block->ports, dev->ifindex, dev, GFP_KERNEL);
> +		if (err) {
> +			netdev_put(dev, &sch->eg_block_tracker);
> +			NL_SET_ERR_MSG(extack, "Egress block dev insert failed");
> +			goto err_out;
> +		}
> +		netdev_hold(dev, &sch->eg_block_tracker, GFP_KERNEL);
> +	}
> +
> +	return 0;
> +err_out:
> +	if (in_block) {
> +		xa_erase(&in_block->ports, dev->ifindex);
> +		netdev_put(dev, &sch->in_block_tracker);
> +		NL_SET_ERR_MSG(extack, "ingress block dev insert failed");
> +	}
> +	return err;
> +}
> +
>  static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
>  				   struct netlink_ext_ack *extack)
>  {
> @@ -1270,7 +1335,7 @@ static struct Qdisc *qdisc_create(struct net_device *dev,
>  	sch = qdisc_alloc(dev_queue, ops, extack);
>  	if (IS_ERR(sch)) {
>  		err = PTR_ERR(sch);
> -		goto err_out2;
> +		goto err_out1;
>  	}
>  
>  	sch->parent = parent;
> @@ -1289,7 +1354,7 @@ static struct Qdisc *qdisc_create(struct net_device *dev,
>  			if (handle == 0) {
>  				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
>  				err = -ENOSPC;
> -				goto err_out3;
> +				goto err_out2;
>  			}
>  		}
>  		if (!netif_is_multiqueue(dev))
> @@ -1311,7 +1376,7 @@ static struct Qdisc *qdisc_create(struct net_device *dev,
>  
>  	err = qdisc_block_indexes_set(sch, tca, extack);
>  	if (err)
> -		goto err_out3;
> +		goto err_out2;
>  
>  	if (tca[TCA_STAB]) {
>  		stab = qdisc_get_stab(tca[TCA_STAB], extack);
> @@ -1350,6 +1415,10 @@ static struct Qdisc *qdisc_create(struct net_device *dev,
>  	qdisc_hash_add(sch, false);
>  	trace_qdisc_create(ops, dev, parent);
>  
> +	err = qdisc_block_add_dev(sch, dev, tca, extack);
> +	if (err)
> +		goto err_out4;
> +
>  	return sch;
>  
>  err_out4:
> @@ -1360,9 +1429,11 @@ static struct Qdisc *qdisc_create(struct net_device *dev,
>  		ops->destroy(sch);
>  	qdisc_put_stab(rtnl_dereference(sch->stab));
>  err_out3:
> +	qdisc_block_undo_set(sch, tca);

Is this a bugfix? This new call is for all sites that jump to
err_out{3|4} even though you only added new code to the end of the
function.

> +err_out2:
>  	netdev_put(dev, &sch->dev_tracker);
>  	qdisc_free(sch);
> -err_out2:
> +err_out1:
>  	module_put(ops->owner);
>  err_out:
>  	*errp = err;
> diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
> index 5d7e23f4cc0e..0fb51fd6f01e 100644
> --- a/net/sched/sch_generic.c
> +++ b/net/sched/sch_generic.c
> @@ -1048,7 +1048,12 @@ static void qdisc_free_cb(struct rcu_head *head)
>  
>  static void __qdisc_destroy(struct Qdisc *qdisc)
>  {
> -	const struct Qdisc_ops  *ops = qdisc->ops;
> +	struct net_device *dev = qdisc_dev(qdisc);
> +	const struct Qdisc_ops *ops = qdisc->ops;
> +	const struct Qdisc_class_ops *cops;
> +	struct tcf_block *block;
> +	unsigned long cl;
> +	u32 block_index;
>  
>  #ifdef CONFIG_NET_SCHED
>  	qdisc_hash_del(qdisc);
> @@ -1059,11 +1064,36 @@ static void __qdisc_destroy(struct Qdisc *qdisc)
>  
>  	qdisc_reset(qdisc);
>  
> +	cops = ops->cl_ops;
> +	if (ops->ingress_block_get) {
> +		block_index = ops->ingress_block_get(qdisc);
> +		if (block_index) {
> +			cl = TC_H_MIN_INGRESS;
> +			block = cops->tcf_block(qdisc, cl, NULL);
> +			if (block) {
> +				if (xa_erase(&block->ports, dev->ifindex))
> +					netdev_put(dev, &qdisc->in_block_tracker);
> +			}
> +		}
> +	}
> +
> +	if (ops->egress_block_get) {
> +		block_index = ops->egress_block_get(qdisc);
> +		if (block_index) {
> +			cl = TC_H_MIN_EGRESS;
> +			block = cops->tcf_block(qdisc, cl, NULL);
> +			if (block) {
> +				if (xa_erase(&block->ports, dev->ifindex))
> +					netdev_put(dev, &qdisc->eg_block_tracker);
> +			}
> +		}
> +	}
> +
>  	if (ops->destroy)
>  		ops->destroy(qdisc);
>  
>  	module_put(ops->owner);
> -	netdev_put(qdisc_dev(qdisc), &qdisc->dev_tracker);
> +	netdev_put(dev, &qdisc->dev_tracker);
>  
>  	trace_qdisc_destroy(qdisc);

Powered by blists - more mailing lists