[PATCH net 0/2] net: propagate max_gso_segs and max_gso_size

[PATCH net 0/2] net: propagate max_gso_segs and max_gso_size

[Eric Dumazet]

bridge code does not properly update max_gso_segs and max_gso_size.

Since this was not really obvious, first patch adds two new rtnetlink
attributes to help debugging this kind of issues (ip -d link)

Second patch fixes bridge code.

Eric Dumazet (2):
  net/rtnetlink: add IFLA_GSO_MAX_SEGS and IFLA_GSO_MAX_SIZE attributes
  bridge: update max_gso_segs and max_gso_size

 include/uapi/linux/if_link.h |  2 ++
 net/bridge/br_if.c           | 16 ++++++++++++++++
 net/core/rtnetlink.c         |  4 ++++
 3 files changed, 22 insertions(+)

-- 
2.8.0.rc3.226.g39d4020

[PATCH net 1/2] net/rtnetlink: add IFLA_GSO_MAX_SEGS and IFLA_GSO_MAX_SIZE attributes

[Eric Dumazet]

It can be useful to report dev->gso_max_segs and dev->gso_max_size
so that "ip -d link" can display them to help debugging.

For the moment, these attributes are read-only.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Petri Gynther <pgynther@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
---
 include/uapi/linux/if_link.h | 2 ++
 net/core/rtnetlink.c         | 4 ++++
 2 files changed, 6 insertions(+)

diff --git a/include/uapi/linux/if_link.h b/include/uapi/linux/if_link.h
index 8e3f88fa5b59..76d315ca597d 100644
--- a/include/uapi/linux/if_link.h
+++ b/include/uapi/linux/if_link.h
@@ -153,6 +153,8 @@ enum {
 	IFLA_LINK_NETNSID,
 	IFLA_PHYS_PORT_NAME,
 	IFLA_PROTO_DOWN,
+	IFLA_GSO_MAX_SEGS,
+	IFLA_GSO_MAX_SIZE,
 	__IFLA_MAX
 };
 
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index d2d9e5ebf58e..a69cd0c097b5 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -895,6 +895,8 @@ static noinline size_t if_nlmsg_size(const struct net_device *dev,
 	       + nla_total_size(4) /* IFLA_PROMISCUITY */
 	       + nla_total_size(4) /* IFLA_NUM_TX_QUEUES */
 	       + nla_total_size(4) /* IFLA_NUM_RX_QUEUES */
+	       + nla_total_size(4) /* IFLA_MAX_GSO_SEGS */
+	       + nla_total_size(4) /* IFLA_MAX_GSO_SIZE */
 	       + nla_total_size(1) /* IFLA_OPERSTATE */
 	       + nla_total_size(1) /* IFLA_LINKMODE */
 	       + nla_total_size(4) /* IFLA_CARRIER_CHANGES */
@@ -1223,6 +1225,8 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, struct net_device *dev,
 	    nla_put_u32(skb, IFLA_GROUP, dev->group) ||
 	    nla_put_u32(skb, IFLA_PROMISCUITY, dev->promiscuity) ||
 	    nla_put_u32(skb, IFLA_NUM_TX_QUEUES, dev->num_tx_queues) ||
+	    nla_put_u32(skb, IFLA_GSO_MAX_SEGS, dev->gso_max_segs) ||
+	    nla_put_u32(skb, IFLA_GSO_MAX_SIZE, dev->gso_max_size) ||
 #ifdef CONFIG_RPS
 	    nla_put_u32(skb, IFLA_NUM_RX_QUEUES, dev->num_rx_queues) ||
 #endif
-- 
2.8.0.rc3.226.g39d4020

[PATCH net 2/2] bridge: update max_gso_segs and max_gso_size

[Eric Dumazet]

It can be useful to lower max_gso_segs on NIC with very low
number of TX descriptors like bcmgenet.

However, this is defeated by bridge since it does not propagate
the lower value of max_gso_segs and max_gso_size.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Petri Gynther <pgynther@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
---
 net/bridge/br_if.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/net/bridge/br_if.c b/net/bridge/br_if.c
index a73df3315df9..8217aecf025b 100644
--- a/net/bridge/br_if.c
+++ b/net/bridge/br_if.c
@@ -437,6 +437,20 @@ int br_min_mtu(const struct net_bridge *br)
 	return mtu;
 }
 
+static void br_set_gso_limits(struct net_bridge *br)
+{
+	unsigned int gso_max_size = GSO_MAX_SIZE;
+	u16 gso_max_segs = GSO_MAX_SEGS;
+	const struct net_bridge_port *p;
+
+	list_for_each_entry(p, &br->port_list, list) {
+		gso_max_size = min(gso_max_size, p->dev->gso_max_size);
+		gso_max_segs = min(gso_max_segs, p->dev->gso_max_segs);
+	}
+	br->dev->gso_max_size = gso_max_size;
+	br->dev->gso_max_segs = gso_max_segs;
+}
+
 /*
  * Recomputes features using slave's features
  */
@@ -564,6 +578,7 @@ int br_add_if(struct net_bridge *br, struct net_device *dev)
 		call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 
 	dev_set_mtu(br->dev, br_min_mtu(br));
+	br_set_gso_limits(br);
 
 	kobject_uevent(&p->kobj, KOBJ_ADD);
 
@@ -610,6 +625,7 @@ int br_del_if(struct net_bridge *br, struct net_device *dev)
 	del_nbp(p);
 
 	dev_set_mtu(br->dev, br_min_mtu(br));
+	br_set_gso_limits(br);
 
 	spin_lock_bh(&br->lock);
 	changed_addr = br_stp_recalculate_bridge_id(br);
-- 
2.8.0.rc3.226.g39d4020

Re: [PATCH net 0/2] net: propagate max_gso_segs and max_gso_size

[David Miller]

From: Eric Dumazet <edumazet@google.com>
Date: Mon, 21 Mar 2016 09:55:09 -0700

> bridge code does not properly update max_gso_segs and max_gso_size.
> 
> Since this was not really obvious, first patch adds two new rtnetlink
> attributes to help debugging this kind of issues (ip -d link)
> 
> Second patch fixes bridge code.

Series applied, thanks Eric.