From: Keefe LIU liuqifa@huawei.com
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8KDQE CVE: NA
-------------------------------------------------
Consider two IPVlan devices are set up on the same master, when they communicate with each other by TCP, the receive part is too fast to make the send packets coalesced, so in this case, the performace is not as good as we expect.
This patch introduces a local xmit queue for l2e mode, when the packets are sent to the IPVlan devices of the same master, the packets will be cloned and added to the local xmit queue, this operation can make the send packets coalesced and improve the TCP performace in this case.
Signed-off-by: Keefe LIU liuqifa@huawei.com Signed-off-by: Zhengchao Shao shaozhengchao@huawei.com --- drivers/net/ipvlan/ipvlan.h | 11 ++++ drivers/net/ipvlan/ipvlan_core.c | 34 +++++++++++- drivers/net/ipvlan/ipvlan_main.c | 89 ++++++++++++++++++++++++++++++-- include/uapi/linux/if_link.h | 2 +- 4 files changed, 131 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ipvlan/ipvlan.h b/drivers/net/ipvlan/ipvlan.h index 025e0c19ec25..4cf12fbe8f62 100644 --- a/drivers/net/ipvlan/ipvlan.h +++ b/drivers/net/ipvlan/ipvlan.h @@ -39,6 +39,11 @@
#define IPVLAN_QBACKLOG_LIMIT 1000
+#if IS_ENABLED(CONFIG_IPVLAN_L2E) +extern int sysctl_ipvlan_loop_qlen; +extern int sysctl_ipvlan_loop_delay; +#endif + typedef enum { IPVL_IPV6 = 0, IPVL_ICMPV6, @@ -70,6 +75,12 @@ struct ipvl_dev { netdev_features_t sfeatures; u32 msg_enable; spinlock_t addrs_lock; +#if IS_ENABLED(CONFIG_IPVLAN_L2E) + int local_packets_cached; + unsigned long local_timeout; + struct timer_list local_free_timer; + struct sk_buff_head local_xmit_queue; +#endif };
struct ipvl_addr { diff --git a/drivers/net/ipvlan/ipvlan_core.c b/drivers/net/ipvlan/ipvlan_core.c index 4e66eec7e65b..0614fe7027b0 100644 --- a/drivers/net/ipvlan/ipvlan_core.c +++ b/drivers/net/ipvlan/ipvlan_core.c @@ -757,9 +757,37 @@ static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev) }
#if IS_ENABLED(CONFIG_IPVLAN_L2E) +static int ipvlan_l2e_local_xmit_event(struct ipvl_dev *ipvlan, + struct sk_buff **pskb) +{ + struct sk_buff *nskb, *tskb; + + while ((ipvlan->local_packets_cached >= sysctl_ipvlan_loop_qlen) && + (tskb = skb_dequeue(&ipvlan->local_xmit_queue))) { + ipvlan->local_packets_cached -= tskb->truesize; + if (ipvlan->local_packets_cached < 0 || + skb_queue_empty(&ipvlan->local_xmit_queue)) + ipvlan->local_packets_cached = 0; + kfree_skb(tskb); + } + + nskb = skb_clone(*pskb, GFP_ATOMIC); + if (!nskb) + return NET_XMIT_DROP; + + ipvlan->local_timeout = jiffies + + (sysctl_ipvlan_loop_delay * HZ) / 1000; + mod_timer(&ipvlan->local_free_timer, ipvlan->local_timeout); + skb_queue_tail(&ipvlan->local_xmit_queue, *pskb); + ipvlan->local_packets_cached += (*pskb)->truesize; + *pskb = nskb; + + return 0; +} + static int ipvlan_xmit_mode_l2e(struct sk_buff *skb, struct net_device *dev) { - const struct ipvl_dev *ipvlan = netdev_priv(dev); + struct ipvl_dev *ipvlan = netdev_priv(dev); struct ethhdr *eth = eth_hdr(skb); struct ipvl_addr *addr; void *lyr3h; @@ -776,6 +804,10 @@ static int ipvlan_xmit_mode_l2e(struct sk_buff *skb, struct net_device *dev) consume_skb(skb); return NET_XMIT_DROP; } + + if (unlikely(ipvlan_l2e_local_xmit_event(ipvlan, + &skb))) + return NET_XMIT_DROP; return ipvlan_rcv_frame(addr, &skb, true); } } diff --git a/drivers/net/ipvlan/ipvlan_main.c b/drivers/net/ipvlan/ipvlan_main.c index 448f21a1c23e..c273b0ff1f9d 100644 --- a/drivers/net/ipvlan/ipvlan_main.c +++ b/drivers/net/ipvlan/ipvlan_main.c @@ -7,9 +7,52 @@ #include "ipvlan.h"
#if IS_ENABLED(CONFIG_IPVLAN_L2E) +static int one = 1; +static int delay_max = 100; +/* set loop queue length from 0 to 10 big packets(65536) */ +static int qlen_min; +static int qlen_max = 655360; + +int sysctl_ipvlan_loop_qlen = 131072; +int sysctl_ipvlan_loop_delay = 10; static int ipvlan_default_mode = IPVLAN_MODE_L3; module_param(ipvlan_default_mode, int, 0400); -MODULE_PARM_DESC(ipvlan_default_mode, "set ipvlan default mode: 0 for l2, 1 for l3, 2 for l3s, 3 for l2e, others invalid now"); +MODULE_PARM_DESC(ipvlan_default_mode, "set ipvlan default mode: 0 for l2, 1 for l3, 2 for l2e, 3 for l3s, others invalid now"); + +static struct ctl_table_header *ipvlan_table_hrd; +static struct ctl_table ipvlan_table[] = { + { + .procname = "loop_delay", + .data = &sysctl_ipvlan_loop_delay, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = &one, + .extra2 = &delay_max, + }, + { + .procname = "loop_qlen", + .data = &sysctl_ipvlan_loop_qlen, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = &qlen_min, + .extra2 = &qlen_max, + }, + { } +}; + +static int ipvlan_sysctl_init(void) +{ + ipvlan_table_hrd = register_net_sysctl(&init_net, + "net/ipvlan", ipvlan_table); + return !ipvlan_table_hrd ? -ENOMEM : 0; +} + +static void ipvlan_sysctl_exit(void) +{ + unregister_net_sysctl_table(ipvlan_table_hrd); +} #endif
static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval, @@ -168,6 +211,34 @@ static int ipvlan_init(struct net_device *dev) return 0; }
+#if IS_ENABLED(CONFIG_IPVLAN_L2E) +static void ipvlan_local_free_handler(struct timer_list *t) +{ + struct ipvl_dev *ipvlan = from_timer(ipvlan, t, local_free_timer); + + skb_queue_purge(&ipvlan->local_xmit_queue); + ipvlan->local_packets_cached = 0; +} + +static inline void ipvlan_local_init(struct net_device *dev) +{ + struct ipvl_dev *ipvlan = netdev_priv(dev); + + ipvlan->local_packets_cached = 0; + skb_queue_head_init(&ipvlan->local_xmit_queue); + timer_setup(&ipvlan->local_free_timer, + ipvlan_local_free_handler, 0); +} + +static inline void ipvlan_local_uninit(struct net_device *dev) +{ + struct ipvl_dev *ipvlan = netdev_priv(dev); + + del_timer(&ipvlan->local_free_timer); + skb_queue_purge(&ipvlan->local_xmit_queue); +} +#endif + static void ipvlan_uninit(struct net_device *dev) { struct ipvl_dev *ipvlan = netdev_priv(dev); @@ -193,6 +264,9 @@ static int ipvlan_open(struct net_device *dev) else dev->flags &= ~IFF_NOARP;
+#if IS_ENABLED(CONFIG_IPVLAN_L2E) + ipvlan_local_init(dev); +#endif rcu_read_lock(); list_for_each_entry_rcu(addr, &ipvlan->addrs, anode) ipvlan_ht_addr_add(ipvlan, addr); @@ -209,7 +283,9 @@ static int ipvlan_stop(struct net_device *dev)
dev_uc_unsync(phy_dev, dev); dev_mc_unsync(phy_dev, dev); - +#if IS_ENABLED(CONFIG_IPVLAN_L2E) + ipvlan_local_uninit(dev); +#endif rcu_read_lock(); list_for_each_entry_rcu(addr, &ipvlan->addrs, anode) ipvlan_ht_addr_del(addr); @@ -1061,7 +1137,11 @@ static int __init ipvlan_init_module(void) ipvlan_l3s_cleanup(); goto error; } - +#if IS_ENABLED(CONFIG_IPVLAN_L2E) + err = ipvlan_sysctl_init(); + if (err < 0) + pr_err("ipvlan proc init failed, continue\n"); +#endif return 0; error: unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block); @@ -1089,6 +1169,9 @@ static void __exit ipvlan_cleanup_module(void) unregister_inet6addr_validator_notifier( &ipvlan_addr6_vtor_notifier_block); #endif +#if IS_ENABLED(CONFIG_IPVLAN_L2E) + ipvlan_sysctl_exit(); +#endif }
module_init(ipvlan_init_module); diff --git a/include/uapi/linux/if_link.h b/include/uapi/linux/if_link.h index 0c965b941d14..8d03e491f5c2 100644 --- a/include/uapi/linux/if_link.h +++ b/include/uapi/linux/if_link.h @@ -741,10 +741,10 @@ enum { enum ipvlan_mode { IPVLAN_MODE_L2 = 0, IPVLAN_MODE_L3, - IPVLAN_MODE_L3S, #if IS_ENABLED(CONFIG_IPVLAN_L2E) IPVLAN_MODE_L2E, #endif + IPVLAN_MODE_L3S, IPVLAN_MODE_MAX };