ipv6: avoid taking refcnt on dst during route lookup
Subsystems (5)
v3 cover ipv6: avoid taking refcnt on dst during route lookup
From: Wei Wang <[email protected]>
Ipv6 route lookup code always grabs refcnt on the dst for the caller. But for certain cases, grabbing refcnt is not always necessary if the call path is rcu protected and the caller does not cache the dst. Another issue in the route lookup logic is:
When there are multiple custom rules, we have to do the lookup into each table associated to each rule individually. And when we can't find the route in one table, we grab and release refcnt on net->ipv6.ip6_null_entry before going to the next table. This operation is completely redundant, and causes false issue because net->ipv6.ip6_null_entry is a shared object.
This patch set introduces a new flag RT6_LOOKUP_F_DST_NOREF for route lookup callers to set, to avoid any manipulation on the dst refcnt. And it converts the major input and output path to use it.
The performance gain is noticable.
I ran synflood tests between 2 hosts under the same switch. Both hosts have 20G mlx NIC, and 8 tx/rx queues.
Sender sends pure SYN flood with random src IPs and ports using trafgen. Receiver has a simple TCP listener on the target port.
Both hosts have multiple custom rules:
- For incoming packets, only local table is traversed.
- For outgoing packets, 3 tables are traversed to find the route. The packet processing rate on the receiver is as follows:
- Before the fix: 3.78Mpps
- After the fix: 5.50Mpps
v2->v3:
- Handled fib6_rule_lookup() when CONFIG_IPV6_MULTIPLE_TABLES is not configured in patch 03 (suggested by David Ahern)
- Removed the renaming of l3mdev_link_scope_lookup() in patch 05 (suggested by David Ahern)
- Moved definition of ip6_route_output_flags() from an inline function in /net/ipv6/route.c to net/ipv6/route.c in order to address kbuild error in patch 05
v1->v2:
- Added a helper ip6_rt_put_flags() in patch 3 suggested by David Miller
Wei Wang (5):
ipv6: introduce RT6_LOOKUP_F_DST_NOREF flag in ip6_pol_route()
ipv6: initialize rt6->rt6i_uncached in all pre-allocated dst entries
ipv6: honor RT6_LOOKUP_F_DST_NOREF in rule lookup logic
ipv6: convert rx data path to not take refcnt on dst
ipv6: convert major tx path to use RT6_LOOKUP_F_DST_NOREF
drivers/net/vrf.c | 5 +- include/net/ip6_route.h | 15 ++++++ net/ipv6/fib6_rules.c | 12 +++-- net/ipv6/ip6_fib.c | 5 +- net/ipv6/route.c | 112 +++++++++++++++++++++++----------------- net/l3mdev/l3mdev.c | 7 ++- 6 files changed, 95 insertions(+), 61 deletions(-)
Signature · 1 line
2.22.0.410.gd8fdbe21b5-goog
PATCH v3 1/5 ipv6: introduce RT6_LOOKUP_F_DST_NOREF flag in ip6_pol_route()
From: Wei Wang <[email protected]>
This new flag is to instruct the route lookup function to not take refcnt on the dst entry. The user which does route lookup with this flag must properly use rcu protection.ip6_pol_route() is the major route lookup function for both tx and rx path.
In this function:
Do not take refcnt on dst if RT6_LOOKUP_F_DST_NOREF flag is set, and directly return the route entry. The caller should be holding rcu lock when using this flag, and decide whether to take refcnt or not.
One note on the dst cache in the uncached_list:
As uncached_list does not consume refcnt, one refcnt is always returned back to the caller even if RT6_LOOKUP_F_DST_NOREF flag is set. Uncached dst is only possible in the output path. So in such call path, caller MUST check if the dst is in the uncached_list before assuming that there is no refcnt taken on the returned dst.
--- include/net/ip6_route.h | 1 + net/ipv6/route.c | 73 +++++++++++++++++------------------------ 2 files changed, 31 insertions(+), 43 deletions(-)
include/net/ip6_route.h +1 -0
@@ -36,6 +36,7 @@ struct route_info {#define RT6_LOOKUP_F_SRCPREF_PUBLIC 0x00000010#define RT6_LOOKUP_F_SRCPREF_COA 0x00000020#define RT6_LOOKUP_F_IGNORE_LINKSTATE 0x00000040 +#define RT6_LOOKUP_F_DST_NOREF 0x00000080/* We do not (yet ?) support IPv6 jumbograms (RFC 2675) * Unlike IPv4, hdr->seg_len doesn't include the IPv6 header
net/ipv6/route.c +30 -43
123 lines not on this page · raw message
Signature · 1 line
2.22.0.410.gd8fdbe21b5-goog
PATCH v3 2/5 ipv6: initialize rt6->rt6i_uncached in all pre-allocated dst entries
In reply to Wei WangFrom: Wei Wang <[email protected]>
Initialize rt6->rt6i_uncached on the following pre-allocated dsts:net->ipv6.ip6_null_entrynet->ipv6.ip6_prohibit_entry net->ipv6.ip6_blk_hole_entry
This is a preparation patch for later commits to be able to distinguish dst entries in uncached list by doing:
!list_empty(rt6->rt6i_uncached)
--- net/ipv6/route.c | 3 +++ 1 file changed, 3 insertions(+)
net/ipv6/route.c +3 -0
@@ -6010,6 +6010,7 @@ static int __net_init ip6_route_net_init(struct net *net) net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;dst_init_metrics (&net->ipv6.ip6_null_entry->dst, ip6_template_metrics, true); +INIT_LIST_HEAD (&net->ipv6.ip6_null_entry->rt6i_uncached);#ifdef CONFIG_IPV6_MULTIPLE_TABLES net->ipv6.fib6_has_custom_rules = false;
@@ -6021,6 +6022,7 @@ static int __net_init ip6_route_net_init(struct net *net) net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;dst_init_metrics (&net->ipv6.ip6_prohibit_entry->dst, ip6_template_metrics, true); +INIT_LIST_HEAD (&net->ipv6.ip6_prohibit_entry->rt6i_uncached); net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template, sizeof(*net->ipv6.ip6_blk_hole_entry),
@@ -6030,6 +6032,7 @@ static int __net_init ip6_route_net_init(struct net *net) net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;dst_init_metrics (&net->ipv6.ip6_blk_hole_entry->dst, ip6_template_metrics, true); +INIT_LIST_HEAD (&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);#endif net->ipv6.sysctl.flush_delay =0 ;
Signature · 1 line
2.22.0.410.gd8fdbe21b5-goog
PATCH v3 3/5 ipv6: honor RT6_LOOKUP_F_DST_NOREF in rule lookup logic
In reply to Wei WangFrom: Wei Wang <[email protected]>
This patch specifically converts the rule lookup logic to honor this flag and not release refcnt when traversing each rule and calling lookup() on each routing table.
Similar to previous patch, we also need some special handling of dst entries in uncached list because there is always 1 refcnt taken for them even if RT6_LOOKUP_F_DST_NOREF flag is set.
--- include/net/ip6_route.h | 10 ++++++++++ net/ipv6/fib6_rules.c | 12 +++++++----- net/ipv6/ip6_fib.c | 5 +++-- 3 files changed, 20 insertions(+), 7 deletions(-)
include/net/ip6_route.h +10 -0
@@ -94,6 +94,16 @@ static inline struct dst_entry *ip6_route_output(struct net *net,return ip6_route_output_flags (net, sk, fl6,0 ); } +/* Only conditionally release dst if flags indicates +* !RT6_LOOKUP_F_DST_NOREF or dst is in uncached_list. +*/ +static inline void ip6_rt_put_flags (struct rt6_info *rt,int flags) +{ +if (!(flags &RT6_LOOKUP_F_DST_NOREF ) || + !list_empty (&rt->rt6i_uncached)) +ip6_rt_put (rt); +} +struct dst_entry *ip6_route_lookup (struct net *net,struct flowi6 *fl6,const struct sk_buff *skb,int flags); struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
net/ipv6/fib6_rules.c +7 -5
@@ -113,14 +113,15 @@ struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, rt =lookup (net, net->ipv6.fib6_local_tbl, fl6, skb, flags);if (rt != net->ipv6.ip6_null_entry && rt->dst.error != -EAGAIN )return &rt->dst; -ip6_rt_put (rt); +ip6_rt_put_flags (rt, flags); rt =lookup (net, net->ipv6.fib6_main_tbl, fl6, skb, flags);if (rt->dst.error != -EAGAIN )return &rt->dst; -ip6_rt_put (rt); +ip6_rt_put_flags (rt, flags); } -dst_hold (&net->ipv6.ip6_null_entry->dst); +if (!(flags &RT6_LOOKUP_F_DST_NOREF )) +dst_hold (&net->ipv6.ip6_null_entry->dst);return &net->ipv6.ip6_null_entry->dst; }
@@ -237,13 +238,14 @@ static int __fib6_rule_action(struct fib_rule *rule, struct flowi *flp, goto out; } again: -ip6_rt_put (rt); +ip6_rt_put_flags (rt, flags); err = -EAGAIN ; rt =NULL ; goto out; discard_pkt: -dst_hold (&rt->dst); +if (!(flags &RT6_LOOKUP_F_DST_NOREF )) +dst_hold (&rt->dst); out: res->rt6 = rt;return err;
net/ipv6/ip6_fib.c +3 -2
@@ -316,9 +316,10 @@ struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, rt =lookup (net, net->ipv6.fib6_main_tbl, fl6, skb, flags);if (rt->dst.error == -EAGAIN ) { -ip6_rt_put (rt); +ip6_rt_put_flags (rt, flags); rt = net->ipv6.ip6_null_entry; -dst_hold (&rt->dst); +if (!(flags |RT6_LOOKUP_F_DST_NOREF )) +dst_hold (&rt->dst); }return &rt->dst;
Signature · 1 line
2.22.0.410.gd8fdbe21b5-goog
PATCH v3 4/5 ipv6: convert rx data path to not take refcnt on dst
In reply to Wei WangFrom: Wei Wang <[email protected]>
ip6_route_input() is the key function to do the route lookup in the rx data path. All the callers to this function are already holding rcu lock. So it is fairly easy to convert it to not take refcnt on the dst:
We pass in flag RT6_LOOKUP_F_DST_NOREF and do skb_dst_set_noref(). This saves a few atomic inc or dec operations and should boost performance overall.
This also makes the logic more aligned with v4.
--- net/ipv6/route.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
net/ipv6/route.c +4 -3
@@ -2375,11 +2375,12 @@ u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,return mhash >>1 ; } +/* Called with rcu held */ void ip6_route_input (struct sk_buff *skb) {const struct ipv6hdr *iph =ipv6_hdr (skb);struct net *net =dev_net (skb->dev); -int flags =RT6_LOOKUP_F_HAS_SADDR ; +int flags =RT6_LOOKUP_F_HAS_SADDR |RT6_LOOKUP_F_DST_NOREF ;struct ip_tunnel_info *tun_info;struct flowi6 fl6 = { .flowi6_iif = skb->dev->ifindex,
@@ -2401,8 +2402,8 @@ void ip6_route_input(struct sk_buff *skb)if (unlikely (fl6.flowi6_proto ==IPPROTO_ICMPV6 )) fl6.mp_hash =rt6_multipath_hash (net, &fl6, skb, flkeys);skb_dst_drop (skb); -skb_dst_set (skb, -ip6_route_input_lookup (net, skb->dev, &fl6, skb, flags)); +skb_dst_set_noref (skb,ip6_route_input_lookup (net, skb->dev, + &fl6, skb, flags)); } static struct rt6_info *ip6_pol_route_output(struct net *net,
Signature · 1 line
2.22.0.410.gd8fdbe21b5-goog
PATCH v3 5/5 ipv6: convert major tx path to use RT6_LOOKUP_F_DST_NOREF
In reply to Wei WangFrom: Wei Wang <[email protected]>
For tx path, in most cases, we still have to take refcnt on the dst cause the caller is caching the dst somewhere. But it still is beneficial to make use of RT6_LOOKUP_F_DST_NOREF flag while doing the route lookup. It is cause this flag prevents manipulating refcnt on net->ipv6.ip6_null_entry when doing fib6_rule_lookup() to traverse each routing table. The null_entry is a shared object and constant updates on it cause false sharing.
We converted the current major lookup function ip6_route_output_flags() to make use of RT6_LOOKUP_F_DST_NOREF.
Together with the change in the rx path, we see noticable performance boost:
I ran synflood tests between 2 hosts under the same switch. Both hosts have 20G mlx NIC, and 8 tx/rx queues.
Sender sends pure SYN flood with random src IPs and ports using trafgen. Receiver has a simple TCP listener on the target port.
Both hosts have multiple custom rules:
- For incoming packets, only local table is traversed.
- For outgoing packets, 3 tables are traversed to find the route. The packet processing rate on the receiver is as follows:
- Before the fix: 3.78Mpps
- After the fix: 5.50Mpps
--- drivers/net/vrf.c | 5 +++-- include/net/ip6_route.h | 4 ++++ net/ipv6/route.c | 29 +++++++++++++++++++++++++++-- net/l3mdev/l3mdev.c | 7 +++---- 4 files changed, 37 insertions(+), 8 deletions(-)
drivers/net/vrf.c +3 -2
@@ -1072,12 +1072,14 @@ static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,#if IS_ENABLED (CONFIG_IPV6 )/* send to link-local or multicast address via interface enslaved to * VRF device. Force lookup to VRF table without changing flow struct +* Note: Caller to this function must hold rcu_read_lock() and no refcnt +* is taken on the dst by this function. */ static struct dst_entry *vrf_link_scope_lookup (const struct net_device *dev,struct flowi6 *fl6) {struct net *net =dev_net (dev); -int flags =RT6_LOOKUP_F_IFACE ; +int flags =RT6_LOOKUP_F_IFACE |RT6_LOOKUP_F_DST_NOREF ;struct dst_entry *dst =NULL ;struct rt6_info *rt;
@@ -1087,7 +1089,6 @@ static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,*/ if (fl6->flowi6_oif == dev->ifindex) { dst = &net->ipv6.ip6_null_entry->dst; -dst_hold (dst);return dst; }
include/net/ip6_route.h +4 -0
@@ -84,6 +84,10 @@ struct dst_entry *ip6_route_input_lookup(struct net *net,struct flowi6 *fl6, const struct sk_buff *skb, int flags); +struct dst_entry *ip6_route_output_flags_noref (struct net *net, +const struct sock *sk, +struct flowi6 *fl6,int flags); +struct dst_entry *ip6_route_output_flags (struct net *net,const struct sock *sk,struct flowi6 *fl6,int flags);
net/ipv6/route.c +27 -2
@@ -2415,8 +2415,9 @@ static struct rt6_info *ip6_pol_route_output(struct net *net,return ip6_pol_route (net, table, fl6->flowi6_oif, fl6, skb, flags); } -struct dst_entry *ip6_route_output_flags (struct net *net,const struct sock *sk, -struct flowi6 *fl6,int flags) +struct dst_entry *ip6_route_output_flags_noref (struct net *net, +const struct sock *sk, +struct flowi6 *fl6,int flags) {bool any_src;
@@ -2424,6 +2425,7 @@ struct dst_entry *ip6_route_output_flags(struct net *net, const struct sock *sk, (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {struct dst_entry *dst; +/* This function does not take refcnt on the dst */ dst =l3mdev_link_scope_lookup (net, fl6);if (dst)return dst;
@@ -2431,6 +2433,7 @@ struct dst_entry *ip6_route_output_flags(struct net *net, const struct sock *sk, fl6->flowi6_iif =LOOPBACK_IFINDEX ; + flags |=RT6_LOOKUP_F_DST_NOREF ; any_src =ipv6_addr_any (&fl6->saddr);if ((sk && sk->sk_bound_dev_if) ||rt6_need_strict (&fl6->daddr) || (fl6->flowi6_oif && any_src))
@@ -2443,6 +2446,28 @@ struct dst_entry *ip6_route_output_flags(struct net *net, const struct sock *sk,return fib6_rule_lookup (net, fl6,NULL , flags, ip6_pol_route_output); } +EXPORT_SYMBOL_GPL (ip6_route_output_flags_noref); + +struct dst_entry *ip6_route_output_flags (struct net *net, +const struct sock *sk, +struct flowi6 *fl6, +int flags) +{ +struct dst_entry *dst; +struct rt6_info *rt6; + +rcu_read_lock (); + dst =ip6_route_output_flags_noref (net, sk, fl6, flags); + rt6 = (struct rt6_info *)dst; +/* For dst cached in uncached_list, refcnt is already taken. */ +if (list_empty (&rt6->rt6i_uncached) && !dst_hold_safe (dst)) { + dst = &net->ipv6.ip6_null_entry->dst; +dst_hold (dst); + } +rcu_read_unlock (); + +return dst; +}EXPORT_SYMBOL_GPL (ip6_route_output_flags);struct dst_entry *ip6_blackhole_route (struct net *net,struct dst_entry *dst_orig)
net/l3mdev/l3mdev.c +3 -4
@@ -118,6 +118,8 @@ EXPORT_SYMBOL_GPL(l3mdev_fib_table_by_index);* local and multicast addresses * @net: network namespace for device index lookup * @fl6: IPv6 flow struct for lookup +* This function does not hold refcnt on the returned dst. +* Caller must hold rcu_read_lock(). */ struct dst_entry *l3mdev_link_scope_lookup(struct net *net,
@@ -126,9 +128,8 @@ struct dst_entry *l3mdev_link_scope_lookup(struct net *net,struct dst_entry *dst =NULL ;struct net_device *dev; +WARN_ON_ONCE (!rcu_read_lock_held ());if (fl6->flowi6_oif) { -rcu_read_lock (); - dev =dev_get_by_index_rcu (net, fl6->flowi6_oif);if (dev &&netif_is_l3_slave (dev)) dev =netdev_master_upper_dev_get_rcu (dev);
@@ -136,8 +137,6 @@ struct dst_entry *l3mdev_link_scope_lookup(struct net *net,if (dev &&netif_is_l3_master (dev) && dev->l3mdev_ops->l3mdev_link_scope_lookup) dst = dev->l3mdev_ops->l3mdev_link_scope_lookup (dev, fl6); - -rcu_read_unlock (); }return dst;
Signature · 1 line
2.22.0.410.gd8fdbe21b5-goog
Quoted 17 lines · Wei Wang
From: Wei Wang <[email protected]>
Ipv6 route lookup code always grabs refcnt on the dst for the caller. But for certain cases, grabbing refcnt is not always necessary if the call path is rcu protected and the caller does not cache the dst. Another issue in the route lookup logic is:
When there are multiple custom rules, we have to do the lookup into each table associated to each rule individually. And when we can't find the route in one table, we grab and release refcnt onnet->ipv6.ip6_null_entrybefore going to the next table. This operation is completely redundant, and causes false issue becausenet->ipv6.ip6_null_entryis a shared object.This patch set introduces a new flag
RT6_LOOKUP_F_DST_NOREFfor route lookup callers to set, to avoid any manipulation on the dst refcnt. And it converts the major input and output path to use it.
Quoted 12 lines
The performance gain is noticable.
I ran synflood tests between 2 hosts under the same switch. Both hosts have 20G mlx NIC, and 8 tx/rx queues.
Sender sends pure SYN flood with random src IPs and ports using trafgen. Receiver has a simple TCP listener on the target port.
Both hosts have multiple custom rules:
- For incoming packets, only local table is traversed.
- For outgoing packets, 3 tables are traversed to find the route. The packet processing rate on the receiver is as follows:
- Before the fix: 3.78Mpps
- After the fix: 5.50Mpps
LGTM. Thanks for doing this - big improvement.
Eric Dumazet
Reviewed-byQuoted 32 lines · David Ahern
Quoted 17 lines · Wei Wang
From: Wei Wang <[email protected]>
Ipv6 route lookup code always grabs refcnt on the dst for the caller. But for certain cases, grabbing refcnt is not always necessary if the call path is rcu protected and the caller does not cache the dst. Another issue in the route lookup logic is:
When there are multiple custom rules, we have to do the lookup into each table associated to each rule individually. And when we can't find the route in one table, we grab and release refcnt onnet->ipv6.ip6_null_entrybefore going to the next table. This operation is completely redundant, and causes false issue becausenet->ipv6.ip6_null_entryis a shared object.This patch set introduces a new flag
RT6_LOOKUP_F_DST_NOREFfor route lookup callers to set, to avoid any manipulation on the dst refcnt. And it converts the major input and output path to use it.Quoted 12 lines
The performance gain is noticable.
I ran synflood tests between 2 hosts under the same switch. Both hosts have 20G mlx NIC, and 8 tx/rx queues.
Sender sends pure SYN flood with random src IPs and ports using trafgen. Receiver has a simple TCP listener on the target port.
Both hosts have multiple custom rules:
- For incoming packets, only local table is traversed.
- For outgoing packets, 3 tables are traversed to find the route. The packet processing rate on the receiver is as follows:
- Before the fix: 3.78Mpps
- After the fix: 5.50Mpps
Quoted 3 lines
LGTM. Thanks for doing this - big improvement.
Reviewed-by David Ahern <[email protected]>
David Miller
maintainerFrom: Wei Wang <[email protected]>
Date: Thu, 20 Jun 2019 17:36:36 -0700
Quoted 8 lines
v2->v3:
- Handledfib6_rule_lookup()whenCONFIG_IPV6_MULTIPLE_TABLESis not configured in patch 03 (suggested by David Ahern)
- Removed the renaming ofl3mdev_link_scope_lookup()in patch 05 (suggested by David Ahern)
- Moved definition ofip6_route_output_flags()from an inline function in /net/ipv6/route.c tonet/ipv6/route.cin order to address kbuild error in patch 05
I'll give David A. a chance to review this before applying.
Quoted 12 lines · David Miller
From: Wei Wang <[email protected]>
Date: Thu, 20 Jun 2019 17:36:36 -0700Quoted 8 lines
v2->v3:
- Handledfib6_rule_lookup()whenCONFIG_IPV6_MULTIPLE_TABLESis not configured in patch 03 (suggested by David Ahern)
- Removed the renaming ofl3mdev_link_scope_lookup()in patch 05 (suggested by David Ahern)
- Moved definition ofip6_route_output_flags()from an inline function in /net/ipv6/route.c tonet/ipv6/route.cin order to address kbuild error in patch 05
Quoted 2 lines
I'll give David A. a chance to review this before applying.
Hey Dave: I responded to the cover-letter on Friday.
David Miller
maintainerFrom: David Ahern <[email protected]>
Date: Sun, 23 Jun 2019 13:29:27 -0600
Quoted 13 lines · David Miller
Quoted 12 lines
From: Wei Wang <[email protected]>
Date: Thu, 20 Jun 2019 17:36:36 -0700Quoted 8 lines
v2->v3:
- Handledfib6_rule_lookup()whenCONFIG_IPV6_MULTIPLE_TABLESis not configured in patch 03 (suggested by David Ahern)
- Removed the renaming ofl3mdev_link_scope_lookup()in patch 05 (suggested by David Ahern)
- Moved definition ofip6_route_output_flags()from an inline function in /net/ipv6/route.c tonet/ipv6/route.cin order to address kbuild error in patch 05
Quoted 4 lines
Quoted 2 lines
I'll give David A. a chance to review this before applying.
Hey Dave: I responded to the cover-letter on Friday.
Indeed, and so did tractor man.
Series applied, thanks everyone.