| Message ID | 20241104083545.114-1-gnaaman@drivenets.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 702c290a1cb16f4a64567cae0bedb848399f7915 |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net-next,v2] sctp: Avoid enqueuing addr events redundantly | expand |
On Mon, Nov 4, 2024 at 3:36 AM Gilad Naaman <gnaaman@drivenets.com> wrote: > > > Avoid modifying or enqueuing new events if it's possible to tell that no > one will consume them. > > Since enqueueing requires searching the current queue for opposite > events for the same address, adding addresses en-masse turns this > inetaddr_event into a bottle-neck, as it will get slower and slower > with each address added. > > Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Acked-by: Xin Long <lucien.xin@gmail.com> Thanks. > --- > Changes in v2: > - Reorder list removal to avoid race with new sessions > --- > net/sctp/ipv6.c | 2 +- > net/sctp/protocol.c | 16 +++++++++++++++- > 2 files changed, 16 insertions(+), 2 deletions(-) > > diff --git a/net/sctp/ipv6.c b/net/sctp/ipv6.c > index f7b809c0d142..b96c849545ae 100644 > --- a/net/sctp/ipv6.c > +++ b/net/sctp/ipv6.c > @@ -103,10 +103,10 @@ static int sctp_inet6addr_event(struct notifier_block *this, unsigned long ev, > ipv6_addr_equal(&addr->a.v6.sin6_addr, > &ifa->addr) && > addr->a.v6.sin6_scope_id == ifa->idev->dev->ifindex) { > - sctp_addr_wq_mgmt(net, addr, SCTP_ADDR_DEL); > found = 1; > addr->valid = 0; > list_del_rcu(&addr->list); > + sctp_addr_wq_mgmt(net, addr, SCTP_ADDR_DEL); > break; > } > } > diff --git a/net/sctp/protocol.c b/net/sctp/protocol.c > index 39ca5403d4d7..8b9a1b96695e 100644 > --- a/net/sctp/protocol.c > +++ b/net/sctp/protocol.c > @@ -738,6 +738,20 @@ void sctp_addr_wq_mgmt(struct net *net, struct sctp_sockaddr_entry *addr, int cm > */ > > spin_lock_bh(&net->sctp.addr_wq_lock); > + > + /* Avoid searching the queue or modifying it if there are no consumers, > + * as it can lead to performance degradation if addresses are modified > + * en-masse. > + * > + * If the queue already contains some events, update it anyway to avoid > + * ugly races between new sessions and new address events. > + */ > + if (list_empty(&net->sctp.auto_asconf_splist) && > + list_empty(&net->sctp.addr_waitq)) { > + spin_unlock_bh(&net->sctp.addr_wq_lock); > + return; > + } > + > /* Offsets existing events in addr_wq */ > addrw = sctp_addr_wq_lookup(net, addr); > if (addrw) { > @@ -808,10 +822,10 @@ static int sctp_inetaddr_event(struct notifier_block *this, unsigned long ev, > if (addr->a.sa.sa_family == AF_INET && > addr->a.v4.sin_addr.s_addr == > ifa->ifa_local) { > - sctp_addr_wq_mgmt(net, addr, SCTP_ADDR_DEL); > found = 1; > addr->valid = 0; > list_del_rcu(&addr->list); > + sctp_addr_wq_mgmt(net, addr, SCTP_ADDR_DEL); > break; > } > } > -- > 2.34.1 >
Hello: This patch was applied to netdev/net-next.git (main) by Paolo Abeni <pabeni@redhat.com>: On Mon, 4 Nov 2024 08:35:44 +0000 you wrote: > Avoid modifying or enqueuing new events if it's possible to tell that no > one will consume them. > > Since enqueueing requires searching the current queue for opposite > events for the same address, adding addresses en-masse turns this > inetaddr_event into a bottle-neck, as it will get slower and slower > with each address added. > > [...] Here is the summary with links: - [net-next,v2] sctp: Avoid enqueuing addr events redundantly https://git.kernel.org/netdev/net-next/c/702c290a1cb1 You are awesome, thank you!
diff --git a/net/sctp/ipv6.c b/net/sctp/ipv6.c index f7b809c0d142..b96c849545ae 100644 --- a/net/sctp/ipv6.c +++ b/net/sctp/ipv6.c @@ -103,10 +103,10 @@ static int sctp_inet6addr_event(struct notifier_block *this, unsigned long ev, ipv6_addr_equal(&addr->a.v6.sin6_addr, &ifa->addr) && addr->a.v6.sin6_scope_id == ifa->idev->dev->ifindex) { - sctp_addr_wq_mgmt(net, addr, SCTP_ADDR_DEL); found = 1; addr->valid = 0; list_del_rcu(&addr->list); + sctp_addr_wq_mgmt(net, addr, SCTP_ADDR_DEL); break; } } diff --git a/net/sctp/protocol.c b/net/sctp/protocol.c index 39ca5403d4d7..8b9a1b96695e 100644 --- a/net/sctp/protocol.c +++ b/net/sctp/protocol.c @@ -738,6 +738,20 @@ void sctp_addr_wq_mgmt(struct net *net, struct sctp_sockaddr_entry *addr, int cm */ spin_lock_bh(&net->sctp.addr_wq_lock); + + /* Avoid searching the queue or modifying it if there are no consumers, + * as it can lead to performance degradation if addresses are modified + * en-masse. + * + * If the queue already contains some events, update it anyway to avoid + * ugly races between new sessions and new address events. + */ + if (list_empty(&net->sctp.auto_asconf_splist) && + list_empty(&net->sctp.addr_waitq)) { + spin_unlock_bh(&net->sctp.addr_wq_lock); + return; + } + /* Offsets existing events in addr_wq */ addrw = sctp_addr_wq_lookup(net, addr); if (addrw) { @@ -808,10 +822,10 @@ static int sctp_inetaddr_event(struct notifier_block *this, unsigned long ev, if (addr->a.sa.sa_family == AF_INET && addr->a.v4.sin_addr.s_addr == ifa->ifa_local) { - sctp_addr_wq_mgmt(net, addr, SCTP_ADDR_DEL); found = 1; addr->valid = 0; list_del_rcu(&addr->list); + sctp_addr_wq_mgmt(net, addr, SCTP_ADDR_DEL); break; } }
Avoid modifying or enqueuing new events if it's possible to tell that no one will consume them. Since enqueueing requires searching the current queue for opposite events for the same address, adding addresses en-masse turns this inetaddr_event into a bottle-neck, as it will get slower and slower with each address added. Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> --- Changes in v2: - Reorder list removal to avoid race with new sessions --- net/sctp/ipv6.c | 2 +- net/sctp/protocol.c | 16 +++++++++++++++- 2 files changed, 16 insertions(+), 2 deletions(-)