LVS
lvs-devel
Google
 
Web LinuxVirtualServer.org

Re: [PATCH] ipvs: Fix reuse connection if real server is dead

To: Julian Anastasov <ja@xxxxxx>
Subject: Re: [PATCH] ipvs: Fix reuse connection if real server is dead
Cc: Grzegorz Lyczba <grzegorz.lyczba@xxxxxxxxx>, lvs-devel@xxxxxxxxxxxxxxx, Simon Horman <horms@xxxxxxxxxxxx>
From: Hans Schillstrom <hans@xxxxxxxxxxxxxxx>
Date: Wed, 08 May 2013 10:07:55 +0200

On Wed, 2013-05-08 at 10:35 +0300, Julian Anastasov wrote:
>       Hello,
> 
> On Tue, 7 May 2013, Grzegorz Lyczba wrote:
> 
> > Expire cached connection for new TCP connection if real server is down
> > 
> > Signed-off-by: Grzegorz Lyczba <grzegorz.lyczba@xxxxxxxxx>
> > ---
> >  net/netfilter/ipvs/ip_vs_core.c |   20 ++++++++++++++++++++
> >  1 files changed, 20 insertions(+), 0 deletions(-)
> > 
> > diff --git a/net/netfilter/ipvs/ip_vs_core.c 
> > b/net/netfilter/ipvs/ip_vs_core.c
> > index 47edf5a..7c5f083 100644
> > --- a/net/netfilter/ipvs/ip_vs_core.c
> > +++ b/net/netfilter/ipvs/ip_vs_core.c
> > @@ -1002,6 +1002,17 @@ static inline int is_tcp_reset(const struct sk_buff 
> > *skb, int nh_len)
> >     return th->rst;
> >  }
> >  
> > +static inline int is_tcp_syn(const struct sk_buff *skb, int nh_len)
> > +{
> > +    struct tcphdr _tcph, *th;
> > +
> > +    th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
> 
>       As Hans notes, it is a good idea to support SCTP by
> using the iph address as second argument. May be the func can
> be named is_new_conn:
> 
> static inline bool is_new_conn(const struct sk_buff *skb,
>                              struct ip_vs_iphdr *iph)
> ...
> 
> > +    if (th == NULL)
> > +        return 0;
> > +    return th->syn;
> > +}
> > +
> > +
> >  /* Handle response packets: rewrite addresses and send away...
> >   */
> >  static unsigned int
> > @@ -1627,6 +1638,15 @@ ip_vs_in(unsigned int hooknum, struct sk_buff *skb, 
> > int af)
> >      * Check if the packet belongs to an existing connection entry
> >      */
> >     cp = pp->conn_in_get(af, skb, &iph, 0);
> > +
> > +   if (likely(cp) && likely(cp->dest) &&
> 
>       I hope it is appropriate to use the expire_nodest_conn flag:
> 
>       if (unlikely(sysctl_expire_nodest_conn(ipvs) && cp && cp->dest &&
> ...
> 
> > +       unlikely(!atomic_read(&cp->dest->weight)) &&
> 
>       We should add '!iph.fragoffs &&' here.

Do we really need that?
I don't think a SYN can be fragmented.


Regards
Hans
> 
> > +       iph.protocol == IPPROTO_TCP && is_tcp_syn(skb, iph.len)) {
> > +           ip_vs_conn_expire_now(cp);
> > +           __ip_vs_conn_put(cp);
> > +           cp = NULL;
> > +   }
> > +
> >     if (unlikely(!cp) && !iph.fragoffs) {
> >             /* No (second) fragments need to enter here, as nf_defrag_ipv6
> >              * replayed fragment zero will already have created the cp
> 
> Regards
> 
> --
> Julian Anastasov <ja@xxxxxx>
> --
> To unsubscribe from this list: send the line "unsubscribe lvs-devel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Attachment: smime.p7s
Description: S/MIME cryptographic signature

<Prev in Thread] Current Thread [Next in Thread>