| Message ID | CAPWQB7F-TGmcTNc0mC2TbTx_sFm33jWD_UiBkEa9vfoWdOOqBg@mail.gmail.com |
|---|---|
| State | Not Applicable |
| Headers | show |
> Thanks all, I'm considering rolling the following incremental into > this patch, does this make sense? looks good to me ~~ > diff --git a/datapath/linux/compat/ip_fragment.c > b/datapath/linux/compat/ip_fragment.c > index efa86fcfae1b..de08f6c6744e 100644 > --- a/datapath/linux/compat/ip_fragment.c > +++ b/datapath/linux/compat/ip_fragment.c > @@ -732,9 +732,7 @@ int rpl_ip_defrag(struct net *net, struct sk_buff > *skb, u32 user) > #if LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0) > static int __net_init ipv4_frags_init_net(struct net *net) > { > - nf_defrag_ipv4_enable(net); > - > - return 0; > + return nf_defrag_ipv4_enable(net); > } > #endif > > diff --git a/datapath/linux/compat/nf_conntrack_reasm.c > b/datapath/linux/compat/nf_conntrack_reasm.c > index cb6da6c64e05..5832fead800e 100644 > --- a/datapath/linux/compat/nf_conntrack_reasm.c > +++ b/datapath/linux/compat/nf_conntrack_reasm.c > @@ -561,9 +561,7 @@ out_unlock: > #if LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0) > static int nf_ct_net_init(struct net *net) > { > - nf_defrag_ipv6_enable(net); > - > - return 0; > + return nf_defrag_ipv6_enable(net); > } > #endif
On Wed, Apr 26, 2017 at 6:12 PM, Joe Stringer <joe@ovn.org> wrote: > On 25 April 2017 at 10:00, Greg Rose <gvrose8192@gmail.com> wrote: >> On Tue, 2017-04-25 at 23:26 +0800, Guoshuai Li wrote: >>> In kernel 4.10.0, the function "nf_defrag_ipv6_enable" need >>> parameters "struct net *", so we need call it for each namespace init >>> to load netfilter fragment kmod. >>> >>> Reported-by: Raymond Burkholder <ray@oneunified.net> >>> Reported-at: https://mail.openvswitch.org/pipermail/ovs-dev/2017-April/331411.html >>> Signed-off-by: Guoshuai Li <ligs@dtdream.com> >> >> Compile tested and looks OK to me. >> >> Reviewed-by: Greg Rose <gvrose8192@gmail.com> > > Thanks all, I'm considering rolling the following incremental into > this patch, does this make sense? > In general, we try not to rely on kernel version number. In this case, does it make sense to look at the signature of nf_defrag_ipv{4|6}_enable instead?
I try remove "nf_defrag_ipv{4|6}_enabl", and then test by lsmod:
stack@devstack:~$ lsmod | grep defrag
nf_defrag_ipv6 36864 1 nf_conntrack_ipv6
nf_defrag_ipv4 16384 1 nf_conntrack_ipv4
stack@devstack:~$ lsmod | grep conntrack
nf_conntrack_ipv6 20480 1
nf_defrag_ipv6 36864 1 nf_conntrack_ipv6
nf_conntrack_ipv4 16384 5
nf_defrag_ipv4 16384 1 nf_conntrack_ipv4
without remove, nf_defrag_ipv6 is used by openvswitch:
stack@devstack:~/neutron$ lsmod | grep defrag
nf_defrag_ipv6 36864 2 openvswitch,nf_conntrack_ipv6
nf_defrag_ipv4 16384 1 nf_conntrack_ipv4
Normally, defrag will be loaded, but is not sure whether it is absolute.
> In general, we try not to rely on kernel version number. In this case,
> does it make sense to look at the signature of
> nf_defrag_ipv{4|6}_enable instead?
diff --git a/datapath/linux/compat/ip_fragment.c b/datapath/linux/compat/ip_fragment.c index efa86fcfae1b..de08f6c6744e 100644 --- a/datapath/linux/compat/ip_fragment.c +++ b/datapath/linux/compat/ip_fragment.c @@ -732,9 +732,7 @@ int rpl_ip_defrag(struct net *net, struct sk_buff *skb, u32 user) #if LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0) static int __net_init ipv4_frags_init_net(struct net *net) { - nf_defrag_ipv4_enable(net); - - return 0; + return nf_defrag_ipv4_enable(net); } #endif diff --git a/datapath/linux/compat/nf_conntrack_reasm.c b/datapath/linux/compat/nf_conntrack_reasm.c index cb6da6c64e05..5832fead800e 100644 --- a/datapath/linux/compat/nf_conntrack_reasm.c +++ b/datapath/linux/compat/nf_conntrack_reasm.c @@ -561,9 +561,7 @@ out_unlock: #if LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0) static int nf_ct_net_init(struct net *net) { - nf_defrag_ipv6_enable(net); - - return 0; + return nf_defrag_ipv6_enable(net); } #endif _______________________________________________