| Message ID | X6rJ7c1C95uNZ/xV@santucci.pierpaolo (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | BPF |
| Headers | show |
| Series | selftest/bpf: fix IPV6FR handling in flow dissector | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Not a local patch |
On Tue, Nov 10, 2020 at 9:12 AM Santucci Pierpaolo <santucci@epigenesys.com> wrote: > > From second fragment on, IPV6FR program must stop the dissection of IPV6 > fragmented packet. This is the same approach used for IPV4 fragmentation. > Jakub, can you please take a look as well? > Signed-off-by: Santucci Pierpaolo <santucci@epigenesys.com> > --- > tools/testing/selftests/bpf/progs/bpf_flow.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/tools/testing/selftests/bpf/progs/bpf_flow.c b/tools/testing/selftests/bpf/progs/bpf_flow.c > index 5a65f6b51377..95a5a0778ed7 100644 > --- a/tools/testing/selftests/bpf/progs/bpf_flow.c > +++ b/tools/testing/selftests/bpf/progs/bpf_flow.c > @@ -368,6 +368,8 @@ PROG(IPV6FR)(struct __sk_buff *skb) > */ > if (!(keys->flags & BPF_FLOW_DISSECTOR_F_PARSE_1ST_FRAG)) > return export_flow_keys(keys, BPF_OK); > + } else { > + return export_flow_keys(keys, BPF_OK); > } > > return parse_ipv6_proto(skb, fragh->nexthdr); > -- > 2.29.2 >
On Wed, Nov 11, 2020 at 05:48 AM CET, Andrii Nakryiko wrote: > On Tue, Nov 10, 2020 at 9:12 AM Santucci Pierpaolo > <santucci@epigenesys.com> wrote: >> >> From second fragment on, IPV6FR program must stop the dissection of IPV6 >> fragmented packet. This is the same approach used for IPV4 fragmentation. >> > > Jakub, can you please take a look as well? I'm not initimately familiar with this test, but looking at the change I'd consider that Destinations Options and encapsulation headers can follow the Fragment Header. With enough of Dst Opts or levels of encapsulation, transport header could be pushed to the 2nd fragment. So I'm not sure if the assertion from the IPv4 dissector that 2nd fragment and following doesn't contain any parseable header holds. Taking a step back... what problem are we fixing here? > >> Signed-off-by: Santucci Pierpaolo <santucci@epigenesys.com> >> --- >> tools/testing/selftests/bpf/progs/bpf_flow.c | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/tools/testing/selftests/bpf/progs/bpf_flow.c b/tools/testing/selftests/bpf/progs/bpf_flow.c >> index 5a65f6b51377..95a5a0778ed7 100644 >> --- a/tools/testing/selftests/bpf/progs/bpf_flow.c >> +++ b/tools/testing/selftests/bpf/progs/bpf_flow.c >> @@ -368,6 +368,8 @@ PROG(IPV6FR)(struct __sk_buff *skb) >> */ >> if (!(keys->flags & BPF_FLOW_DISSECTOR_F_PARSE_1ST_FRAG)) >> return export_flow_keys(keys, BPF_OK); >> + } else { >> + return export_flow_keys(keys, BPF_OK); >> } >> >> return parse_ipv6_proto(skb, fragh->nexthdr); >> -- >> 2.29.2 >>
Hi Jakub,
thanks for your reply.
Let me explain the problem with an example.
Please consider the PCAP file:
https://github.com/named-data/ndn-tools/blob/master/tests/dissect-wireshark/ipv6-udp-fragmented.pcap
Let's assume that the dissector is invoked without the flag:
BPF_FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL.
Without the proposed patch, the flow keys for the second fragment (packet
timestamp 0.256997) will contain the value 0x6868 for the source and
destination port fields: this is obviously wrong.
The same happens for the third fragment (packet timestamp 0.256998) and for
the fourth fragment (packet timestamp 0.257001).
So it seems that the correct thing to do is to stop the dissector after the
IPV6 fragmentation header for all fragments from the second on.
Regards,
Pierpaolo Santucci
On Wed, Nov 11, 2020 at 12:17:06PM +0100, Jakub Sitnicki wrote:
> On Wed, Nov 11, 2020 at 05:48 AM CET, Andrii Nakryiko wrote:
> > On Tue, Nov 10, 2020 at 9:12 AM Santucci Pierpaolo
> > <santucci@epigenesys.com> wrote:
> >>
> >> From second fragment on, IPV6FR program must stop the dissection of IPV6
> >> fragmented packet. This is the same approach used for IPV4 fragmentation.
> >>
> >
> > Jakub, can you please take a look as well?
>
> I'm not initimately familiar with this test, but looking at the change
> I'd consider that Destinations Options and encapsulation headers can
> follow the Fragment Header.
>
> With enough of Dst Opts or levels of encapsulation, transport header
> could be pushed to the 2nd fragment. So I'm not sure if the assertion
> from the IPv4 dissector that 2nd fragment and following doesn't contain
> any parseable header holds.
>
> Taking a step back... what problem are we fixing here?
>
> >
> >> Signed-off-by: Santucci Pierpaolo <santucci@epigenesys.com>
> >> ---
> >> tools/testing/selftests/bpf/progs/bpf_flow.c | 2 ++
> >> 1 file changed, 2 insertions(+)
> >>
> >> diff --git a/tools/testing/selftests/bpf/progs/bpf_flow.c b/tools/testing/selftests/bpf/progs/bpf_flow.c
> >> index 5a65f6b51377..95a5a0778ed7 100644
> >> --- a/tools/testing/selftests/bpf/progs/bpf_flow.c
> >> +++ b/tools/testing/selftests/bpf/progs/bpf_flow.c
> >> @@ -368,6 +368,8 @@ PROG(IPV6FR)(struct __sk_buff *skb)
> >> */
> >> if (!(keys->flags & BPF_FLOW_DISSECTOR_F_PARSE_1ST_FRAG))
> >> return export_flow_keys(keys, BPF_OK);
> >> + } else {
> >> + return export_flow_keys(keys, BPF_OK);
> >> }
> >>
> >> return parse_ipv6_proto(skb, fragh->nexthdr);
> >> --
> >> 2.29.2
> >>
On 11/11/20 3:12 PM, Santucci Pierpaolo wrote: > Hi Jakub, > > thanks for your reply. (Santucci, please do not top-post but always reply inline which makes it easier for discussions to follow.) > Let me explain the problem with an example. > > Please consider the PCAP file: > https://github.com/named-data/ndn-tools/blob/master/tests/dissect-wireshark/ipv6-udp-fragmented.pcap > Let's assume that the dissector is invoked without the flag: > BPF_FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL. > > Without the proposed patch, the flow keys for the second fragment (packet > timestamp 0.256997) will contain the value 0x6868 for the source and > destination port fields: this is obviously wrong. > The same happens for the third fragment (packet timestamp 0.256998) and for > the fourth fragment (packet timestamp 0.257001). > > So it seems that the correct thing to do is to stop the dissector after the > IPV6 fragmentation header for all fragments from the second on. > [...] >> >> I'm not initimately familiar with this test, but looking at the change >> I'd consider that Destinations Options and encapsulation headers can >> follow the Fragment Header. >> >> With enough of Dst Opts or levels of encapsulation, transport header >> could be pushed to the 2nd fragment. So I'm not sure if the assertion >> from the IPv4 dissector that 2nd fragment and following doesn't contain >> any parseable header holds. Hm, staring at rfc8200, it says that the first fragment packet must include the upper-layer header (e.g. tcp, udp). The patch here should probably add a comment wrt to the rfc. Thanks, Daniel
On Thu, Nov 12, 2020 at 12:06 AM CET, Daniel Borkmann wrote: [...] >>> I'm not initimately familiar with this test, but looking at the change >>> I'd consider that Destinations Options and encapsulation headers can >>> follow the Fragment Header. >>> >>> With enough of Dst Opts or levels of encapsulation, transport header >>> could be pushed to the 2nd fragment. So I'm not sure if the assertion >>> from the IPv4 dissector that 2nd fragment and following doesn't contain >>> any parseable header holds. > > Hm, staring at rfc8200, it says that the first fragment packet must include > the upper-layer header (e.g. tcp, udp). The patch here should probably add a > comment wrt to the rfc. You're right, it clearly says so. Nevermind my worries about malformed packets then. Change LGTM: Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Jakub Sitnicki wrote: > On Thu, Nov 12, 2020 at 12:06 AM CET, Daniel Borkmann wrote: > > [...] > > >>> I'm not initimately familiar with this test, but looking at the change > >>> I'd consider that Destinations Options and encapsulation headers can > >>> follow the Fragment Header. > >>> > >>> With enough of Dst Opts or levels of encapsulation, transport header > >>> could be pushed to the 2nd fragment. So I'm not sure if the assertion > >>> from the IPv4 dissector that 2nd fragment and following doesn't contain > >>> any parseable header holds. > > > > Hm, staring at rfc8200, it says that the first fragment packet must include > > the upper-layer header (e.g. tcp, udp). The patch here should probably add a > > comment wrt to the rfc. > > You're right, it clearly says so. Nevermind my worries about malformed > packets then. Change LGTM: > > Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com> Also please add some of the details discussed here to the commit msg so we can remember this next time. Thanks!
diff --git a/tools/testing/selftests/bpf/progs/bpf_flow.c b/tools/testing/selftests/bpf/progs/bpf_flow.c index 5a65f6b51377..95a5a0778ed7 100644 --- a/tools/testing/selftests/bpf/progs/bpf_flow.c +++ b/tools/testing/selftests/bpf/progs/bpf_flow.c @@ -368,6 +368,8 @@ PROG(IPV6FR)(struct __sk_buff *skb) */ if (!(keys->flags & BPF_FLOW_DISSECTOR_F_PARSE_1ST_FRAG)) return export_flow_keys(keys, BPF_OK); + } else { + return export_flow_keys(keys, BPF_OK); } return parse_ipv6_proto(skb, fragh->nexthdr);