Hi Josh,

Why did you need to move the resubmit label?


On Fri, May 29, 2015 at 2:43 PM, Josh Hunt [off-list ref] wrote:

On 05/29/2015 03:27 PM, Sergei Shtylyov wrote:

quoted

Hello.

On 05/29/2015 11:04 PM, Josh Hunt wrote:

quoted

I came across this problem while trying to use UDP encapsulation with
IPv6. The
change below fixes that, but it was not immediately apparent if there
are any
other protocols relying on this broken behavior. FWIW the behavior
below now
matches IPv4.

quoted

Josh

    Need "Signed-off-by:" line insted of this.

quoted

v2: Actually sets nexthdr so we can use it ;)

    Normally goes after ---, not before.

quoted

---

quoted

UDP encapsulation is broken on IPv6 because it expects when it returns
a negative
value that the packet will be resubmitted to the stack with the
handler corresponding
to the return value. The check currently looks for return values > 0
and then resubmits.

quoted

This patch fixes that check and also moves the resubmit label to take
advantage of
the return code identifying the next protocol we want to process.

quoted

Signed-off-by: Josh Hunt <redacted>
---
  net/ipv6/ip6_input.c |    7 ++++---
  1 file changed, 4 insertions(+), 3 deletions(-)

quoted

diff --git a/net/ipv6/ip6_input.c b/net/ipv6/ip6_input.c
index f2e464e..e16c289 100644
--- a/net/ipv6/ip6_input.c
+++ b/net/ipv6/ip6_input.c

[...]

quoted

@@ -246,9 +246,10 @@ resubmit:
              goto discard;

          ret = ipprot->handler(skb);
-        if (ret > 0)
+        if (ret < 0) {
+            nexthdr = -ret;
              goto resubmit;
-        else if (ret == 0)
+        } else if (ret == 0)
              IP6_INC_STATS_BH(net, idev, IPSTATS_MIB_INDELIVERS);

    The CodingStyle dictates that all branches must have {} if at least
one has them.

Thanks Sergei, I appreciate the review. I will make these changes in a v3
after I get feedback this is the correct fix for the problem.

Josh