svn commit: r353635 - in head/sys: netinet netinet6

[Hans Petter Selasky]

On 2019-10-17 17:08, Gleb Smirnoff wrote:
> On Wed, Oct 16, 2019 at 10:46:44PM +0200, Hans Petter Selasky wrote:
> H> > as far as I remember I was against this changeset and I had
> H> > several other developers agreed that this should be fixed in
> H> > different way. Why did you proceed with checking it in? :(
> H>
> H> Hi Gleb,
> H>
> H> This issue has been discussed in-depth at various transport meetings and
> H> we have agreed on a solution.
> 
> Is the list of people who agreed longer than "Reviewed by" list?

Yes.

> 
> H> Are you seeing something broken as of this patch?
> 
> As I already explained, first, we are dropping absolutely legitimate
> packets. At the time of arrival there were nothing wrong about them.
> This is idelogically wrong from viewpoint of abstract network stack.

No packets are dropped. This was the initial version of my patch. Please 
re-read the history of the differential revision.

> Second, the problem should be fixed in a different way: when we put
> packets on the queue, we should take all important values out of the
> ifnet and store them on queue entry.

No, this won't work. Sometimes you need to send an ICMP error message 
back, but to which interface? You cannot use unit-numbers (risking the 
packet goes to wrong interface) nor pointers, which then can point to 
freed memory.

--HPS