On Tue, 2 Sep 2025 20:10:21 +1000, Darren Tucker <[email protected]>
wrote:

> On Sat, Jun 07, 2025 at 08:59:38PM -0400, Daniel Jakots wrote:
> > tl;dr: when I download base77.tgz, I have
> > $ ftp https://whatever/pub/OpenBSD/snapshots/amd64/base77.tgz 
> > Trying 10.10.10.46...
> > Requesting https://whatever/pub/OpenBSD/snapshots/amd64/base77.tgz
> >   2% |*                | 11648 KB  - stalled -
> > ftp: Reading from socket: read failed: error:06FFF064:digital
> > envelope routines:CRYPTO_internal:bad decrypt  
> 
> So the connection stalled?  You don't happen to be using jumbo frames
> do you?

I don't think I've configured anything related to jumbo frames. Can
they be used with default settings? How can I check?

$ ifconfig em | grep UP | awk '{print $3 " " $4}'
mtu 1500
mtu 1500
mtu 1500

Does that mean I'm not using jumbo frames?

> If so, it might be this, wherein em(4) would truncate jumbo frames
> whose length fell on 2k boundaries:
> https://cvsweb.openbsd.org/src/sys/dev/pci/if_em.c?rev=1.379&content-type=text/x-cvsweb-markup
> 
> Such frames would get dropped and retransmitted, but they'd still be
> the same size to they'd just get dropped again, and ultimately the TCP
> connection would terminate.  Standard MTU is 1500, so this can't be
> the problem if you're not using jumbos.

When I saw the jmatthew's diff on tech@, I reached out to him offlist
mentioning my issue. He answered that he didn't think the diff would
fix the problem. He advised to check counters with `kstat
em0::em-stats:` (but I didn't seen anything), and hacking the driver to
check some theories.

> > OpenBSD 7.7-current (GENERIC.MP) #703: Sat Jun  7 08:24:55 MDT 2025
> >     [email protected]:/usr/src/sys/arch/amd64/compile/GENERIC.MP
> >  
> 
> This snapshot is from a week prior to the fix going in, so if this was
> your problem it should be fixed now, at least in -current.
> 

iirc, each time I update to a newer snapshot I check if the problem
still happens, and I don't remember seeing it fixed. However I'm
checking now and I can't reproduce the problem. (I'm running a snapshot
from 2025-08-11.)

Thanks for checking!
Daniel

Reply via email to