kern/57049: large TCP transfers NetBSD-Xen-Guest -> NetBSD-Xen-DOM0 abort with EHOSTDOWN

[kardel%netbsd.org@localhost]

>Number:         57049
>Category:       kern
>Synopsis:       large TCP transfers NetBSD-Xen-Guest -> NetBSD-Xen-DOM0 abort with EHOSTDOWN
>Confidential:   no
>Severity:       critical
>Priority:       high
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Oct 07 11:55:00 +0000 2022
>Originator:     kardel%netbsd.org@localhost
>Release:        NetBSD 9.99.100
>Organization:
>Environment:
	
	
System: NetBSD TEMPLATE-fix-hostname 9.99.100 NetBSD 9.99.100 (GENERIC) #6: Fri Oct 7 13:19:45 CEST 2022 kardel@gaiatest:/src/NetBSD/current/src/obj.amd64/sys/arch/amd64/compile/GENERIC amd64
Architecture: x86_64
Machine: amd64
>Description:
	When copying large files (e.g. 20GB) via scp from a Xen guest to a Xen DOM0 the transfers often fail with EHOSTDOWN.
	The errorcode comes from sys/net/nd.c:nd_resolve() (nd.c:384)
	The error can be replicated with a simple ttcp test - see below:
	Also, during the transfer, following routing messages can be observed on the guest:
got message of size 152 on Fri Oct  7 11:41:36 2022
RTM_MISS: Lookup failed on this address: len 152, pid 0, seq 0, errno 0, flags: 0x40<DONE>
locks: 0 inits: 0
sockaddrs: 0x3<DST,GATEWAY>
 10.0.2.16 link#1
got message of size 152 on Fri Oct  7 11:41:39 2022
RTM_MISS: Lookup failed on this address: len 152, pid 0, seq 0, errno 0, flags: 0x40<DONE>
locks: 0 inits: 0
sockaddrs: 0x3<DST,GATEWAY>
 10.0.2.16 link#1
got message of size 152 on Fri Oct  7 11:41:39 2022
RTM_ADD: Add Route: len 152, pid 0, seq 0, errno 0, flags: 0x2445<UP,HOST,DONE,LLINFO,CLONED>
locks: 0 inits: 0
sockaddrs: 0x3<DST,GATEWAY>
 10.0.2.16 aa:bb:cc:dd:ee:ff

>How-To-Repeat:
	on DOM0:
	Zugspitze# ttcp -s -r
	ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
	ttcp-r: socket

	on guest:
	Guest# ttcp -s -t -n 1000000 zugspitze
	ttcp-t: socket
	ttcp-t: connect
	ttcp-t: IO: Host is down
	errno=64
>Fix:
	no workaround so far

>Unformatted: