In message <20010625012109.559511A5@proven.weird.com>, Greg A. Woods writes:
>[ On Sunday, June 24, 2001 at 19:17:13 (-0400), Steven M. Bellovin wrote: ]
>> Subject: Re: no buffer space available 
>>
>> You can get ENOBUFS when the output queue for the device is full, I 
>> believe.  Note the following code fragment from if_ppp.c:
>
>That might explain why I didn't ever get any similar behaviour when I
>was running PPP, because that was on BSD/OS 1.1.
>
>But I don't think it can explain how an ethernet-only router like mine
>can get clogged....
>

A lost interrupt could probably produce the same symptoms, if the 
driver doesn't have adequate recovery code, such as a watchdog timer.
I haven't looked at any of the NetBSD Ethernet drivers to see what they 
do.  But I remember the first time I saw a device driver that could 
handle such situations -- it was amazing how much of a difference it 
made to system stability.  (Anyone else remember having to pop out 
the unit number plug on IBM 2314 disk drives, or do a 'vary ...,offline'/
'vary ...,online' sequence on IBM mainframes to recover from lost disk 
drive interrupts?)

		--Steve Bellovin, http://www.research.att.com/~smb