[9fans] fs arp-related fix

[9fans] fs arp-related fix

[Geoff Collyer]

In the course of getting forsyth's 83815 (Netgear FA31[12])
driver running in the file server kernel, I tripped over a bug
that permits packets smaller than ETHERMINTU (60 bytes) to be
transmitted.  Such packets are typically rejected by Ethernet
interfaces or drivers upon input and thus are not seen by other systems.
ARP requests are 42 bytes long and thus need to be padded, otherwise
the system sending the ARP requests will never get any answers and thus
be unable to communicate.  Apparently some cards automatically pad outgoing
packets, but the FA311 apparently doesn't.  /sys/src/fs/pc/etherif.c
attempts to pad short packets, but forgot to set mb->count to the
new, longer length.  Here's my fix:

: cpu; history -D etherif.c
Sep 24 21:33:50 PDT 2002 etherif.c 6324 [geoff]
etherif.c:140 c /n/dump/2002/0924/sys/src/fs/pc/etherif.c:140
< 			mb->count = len = ETHERMINTU;
---
> 			len = ETHERMINTU;

I should acknowledge that Norman Wilson recently described a similar
bug that he found in Tenth Edition Unix, so the problem was on my mind
when I started debugging.

Re: [9fans] fs arp-related fix

[YAMANASHI Takeshi]

[-- Attachment #1: Type: text/plain, Size: 281 bytes --]

This is my understanding that the user `none'
can always mount a fs(4) without any authentication.
Is this correct?  And if so, can I toggle this
feature on and off?

sources.cs.bell-labs.com seems not allow `none' to
attach its root.

Thank you.
--
YAMANASHI Takeshi


[-- Attachment #2: Type: text/plain, Size: 57 bytes --]

Content-Type: message/rfc822
Content-Disposition: inline