[Freeipmi-devel] RE: libfreeipmi - sequence numbers

[Cress, Andrew R]

Al/Ian,

I do understand the rq_seq being output and used in the lan header, and
the other is in the session header.  And I see that the lan header in
section 12.4 defines rq_seq as 6 bits (8 - 2 for lun).  So, we may be
dealing with something more subtle here.  However, the empirical fact is
that it doesn't work as coded (I'm calling ipmi_lan_cmd), and with the
patch below, it does work for up to 0x10B packets in a test session. 

If you have a better solution, I'm open to it.
Without a fix of some kind, I can't use the FreeIPMI LAN interface.

Attached is my test version of showsel (i386 Linux), to exhibit the
problem.  It looks for libfreeipmi.so.0.
"showsel -N ipminodename -x"  
The -x option shows debug messages with the rq_seq listed.
As long as there are more than 63 SEL records, the 0.1.0 library will
abort, and the patched one works fine.  If the app attempts to conform
by wrapping the rq_seq, the session will hang (BMC silently discards
packets).  

Andy

--- libfreeipmi/src/ipmi-lan-interface.c.orig   2004-10-28
10:38:04.000000000 -0400
+++ libfreeipmi/src/ipmi-lan-interface.c        2004-10-28
17:26:58.000000000 -0400
@@ -108,7 +108,7 @@
 {
   if ((net_fn > IPMI_NET_FN_TRANSPORT_RS)
       || (rs_lun > IPMI_BMC_IPMB_LUN_OEM_LUN2)
-      || (rq_seq > IPMI_LAN_SEQ_NUM_MAX)
+     /*  || (rq_seq > IPMI_LAN_SEQ_NUM_MAX)   /* ARCress +++*/
       || (obj_msg == NULL))
     {
       errno = EINVAL;

-----Original Message-----
From: Albert Chu [mailto:address@hidden 
Sent: Friday, October 29, 2004 12:05 PM
To: Cress, Andrew R
Cc: address@hidden
Subject: Re: RE: RE: RE: [Freeipmi-devel] Using libfreeipmi interface


Hi Andrew,

> Where did the 0x3f arbitrary sequence limit come from anyway?  I
> couldn't find any reference to sequence number limits in the IPMI spec
> or platform specs.

The 0x3f comes from the fact the requester sequence number is 6 bits
long.  You can see this in section 7.3, figure 7-1 of the IPMI 1.5 spec.
 (That section is pointed to by section 12.4, which discusses IPMI over
Lan).  

I get the feeling you are passing in session sequence numbers, which are
32 bits, in place of the requester sequence numbers, which are 6 bits
long.  The session sequence number is passed to the fill_hdr_session()
function when building a packet and the requester sequence number is
passed to fill_lan_msg_hdr().

Al

--
Albert Chu
address@hidden
Lawrence Livermore National Laboratory

----- Original Message -----
From: "Cress, Andrew R" <address@hidden>
Date: Friday, October 29, 2004 7:44 am
Subject: RE: RE: RE: [Freeipmi-devel] Using libfreeipmi interface

> 
> re: IPMI_LAN_SEQ_NUM_MAX (=0x3F)
> 
> As a use case, to understand the problem this causes, one of my
> applications 'showsel' allows reading the IPMI SEL.  This can be quite
> long on Intel platforms (e.g. 4032 record capacity on one test system
> here).  On one test system, with 257 SEL records, the sequence number
> increments to 267 (0x10B), and wrapping the sequence number during 
> thissession causes the IPMI chipset to ignore the command with a
> duplicate/lower sequence number (and the app/lib would hang).  This
> would also affect the fish/sel script/app similarly.  The sequence
> number is 32 bits, and incrementing up through that capacity is 
> allowed.My Tiger (ia64) system isn't up right now, but as I recall, 
> it behaves
> the same way.
> 
> Where did the 0x3f arbitrary sequence limit come from anyway?  I
> couldn't find any reference to sequence number limits in the IPMI spec
> or platform specs.
> 
> So, my position is that it is not just inappropriate to check sequence
> number bounds at 0x3f, but invalid for IPMI 1.5 platforms with LAN
> support (a bug).
> 
> Andy
> 
> -----Original Message-----
> From: Albert Chu [mailto:address@hidden 
> Sent: Thursday, October 28, 2004 8:04 PM
> To: Cress, Andrew R
> Cc: address@hidden
> Subject: Re: RE: RE: [Freeipmi-devel] Using libfreeipmi interface
> 
> 
> > I also found a problem with the  IPMI_LAN_SEQ_NUM_MAX (=0x3F).
> > There isn't any precedent in IPMI 1.5 for this, and limiting the 
> 32-
> > bitsequence number to 6 bits causes some problems, since the BMC 
> > LAN won't handle wrapping the sequence number back to 0 or 1 
> after 63 
> > (0x3f).
> 
> I'm not quite convinced this is a problem.  The bounds check makes the
> user knowledgeable to the fact that arbitrary sequence numbers aren't
> allowed.  I spoke to a few other developers I work with, and they feel
> that the bounds checking is more appropriate.  Ab, Ian, Bala, what are
> your thoughts??
> 
>

showsel
Description: showsel