[AGENT++] Notify Filter Mask Clarification

[Frank Fock]

Hi Dave,

You are absolutely right, the current implementation of the mask
ist wrong, it will be changed in the next AGENT++ release!
It is remarkable, that no one has found the bug yet (including mine).
:-(

Best regards,
Frank

Dave White wrote:

>I have been experimenting with traps/notifications lately and it has raised a
>question about how masks are supposed to work.  While reading RFC 3413 (page
>51), it describes the operation of the snmpNotifyFilterMask.  (The operation
>of the notify filter mask is exactly the same as the vacmViewTreeFamilyMask
>from RFC 3415 (page 24).  Agent++ implements the masking operation with the
>Oidx::mask method but the way the code is implemented seems wrong to me.
>Please help me understand where my thinking goes wrong on this.
>
> 
>
>The object ID masking in the code works starting from the least significant
>bit to the most significant bit, but my reading of the RFCs indicate that
>they are supposed to work starting from the most significant bit.  Here is
>the text from the RFCs:
>
> 
>
>"Each bit of this bit mask corresponds to a sub-identifier of
>snmpNotifyFilterSubtree, with the most significant bit of the i-th octet of
>this octet string value (extended if necessary, see below) corresponding to
>the (8*i - 7)-th sub-identifier, and the least significant bit of the i-th
>octet of this octet string corresponding to the (8*i)-th sub-identifier,
>where i is in the range 1 through 16."
>
> 
>
>For this discussion, I'll refer to the most significant bit (msb) of an octet
>as bit-7 and the least significant bit (lsb) of an octet as bit-0.
>
> 
>
>So for example, when i = 1 (or first) octet of the mask octet string, bit-7
>(msb) corresponds to the (8*1 - 7)-th sub-identifier, or the 1-th (or first)
>sub-identifier.  Likewise, when i = 1, bit-0 corresponds to the (8*1)-th
>sub-identifier, or the 8-th sub-identifier.
>
> 
>
>The relevant code from Oidx::mask(const OctetStr& mask) follows:
>
> 
>
>    for (unsigned int i=0; (i<len()) && (i<cmask.len()*8); i++) {
>
>        char m = 0x01 << (i%8);
>
>        if (!(cmask[i/8] & m)) {
>
>            (*this)[i] = 0ul;
>
>        }
>
>    }
>
> 
>
>In my thinking, the value of 'm' should be set as follows:
>
>            char m = 0x80 >> (i%8);
>
> 
>
>If you have a sub-tree of 1.3.6.1.6.3.1.1.5.0 and wanted to include all but
>the last sub-identifier, the octet string for Agent++ would have to be FF.01
>(hex).  The octet string for the way I read the RFC would be FF.80 (hex).
>Can someone please explain how I'm misreading the RFC?  Has anyone ever used
>a mask other than the zero-length string resulting in a mask of all ones?
>
> 
>
>BTW: I have also been using the book "A Practical Guide to SNMPv3 and Network
>Management" by David Zeltserman (1999), but its use of snmpNotifyFilterMask
>seems different from both the RFC and from the way the code is implemented in
>Agent++.  It suggests that the octet string for the sub-tree above would be
>03.FE (hex) causing even more confusion.
>
> 
>
> 
>
>Thanks,
>
>Dave
>
> 
>
>_______________________________________________
>AGENTPP mailing list
>AGENTPP at agentpp.org
>http://agentpp.org/mailman/listinfo/agentpp
>
>  
>