This document is obsolete. Please
refer to RFC 2576.
Network Working Group B. Wijnen Request for Comments: 2089 IBM Category: Informational D. Levi SNMP Research, Inc January 1997
V2ToV1 Mapping SNMPv2 onto SNMPv1 within a bi-lingual SNMP agent
Status of this Memo
This memo provides information for the Internet community. This memo does not specify an Internet standard of any kind. Distribution of this memo is unlimited.
Abstract
The goal of this memo is to document a common way of mapping an SNMPv2 response into an SNMPv1 response within a bi-lingual SNMP agent (one that supports both SNMPv1 and SNMPv2).
We now have the SNMPv1 protocol (RFC1157 [1]) as a full standard and the SNMPv2 protocol (RFC1905 [1]) as a DRAFT standard. It can be expected that many agent implementations will support both SNMPv1 and SNMPv2 requests coming from SNMP management entities. In many cases the underlying instrumentation will be implemented using the new SNMPv2 SMI and SNMPv2 protocol. The SNMP engine then gets the task to ensure that any SNMPv2 response data coming from such SNMPv2 compliant instrumentation gets converted to a proper SNMPv1 response if the original request came in as an SNMPv1 request. The SNMP engine should also deal with mapping SNMPv2 traps which are generated by an application or by the SNMPv2 compliant instrumentation into SNMPv1 traps if the agent has been configured to send traps to an SNMPv1 manager.
It seems beneficial if all such agents do this mapping in the same way. This document describes such a mapping based on discussions and perceived consensus on the various mailing lists. The authors of this document have also compared their own implementations of these mappings. They had a few minor differences and decided to make their implementation behave the same and document this mapping so others can benefit from it.
We recommend that all agents implement this same mapping.
Note that the mapping described in this document should also be followed by SNMP proxies that provide a mapping between SNMPv1 management applications and SNMPv2 agents.
o Mapping of the SNMPv2 error-status into SNMPv1 error-status
o Mapping of the SNMPv2 exceptions into SNMPv1 error-status
o Skipping object instances that have a non-SNMPv1 Syntax (specifically Counter64)
o Mapping of SNMPv2 traps into SNMPv1 traps
Wijnen & Levi Informational [Page 2]
RFC 2089 V2toV1 January 1997
2.1 Mapping SNMPv2 error-status into SNMPv1 error-status
With the new SNMPv2 protocol (RFC1905 [1]) we get a set of error- status values that return the cause of an error in much more detail. But an SNMPv1 manager does not understand such error-status values.
So, when the instrumentation code returns response data and uses an SNMPv2 error-status to report on the success or failure of the requested operation and if the original SNMP request is an SNMPv1 request, then we must map such an error-status into an SNMPv1 error- status when composing the SNMP response PDU.
The SNMPv2 error status is mapped to an SNMPv1 error-status using this table:
In SNMPv2 we have so called exception values. These will allow an SNMPv2 response PDU to return as much management information as possible, even if one or more of the requested variables do not exist. SNMPv1 does not support exception values, and thus does not return the value of management information when any error occurs.
When multiple variables do not exist, an SNMPv1 agent can return only a single error and index of a single variable. The agent determines by its implementation strategy which variable to identify as the
Wijnen & Levi Informational [Page 3]
RFC 2089 V2toV1 January 1997
cause of the error via the value of the error-index field. Thus, an SNMPv1 manager may make no assumption on the validity of the other variables in the request.
So, when an SNMPv1 request is received, we must check the varBinds returned from SNMPv2 compliant instrumentation for exception values, and convert these exception values into SNMPv1 error codes.
The type of exception we can get back and the action we must take depends on the SNMP operation that is requested.
o For SNMP GET requests we can get back noSuchObject and noSuchInstance.
o For SNMP GETNEXT requests we can get back endOfMibView.
o For SNMP SET requests we cannot get back any exceptions.
o For SNMP GETBULK requests we can get back endOfMibView, but such a request should only come in as an SNMPv2 request, so we do not have to worry about any mapping onto SNMPv1. If a GETBULK comes in as an SNMPv1 request, it is treated as an error and the packet is dropped.
A noSuchObject or noSuchInstance exception generated by SNMPv2 compliant instrumentation indicates that the requested object instance can not be returned. The SNMPv1 error code for this condition is noSuchName, and so the error-status field of the response PDU should be set to noSuchName. Also, the error-index field is set to the index of the varBind for which an exception occurred, and the varBind list from the original request is returned with the response PDU.
Note that when the response contains multiple exceptions, that the agent may pick any one to be returned.
When SNMPv2 compliant instrumentation returns a varBind containing an endOfMibView exception in response to a GETNEXT request, it indicates that there are no object instances available which lexicographically follow the object in the request. In an SNMPv1 agent, this condition normally results in a noSuchName error, and so the error-status field of the response PDU should be set to noSuchName. Also, the error- index field is set to the index of the varBind for which an exception occurred, and the varBind list from the original request is returned with the response PDU.
Note that when the response contains multiple exceptions, that the agent may pick any one to be returned.
The SNMPv2 SMI (RFC1902 [2]) defines basically one new syntax that is problematic for SNMPv1 managers. That is the syntax Counter64. All the others can be handled by SNMPv1 managers.
The net impact on bi-lingual agents is that they should make sure that they never return a varBind with a Counter64 value to an SNMPv1 manager.
The best accepted practice is to consider such object instances implicitly excluded from the view. So:
o On an SNMPv1 GET request, we return an error-status of noSuchName and the error-index is set to the varBind that causes this error.
o On an SNMPv1 GETNEXT request, we skip the object instance and fetch the next object instance that follows the one with a syntax of Counter64.
o Any SET request that has a varBind with a Counter64 value must have come from a SNMPv2 manager, and so it should not cause a problem. If we do receive a Counter64 value in an SNMPv1 SET packet, it should result in an ASN.1 parse error since Counter64 is not valid in the SNMPv1 protocol. When an ASN.1 parse error occurs, the counter snmpInASNParseErrs is incremented and no response is returned.
o The GETBULK is an SNMPv2 operation, so it should never come from an SNMPv1 manager. If we do receive a GETBULK PDU from in an SNMPv1 packet, then we consider it an invalid PDU-type and we drop the packet.
This sections contains a step by step description of how to handle SNMPv1 requests in an agent where the underlying instrumentation code is SNMPv2 compliant.
First, the request is converted into a call to the underlying instrumentation. This is implementation specific.
When such instrumentation returns response data using SNMPv2 syntax and error-status values, then:
1. If the error-status is anything other than noError,
a. The error status is translated to an SNMPv1 error-status using the table from 2.1, "Mapping SNMPv2 error-status into SNMPv1 error-status" on page 2
b. The error-index is set to the position (in the original request) of the varBind that caused the error-status.
c. The varBindList of the response PDU is made exactly the same as the varBindList that was received in the original request.
2. If the error-status is noError, then find any varBind that contains an SNMPv2 exception (noSuchObject or noSuchInstance) or an SNMPv2 syntax that is unknown to SNMPv1 (Counter64). (Note that if there are more than one, the agent may choose any such varBind.) If there are any such varBinds, then for the one chosen:
a. Set the error-status to noSuchName
b. Set the error-index to the position (in the varBindList of the original request) of the varBind that returned such an SNMPv2 exception or syntax.
c. Make the varBindList of the response PDU exactly the same as the varBindList that was received in the original request.
First, the request is converted into a call to the underlying instrumentation. This is implementation specific. There may be repetitive calls to (possibly different pieces of) instrumentation code to try to find the first object which lexicographically follows each of the objects in the request. Again, this is implementation specific.
When the instrumentation finally returns response data using SNMPv2 syntax and error-status values, then:
1. If the error-status is anything other than noError,
a. The error status is translated to an SNMPv1 error-status using the table from 2.1, "Mapping SNMPv2 error-status into SNMPv1 error-status" on page 2
b. The error-index is set to the position (in the original request) of the varBind that caused the error-status.
c. The varBindList of the response PDU is made exactly the same as the varBindList that was received in the original request.
a. If there are any varBinds containing an SNMPv2 syntax of Counter64, then consider these varBinds to be not in view and repeat the call to the instrumentation code as often as needed till a value other than Counter64 is returned.
b. Find any varBind that contains an SNMPv2 exception endOfMibView. (Note that if there are more than one, the agent may choose any such varBind.) If there are any such varBinds, then for the one chosen:
1) Set the error-status to noSuchName
Wijnen & Levi Informational [Page 7]
RFC 2089 V2toV1 January 1997
2) Set the error-index to the position (in the varBindList of the original request) of the varBind that returned such an SNMPv2 exception.
3) Make the varBindList of the response PDU exactly the same as the varBindList that was received in the original request.
c. If there are no such varBinds, then:
1) Set the error-status to noError
2) Set the error-index to zero
3) Compose the varBindList of the response, using the data as it is returned by the instrumentation code.
If SNMPv2 compliant instrumentation presents an SNMPv2 trap to the SNMP engine and such a trap passes all regular checking and then is to be sent to an SNMPv1 destination, then the following steps must be followed to convert such a trap to an SNMPv1 trap. This is basically the reverse of the SNMPv1 to SNMPv2 mapping as described in RFC1908 [3].
1. If any of the varBinds in the varBindList has an SNMPv2 syntax of Counter64, then such varBinds are implicitly considered to be not in view, and so they are removed from the varBindList to be sent with the SNMPv1 trap.
2. The 3 special varBinds in the varBindList of an SNMPv2 trap (sysUpTime.0 (TimeTicks), snmpTrapOID.0 (OBJECT IDENTIFIER) and optionally snmpTrapEnterprise.0 (OBJECT IDENTIFIER)) are removed from the varBindList to be sent with the SNMPv1 trap. These 2 (or 3) varBinds are used to decide how to set other fields in the SNMPv1 trap PDU as follows:
a. The value of sysUpTime.0 is copied into the timestamp field of the SNMPv1 trap.
Wijnen & Levi Informational [Page 8]
RFC 2089 V2toV1 January 1997
b. If the snmpTrapOID.0 value is one of the standard traps the specific-trap field is set to zero and the generic trap field is set according to this mapping:
The enterprise field is set to the value of snmpTrapEnterprise.0 if this varBind is present, otherwise it is set to the value snmpTraps as defined in RFC1907 [4].
c. If the snmpTrapOID.0 value is not one of the standard traps, then the generic-trap field is set to 6 and the specific-trap field is set to the last subid of the snmpTrapOID.0 value.
o If the next to last subid of snmpTrapOID.0 is zero, then the enterprise field is set to snmpTrapOID.0 value and the last 2 subids are truncated from that value. o If the next to last subid of snmpTrapOID.0 is not zero, then the enterprise field is set to snmpTrapOID.0 value and the last 1 subid is truncated from that value.
In any event, the snmpTrapEnterprise.0 varBind (if present) is ignored in this case.
3. The agent-addr field is set with the appropriate address of the the sending SNMP entity, which is the IP address of the sending entity of the trap goes out over UDP; otherwise the agent-addr field is set to address 0.0.0.0.
The authors wish to thank the contributions of the SNMPv2 Working Group in general. Special thanks for their detailed review and comments goes to these individuals:
Mike Daniele (DEC) Dave Harrington (Cabletron) Brian O'Keefe (Hewlett Packard) Keith McCloghrie (Cisco Systems) Dave Perkins (independent) Shawn Routhier (Epilogue) Juergen Schoenwaelder (University of Twente)
[1] Jeffrey D. Case, Mark Fedor, Martin Lee Schoffstall and James R. Davin, Simple Network Management Protocol (SNMP), SNMP Research, Performance Systems International, MIT Laboratory for Computer Science, RFC 1157, May 1990.
[2] Jeffrey D. Case, Keith McCloghrie, Marshall T. Rose and Steven Waldbusser, Structure of Managment Information for Version 2 of the Simple Network Management Protocol (SNMPv2), SNMP Research Inc, Cisco Systems Inc, Dover Beach Consulting Inc, International Network Services, RFC1902, January 1996.
[3] Jeffrey D. Case, Keith McCloghrie, Marshall T. Rose and Steven Waldbusser, Coexistence between Version 1 and Version 2 of the Internet-standard Network Management Framework, SNMP Research Inc, Cisco Systems Inc, Dover Beach Consulting Inc, International Network Services, RFC1908, January 1996.
[4] Jeffrey D. Case, Keith McCloghrie, Marshall T. Rose and Steven Waldbusser, Management Information Base for Version 2 of the Simple Network Management Protocol (SNMPv2), SNMP Research Inc, Cisco Systems Inc, Dover Beach Consulting Inc, International Network Services, RFC1907, January 1996.
The mapping of SNMPv2 error-status values to SNMPv1 error-status values is based on the common interpretation of how an SNMPv1 entity should create an error-status value based on the elements of procedure defined in RFC1157 [1].
There was a suggestion to map wrongEncoding into genErr, because it could be caused by an ASN.1 parsing error. Such maybe true, but in most cases when we detect the ASN.1 parsing error, we do not yet know about the PDU data yet. Most people who responded to our queries have implemented the mapping to a badValue. So we "agreed" on the mapping to badValue.
RFC1448 says that if one of the objects in the varBindList is not included in the view, then the trap is NOT sent. Current SNMPv2u and SNMPv2* documents make the same statement. However, the "rough consensus" is that it is better to send partial information than no information at all. Besides:
o RFC1448 does not allow for a TRAP to be sent with the varBinds that are not included in the view removed, so it is an all or nothing decision.
o We do NOT include the Counter64 varBinds... so the "not in view" varBinds are not sent to the trap destination.
o The Counter64 objects are "implicit" not in view. If any objects are explicit not in view, then this is checked before we do the conversion from an SNMPv2 trap to an SNMPv1 trap, and so the trap is not sent at all.