Network Working Group M. Foschiano Request for Comments: 5171 Cisco Systems Category: Informational April 2008
Cisco Systems UniDirectional Link Detection (UDLD) Protocol
Status of This Memo
This memo provides information for the Internet community. It does not specify an Internet standard of any kind. Distribution of this memo is unlimited.
This RFC is not a candidate for any level of Internet Standard. The IETF disclaims any knowledge of the fitness of this RFC for any purpose and in particular notes that the decision to publish is not based on IETF review for such things as security, congestion control, or inappropriate interaction with deployed protocols. The RFC Editor has chosen to publish this document at its discretion. Readers of this document should exercise caution in evaluating its value for implementation and deployment. See RFC 3932 for more information.
This document describes a Cisco Systems protocol that can be used to detect and disable unidirectional Ethernet fiber or copper links caused, for instance, by mis-wiring of fiber strands, interface malfunctions, media converters' faults, etc. It operates at Layer 2 in conjunction with IEEE 802.3's existing Layer 1 fault detection mechanisms.
This document explains the protocol objectives and applications, illustrates the specific premises the protocol was based upon, and describes the protocol architecture and related deployment issues to serve as a possible base for future standardization.
Today's Ethernet-based switched networks often rely on the Spanning Tree Protocol (STP) defined in the IEEE 802.1D standard  to create a loop-free topology that is used to forward the traffic from a source to a destination based on the Layer 2 packet information learned by the switches and on the knowledge of the status of the physical links along the path.
Issues arise when, due to mis-wirings or to hardware faults, the communication path behaves abnormally and generates forwarding anomalies. The simplest example of such anomalies is the case of a bidirectional link that stops passing traffic in one direction and therefore breaks one of the most basic assumptions that high-level protocols typically depend upon: reliable two-way communication between peers.
The purpose of the UDLD protocol is to detect the presence of anomalous conditions in the Layer 2 communication channel, while relying on the mechanisms defined by the IEEE in the 802.3 standard  to properly handle conditions inherent to the physical layer.
The UniDirectional Link Detection protocol (often referred to in short as "UDLD") is a lightweight protocol that can be used to detect and disable one-way connections before they create dangerous situations such as Spanning Tree loops or other protocol malfunctions.
The protocol's main goal is to advertise the identities of all the capable devices attached to the same LAN segment and to collect the information received on the ports of each device to determine if the Layer 2 communication is happening in the appropriate fashion.
In a network that has an extensive fiber cabling plant, problems may arise when incorrect patching causes a switch port to have its RX fiber strand connected to one neighbor port and its TX fiber strand connected to another. In these cases, a port may be deemed active if it is receiving an optical signal on its RX strand. However, the problem is that this link does not provide a valid communication path at Layer 2 (and above).
If this scenario of wrongly connected fiber strands is applied to multiple ports to create a fiber loop, each device in the loop could directly send packets to a neighbor but would not be able to receive from that neighbor.
Albeit the above scenario is rather extreme, it exemplifies how the lack of mutual identification of the neighbors can bring protocols to the wrong assumption that during a transmission the sender and the receiver are always properly matched. Another equally dangerous incorrect assumption is that the lack of reception of protocol messages on a port unmistakably indicates the absence of transmitting protocol entities at the other end of the link.
The UDLD protocol was implemented to help correct certain assumptions made by other protocols, and in particular to help the Spanning Tree Protocol to function properly so as to avoid the creation of dangerous Layer 2 loops. It has been available on most Cisco Systems switches for several years and is now part of numerous network design best practices.
The current implementation of UDLD is based on the following considerations/presuppositions:
o The protocol would have to be run in the control plane of a network device to be flexible enough to support upgrades and bug fixes. The control plane speed would ultimately be the limiting factor to the capability of fast fault detection of the protocol (CPU speed, task switching speed, event processing speed, etc.). The transmission medium's propagation delay at 10 Mbps speed (or higher) would instead be considered a negligible factor.
o Network events typically do not happen with optimal timing, but rather at the speed determined by the software/firmware infrastructure that controls them. (For psychological and practical reasons, developers tend to choose round timer values rather than determine the optimal value for the specific software architecture in use. Also, software bugs, coding oversights, slow process switching, implementation overhead can all affect the control plane responsiveness and event timings.) Hence it was deemed necessary to adopt a conservative protocol design to minimize false positives during the detection process.
o If a fault were discovered, it was assumed that the user would want to keep the faulty port down for a predetermined amount of time to avoid unnecessary port state flapping. For that reason, a time-based fault recovery mechanism was provided (although alternative recovery mechanisms are not implicitly precluded by the protocol itself).
UDLD is meant to be a Layer 2 detection protocol that works on top of the existing Layer 1 detection mechanisms defined by the IEEE standards. For example, the Far End Fault Indication (FEFI) function for 100BaseFX interfaces and the Auto-Negotiation function for 100BaseTX/1000BaseX interfaces represent standard physical-layer mechanisms to determine if the transmission media is bidirectional. (Please see sections 184.108.40.206 and 220.127.116.11 of  for more details.) The typical case of a Layer 1 "fault" indication is the "loss of light" indication.
UDLD differs from the above-mentioned mechanisms insofar as it performs mutual neighbor identification; in addition, it performs neighbor acknowledgement on top of the Logical Link Control (LLC)
Foschiano Informational [Page 4]
RFC 5171 UDLD April 2008
layer and thus is able to discover logical one-way miscommunication between neighbors even when either one of the said PHY layer mechanisms has deemed the transmission medium bidirectional.
a. It advertises a port's identity and learns about its neighbors on a specific LAN segment; it keeps the acquired information on the neighbors in a cache table.
b. It sends a train of echo messages in certain circumstances that require fast notifications or fast resynchronization of the cached information.
Because of the above, the algorithm run by UDLD requires that all the devices connected to the same LAN segment be running the protocol in order for a potential misconfiguration to be detected and for a prompt corrective action to be taken.
UDLD sends periodical "hello" packets (also called "advertisements" or "probes") on every active interface to keep each device informed about its neighbors. When a hello message is received, it is cached and kept in memory at most for a defined time interval, called "holdtime" or "time-to-live", after which the cache entry is considered stale and is aged out.
If a new hello message is received when a correspondent old cache entry has not been aged out yet, then the old entry is dropped and is replaced by the new one with a reset time-to-live timer. Whenever an interface gets disabled and UDLD is running, or whenever UDLD is disabled on an interface, or whenever the device is reset, all existing cache entries for the interfaces affected by the configuration change are cleared, and UDLD sends at least one message to inform the neighbors to flush the part of their caches also affected by the status change. This mechanism is meant to keep the caches coherent on all the connected devices.
The echoing mechanism is the base of UDLD's detection algorithm: whenever a UDLD device learns about a new neighbor or receives a resynchronization request from an out-of-synch neighbor, it (re)starts the detection process on its side of the connection and sends N echo messages in reply. (This mechanism implicitly assumes that N packets are sufficient to get through a link and reach the other end, even though some of them might get dropped during the transmission.)
Since this behavior must be the same on all the neighbors, the sender of the echoes expects to receive (after some time) an echo in reply. If the detection process ends without the proper echo information being received, and under specific conditions, the link is considered to be unidirectional.
UDLD can function in two modes: normal mode and aggressive mode.
In normal mode, a protocol determination at the end of the detection process is always based on information received in UDLD messages: whether it's the information about the exchange of proper neighbor identifications or the information about the absence of such proper identifications. Hence, albeit bound by a timer, normal mode determinations are always based on gleaned information, and as such are "event-based". If no such information can be obtained (e.g., because of a bidirectional loss of connectivity), UDLD follows a conservative approach based on the considerations in Section 3 and deems a port to be in "undetermined" state. In other words, normal mode will shut down a port only if it can explicitly determine that the associated link is faulty for an extended period of time.
In contrast, in aggressive mode, UDLD will also shut down a port if it loses bidirectional connectivity with the neighbor for the same extended period of time mentioned above and subsequently fails repeated last-resort attempts to re-establish communication with the other end of the link. This mode of operation assumes that loss of communication with the neighbor is a meaningful network event in itself and is a symptom of a serious connectivity problem. Because this type of detection can be event-less, and lack of information cannot always be associated to an actual malfunction of the link, this mode is optional and is recommended only in certain scenarios (typically only on point-to-point links where no communication failure between two neighbors is admissible).
The UDLD protocol runs on top of the LLC sub-layer of the data link layer of the OSI model. It uses a specially assigned multicast destination MAC address and encapsulates its messages using the standard Subnetwork Access Protocol (SNAP) format as described in the following:
Destination MAC address 01-00-0C-CC-CC-CC
UDLD SNAP format: LLC value 0xAAAA03 Org Id 0x00000C HDLC protocol type 0x0111
UDLD's Protocol Data Unit (PDU) format is as follows:
Type (16 bits): If an implementation does not understand a Type value, it should skip over it using the length field.
Length (16 bits): Length in bytes of the Type, Length, and Value fields. In order for this field value to be valid, it should be greater than or equal to the minimum allowed length, 4 bytes. If the value is less than the minimum, the whole packet is to be considered corrupted and therefore it must be discarded right away during the parsing process. TLVs should not be split across packet boundaries.
Foschiano Informational [Page 7]
RFC 5171 UDLD April 2008
Value (variable length): Object contained in the TLV.
The protocol header fields are defined as follows:
Ver (3 bits): 0x01: UDLD PDU version number
Opcode (5 bits): 0x00: Reserved 0x01: Probe message 0x02: Echo message 0x03: Flush message 0x04-0x1F: Reserved for future use
Flags (8 bits): bit 0: Recommended timeout flag (RT) bit 1: ReSynch flag (RSY) bit 2-7: Reserved for future use
PDU Checksum (16 bits): IP-like checksum. Take the one's complement of the one's complement sum (with the modification that the odd byte at the end of an odd length message is used as the low 8 bits of an extra word, rather than as the high 8 bits.) NB: All UDLD implementations must comply with this specification.
The list of currently defined TLVs comprises:
Name Type Value format
Device-ID TLV 0x0001 ASCII character string Port-ID TLV 0x0002 ASCII character string Echo TLV 0x0003 List of ID pairs Message Interval TLV 0x0004 8-bit unsigned integer Timeout Interval TLV 0x0005 8-bit unsigned integer Device Name TLV 0x0006 ASCII character string Sequence Number TLV 0x0007 32-bit unsigned integer Reserved TLVs > 0x0007 Format unknown. To be skipped by parsing routine.
This TLV uniquely identifies the device that is sending the UDLD packet. The TLV length field determines the length of the carried identifier and must be greater than zero. In version 1 of the protocol, the lack of this ID is considered a symptom of packet corruption that implies that the message is invalid and must be discarded.
This TLV uniquely identifies the physical port the UDLD packet is sent on. The TLV length field determines the length of the carried identifier and must be greater than zero. In version 1 of the protocol, the lack of this ID is considered a symptom of packet corruption that implies that the message is invalid and must be discarded.
This TLV contains the list of valid DeviceID/PortID pairs received by a port from all its neighbors. If either one of the identifiers in a pair is corrupted, the message will be ignored. This list includes only DeviceIDs and PortIDs extracted from UDLD messages received and cached on the same interface on which this TLV is sent. If no UDLD messages are received, then this TLV is sent containing zero pairs. Despite its name, this TLV must be present in both probe and echo messages, whereas in flush messages it's not required.
Message Interval TLV:
This required TLV contains the 8-bit time interval value used by a neighbor to send UDLD probes after the linkup or detection phases. Its time unit is 1 second. The holdtime of a cache item for a received message is calculated as (advertised-message-interval x R), where R is a constant called "TTL to message interval ratio".
Timeout Interval TLV:
This optional TLV contains the 8-bit timeout interval value (T) used by UDLD to decide the basic length of the detection phase. Its time unit is 1 second. If it's not present in an advertisement, T is assumed to be a hard-coded constant.
Foschiano Informational [Page 9]
RFC 5171 UDLD April 2008
Device Name TLV:
This required TLV is meant to be used by the CLI or SNMP and typically contains the user-readable device name string.
Sequence Number TLV:
The purpose of this optional TLV is to inform the neighbors of the sequence number of the current message being transmitted. A counter from 1 to 2^32-1 is supposed to keep track of the sequence number; it is reset whenever a transition of phase occurs so that it will restart counting from one, for instance, whenever an echo message sequence is initiated, or whenever a linkup message train is triggered.
No wraparound of the counter is supposed to happen.
The zero value is reserved and can be used as a representation of a missing or invalid sequence number by the user interface. Therefore, the TLV should never contain zero. (NB: The use of this TLV is currently limited only to informational purposes.)
UDLD timer values could vary within certain "safety" ranges based on the considerations in Section 3. However, in practice, in the current implementation, timers use only certain values verified during testing. Their time unit is one second.
During the detection phase, messages are exchanged at the maximum possible rate of one per second. After that, if the protocol reaches
Foschiano Informational [Page 10]
RFC 5171 UDLD April 2008
a stable state and can make a certain determination on the "bidirectionality" of the link, the message interval is increased to a configurable value based on a curve known as M1(t), a time-based function.
In case the link is deemed anything other than bidirectional at the end of the detection, this curve is a flat line with a fixed value of Mfast (7 seconds in the current implementation).
In case the link is instead deemed bidirectional, the curve will use Mfast for the first 4 subsequent message transmissions and then will transition to an Mslow value for all other steady-state transmissions. Mslow can be either a fixed value (60 seconds in some obsolete implementations) or a user-configurable value (between Mfast and 90 seconds with a default of 15 seconds in the current implementations).
8. Comparison with Bidirectional Forwarding Detection
Similarly to UDLD, the Bidirectional Forwarding Detection (BFD)  protocol is intended to detect faults in the path between two network nodes. However, BFD is supposed to operate independently of media, data protocols, and routing protocols. There is no address discovery mechanism in BFD, which is left to the application to determine.
On the other hand, UDLD works exclusively on top of a L2 transport supporting the SNAP encapsulation and operates independently of the other bridge protocols (UDLD's main "applications"), with which it has limited interaction. It also performs full neighbor discovery on point-to-point and point-to-multipoint media.
In a heterogeneous Layer 2 network that is built with different models of network devices or with devices running different software images, the UDLD protocol should be supported and configured on all ports interconnecting said devices in order to achieve a complete coverage of its detection process. Note that UDLD is not supposed to be used on ports connected to untrusted devices or incapable devices; hence, it should be disabled on such ports.
 IEEE 802.1D-2004 Standard -- Media access control (MAC) Bridges
 IEEE 802.3-2002 IEEE Standard -- Local and metropolitan area networks Specific requirements--Part 3: Carrier Sense Multiple Access with Collision Detection (CSMA/CD) Access Method and Physical Layer Specifications
This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at firstname.lastname@example.org.