Packet Path/

OSPF Neighbor States Explained: Down to Full, and Where Each One Sticks

When two OSPF routers meet, they do not swap routes straight away. They climb a ladder of states — Down, Init, 2-Way, ExStart, Exchange, Loading, Full — and each rung proves one thing before the next is allowed. This guide walks the whole climb, shows one clean adjacency end to end, and explains exactly where each stuck state points: a neighbour stuck in Init, in 2-Way, or in ExStart is telling you three different stories.

Routing · Intermediate · 10 min · November 11, 2026

Editorial illustration on a dark teal background: two routers connected by a glowing green link, above a row of seven circles that light up step by step from dim to bright green.

The short answer

What are the OSPF neighbor states? They are the steps two OSPF routers climb, in order, before they trust each other with routes: Down, then Init, then 2-Way, then ExStart, then Exchange, then Loading, then Full. Each state proves one thing. If a neighbour is stuck, the state it is stuck in tells you which proof failed — which is why engineers read the state first and guess second.

Think of it like two neighbours meeting over a fence. Down is silence: you have heard nothing. Init is hearing a voice, but not hearing your own name in what they said. 2-Way is the moment you both name each other — a real two-way conversation. Only then does the serious part start: ExStart is agreeing who will lead the conversation, Exchange is comparing lists of what you each know, Loading is asking for the full details of the items you are missing, and Full is knowing you both hold the same complete picture.

In plain EnglishNeighbor: a router on the same link that you have heard OSPF Hellos from. Adjacency: a neighbour you have fully synchronised databases with — a stronger relationship than just being neighbours. Hello: the small OSPF packet routers send regularly to announce themselves and list the neighbours they can already hear. Link-state database (LSDB): the map of the network each router builds. LSA: one advertisement — one piece of that map. Database Description (DBD or DD): a summary list of LSAs, swapped before the full LSAs. DR / BDR: the designated router and its backup on a shared network, elected so not every pair has to synchronise with every other pair.

The eight states, in order

OSPF version 2 defines these states in RFC 2328, section 10.1. There are eight, because one — Attempt — only appears on a special kind of network. In everyday work you will mostly watch the other seven.

1. Down. No recent information has been received from this neighbour. No Hellos have arrived. On most links this is simply where every neighbour starts, and where one returns after it goes quiet.

2. Attempt. This state is only valid on non-broadcast multi-access (NBMA) networks, where neighbours do not discover each other by multicast and must be contacted directly. It means: no recent information received, but a more concerted effort is being made to contact the neighbour anyway. You will rarely meet it on ordinary Ethernet links.

3. Init. A Hello packet has recently been seen from the neighbour, but two-way communication has not yet been established: your own router ID did not appear in that Hello. In fence terms, you heard them talking, but they have not yet shown they can hear you.

4. 2-Way. Communication is bidirectional. Each router has seen itself listed in the other router's Hello. This is the most advanced state before database exchange begins, and it is where the DR and BDR are chosen from. No routes have been swapped yet.

5. ExStart. The first step in creating an adjacency. The two routers negotiate who is master and who is slave for the database exchange, and agree an initial sequence number for the DBD packets. The router with the higher router ID becomes master. Nothing about the network map has been exchanged yet — this is agreeing how the exchange will run.

6. Exchange. The routers exchange DBD packets describing their link-state databases — the table of contents, not the book. Each side can now work out which LSAs the other has that it is missing.

7. Loading. The actual exchange of link-state information. Based on the DBDs, each router sends Link State Request packets for the newer or missing LSAs, and receives them in Link State Update packets.

8. Full. The routers are fully adjacent. Their databases are fully synchronised, and the adjacency is advertised in LSAs. Full is the normal, healthy end state — with one important exception, covered below.

One adjacency, seven everyday stepsDownInit2-WayExStartExchangeLoadingFullHello proof first, then database exchange, then Full

The state chain in order. ExStart is highlighted because it is the state engineers meet most often when an adjacency refuses to form — usually for one well-known reason, below.

One clean walk, end to end

Two routers, R1 and R2, on a point-to-point link, both correctly configured. R1 starts in Down for R2: it has heard nothing. A Hello arrives from R2, so R1 moves to Init — a Hello was seen, but R1 is not in R2's neighbour list yet. R2's next Hello lists R1's router ID, so the state moves to 2-Way: both sides have now proved they can hear each other. Because this link needs a full adjacency, they move straight into ExStart, agree that the higher router ID leads, and set the starting DBD sequence number. In Exchange they swap DBD summaries and spot the LSAs each is missing. In Loading they request and receive those LSAs in full. When there is nothing left to request, the state becomes Full: the databases match, and routes through each other can now be calculated.

On a healthy link that whole climb usually takes seconds. That is exactly why a state that does not move is such good evidence: the climb has stopped at the first proof it cannot complete.

Where each stuck state points

Stuck in Init means Hellos are arriving one way only, or arriving but not being accepted as two-way. Check the plain things first: is OSPF traffic (IP protocol 89, not a TCP or UDP port) being blocked in one direction by a filter? Do the Hello and Dead timers, area, and authentication match on both ends? A neighbour that flaps between Down and Init is usually losing Hellos, not failing maths.

Stuck in 2-Way needs a careful read, because 2-Way is often correct. On a broadcast network, routers that are neither DR nor BDR — DROTHERs — stay at 2-Way with each other by design, and go Full only with the DR and BDR. That saves a storm of duplicate synchronisation. 2-Way is only a problem when the pair should go Full — for example, one of them is the DR, or the link is point-to-point — and does not move on.

Stuck in ExStart is the famous one. The leading cause is an interface MTU mismatch: the DBD packets carry the interface MTU, and a router can reject a DBD whose MTU is larger than it can accept, so the negotiation never completes. One side set to 9000 and the other to 1500 is the classic production case. Check MTU on both ends first. There is a command that tells OSPF to ignore the MTU check, but treat it as a diagnostic clue, not a fix — mismatched MTUs will cause real trouble later with large packets, even if the adjacency is forced up.

Stuck in Exchange or Loading means the summary exchange started but the full information is not flowing: requests are not being answered, or large updates are not getting through. Work it in the same spirit: what changed, is anything filtering or dropping larger packets on the link, and do both sides still agree on MTU?

For the full mid-incident version of the ExStart case — checks in senior order, what each result rules in or out, and a safe fix — work through the companion case: OSPF neighbor stuck in ExStart. If the log line you are staring at is the adjacency dropping, the decoder entry for %OSPF-5-ADJCHG explains what that message does and does not tell you.

Four mistakes that waste the evidence

1. Reading 2-Way as broken, every time. Between two DROTHERs it is the designed end state. Check the DR and BDR roles before changing anything.

2. Fixing ExStart with the MTU-ignore command and walking away. The adjacency may come up, but the underlying mismatch remains, waiting for a large packet. Match the MTUs properly; use ignore only to prove the diagnosis.

3. Restarting OSPF before reading the state. A restart throws away exactly the evidence — the stuck state — that names the failing proof. Read first, record the state and the neighbour, then change one thing.

4. Assuming stuck means a routing problem downstream. Until a neighbour reaches Full, there is no synchronised database and no reliable route calculation through it. Fix the adjacency first; many apparent route problems are an unfinished climb.

Try it yourself

Run a config pair through the Network Config Verifier and look for the mismatches that stop adjacencies before they start, then read the worked ExStart case and the decoder entry linked above. The verifier checks the configuration; the state machine tells you what the routers managed to prove live — you want both views.

Fact-check sources: the state definitions, the Attempt state's NBMA-only scope, the Init and 2-Way meanings, master and slave negotiation in ExStart, DBD exchange in Exchange, link-state requests in Loading, and Full meaning fully synchronised databases are all from RFC 2328 section 10.1 and Cisco's Understand OSPF Neighbor States document, which is based on that section. The MTU cause for stuck ExStart is documented in Cisco's OSPF troubleshooting guidance for ExStart and Exchange problems.

Related guides

Key takeaways

Keep reading