Skip to content

linux-nic-link-is-down.md

Latest commit

 

History

History
161 lines (123 loc) · 5.57 KB

linux-nic-link-is-down.md

File metadata and controls

161 lines (123 loc) · 5.57 KB
Raw
title Linux NIC Link is Down
slug linux-nic-link-is-down
technologies
linux
severity high
tags
linux
networking
nic
link
production
related
linux-systemd-failed-to-start-unit
linux-ssh-permission-denied-publickey
last_reviewed 2026-06-27

Linux NIC Link is Down

Error Message

kernel: e1000e 0000:00:1f.6 eth0: NIC Link is Down

kernel: igb 0000:03:00.0 eno1: Link is Down
NetworkManager: <warn>  [...] device (eth0): carrier: link disconnected

Description

NIC Link is Down (also "Link is Down" / "carrier: link disconnected") is a kernel/driver message reporting that the network interface has lost physical layer (L1) carrier — the NIC no longer sees a link partner on the wire. This is distinct from an IP/routing misconfiguration: at this level the cable, switch port, or transceiver is the issue, and the interface's LOWER_UP flag is cleared. No traffic can flow on that interface until carrier returns. When it flaps (Down/Up repeatedly) you'll see paired messages and intermittent connectivity.

Technologies

  • linux (network device drivers, link layer)

Severity

high — the host loses connectivity on that interface. If it's the primary or only NIC, the machine goes off the network entirely (you may lose SSH and depend on console/OOB). For bonded/redundant setups a single link-down is degraded rather than down, but a flapping link can be worse than a clean failure.

Common Causes

  1. Physical layer: unplugged/faulty cable, bad switch port, dead SFP/transceiver, or a powered-off switch.
  2. The interface is administratively down (ip link set ... down) or never brought up by the network config.
  3. Speed/duplex auto-negotiation mismatch with the switch (link won't establish or flaps).
  4. Driver/firmware bug or a NIC power-management quirk (EEE/ASPM) dropping link.
  5. Hardware failure of the NIC itself, or a recent kernel/driver upgrade regression.

Root Cause Analysis

The NIC's PHY continuously negotiates a link with its partner (switch port). When carrier is lost, the driver clears IFF_RUNNING/LOWER_UP and emits "Link is Down"; the kernel then marks routes through that interface unusable and NetworkManager/networkd report carrier: down. The driver also records negotiated speed/duplex — a half/full-duplex or speed mismatch with the switch prevents a stable link and shows up as flapping with rising error/CRC counters. Because this is L1, no amount of IP reconfiguration helps until the driver reports Link is Up; the fix is on the wire, the switch, or the driver, not in the routing table.

Diagnostic Commands

# Operational state and flags (look for state DOWN / NO-CARRIER)
ip -br link show
ip link show eth0

# Driver link events: up/down, negotiated speed/duplex, flapping
dmesg -T | grep -iE 'link is down|link is up|nic link|carrier'

# Carrier, negotiated speed/duplex, and whether link is detected
sudo ethtool eth0 | grep -iE 'speed|duplex|link detected|auto-negotiation'

# Interface error/drop/CRC counters (rising = bad cable/duplex mismatch)
ip -s link show eth0
cat /sys/class/net/eth0/carrier   # 1 = link up, 0 = link down

Expected Results

$ ip -br link show
eth0  DOWN  00:1b:21:aa:bb:cc <NO-CARRIER,BROADCAST,MULTICAST,UP>

$ ethtool eth0
        Speed: Unknown!
        Duplex: Unknown! (255)
        Link detected: no

$ cat /sys/class/net/eth0/carrier
0

NO-CARRIER with Link detected: no and carrier 0 confirms L1 link loss. If ethtool shows e.g. Speed: 100Mb/s, Duplex: Half on a gigabit link, suspect an auto-negotiation/duplex mismatch; rising errors/dropped in ip -s link points at a marginal cable.

Resolution

  1. Check and reseat/replace the physical path first: cable, switch port, and transceiver. Move to a known-good port to isolate.

  2. If the interface is administratively down, bring it up:

    sudo ip link set eth0 up

    and ensure the persistent config (netplan / NetworkManager / systemd-networkd) actually enables it on boot.

  3. For a duplex/speed mismatch, set both ends consistently; prefer auto-negotiation, but if the switch is fixed, match it:

    sudo ethtool -s eth0 autoneg off speed 1000 duplex full   # match the switch

    (persist via your network config). Risk: a mismatch here causes a hard link failure — change one variable at a time.

  4. If a driver power feature flaps the link, disable EEE/ASPM for that NIC, or roll back a regressing kernel/driver.

  5. If the NIC is dead, fail over to a redundant interface/bond and replace the hardware.

Validation

ip -br link show eth0                 # expect state UP, no NO-CARRIER
ethtool eth0 | grep 'Link detected'   # expect: Link detected: yes
ping -c3 <gateway>                    # connectivity restored
dmesg -T | grep -i 'link is up' | tail -1

Prevention

  • Use bonded/redundant NICs (LACP) so one link-down is non-disruptive.
  • Monitor carrier, interface error/CRC counters, and link flaps; alert on them.
  • Standardize on auto-negotiation across hosts and switch ports.
  • Keep OOB/console access so a primary-NIC failure isn't a lockout.
  • Test NIC driver/firmware updates in staging before fleet rollout.

Related Errors

References

Tags

linux · networking · nic · link · production