Engineer working on a laptop connected to a device test rig, surrounded by wiring harnesses and test hardware

Firmware vulnerabilities & fixes

Avahi and mDNS/DNS-SD vulnerabilities in embedded Linux: find the responder and version, rank the CVEs, keep port 5353 on the local link

Last updated:

The Avahi and mDNS/DNS-SD vulnerabilities in embedded Linux that matter most as of 2026-11-14 are the ones reachable over UDP port 5353: CVE-2017-6519, where avahi-daemon answered queries from off-link addresses (fixed in 0.8), and CVE-2025-68468 and CVE-2026-24401, where crafted CNAME announcements from the local link crash the daemon (fixed in v0.9-rc3). Local socket and D-Bus flaws such as CVE-2021-3468 and CVE-2023-38469 to 38473 follow. Find the responder and version, patch or backport, and keep port 5353 on the local link.

Key takeaways

  • Avahi (avahi-daemon), the common choice on Linux distributions and on build systems such as Buildroot and Yocto.
  • The Avahi CVEs on embedded images fall into two groups: flaws an attacker can reach with mDNS packets, and flaws that need a local process to talk to the daemon over its Unix socket or D-Bus.
  • Multicast DNS assumes that everyone who can send to port 5353 is on the same link, and the protocol defines that boundary itself.
  • Start by establishing which responder is present, then its version, then the configuration it actually runs with.

Where do avahi-daemon and other mDNS responders ship?

An mDNS responder ships wherever a device wants to be found on a LAN without configuration: NAS boxes announcing file shares and Time Machine targets, printers announcing IPP and AirPrint, IP cameras and NVRs announcing their web interface, smart-home hubs, media players and home or industrial gateways. Multicast DNS answers name queries for .local names on UDP port 5353, and DNS-SD uses the same records to list services such as _ipp._tcp or _smb._tcp.

On embedded Linux, the responder is usually one of three:

  • Avahi (avahi-daemon), the common choice on Linux distributions and on build systems such as Buildroot and Yocto. It is LGPL-2.1-or-later, and OpenWrt's avahi package Makefile builds it from the v0.9-rc5 tag, in a D-Bus variant (avahi-dbus-daemon) and a smaller one without D-Bus (avahi-nodbus-daemon).
  • umdns, OpenWrt's own lightweight responder, built from the OpenWrt project's mdnsd repository according to its package Makefile. It is a separate code base with its own history, so Avahi CVEs do not apply to it.
  • mDNSResponder, Apple's implementation, published as open source with a mDNSPosix directory for POSIX systems. Device makers port it as mdnsd, especially in printers and media devices that need Apple compatibility.

Small images can also carry a vendor responder inside a larger binary, such as a camera's main application. A file-system scan that looks only for avahi-daemon misses those, so search for port 5353 and the multicast addresses 224.0.0.251 and ff02::fb in binaries as well. This guide covers the responder; our dnsmasq guide covers the unicast DNS forwarder that often sits next to it, and our MiniUPnPd and UPnP guide covers SSDP, the other discovery protocol on the same devices.

Which Avahi CVEs matter most for device makers?

The Avahi CVEs on embedded images fall into two groups: flaws an attacker can reach with mDNS packets, and flaws that need a local process to talk to the daemon over its Unix socket or D-Bus. Scores below are NVD's, with the GitHub advisory's score where NVD has none. EPSS values are FIRST's for 2026-11-14, and none of these CVEs is in CISA's KEV catalog version 2026.11.13.

CVE What it is Reach Affected Fixed upstream CVSS EPSS
CVE-2017-6519 Answers IPv6 unicast queries from off-link sources: traffic amplification and information leakage Network, UDP 5353 Through 0.6.32 and 0.7 0.8 (2020-02-18) 9.1 0.032
CVE-2025-68468 CNAME records pointing to short-TTL records crash the daemon when they expire Local link, mDNS 0.9-rc2 and earlier v0.9-rc3 6.5 0.004
CVE-2026-24401 A CNAME record pointing to itself causes unbounded recursion and a segmentation fault Local link, mDNS 0.9-rc2 and earlier v0.9-rc3 6.5 0.004
CVE-2024-52616 Sequential transaction IDs in wide-area (unicast) DNS-SD queries make spoofing easier Network, only with wide-area on Before 0.9 v0.9-rc2 5.3 (advisory) 0.007
CVE-2021-3468 A closed client connection on the Unix socket sends the daemon into an infinite loop Local 0.6 up to 0.8 v0.9-rc1 5.5 0.004
CVE-2023-38469 to CVE-2023-38473 Five reachable assertions (packet append, label escaping, host name setting, rdata parsing, alternative host name) abort the daemon Local, D-Bus Before 0.9 v0.9-rc1 5.5 each 0.003

Every entry is a denial of service or a spoofing aid, not code execution, which is why scores stay moderate and none is in KEV. On a device, though, a crashed avahi-daemon means a printer, camera or NAS disappears from discovery until something restarts it, and CVE-2017-6519 turns the device into a traffic source for others. The flow of advisories continues: Avahi's security advisories page lists CVE-2026-34933, a local D-Bus call with conflicting publish flags that crashes the daemon, fixed in v0.9-rc4, and two July 2026 advisories fixed in v0.9-rc5.

Why is mDNS on a WAN or untrusted VLAN an exposure?

Multicast DNS assumes that everyone who can send to port 5353 is on the same link, and the protocol defines that boundary itself. RFC 6762 section 11 says a querier "MUST only accept responses to that query that originate from the local link", so that remote hosts cannot send spoofed answers, and that a responder should avoid unicast replies to sources outside its subnet. CVE-2017-6519 is the responder-side failure: according to its NVD record, avahi-daemon answered IPv6 unicast queries from addresses that were not on-link, which let remote attackers use it for traffic amplification and read information from the responses.

The Avahi 0.8 release notes list the fix as dropping legacy unicast queries from addresses not on the local link. Version alone does not remove the exposure, though. Any responder that listens on a WAN or point-to-point interface publishes the device's host name, services and addresses to that network, and the avahi-daemon.conf manual warns that enabling point-to-point interfaces "opens a potential security hole by allowing mDNS access from Internet connections".

Untrusted VLANs bring the network-reachable crashes into play. The advisory for CVE-2025-68468 shows that unsolicited announcements are enough to abort avahi-daemon while a client browses for a service such as _ipp._tcp, which a print or media application does all the time. A guest Wi-Fi network, a shared building network or a flat OT segment gives every attached device that position. Reflectors widen it further: enable-reflector=yes forwards mDNS between all interfaces of a gateway, so a crash or leak on one segment reaches the others.

How do you find the responder and its version in a firmware image?

Start by establishing which responder is present, then its version, then the configuration it actually runs with.

  1. Binaries and librariesIn the unpacked root file system, look for avahi-daemon, libavahi-core, libavahi-common and libavahi-client, for umdns, and for mdnsd or libdns_sd. Our BusyBox guide explains the userland these images are usually built on.
  2. Package metadataOpenWrt's opkg database records avahi-dbus-daemon or avahi-nodbus-daemon with versions such as 0.9_rc5-1; Buildroot and Yocto record avahi in their manifests. Our embedded Linux build systems and SBOM guide shows where those records live.
  3. Version stringsSearch the daemon and libavahi-core for the version string, or run avahi-daemon --version on a device you control. Treat a release-candidate version such as 0.9-rc3 carefully: some scanners compare it as lower than 0.9 and others as equal.
  4. ConfigurationRead the avahi-daemon.conf the device ships and any service files in the services directory next to it, which list what the device announces. Note allow-interfaces, deny-interfaces, allow-point-to-point, publish-hinfo, enable-reflector, enable-wide-area and disable-user-service-publishing.
  5. Runtime checkOn a device you control, list listening sockets to see which addresses bind UDP 5353, and send an mDNS query from the WAN side to confirm the firewall drops it.
  6. Vendor patchesCheck the patch directory of the vendor's build tree and the distribution changelog for backported fixes before calling a CVE open.

Where are Avahi fixes published?

Avahi publishes fixes in three places, and a device maker needs all of them. The GitHub releases page holds the last final release, 0.8 from 2020-02-18, whose notes include the CVE-2017-6519 fix. Every later fix landed on the development branch and in the v0.9-rc1 to v0.9-rc5 tags, the newest of which dates from June 2026. Avahi's security advisories give the affected and patched tag for each CVE, for example v0.9-rc1 for CVE-2021-3468 and v0.9-rc3 for CVE-2026-24401.

Distributions fill the gap for products that stay on 0.8. The Debian tracker for CVE-2023-38469 shows the v0.9-rc1 fix backported to 0.8-5+deb11u3 in bullseye (DLA-3990-1) and 0.8-10+deb12u1 in bookworm, and records disable-user-service-publishing as a mitigation. The same tracker shows the limits of that approach: for CVE-2025-68468 it lists bookworm and trixie as vulnerable, marked as a minor issue, with only unstable fixed in 0.8-18. For CVE-2017-6519, Debian rates the issue unimportant and lists the fix commit, e111def, which is the one to cherry-pick into an older tree.

How should you harden avahi-daemon on a device?

Most Avahi exposure on devices comes from running it where it is not needed. The upstream default avahi-daemon.conf leaves allow-interfaces unset, so the daemon uses every local interface except loopback and point-to-point. OpenWrt's default configuration also sets publish-hinfo=yes, which the manual says announces "the local operating system and CPU". A device baseline looks like this:

[server]
# Only the LAN bridge; never WAN, VPN or guest interfaces
allow-interfaces=br-lan
allow-point-to-point=no
use-ipv4=yes
use-ipv6=yes

[wide-area]
# Unicast DNS-SD off (CVE-2024-52615, CVE-2024-52616)
enable-wide-area=no

[publish]
# No OS/CPU or workstation records
publish-hinfo=no
publish-workstation=no
# Local applications cannot publish services over D-Bus
disable-user-service-publishing=yes
# For a device that only browses, set this instead:
# disable-publishing=yes

[reflector]
enable-reflector=no

Adjust it to the product. Disable the daemon entirely if no feature uses discovery; a service that does not run has no CVEs to triage. deny-interfaces takes precedence over allow-interfaces and is useful for excluding a guest bridge. Leave disable-user-service-publishing at no only if a local application really needs to publish over D-Bus, because that is the path to the CVE-2023-38469 to 38473 assertions. Turn the reflector on only on a gateway that must bridge discovery between two trusted segments, and then use reflect-filters to limit it to the service types the product needs.

Configuration is not a firewall. Drop UDP port 5353 on every WAN, VPN and guest interface in the device firewall, and accept mDNS only from the LAN to 224.0.0.251 and ff02::fb. That also covers a responder that ignores its own interface settings after an update, and it is the only control that applies equally to Avahi, umdns and a vendor responder.

How does a device maker patch or backport Avahi?

  1. Move to the newest tag the build system packagesAs of 2026-11-14 that is v0.9-rc5, which OpenWrt's packages feed already builds. It closes every CVE in the table above.
  2. Backport when you stay on 0.8Take the fix commits from Avahi's advisories or from the Debian tracker's notes, for example a337a1b for CVE-2023-38469 and f66be13 for CVE-2025-68468, and apply them as patches in the Buildroot or Yocto recipe. Debian's 0.8 packages are a tested reference for which patches apply cleanly. Our firmware diffing and patch analysis guide shows how to confirm a backport landed in the shipped binary.
  3. Mitigate in configuration meanwhileRestrict interfaces and block port 5353 on untrusted networks for CVE-2017-6519 and the CNAME crashes, set disable-user-service-publishing=yes for the D-Bus assertions, and keep wide-area off for CVE-2024-52615 and CVE-2024-52616.
  4. Restart on failureRun avahi-daemon under a supervisor that restarts it, so a crash costs seconds of discovery instead of a support call. That limits the impact of a denial of service; it does not fix it.
  5. Release with evidenceRegenerate the SBOM, record the Avahi version and the backported patch list, and state per CVE whether the product is affected. Our CSAF and VEX for device makers guide shows the format, for example "not affected" for CVE-2024-52616 when wide-area is compiled out or disabled.

Worked example: how would you triage Avahi in a camera image?

Synthetic example, not a real product. A camera maker scans a Yocto-based image whose manifest shows avahi 0.8 with no vendor patches.

  1. Confirm the baselibavahi-core contains the 0.8 version string, and the recipe's patch list is empty.
  2. Match the CVEsCVE-2017-6519 does not apply (fixed in 0.8). CVE-2021-3468 and CVE-2023-38469 to 38473 apply (fixed in v0.9-rc1), as do CVE-2025-68468 and CVE-2026-24401 (fixed in v0.9-rc3). CVE-2024-52616 applies on version but depends on enable-wide-area.
  3. Check configuration and exposureThe shipped configuration sets no allow-interfaces, the camera's main application browses for recorders through avahi-daemon, and its own service is announced from a static service file, so no local application needs to publish over D-Bus. Wide-area is set to no.
  4. RankThe CNAME crashes come first: anyone on the camera's network can stop discovery, and cameras often sit on shared or guest segments. The D-Bus assertions follow, mitigated by disable-user-service-publishing=yes. CVE-2024-52616 is recorded as not affected by configuration.
  5. FixShip an update with the v0.9-rc3 fix commits backported or a move to v0.9-rc5, restrict allow-interfaces to the camera's LAN interface, drop port 5353 on other interfaces, then rebuild, re-scan and publish a VEX statement per CVE.

How does AI-assisted analysis help with Avahi findings?

SoftWSP's automated scan unpacks a firmware image, catalogs the software packages in the extracted file system with their versions (the inventory our firmware SBOM guide describes), and matches each one against CVE records, CISA's KEV catalog and FIRST's EPSS scores. This is deterministic code: the code scores, and AI narrates. You can inspect the resulting CycloneDX file in the free SBOM viewer.

Version matching follows NVD's records and inherits their limits, which matter for Avahi. NVD files these CVEs under cpe:2.3:a:avahi:avahi, OpenWrt declares cpe:/a:avahi:avahi for its packages, and recent records list 0.9-rc1 and 0.9-rc2 as separate affected versions, so a release-candidate build has to be compared with care. A version also cannot show which interfaces the daemon listens on, whether D-Bus publishing is open or whether a distribution or vendor backport closed the bug. AI's part is the written explanation of what matched, citing the component, version and record behind each sentence. An engineer reviews every finding before the report is delivered, and that review is where interface settings, firewall rules, reflector use and any backport become a documented "affected" or "fixed" decision. Kernel-level networking flaws on the same devices are covered in our Linux kernel CVEs in embedded devices guide.

For where discovery services fit in a full assessment, see the OWASP Firmware Security Testing Methodology guide, our KEV and EPSS prioritisation guide for weighing the scores above, or device & firmware security services to have a specific image reviewed.

This guide draws on Avahi's GitHub releases and security advisories, the avahi-daemon.conf manual and default configuration, RFC 6762, the OpenWrt avahi and umdns package files, Apple's mDNSResponder distribution, the Debian Security Tracker, NVD, CISA's KEV catalog (version 2026.11.13) and FIRST EPSS data dated 2026-11-14, all read directly on 2026-11-15.

Put this guide to work

Frequently asked questions

Which Avahi vulnerabilities matter most for embedded devices?

As of 2026-11-14, start with the flaws an attacker can reach over the network. CVE-2017-6519 made avahi-daemon answer IPv6 unicast queries from addresses that are not on the local link, which allows traffic amplification and information leakage over UDP port 5353; Avahi 0.8 fixed it. CVE-2025-68468 and CVE-2026-24401 let anyone on the link crash avahi-daemon with crafted CNAME announcements, fixed in v0.9-rc3. CVE-2021-3468 and the CVE-2023-38469 to CVE-2023-38473 reachable assertions need local access, through the daemon's Unix socket or D-Bus, and were fixed in v0.9-rc1. None of them is in CISA's KEV catalog (version 2026.11.13).

How do I find the Avahi version in a firmware image?

Look for avahi-daemon, its libraries (libavahi-core, libavahi-common, libavahi-client) and its configuration in the avahi directory under etc in the unpacked file system. Package records give the version most reliably: OpenWrt's opkg database lists avahi-dbus-daemon or avahi-nodbus-daemon, and Buildroot and Yocto record avahi in their build manifests. On a device you control, avahi-daemon --version prints it. Then check the vendor patch list, because distributions backport fixes into 0.8 packages.

Why is mDNS on a WAN or untrusted VLAN an exposure?

Multicast DNS is designed for one local link. RFC 6762 requires a querier to accept only responses that originate from the local link, and a responder that answers queries from other networks becomes an amplifier and an information source: CVE-2017-6519 is exactly that case. On an untrusted VLAN, anyone on the segment can send the crafted announcements behind CVE-2025-68468 and CVE-2026-24401. Published records also reveal host names, services and, if HINFO is on, the operating system and CPU.

How do I harden avahi-daemon on a device?

Disable it if no product feature uses service discovery. Otherwise set allow-interfaces to the LAN interfaces only, keep allow-point-to-point at no, turn off publish-hinfo and publish-workstation, set disable-user-service-publishing to yes if no local application needs to publish services, keep enable-wide-area and enable-reflector at no unless needed, and drop UDP port 5353 on WAN interfaces in the firewall. Use disable-publishing for a device that only needs to browse.

Does SoftWSP find Avahi in my firmware automatically?

SoftWSP's automated scan catalogs the software packages in an unpacked firmware file system with their versions and matches them against CVE records, CISA's KEV catalog and FIRST EPSS scores. AI then writes a plain-language summary that cites the component, version and record behind each finding, and an engineer reviews it. Which interfaces avahi-daemon listens on, whether D-Bus publishing is open to local users, whether the firewall drops port 5353 on the WAN and whether a backport already closes a matched CVE are part of that engineer review, not an automatic verdict.

Sources

  1. Avahi: GitHub releases (0.8 release notes, CVE-2017-6519 fix)
  2. Avahi: GitHub security advisories
  3. Avahi advisory GHSA-mrwc-c54v-3rg8 (CVE-2021-3468)
  4. Avahi advisory GHSA-cp79-r4x9-vf52 (CVE-2025-68468)
  5. Avahi advisory GHSA-h4vp-5m8j-f6w3 (CVE-2026-24401)
  6. Avahi advisory GHSA-r9j3-vjjh-p8vm (CVE-2024-52616)
  7. Avahi advisory GHSA-w65r-6gxh-vhvc (CVE-2026-34933)
  8. Avahi manual source: avahi-daemon.conf(5)
  9. Avahi: default avahi-daemon.conf
  10. IETF RFC 6762: Multicast DNS (section 11, Source Address Check)
  11. OpenWrt packages: avahi package Makefile
  12. OpenWrt packages: avahi default avahi-daemon.conf
  13. OpenWrt: umdns package Makefile
  14. Apple: mDNSResponder open source distribution
  15. Debian Security Tracker: CVE-2017-6519
  16. Debian Security Tracker: CVE-2023-38469
  17. Debian Security Tracker: CVE-2025-68468
  18. NIST NVD: CVE-2017-6519
  19. NIST NVD: CVE-2021-3468
  20. NIST NVD: CVE-2023-38469
  21. NIST NVD: CVE-2025-68468
  22. NIST NVD: CVE-2026-24401
  23. CISA: Known Exploited Vulnerabilities (KEV) Catalog
  24. FIRST: Exploit Prediction Scoring System (EPSS)
Was this guide helpful?

In this cluster

Avahi and mDNS/DNS-SD vulnerabilities in embedded Linux: find the responder and version, rank the CVEs, keep port 5353 on the local link