The NTP vulnerabilities in embedded firmware that matter most as of 2026-11-15 are servers that answer queries they should not, led by the ntpd monlist amplification flaw (CVE-2013-5211), and remote crashes in NTPsec and BusyBox ntpd. Chrony's pidfile flaw (CVE-2020-14367) needs local access and ranks lower. The bigger risk is often the clock itself: a device with the wrong time fails or skips TLS certificate checks. Identify the time client and version, make it client-only, then upgrade or backport.
Key takeaways
- Five families of time client cover almost every device, and they differ in size, features and attack surface.
- Identify the family first, because a CVE for "ntpd" means different things in ntp.org, NTPsec and BusyBox.
- Six CVEs cover what a firmware scan of time services usually turns up.
- Monlist is still a firmware problem because firmware outlives the fix by years.
Where does each NTP implementation ship in embedded devices?
Five families of time client cover almost every device, and they differ in size, features and attack surface. Which one an image carries usually follows from the build system and the operating system, not from a deliberate security choice.
| Implementation | Typical devices | Server by default? | NTS support |
|---|---|---|---|
| ntp.org reference ntpd (NTP Classic) | Older Linux gateways, industrial PCs, appliances built from desktop distributions | Answers clients unless restrict says otherwise; mode 6 and 7 queries on older builds |
No |
| NTPsec | Newer Debian-based appliances, timing and GNSS boxes | Depends on restrict lines |
Yes |
| chrony | Yocto and Buildroot images, RHEL- and Debian-derived appliances | No: client-only unless an allow directive is set |
Yes, since 4.0 |
| BusyBox ntpd | OpenWrt routers and most small embedded Linux images | No: server only with -l |
No |
| systemd-timesyncd | systemd-based images (Yocto with systemd, Debian-derived) | No, client only | No |
| lwIP SNTP and other RTOS clients | Microcontroller devices: ESP-IDF, FreeRTOS, other RTOS stacks | No, client only | No |
On routers, BusyBox is the default. OpenWrt's NTP configuration page shows its system.ntp section with enable_server set to 0 by default, and notes that setting it to 1 makes the router answer as a time server, with busybox-ntpd listening on UDP 123. Our BusyBox vulnerabilities guide covers the other applets in that binary; this guide covers only the time service and the clock behind it.
systemd-timesyncd is deliberately small. Its manual page says it "implements SNTP only", steps the clock for large offsets and slews it for small ones, and saves the time to disk on each synchronization so the clock moves roughly forward across reboots on boards without a battery-backed RTC. On microcontrollers the client is usually part of the network stack: Espressif's ESP-IDF system time guide explains that the lwIP SNTP library sets the time through settimeofday() or adjtime() when a server responds, re-synchronizes every hour by default, and uses the RFC 2030 convention so its timestamps stay valid until 2104. Our embedded TCP/IP stack vulnerabilities guide covers the stacks those SNTP clients sit on.
How do you identify the NTP implementation and version in a firmware image?
Identify the family first, because a CVE for "ntpd" means different things in ntp.org, NTPsec and BusyBox.
List the binariesIn the unpacked root file system, ntpdwith bothntpqandntpdcpoints to the ntp.org reference build; NTPsec shipsntpdandntpqwithoutntpdc, plus anntpPython package for its tools.chronydwithchronycis chrony. Anntpdthat is a symlink or hard link tobusyboxis the BusyBox applet.systemd-timesyncdlives in the systemd library directory.Read the versionSearch the daemon binary for its version string, such as 4.2.8p15for ntp.org or1.2.2for NTPsec, or run its version option on a device you control. For BusyBox, read the banner in thebusyboxbinary and check whether the ntpd applet and its server feature were compiled in.Read the package recordsOpenWrt's opkg database, Buildroot's manifest and Yocto's build history record the package name and version, which is more reliable than strings. Our embedded Linux build systems and SBOM guide shows where each one keeps them. Find out how it startsRead the init script, systemd unit or OpenWrt UCI configuration. For BusyBox, the question is whether the command line includes -lor-I. For chrony, look forallowdirectives. For ntpd and NTPsec, read everyrestrictline inntp.conf.Check RTOS images for an SNTP clientIn a monolithic microcontroller image, look for the configured NTP server names (for example pool.ntp.orgor a vendor host) and SNTP library symbols near them. These clients do not answer queries, so the questions are which servers they trust and what they do with a large time jump.
Which NTP CVEs matter most for device makers?
Six CVEs cover what a firmware scan of time services usually turns up. Fixed versions and dates come from each project's own release notes and security pages; scores are NVD's unless marked as the CNA's, EPSS values are FIRST's for 2026-11-14, and KEV status is from CISA's catalog version 2026.11.13, in which none of these appear.
| CVE | Implementation | What it is | Affected | Fixed upstream | CVSS | EPSS |
|---|---|---|---|---|---|---|
| CVE-2013-5211 | ntp.org ntpd | monlist (mode 7) replies used for traffic amplification | Before 4.2.7p26 | 4.2.7p26 (2010-04-24) | 5.0 (v2) | 0.975 |
| CVE-2019-6443 | NTPsec | Stack buffer over-read in ctl_getitem while answering a control (mode 6) request |
Before 1.1.3 | 1.1.3 (2019-01-13) | 9.1 | 0.669 |
| CVE-2016-6301 | BusyBox ntpd | A forged NTP packet starts a reply loop, burning CPU and bandwidth | Before the fix, with the server running | 1.25.1 (2016-10-07) | 7.5 | 0.090 |
| CVE-2026-102486 | NTPsec | Out-of-bounds read in NTS extension parsing crashes ntpd | 1.1.4 to 1.2.5, as NTS server | 1.2.6 (2026-10-20) | 7.5 (CNA) | 0.005 |
| CVE-2026-102483 | NTPsec | Too many NTS Cookie Placeholders make ntpd exit | 1.1.4 to 1.2.5, as NTS server | 1.2.6 (2026-10-20) | 7.5 (CNA) | 0.004 |
| CVE-2020-14367 | chrony | chronyd follows a symlink when writing its pidfile as root | Before 3.5.1 | 3.5.1 (2020-08-20) | 6.0 | 0.005 |
NVD status on 2026-11-15 for both 2026 NTPsec records: Awaiting Analysis, no NVD score; the 7.5 is GitLab's score as the CNA. NTPsec's release notes describe the first as an unchecked nonce length that lets "a remote attacker holding an NTS cookie" crash ntpd, and the second as a remotely triggerable exit in the NTS server. The same release fixed two reference-clock driver bugs (CVE-2026-102484 in the JJY driver and CVE-2026-102485 in the Motorola Oncore driver), which matter only on timing devices with those receivers attached. An earlier NTPsec fix in 1.2.2a (2023-08-02) closed CVE-2023-4012, a crash when a server without NTS configured receives an NTS client request.
For the reference implementation, ntp.org's security notice lists 4.2.8p18 (2024-05-25) as the newest release and records no security fixes in it or in 4.2.8p17. The last security release was 4.2.8p16 on 2023-05-30, which fixed four low-severity bugs, among them the mstolfp() bounds check and the out-of-bounds write in the Palisade reference-clock parser. A device on 4.2.8p15 or older therefore has open findings, even if most need a malicious server, a GPS receiver or a local ntpq user to reach.
Why is monlist (CVE-2013-5211) still a firmware problem?
Monlist is still a firmware problem because firmware outlives the fix by years. ntp.org's notice for bug 1532 says unrestricted access to monlist in ntpd before 4.2.7p26 let attackers cause "traffic amplification" with forged REQ_MON_GETLIST requests, "as exploited in the wild in December 2013". The fix shipped on 2010-04-24, more than three years before that wave of attacks, which shows how long old builds stay deployed. Its EPSS of 0.975 on 2026-11-14 reflects that scanning for it has never stopped.
The mitigation for builds that cannot be upgraded is one configuration line. ntp.org advises either noquery in the default restrictions, which blocks all status queries, or disable monitor, which turns off monlist alone. RFC 8633 (BCP 223) goes further: control (mode 6) responses "are much larger than the corresponding queries" and can be abused in high-bandwidth DDoS attacks, so public-facing servers should block control queries from outside their organization, and its appendix for ntpd gives restrict default -4 nomodify notrap nopeer noquery and the matching IPv6 line as the baseline. The same reasoning applies to NTPsec's mode 6 handler, which is where CVE-2019-6443 sat. A device that has no reason to serve time should not answer either mode at all.
What do the chrony, BusyBox and NTPsec advisories mean for a device?
They mean very different things, so read the precondition before the score.
chrony. chrony's news page explains that CVE-2020-14367 applies when the pidfile sits in a directory the chrony user can write to, such as /var/run/chrony, the default since chrony 3.4. An attacker who already controls the chrony account can plant a symlink there, and chronyd, starting as root, writes its process ID through it, causing data loss or denial of service. On a device with no interactive accounts that is a hardening finding, not a remote risk. The newest release on 2026-11-15 is chrony 4.9, published on 27 August 2026.
BusyBox ntpd. CVE-2016-6301 only matters when the applet runs as a server. The BusyBox news page describes the 1.25.1 fix as making ntpd "check type of incoming packet, respond only to client requests", and the current ntpd.c source still carries that check, answering only client and symmetric-active packets, and documents that it does not support mode 6 queries from ntpq. The Debian tracker for CVE-2016-6301 rates it unimportant for Debian's build because the NTP server is not enabled in its configuration, which is exactly the check to make in a vendor image: is -l on the command line, or enable_server set on OpenWrt?
NTPsec. The 2019 bugs, CVE-2019-6442 to CVE-2019-6445, sit in the control (mode 6) and remote configuration paths and were fixed in 1.1.3. The 2026 ones sit in the NTS server. A device that only acts as an NTS client, or that does not answer mode 6 from the network, is outside the main path of both, but should still be upgraded because the code ships.
Why does wrong time break TLS certificate checks, updates and logs?
Wrong time breaks every security check that compares a date with "now". Certificate validation under RFC 5280 requires the current time to lie between a certificate's notBefore and notAfter fields. RFC 8915, the NTS specification, spells out the problem this creates for time clients themselves: a client that starts without a correctly set clock has difficulty checking a certificate's validity period, and it may be operationally necessary to accept a certificate that appears expired or outside its validity window.
On devices this shows up in four ways:
- TLS failures at boot. A board without a battery-backed RTC can start in 1970 or at its build date. Every certificate then looks out of range, so cloud connections, MQTT over TLS and update downloads fail until time is set. Our OpenSSL vulnerabilities in embedded firmware guide covers the library doing those checks.
- Disabled checks. The worse outcome is a developer working around the boot failure by turning certificate time checks off for good. That accepts expired certificates, including ones whose keys have leaked.
- Update rollback and freeze. Signed update metadata often carries expiry dates so a device cannot be fed an old, vulnerable image indefinitely. A clock an attacker can move backwards makes stale metadata look fresh. Our firmware OTA update security guide covers anti-rollback and metadata expiry.
- Untrustworthy logs. Incident response, vulnerability reports and forensic timelines depend on timestamps. A device whose clock jumps or starts at 1970 produces logs that cannot be correlated with the network or the cloud.
RFC 8915 lists mitigations that suit firmware well: check whether some trusted software has already set the time before validating strictly, allow strict validation always on boards with a battery-backed clock, and once synchronized, periodically save the current time to persistent storage and refuse any certificate whose notAfter is earlier than the last saved time. systemd-timesyncd's saved clock file and chrony's nocerttimecheck directive are implementations of that idea. The chrony.conf manual describes nocerttimecheck as disabling certificate time checks for a set number of clock updates, for computers that start with the wrong time, and warns that it "has important security implications and should be used only as a last resort", preferably with a minimal set of trusted certificates; its default of 0 keeps the checks on.
What do NTS (RFC 8915) and authenticated NTP add?
Plain NTP and SNTP replies are unauthenticated UDP, so anyone on the path can shift a device's clock. Two mechanisms close that gap.
Network Time Security. RFC 8915 (September 2020) defines NTS as two parts: NTS Key Establishment over TLS on TCP port 4460, followed by NTP packets protected with AEAD extension fields. The client authenticates the server with an ordinary X.509 certificate, so no keys have to be provisioned per device. chrony added NTS in 4.0 on 7 October 2020; its manual describes the nts option on a server or pool line and the ntsdumpdir directive, which saves NTS cookies so a restart does not need a new key-establishment handshake. NTPsec supports NTS, and its 1.2.5 release on 2026-07-30 added the nts flag to pool lines. The cost on a device is a TLS library, a CA store and a few extra round trips at boot.
Symmetric keys. Older authenticated NTP uses a key shared between client and server. BusyBox ntpd accepts an ntp.keys-compatible file with -k and a key number per peer when built with its authentication feature, chrony uses the key option, and NTPsec 1.2.6 added HMAC mode for shared keys. Shared keys suit a closed fleet talking to the maker's own time servers, but a key baked identically into every unit is a hard-coded secret.
The small clients have neither. systemd-timesyncd's manual describes SNTP only, and its timesyncd.conf reference documents server lists, fallback servers and root-distance limits but no NTS or key options as of 2026-11-15. lwIP SNTP on RTOS devices is plain SNTP too. For those, rely on several independent sources, a sane limit on how far one reply can move the clock, and a saved monotonic time, or move to chrony where the platform allows.
How should you harden NTP on a device?
Most devices only need to be clients. Make that explicit in each implementation and back it with the firewall.
# chrony: client-only with NTS
server time.example.net iburst nts # maker's or a public NTS server
pool pool.example.net iburst nts maxsources 3
ntsdumpdir /var/lib/chrony # keep NTS cookies across restarts
makestep 1 3 # step only in the first 3 updates
maxchange 1000 1 2 # refuse later huge jumps
cmdport 0 # no chronyc UDP port
# no "allow" line: chronyd serves no clients
# ntp.org ntpd / NTPsec when a server is not needed
restrict default -4 nomodify notrap nopeer noquery
restrict default -6 nomodify notrap nopeer noquery
restrict source nomodify notrap noquery
disable monitor # ntp.org ntpd: no monlist
- chrony. The manual states that without an
allowdirective "chronyd operates purely as an NTP client".cmdport 0stops chronyd from opening its UDP command port, while the local Unix socket for chronyc stays.makesteplimits clock steps to the first updates after start, andmaxchangeignores, and eventually exits on, offsets larger than a threshold after start. - ntp.org ntpd and NTPsec. Use RFC 8633's
restrictbaseline so the daemon answers no mode 6 or 7 queries and accepts no remote configuration, and adddisable monitoron ntp.org builds. If the device must serve time to a LAN, allow that subnet explicitly and keepnoqueryon everything else. - BusyBox ntpd. Start it with
-ppeers only, without-lor-I. On OpenWrt keepenable_serverat 0; if the product needs a LAN time server, set itsinterfaceoption so it binds to the LAN only. - systemd-timesyncd and RTOS SNTP. These never answer queries. Configure several servers you trust, prefer the maker's own or a well-run pool, and handle the "first sync" state explicitly in application code instead of trusting whatever time the board booted with.
- Firewall. Drop inbound UDP 123 on WAN interfaces regardless of daemon settings; a stateful firewall still lets replies to the device's own requests back in. If NTS is used, allow outbound TCP 4460. Never forward UDP 123 to the device.
- Least privilege. Run the daemon as its own unprivileged user after start-up, keep its pidfile and state directories owned by root where the daemon allows it, which is the class of problem CVE-2020-14367 shows, and drop the binaries you do not use, such as ntpdc and ntpq on a production image.
How does a device maker patch or backport NTP?
Upgrade to a maintained releaseAs of 2026-11-15 that means chrony 4.9, NTPsec 1.2.6 or ntp.org 4.2.8p18, and a current BusyBox. chrony is usually the easiest move for an embedded Linux image: it is client-only by default, supports NTS and has a short security history. Backport when you cannotEach fix in this guide is small. BusyBox's fix for CVE-2016-6301 is a single upstream commit, and distribution trackers show how others backported. The Debian tracker for CVE-2020-14367 lists buster fixed in chrony 3.4-4+deb10u1, below the upstream 3.5.1, which is exactly the case where a plain version match reports a false positive. Our firmware diffing and patch analysis guide shows how to confirm a backport landed in the shipped binary. Mitigate in configuration meanwhile noqueryanddisable monitorfor monlist and the mode 6 bugs, no-lfor BusyBox ntpd, no NTS server role for the 2026 NTPsec CVEs, and a WAN firewall rule for all of them.Fix the clock at boot, not the checksSave a known-good time to flash, refuse to go backwards past it, use the RTC if the board has one, and never ship firmware with certificate time checks disabled. If a time bootstrap exception is unavoidable, limit it to the first update, as chrony's nocerttimecheck 1does.Release with evidenceRegenerate the SBOM, record the time client, its version and backported patches, 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-2016-6301 on a build whose BusyBox ntpd never runs with -l.
Worked example: how would you triage time sync in a gateway image?
Synthetic example, not a real product. An industrial gateway maker scans a Buildroot image whose package manifest lists BusyBox 1.24.2 and ntp 4.2.6p5, and whose cloud agent connects over TLS.
Identify both clientsThe busyboxbinary has the ntpd applet compiled in, but the init scripts start/usr/sbin/ntpdfrom the ntp package, withntpqandntpdcnext to it. The BusyBox applet is never started.Match the CVEsCVE-2013-5211 applies: 4.2.6p5 is older than 4.2.7p26, and ntp.confhas norestrictlines, so the gateway answers monlist from anywhere. CVE-2016-6301 matches BusyBox 1.24.2 by version, but the applet does not run, so it is recorded as not affected with that justification.Check exposureA port scan of a test unit shows UDP 123 open on the cellular WAN interface, and ntpdc -c monlistanswers from outside.Check the clockThe board has no RTC battery, and the cloud agent's TLS client has certificate time checks turned off so it can connect before NTP has synchronized. FixReplace ntp with chrony 4.9 in client-only mode with NTS, add a WAN rule dropping inbound UDP 123, save the time to flash on each sync and refuse to start the cloud agent's TLS connection with time checks off, then rebuild, re-scan and publish a VEX statement per CVE.
Step 2 shows why version matching alone misleads in both directions: the BusyBox match is a false positive, and the ntp match is real but only becomes urgent once the configuration and the open WAN port are known.
How does AI-assisted analysis help with NTP 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. Our KEV and EPSS prioritisation guide explains how those two signals are weighed, and you can inspect the resulting CycloneDX file in the free SBOM viewer.
Version matching follows NVD's records and inherits their limits, which are significant for time services. NVD files the reference implementation, NTPsec, chrony and BusyBox under different product names, and an "ntpd" binary has to be attributed to the right one before any match means anything. A version also cannot show whether a daemon runs as a server, answers mode 6 or 7 queries, uses NTS, or whether TLS clients on the device check certificate dates. 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 start-up flags, restrict and allow lines, firewall rules and any vendor or distribution backport become a documented "affected" or "not affected" decision.
For the C library underneath these daemons, see our glibc, uClibc and musl vulnerabilities guide; for other network daemons, the OpenSSH and dnsmasq guides; or device & firmware security services to have a specific image reviewed.
This guide draws on ntp.org's security notice and its notice for bug 1532, NTPsec's release notes, chrony's news page and chrony.conf manual, BusyBox's news page and ntpd source, the systemd-timesyncd and timesyncd.conf manuals, Espressif's ESP-IDF system time guide, OpenWrt's NTP configuration page, RFC 8915, RFC 8633, 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 NTP vulnerabilities matter most for embedded devices?
As of 2026-11-15, start with any device that answers NTP queries from the internet. CVE-2013-5211, the ntpd monlist amplification flaw fixed in ntp 4.2.7p26 in 2010, still has an EPSS score of 0.975 because old ntpd builds keep answering it. Next come daemon crashes a remote packet can trigger: CVE-2019-6443 in NTPsec before 1.1.3, CVE-2016-6301 in BusyBox ntpd before 1.25.1, and CVE-2026-102486 and CVE-2026-102483 in NTPsec 1.1.4 to 1.2.5 running as an NTS server, fixed in 1.2.6 on 2026-10-20. Chrony's CVE-2020-14367 needs a compromised chrony account, so it ranks lower on most devices.
How do I find which NTP implementation and version a firmware image uses?
List the time daemons in the unpacked file system: ntpd with ntpq and ntpdc points to the ntp.org reference build, ntpd with ntpq but no ntpdc and an ntpsec Python module points to NTPsec, chronyd with chronyc to chrony, an ntpd link to the busybox binary to BusyBox, and systemd-timesyncd inside the systemd tree. Read the version from the binary's strings or from the opkg, Buildroot or Yocto package record, then read the start-up script and configuration to see whether the daemon also runs as a server.
Why does a wrong clock break TLS on a device?
A TLS client checks that the current time lies between the certificate's notBefore and notAfter dates. A device that boots without a battery-backed clock can start in 1970 or at its build date, so valid certificates look out of range and the client either fails or, worse, the firmware turns time checks off. A clock pushed forward by a spoofed NTP reply can make a certificate look expired; one pushed back can make an expired or revoked certificate look current. Signed update metadata with expiry dates and log timestamps depend on the same clock.
Should an embedded device use NTS?
Use NTS where the time client supports it and the device has a CA store and enough flash for TLS. Chrony has supported NTS since version 4.0 in October 2020, and NTPsec supports it too. systemd-timesyncd implements SNTP only and its configuration reference documents no NTS option, and the lwIP SNTP client used on many RTOS devices sends plain SNTP. On those clients, use several sources, sanity limits on large steps and a monotonic saved time instead.
Does SoftWSP find the NTP daemon 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. Whether the daemon runs as a server, answers mode 6 or 7 queries, uses NTS or keys, and whether a vendor backport already closes a matched CVE is decided in that engineer review, not by the scan.
Sources
- ntp.org: Security Notice (known vulnerabilities by release version)
- ntp.org: NTP Bug 1532, DRDoS/Amplification Attack using ntpdc monlist command (CVE-2013-5211)
- NTPsec: NEWS (release notes and security fixes)
- chrony: News (release history and security fixes)
- chrony 4.9 manual: chrony.conf(5)
- BusyBox: News (1.25.1 release notes)
- BusyBox source: networking/ntpd.c (GitHub mirror)
- systemd manual: systemd-timesyncd.service(8)
- systemd manual: timesyncd.conf(5)
- ESP-IDF Programming Guide: System Time (SNTP Time Synchronization)
- OpenWrt documentation: NTP client / NTP server configuration
- RFC 8915: Network Time Security for the Network Time Protocol
- RFC 8633: Network Time Protocol Best Current Practices (BCP 223)
- Debian Security Tracker: CVE-2020-14367
- Debian Security Tracker: CVE-2016-6301
- NIST NVD: CVE-2013-5211
- NIST NVD: CVE-2016-6301
- NIST NVD: CVE-2019-6443
- NIST NVD: CVE-2020-14367
- NIST NVD: CVE-2026-102486
- NIST NVD: CVE-2026-102483
- CISA: Known Exploited Vulnerabilities (KEV) Catalog
About this guide
SoftWSP Firmware Security Desk
Covers the Firmware vulnerabilities & fixes cluster: 125 guides.

In this cluster
Questions about this page?
