About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
The kernel line audit: backlog limit exceeded means the kernel's queue of waiting audit events was over its limit and an event was discarded. The lost counter in auditctl -s rises each time. Three fixes are to raise the limit with -b, to confirm auditd is running and draining the queue, and to cut audit rules that fire on every system call.
What the error means
The Linux audit system keeps outstanding events in a kernel queue until auditd reads them. The auditctl(8) manual describes -b backlog as the "max number (limit) of outstanding audit buffers allowed (Kernel Default=64)" and adds "If all buffers are full, the failure flag is consulted by the kernel for action."
The message text below comes from the Linux kernel source code, not from a documentation page: kernel/audit.c at tag v6.12. When the queue is longer than the limit, the process that generated the event can wait, up to a wait time (--backlog_wait_time, kernel default 60*HZ), for room. If the wait is exhausted, the source shows three messages. Each is a format string; the part that varies is the %d or %u value, and the audit: prefix is added by the kernel's pr_fmt setting:
audit_backlog=%d > audit_backlog_limit=%d
audit_lost=%u audit_rate_limit=%u audit_backlog_limit=%u
backlog limit exceeded
The first line is printed, rate limited, whatever the failure mode. The second is printed by the routine that counts the lost event, also in every mode, though it is rate limited. Only the third goes through the failure mode, set with -f. The manual says "Set failure mode 0=silent 1=printk 2=panic" and "The default value is 1." In the source, mode 0 prints nothing for the third line, mode 1 prints it as an error, and mode 2 calls panic(). In every mode the lost counter increases. A host in mode 0 therefore still shows the first two lines, but not backlog limit exceeded.
The auditctl -s status shows the counters. The manual says "The lost entry will tell you how many event records that have been discarded due to the kernel audit queue overflowing. The backlog field tells how many event records are currently queued waiting for auditd to read them." A pid of 0 "indicates that the audit daemon is not running".
This is a different queue from the one set by q_depth in /etc/audit/auditd.conf, which is the size of auditd's internal dispatcher queue (default 2000). Raising q_depth does not change the kernel limit.
Likely causes
| Cause | What you see | How to confirm |
|---|---|---|
| Kernel limit too small for the event rate | lost rises in bursts; backlog touches the limit |
auditctl -s: the backlog limit is low (kernel default is 64) |
| auditd not running or stopped reading | pid is 0 and backlog stays high |
auditctl -s shows pid 0; systemctl status auditd |
| Too many or too broad syscall rules | Constant high event volume, slow system calls | auditctl -l lists many syscall rules without arch or dir |
| Log rotation or slow disk delaying auditd | Backlog grows during rotation | auditd.conf has num_logs above 1; backlog rises at rotation |
Rules locked with -e 2 |
auditctl -b is refused |
auditctl -s shows the enabled flag as 2 |
Prove the cause before changing anything
Everything here only reads. Run as root.
1. Read the status and keep it.
auditctl -s
The manual says -s reports "the in-kernel values that can be set by -e, -f, -r, and -b options". Note the pid, lost, backlog and backlog limit values, and the failure mode.
2. Find the kernel messages.
journalctl -k | grep -E 'audit_backlog=|audit_lost=|backlog limit exceeded'
The journalctl manual documents -k for kernel messages. A good result is no output. A bad result is any of the three lines. Search for all three, because on a host in failure mode 0 only the first two appear.
3. Is auditd running?
systemctl status auditd
The systemctl manual says status shows runtime state "followed by most recent log data from the journal".
4. List the rules.
auditctl -l
The manual says -l lists "all rules 1 per line". The audit.rules manual warns that "The more rules, the bigger the performance hit" and says an arch field should be supplied in each rule, because without it every system call is evaluated.
5. Are the rules built from files that will be reloaded?
augenrules --check
The augenrules manual says --check tests "if rules have changed and need updating without overwriting audit.rules". The same manual says the last -b directive is placed on the second line of the merged file.
Fix
Raise the kernel backlog limit
First print and keep the current status, including the current limit:
auditctl -s > /root/auditctl-s-2026-11-22.txt
Risk: the next command changes a running kernel audit setting at once. A higher limit allows more outstanding audit buffers in the kernel (the manual gives no memory figure per buffer), and if the configuration was locked with -e 2 the manual says "Any attempt to change the configuration in this mode will be audited and denied", so only a reboot changes it.
Set a larger limit. The manual gives no recommended value, so raise it in steps and watch the status. For example, to 8192:
auditctl -b 8192
To undo it: read the old limit from the file you saved and set it again.
auditctl -b <old-limit-from-saved-file>
auditctl -s
A limit set this way is not stored in a file. The audit.rules manual says the rules file is loaded "whenever the daemon is started", so a value that must survive a restart belongs in the rules, which the next section covers.
Make the limit permanent
First save the current rules file. On a system that uses augenrules, the rules live in /etc/audit/rules.d/; the path below is an example. Save the file you will edit (the one that holds the -b line, or the one where you will add it), and restore that same file in the back-out:
cp -p /etc/audit/rules.d/audit.rules /etc/audit/rules.d/audit.rules.2026-11-22
The augenrules manual says files in that directory that do not end in .rules are ignored, so the saved copy with a date suffix is not merged into the rules.
If you create a new .rules file instead of editing an existing one (if you edit an existing file, skip this paragraph), there is nothing to save. If you create a new file instead, first run ls -l on the new name (the name below is an example) and stop if it shows a file:
ls -l /etc/audit/rules.d/99-backlog.rules
Disruptive step: the next edit followed by augenrules --load reloads the audit rules in the kernel. Auditing is reconfigured at that moment, and a mistake in any .rules file in that directory changes which rules are loaded.
Edit the -b line in that file to the new limit, for example -b 8192 (add the line if there is none; in a new file, create it with the -b line alone). The augenrules manual says --check tests "if rules have changed and need updating without overwriting audit.rules"; the manual does not say that it checks syntax. Then see whether the rules changed, rebuild and load them:
augenrules --check
augenrules --load
auditctl -s
To back out: put the saved file back under its original name only if the saved copy exists, rebuild and load the rules, and compare with the saved status.
ls -l /etc/audit/rules.d/audit.rules.2026-11-22 && cp -p /etc/audit/rules.d/audit.rules.2026-11-22 /etc/audit/rules.d/audit.rules
augenrules --load
auditctl -s
To undo it when you created a new file (see "If you create a new file instead" above), and only if the ls -l there reported No such file or directory before you created it, delete that one file with its own command, then reload:
rm /etc/audit/rules.d/99-backlog.rules
augenrules --load
If the rules are locked with -e 2, a reload cannot change them; the new file is read at the next boot.
Cut the event volume
The auditctl manual's performance tips say that with ten syscall rules "every program on your system will delay during a syscall while the audit system evaluates each rule", and recommend combining rules whose filter, action, key and fields are identical, for example -S openat -S truncate in one rule, and using file system auditing such as -F dir=/etc where practical. No command is given for rewriting rules here, because the right rewrite depends on your rule set; it is a separate change to the same rules files.
auditd not running
If auditctl -s showed pid 0, journalctl -u auditd shows the daemon's recent messages. Find why it stopped and correct that cause. No command to start the daemon is given here, because starting it is a separate service change that depends on your distribution.
Example (illustrative): db01 has 14 syscall rules and no -b line, so the kernel default of 64 applies. At 09:00 auditctl -s shows lost 4821 and backlog 64, and the kernel log has audit: audit_backlog= and audit: backlog limit exceeded lines. The engineer saves auditctl -s output, runs auditctl -b 8192, and reads the status again: backlog 0. At 17:00 lost is still 4821, a rise of 0 over 8 hours, so no new events were discarded. The engineer then saves the rules file, sets -b 8192 in it, runs augenrules --check to see whether the rules changed, and loads them.
Confirm the fix
auditctl -sshows the new backlog limit and a pid other than 0.- The
lostvalue from the saved status is unchanged after a busy period. Subtract the saved value from the current one: a result of 0 means no new loss. journalctl -k | grep -E 'audit_backlog=|audit_lost=|backlog limit exceeded'shows no new lines. A host in failure mode 0 prints nobacklog limit exceededline even when events are lost, so rely on the first two lines and on thelostcounter.- After a reboot,
auditctl -sshows the limit from the rules file.
What to watch afterwards
- Record
lostandbacklogon a schedule. A risinglostcount means events are being discarded and the audit trail has gaps. - If you increase
num_logsinauditd.conf, the manual says "you may need to adjust the kernel backlog setting upwards since it takes more time to rotate the files." - Treat gaps as a security-record problem as well as a performance one. A lost event cannot be recovered.
Read next
Related articles you may want next.
- TroubleshootingJournal logs filling the disk: how to limit /var/log/journal and vacuum it safely/var/log/journal is huge? Check journalctl --disk-usage, trim with --vacuum-size, and cap it with SystemMaxUse. What is lost, and what vacuum cannot touch.Read article
- TroubleshootingLinux OOM Killer on a VM: Find the Cause, Set Memory LimitsA Linux VM killed a service for lack of memory? Read the kernel or systemd-oomd record, find which limit applied, then set limits from measured use.Read article
- HostingProvider administrator access in managed hosting: who can log in, how it is logged, and break-glass accountsWhat administrator access a managed hosting provider holds to a customer's servers, what Australian and Microsoft guidance says about controlling and logging privileged access, how emergency break-glass accounts should work, and the questions to ask before signing.Read article
Risks and limits
- These commands have not been run or lab-tested.
- Failure mode 2 makes the kernel panic when the backlog limit is exceeded. The auditctl manual says "Secure environments will probably want to set this to 2", so check
-fin your rules before raising event volume. - The manual gives no recommended
-bvalue and no memory cost per buffer. Raise the limit in steps and measure. - Kernel message wording comes from the source code at tag v6.12 and may differ on other kernel versions.
Sources
- auditctl(8) — man7.org, Red Hat Sep 2023, page rendered 2026-09-09, accessed 22 November 2026
- auditd.conf(5) — man7.org, Red Hat June 2025, page rendered 2026-09-09, accessed 22 November 2026
- audit.rules(7) — man7.org, Red Hat Sep 2023, page rendered 2026-09-09, accessed 22 November 2026
- augenrules(8) — Debian Manpages, bookworm, source updated 2023-02-09, accessed 22 November 2026
- kernel/audit.c at v6.12 — Linux kernel source code, tag v6.12, no date shown, accessed 22 November 2026
- systemctl(1) — Debian Manpages, bookworm, source updated 2025-03-06, accessed 22 November 2026
- journalctl(1) — Debian Manpages, bookworm, source updated 2025-03-06, accessed 22 November 2026
This article was originally published on .