Source: dbus
Severity: normal
Tags: upstream security
X-Debbugs-Cc: Debian Security Team <[email protected]>
Forwarded: https://gitlab.freedesktop.org/dbus/dbus/-/work_items/589
AISLE in partnership with Red Hat reported a local denial of service
against `dbus-daemon --system`. A local attacker can prevent service
activation (autostarting) by exhausting the maximum number of autostart
requests that the bus allows, an arbitrary limit that was, ironically,
implemented as an attempt to defend against denial of service. The
attacker could use this to prevent system services from starting.
A mitigation is that the attack is only feasible if one of the
installed, trusted system services (/usr/share/dbus-1/system-services)
does not start successfully and reasonably promptly, or if one of those
services can be made to exit by the attacker so that they can waste
activation "slots" on starting it again.
Another mitigation is that a sysadmin can either start important
services manually (or via their systemd units or init scripts, if any),
or arrange for them to be autostarted before the attacker can block
autostarting.
All versions are believed to be vulnerable. Unlike the other recent
LLM-reported denial of service issues, there is no fixed version
available yet for this one. Unfortunately, it was accidentally disclosed
to the public when the reporter proposed a LLM-generated solution (which
appears to be incorrect and does not prevent the denial of service).
I have added a SECURITY.md to dbus to try to take denial-of-service by
resource exhaustion out-of-scope, because I do not believe that it is
actually feasible to prevent all denial-of-service vulnerabilities of
that type. I realise that this will not affect what the community
demands from dbus' maintainers, so perhaps this will still be treated as
a vulnerability, I don't know.
A CVE ID has been requested from Red Hat's CNA but is not yet available.
smcv