SchedMD Slurm 21.08.x through 20.11.x has Incorrect Access Control that leads to Escalation of Privileges and code execution.

SchedMD Slurm 21.08.x through 20.11.x has Incorrect Access Control that leads to Information Disclosure.

SchedMD Slurm before 20.02.7 and 20.03.x through 20.11.x before 20.11.7 allows remote code execution as SlurmUser because use of a PrologSlurmctld or EpilogSlurmctld script leads to environment mishandling.

Slurm before 19.05.8 and 20.x before 20.02.6 exposes Sensitive Information to an Unauthorized Actor because xauth for X11 magic cookies is affected by a race condition in a read operation on the /proc filesystem.

Slurm before 19.05.8 and 20.x before 20.02.6 has an RPC Buffer Overflow in the PMIx MPI plugin.

SchedMD Slurm before 18.08.9 and 19.x before 19.05.5 executes srun --uid with incorrect privileges.

SchedMD Slurm before 18.08.9 and 19.x before 19.05.5 has weak slurmdbd.conf permissions.

SchedMD Slurm before 17.02.11 and 17.1x.x before 17.11.7 mishandles user names (aka user_name fields) and group ids (aka gid fields).

SchedMD Slurm before 17.02.10 and 17.11.x before 17.11.5 allows SQL Injection attacks against SlurmDBD.

Insecure SPANK environment variable handling exists in SchedMD Slurm before 16.05.11, 17.x before 17.02.9, and 17.11.x before 17.11.0rc2, allowing privilege escalation to root during Prolog or Epilog execution.

The _prolog_error function in slurmd/req.c in Slurm before 15.08.13, 16.x before 16.05.7, and 17.x before 17.02.0-pre4 has a vulnerability in how the slurmd daemon informs users of a Prolog failure on a compute node. That vulnerability could allow a user to assume control of an arbitrary file on the system. Any exploitation of this is dependent on the user being able to cause or anticipate the failure (non-zero return code) of a Prolog script that their job would run on. This issue affects all Slurm versions from 0.6.0 (September 2005) to present. Workarounds to prevent exploitation of this are to either disable your Prolog script, or modify it such that it always returns 0 ("success") and adjust it to set the node as down using scontrol instead of relying on the slurmd to handle that automatically. If you do not have a Prolog set you are unaffected by this issue.