tree ca173653d8e2defc0938a95278fb876c60616a20
parent 508c7f6fba194bb8161fff1636d3880999758200
author Megan Dahl <megan@schedmd.com> 1731689462 -0700
committer Marshall Garey <marshall@schedmd.com> 1757112871 +0000

Prevent invalid read if task count is more than arbitrary nodelist

If the job is using arbitrary distribution and the number of node names in
the arbitrary node list is less then the task count there will be an
invalid memory read in _at_tpn_limit() when laying out the tasks. The task
count needs to be equal to the number of node names in the nodelist. If the
task count is not equal return an ESLURM_BAD_TASK_COUNT error.

This also has to prevent the loss of all jobs when upgrading to 25.11 if
any job would hit this error. Do to the new restriction, the jobs being
loaded from versions 25.05 and lower will still be allowed regardless of
the task count. However, arbitrary_tasks_np will be reallocated to prevent
an invalid read from happening.

Changelog: slurmctld - Prevent an invalid read and a possible crash by
 rejecting any arbitrary distribution jobs that do not specify a task count
 equal to the number of node names in their node list. This does not affect
 srun, salloc, or sbatch if -n is not used since they set the default task
 count.
Ticket: 21444