I won't apply this version, please send v2.

On Mon, 16 Nov 2026 14:45:45 +0400, Mirian Shilakadze wrote:
> +     if (task->mm && ve_is_super(old_ve) && ve->state != VE_STATE_STARTING)
>               task->mm->vps_dumpable = VD_VE_ENTER_TASK;

Item 1 of the cover letter reply: keep a host task hidden when it
moves from one VE into another.

Why: setns() into a VE needs CAP_SYS_ADMIN only, so a host task in
starting VE A can setns() into running VE B. old_ve is then A, not
ve0, so the mark is left alone. With patch 3 the mm stays
VD_VE_SETUP_TASK, B is not pseudosuper, ve_vps_dumpable() returns
VD_PTRACE_COREDUMP, and B's root can ptrace the task and read its
fds. Before the series it got VD_VE_ENTER_TASK on entering B. Only a
hand-made flow does this, vzctl and CRIU always enter from ve0.

> A
> container task gets the mark too when it setns()es into its own VE, and
> the mark goes to the mm, which it may share with another task.

Item 2 of the cover letter reply: say why a container task must not
get the mark, and that a task entering its nested VE is no longer
hidden there.

Why: the reader sees what changes but not what was wrong with it, and
the nested VE case changes behaviour without a word.

-- 
Konstantin Khorenko,
Virtuozzo Linux Kernel Team

_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel

Reply via email to