Description
User Story
I run OpenShell sandboxes with the Podman and Kubernetes drivers and I work on the sandbox isolation core. This came out of the review of #5212. I found it by reading the code and have not hit it in production.
Problem Statement
crates/openshell-sandbox/src/linux/child_seccomp.rs denies ioprio_set for every target. A workload cannot change its own I/O priority, so ionice -p $$ and ionice -c3 <command> fail with EPERM, although they work in a plain container. It is the same gap #5185 covers for sched_set*, prlimit64, and setpriority, and #5212 leaves it out of scope.
Impact / Why This Matters
Build and batch tooling that lowers its own I/O priority (ionice, psutil ionice()) fails only under OpenShell. The only workaround is to drop the call from the workload.
Acceptance Criteria
-
ioprio_set(IOPRIO_WHO_PROCESS, 0, ...)and the caller's own PID or sibling TIDs are allowed. - Other targets,
IOPRIO_WHO_PGRP, andIOPRIO_WHO_USERstay denied withEPERM. - The syscall joins the mediated scheduling table from fix(sandbox): make mediated signal, scheduling, and limit calls behave natively #5212 (its selector is argument 0, like
setpriority), so the notified set, the native pid 0 path, the broker dispatch, and the static deny stay derived from one place. - Unit and e2e tests cover the allowed and denied cases.
Reproduction Steps
Not run; from the code. Inside a sandbox, ionice -c3 -p $$ fails with Operation not permitted.
Environment
- OpenShell:
mainatfea144c45 - OS: Fedora 45, Linux 7.2.15
- Deployment: Podman driver (local gateway)
Activity