Skip to content

Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors

Rust

fix(wrapper): detect rootless and SELinux under podman

Rodrigo Palharese56bdfae0a7e
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 75fd2f49..69189362 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## [Unreleased]
+### Fixed
+- The `bin/ai-memory` wrapper now detects rootless mode and SELinux under
+ podman, so host-file commands stop failing with `Permission denied (os error
+ 13)` on podman-based distros. Both the `-u 0:0` remap and `--security-opt
+ label=disable` were decided from `docker info --format
+ '{{.SecurityOptions}}'`, a Docker-only field that podman cannot evaluate;
+ the swallowed error left both gates off exactly where they were needed.
+ Podman's `.Host.Security.*` keys are consulted when that probe comes back
+ empty. `bootstrap` is now covered as well: it only reads host files, but an
+ unmapped UID blocks reads just as hard, and it degraded silently to "no
+ `.git` found" before dying (#385).
+
## [1.25.0] - 2026-08-07
### Added
diff --git a/bin/ai-memory b/bin/ai-memory
index dd36d621..8cb2d062 100755
--- a/bin/ai-memory
+++ b/bin/ai-memory
@@ -440,6 +440,23 @@ case "${WRAPPER_SUBCOMMAND}" in
WRITES_HOST_FILES=1
;;
esac
+# bootstrap only *reads* host files — the repo bind-mounted at /work and the
+# `.ai-memory.toml` markers under $HOME — but an unmapped UID and a confined
+# SELinux label block reads exactly as hard as writes. The failure is
+# two-stage and misleading: bootstrap first degrades silently to
+# "no .git found at /work; bootstrapping from README/docs/rules only", then
+# dies with "Permission denied (os error 13)" when it reaches a file it
+# cannot skip. So readers need the same treatment as writers.
+READS_HOST_FILES=0
+case "${WRAPPER_SUBCOMMAND}" in
+ bootstrap)
+ READS_HOST_FILES=1
+ ;;
+esac
+TOUCHES_HOST_FILES=0
+if [ "${WRITES_HOST_FILES}" -eq 1 ] || [ "${READS_HOST_FILES}" -eq 1 ]; then
+ TOUCHES_HOST_FILES=1
+fi
# Rootless Docker runs the whole daemon inside its own user namespace, where
# container UID 0 maps back to the invoking host user, but any *non-zero*
@@ -451,15 +468,35 @@ esac
# staging dir, $PWD/CLAUDE.md, skill directories, …) fails with
# EACCES/ENOENT. Running as UID 0 instead hits rootlesskit's primary
# mapping and lands as the real host user. Only do this for commands that
-# write to a host bind mount: other commands only touch the /data named
+# touch a host bind mount: other commands only touch the /data named
# volume, which isn't host-visible and doesn't have this problem, so leave
# their UID mapping alone.
DOCKER_SECURITY_OPTIONS=$("${DOCKER}" info --format '{{.SecurityOptions}}' 2>/dev/null || true)
ROOTLESS_DOCKER=0
+CONTAINER_SELINUX=0
if printf '%s\n' "${DOCKER_SECURITY_OPTIONS}" | grep -q 'name=rootless'; then
ROOTLESS_DOCKER=1
fi
-if [ "${WRITES_HOST_FILES}" -eq 1 ] && [ "${ROOTLESS_DOCKER}" -eq 1 ]; then
+if printf '%s\n' "${DOCKER_SECURITY_OPTIONS}" | grep -q 'name=selinux'; then
+ CONTAINER_SELINUX=1
+fi
+# `.SecurityOptions` is a Docker-only field. Podman — including through the
+# podman-docker `docker` shim — fails the template with "can't evaluate field
+# SecurityOptions in type system.infoReport" and exits 125, which the `|| true`
+# above swallows into an empty string. Both gates then read as "not rootless,
+# no SELinux" and rootless podman runs the helper as an unmapped subordinate
+# UID under a confined label, so every host-file access dies with
+# "Permission denied (os error 13)". Podman exposes the same two facts under
+# different keys, so ask it directly when the Docker probe came back empty.
+if [ -z "${DOCKER_SECURITY_OPTIONS}" ]; then
+ if [ "$("${DOCKER}" info --format '{{.Host.Security.Rootless}}' 2>/dev/null || true)" = "true" ]; then
+ ROOTLESS_DOCKER=1
+ fi
+ if [ "$("${DOCKER}" info --format '{{.Host.Security.SELinuxEnabled}}' 2>/dev/null || true)" = "true" ]; then
+ CONTAINER_SELINUX=1
+ fi
+fi
+if [ "${TOUCHES_HOST_FILES}" -eq 1 ] && [ "${ROOTLESS_DOCKER}" -eq 1 ]; then
USER_ARGS=(-u 0:0)
fi
@@ -472,14 +509,14 @@ case "$(uname -s 2>/dev/null || true)" in
# SELinux blocks the helper's container label from writing normal home
# labels even when its uid/gid match the host user. Relabeling the entire
# $HOME bind with :z/:Z is unsafe, so relax label confinement only for the
- # short-lived, trusted helper commands that actually write host files.
+ # short-lived, trusted helper commands that actually touch host files.
SELINUX_MODE=$(getenforce 2>/dev/null || true)
if [ -z "${SELINUX_MODE}" ] && [ -r /sys/fs/selinux/enforce ]; then
SELINUX_MODE=$(cat /sys/fs/selinux/enforce 2>/dev/null || true)
fi
- if [ "${WRITES_HOST_FILES}" -eq 1 ] \
+ if [ "${TOUCHES_HOST_FILES}" -eq 1 ] \
&& { [ "${SELINUX_MODE}" = "Enforcing" ] || [ "${SELINUX_MODE}" = "enforcing" ] || [ "${SELINUX_MODE}" = "1" ]; } \
- && printf '%s\n' "${DOCKER_SECURITY_OPTIONS}" | grep -q 'name=selinux'; then
+ && [ "${CONTAINER_SELINUX}" -eq 1 ]; then
SELINUX_ARGS=(--security-opt label=disable)
fi
;;