fleet/pkg/secure/secure_test.go
Bash Bandicoot 4fcbb57d23
Fix orbit crash loop on incorrect file permissions (#40887)
## Summary

- `checkPermFile` in `pkg/secure/secure.go` now self-heals incorrect
file permissions via `os.Chmod` instead of returning a fatal error
- Fixes orbit crash-looping indefinitely when
`/opt/orbit/updates-metadata.json` has mode 755 instead of the expected
600

## Problem

Orbit refuses to start when `updates-metadata.json` has wrong
permissions (e.g. 755 instead of 600), entering an infinite restart loop
(`systemd` restart counter observed at 3447+). The manual workaround is
`chmod 600 /opt/orbit/updates-metadata.json`, but the root cause — an
external process changing file permissions — is intermittent and hard to
track.

The `checkPermFile` function in `pkg/secure/secure.go` was designed as a
security check, but its behavior of fatally erroring on any permission
mismatch causes a denial-of-service on the legitimate user. For
comparison, `checkPermPath` (the directory equivalent) already tolerates
permissions that are less permissive than expected.

## Fix

When `checkPermFile` detects a permission mismatch, it now attempts
`os.Chmod` to correct the permissions before proceeding. It only returns
an error if the chmod itself fails (e.g. insufficient privileges). This
preserves the security intent — files end up with correct permissions —
while making orbit resilient to external permission drift.

## Test plan

- [ ] `go test ./pkg/secure/ -v -run TestOpenFile` — verifies
self-healing behavior
- [ ] `go test ./pkg/secure/ -v -run TestMkdirAll` — unchanged, verifies
directory checks still work
- [ ] Manual: create `/opt/orbit/updates-metadata.json` with mode 755,
start orbit, confirm it self-heals and starts normally

---------

Co-authored-by: Bash Bandicoot <bash-bandicoot@users.noreply.github.com>
2026-03-06 17:41:31 -03:00

59 lines
1.7 KiB
Go

package secure
import (
"fmt"
"os"
"path/filepath"
"testing"
"github.com/stretchr/testify/require"
)
func TestMkdirAll(t *testing.T) {
tmpDir := t.TempDir()
basePath := filepath.Join(tmpDir, "test")
require.NoError(t, os.MkdirAll(basePath, 0700))
err := MkdirAll(basePath, 0677)
require.Error(t, err)
expectedErr := fmt.Sprintf(
"Path %s already exists with mode 20000000700 instead of the expected %o", basePath, 0677^os.ModeDir)
require.Equal(t, expectedErr, err.Error())
err = MkdirAll(filepath.Join(basePath, "test2", "test3"), 0677)
require.Error(t, err)
require.Equal(t, expectedErr, err.Error())
err = MkdirAll(filepath.Join(basePath, "test2", "test3"), 0700)
require.NoError(t, err)
}
func TestOpenFile(t *testing.T) {
tmpDir := t.TempDir()
basePath := filepath.Join(tmpDir, "test")
require.NoError(t, os.MkdirAll(basePath, 0755))
filePath := filepath.Join(basePath, "file1")
_, err := OpenFile(filePath, os.O_CREATE|os.O_WRONLY, 0677)
require.Error(t, err)
expectedErr := fmt.Sprintf(
"Path %s already exists with mode 20000000755 instead of the expected %o", basePath, 0677^os.ModeDir)
require.Equal(t, expectedErr, err.Error())
fd, err := OpenFile(filePath, os.O_CREATE|os.O_WRONLY, 0755)
require.NoError(t, err)
require.NotNil(t, fd)
require.NoError(t, fd.Close())
// Opening with a different perm should self-heal via chmod rather than error.
fd, err = OpenFile(filePath, os.O_CREATE|os.O_WRONLY, 0600)
require.NoError(t, err)
fd.Close()
info, err := os.Stat(filePath)
require.NoError(t, err)
require.Equal(t, os.FileMode(0600), info.Mode())
// Re-open with the now-correct mode still works.
fd, err = OpenFile(filePath, os.O_CREATE|os.O_WRONLY, 0600)
require.NoError(t, err)
fd.Close()
}