banger/internal/daemon/doctor_test.go

package daemon

import (
	"context"
	"errors"
	"os"
	"path/filepath"
	"strings"
	"testing"

	"banger/internal/firecracker"
	"banger/internal/model"
	"banger/internal/paths"
	"banger/internal/system"
)

// permissiveRunner satisfies system.CommandRunner by returning a
// configurable response for every call. Doctor tests don't care about
// the exact ip/iptables commands run — they care that the aggregated
// report surfaces each feature check correctly, so a one-size runner
// keeps the test prelude short.
type permissiveRunner struct {
	out []byte
	err error
}

func (r *permissiveRunner) Run(_ context.Context, _ string, _ ...string) ([]byte, error) {
	return r.out, r.err
}

func (r *permissiveRunner) RunSudo(_ context.Context, _ ...string) ([]byte, error) {
	return r.out, r.err
}

// buildDoctorDaemon stands up a Daemon the way doctorReport expects:
// fake PATH with every tool the preflights look for, fake firecracker
// + vsock companion binaries, fake vsock host device file, and a
// permissive runner that claims a default-route via eth0 so NAT's
// defaultUplink call succeeds. Returns the wired *Daemon.
func buildDoctorDaemon(t *testing.T) *Daemon {
	t.Helper()
	binDir := t.TempDir()
	for _, name := range []string{
		"sudo", "ip", "dmsetup", "losetup", "blockdev", "truncate", "pgrep",
		"chown", "chmod", "kill", "e2cp", "e2rm", "debugfs",
		"iptables", "sysctl", "mkfs.ext4", "mount", "umount", "cp",
	} {
		writeFakeExecutable(t, filepath.Join(binDir, name))
	}
	t.Setenv("PATH", binDir)

	firecrackerBin := filepath.Join(t.TempDir(), "firecracker")
	if err := os.WriteFile(firecrackerBin, []byte("#!/bin/sh\nexit 0\n"), 0o755); err != nil {
		t.Fatalf("write firecracker: %v", err)
	}
	vsockHelper := filepath.Join(t.TempDir(), "banger-vsock-agent")
	if err := os.WriteFile(vsockHelper, []byte("#!/bin/sh\nexit 0\n"), 0o755); err != nil {
		t.Fatalf("write vsock helper: %v", err)
	}
	t.Setenv("BANGER_VSOCK_AGENT_BIN", vsockHelper)

	sshKey := filepath.Join(t.TempDir(), "id_ed25519")
	if err := os.WriteFile(sshKey, []byte("unused"), 0o600); err != nil {
		t.Fatalf("write ssh key: %v", err)
	}

	vsockHostDevice := filepath.Join(t.TempDir(), "vhost-vsock")
	if err := os.WriteFile(vsockHostDevice, []byte{}, 0o644); err != nil {
		t.Fatalf("write vsock host device: %v", err)
	}

	runner := &permissiveRunner{out: []byte("default via 10.0.0.1 dev eth0 proto static\n")}

	d := &Daemon{
		layout: paths.Layout{
			ConfigDir: t.TempDir(),
			StateDir:  t.TempDir(),
			DBPath:    filepath.Join(t.TempDir(), "state.db"),
		},
		config: model.DaemonConfig{
			FirecrackerBin:    firecrackerBin,
			SSHKeyPath:        sshKey,
			BridgeName:        model.DefaultBridgeName,
			BridgeIP:          model.DefaultBridgeIP,
			StatsPollInterval: model.DefaultStatsPollInterval,
		},
		runner: runner,
	}
	wireServices(d)
	d.vm.vsockHostDevice = vsockHostDevice
	// HostNetwork defaults its own runner to the one on the struct, but
	// wireServices only copies the Daemon's runner if d.net is nil
	// before that call — in this test we constructed d.net implicitly,
	// so belt-and-braces the permissive runner onto HostNetwork too.
	d.net.runner = runner
	return d
}

// findCheck returns the first CheckResult with the given name, or nil
// if no such check was emitted. The test helper rather than a method
// on Report so the field scope stays tight.
func findCheck(report system.Report, name string) *system.CheckResult {
	for i := range report.Checks {
		if report.Checks[i].Name == name {
			return &report.Checks[i]
		}
	}
	return nil
}

// TestDoctorReport_NonSystemModeEmitsSecurityWarn pins the non-
// system-mode branch: when install.toml is absent the security
// posture check must surface a warn that points at the dev-mode
// caveat in docs/privileges.md. A pass row in this mode would
// imply guarantees the install isn't actually providing. Drives
// the seam variant so the test is independent of whether the host
// happens to have /etc/banger/install.toml.
func TestDoctorReport_NonSystemModeEmitsSecurityWarn(t *testing.T) {
	d := buildDoctorDaemon(t)
	report := system.Report{}
	missingInstall := filepath.Join(t.TempDir(), "install.toml")
	d.addSecurityPostureChecksAt(context.Background(), &report, missingInstall, t.TempDir())

	check := findCheck(report, "security posture")
	if check == nil {
		t.Fatal("security posture check missing from report")
	}
	if check.Status != system.CheckStatusWarn {
		t.Fatalf("security posture status = %q, want warn", check.Status)
	}
	joined := strings.Join(check.Details, " ")
	if !strings.Contains(joined, "outside the system install") {
		t.Fatalf("warn details = %q, want mention of non-system mode", joined)
	}
	if !strings.Contains(joined, "docs/privileges.md") {
		t.Fatalf("warn details = %q, want pointer to docs/privileges.md", joined)
	}
}

func TestAddSocketPermsCheckRejectsWrongMode(t *testing.T) {
	socketPath := filepath.Join(t.TempDir(), "fake.sock")
	if err := os.WriteFile(socketPath, []byte{}, 0o644); err != nil {
		t.Fatalf("write fake socket: %v", err)
	}
	report := system.Report{}
	addSocketPermsCheck(&report, "test socket", socketPath, os.Getuid(), 0o600)
	check := findCheck(report, "test socket")
	if check == nil {
		t.Fatal("expected test socket check")
	}
	if check.Status != system.CheckStatusFail {
		t.Fatalf("status = %q, want fail when mode is 0644 vs 0600 expected", check.Status)
	}
	joined := strings.Join(check.Details, " ")
	if !strings.Contains(joined, "mode is") {
		t.Fatalf("details = %q, want mode-mismatch message", joined)
	}
}

func TestAddSocketPermsCheckPassesWhenModeAndOwnerMatch(t *testing.T) {
	socketPath := filepath.Join(t.TempDir(), "fake.sock")
	if err := os.WriteFile(socketPath, []byte{}, 0o600); err != nil {
		t.Fatalf("write fake socket: %v", err)
	}
	report := system.Report{}
	addSocketPermsCheck(&report, "test socket", socketPath, os.Getuid(), 0o600)
	check := findCheck(report, "test socket")
	if check == nil {
		t.Fatal("expected test socket check")
	}
	if check.Status != system.CheckStatusPass {
		t.Fatalf("status = %q, want pass when mode + uid match; details = %v", check.Status, check.Details)
	}
}

func TestAddUnitHardeningCheckFlagsMissingDirective(t *testing.T) {
	unitPath := filepath.Join(t.TempDir(), "bangerd.service")
	if err := os.WriteFile(unitPath, []byte("[Service]\nUser=alice\nProtectSystem=strict\n"), 0o644); err != nil {
		t.Fatalf("write unit: %v", err)
	}
	report := system.Report{}
	addUnitHardeningCheck(&report, "unit hardening", unitPath, []string{"User=alice", "NoNewPrivileges=yes", "ProtectSystem=strict"})
	check := findCheck(report, "unit hardening")
	if check == nil {
		t.Fatal("expected unit hardening check")
	}
	if check.Status != system.CheckStatusFail {
		t.Fatalf("status = %q, want fail when NoNewPrivileges is missing", check.Status)
	}
	joined := strings.Join(check.Details, " ")
	if !strings.Contains(joined, "NoNewPrivileges=yes") {
		t.Fatalf("details = %q, want mention of the missing directive", joined)
	}
}

func TestAddUnitHardeningCheckPassesWhenAllPresent(t *testing.T) {
	unitPath := filepath.Join(t.TempDir(), "bangerd-root.service")
	body := "[Service]\nNoNewPrivileges=yes\nProtectSystem=strict\nProtectHome=yes\nCapabilityBoundingSet=CAP_CHOWN\n"
	if err := os.WriteFile(unitPath, []byte(body), 0o644); err != nil {
		t.Fatalf("write unit: %v", err)
	}
	report := system.Report{}
	addUnitHardeningCheck(&report, "unit hardening", unitPath, []string{"NoNewPrivileges=yes", "ProtectSystem=strict", "CapabilityBoundingSet="})
	check := findCheck(report, "unit hardening")
	if check == nil {
		t.Fatal("expected unit hardening check")
	}
	if check.Status != system.CheckStatusPass {
		t.Fatalf("status = %q, want pass when every directive is present; details = %v", check.Status, check.Details)
	}
}

func TestAddExecutableOwnershipCheckRejectsSymlink(t *testing.T) {
	dir := t.TempDir()
	real := filepath.Join(dir, "fc")
	if err := os.WriteFile(real, []byte{}, 0o755); err != nil {
		t.Fatalf("write fc: %v", err)
	}
	link := filepath.Join(dir, "fc-symlink")
	if err := os.Symlink(real, link); err != nil {
		t.Fatalf("symlink: %v", err)
	}
	report := system.Report{}
	addExecutableOwnershipCheck(&report, "fc binary", link)
	check := findCheck(report, "fc binary")
	if check == nil {
		t.Fatal("expected fc binary check")
	}
	if check.Status != system.CheckStatusFail {
		t.Fatalf("status = %q, want fail for symlinked binary", check.Status)
	}
	joined := strings.Join(check.Details, " ")
	if !strings.Contains(joined, "symlink") {
		t.Fatalf("details = %q, want symlink rejection message", joined)
	}
}

func TestAddExecutableOwnershipCheckRejectsGroupWritable(t *testing.T) {
	if os.Getuid() == 0 {
		t.Skip("test runs as root; can't construct a non-root-owned check target meaningfully")
	}
	path := filepath.Join(t.TempDir(), "fc")
	if err := os.WriteFile(path, []byte{}, 0o775); err != nil {
		t.Fatalf("write fc: %v", err)
	}
	report := system.Report{}
	addExecutableOwnershipCheck(&report, "fc binary", path)
	check := findCheck(report, "fc binary")
	if check == nil {
		t.Fatal("expected fc binary check")
	}
	if check.Status != system.CheckStatusFail {
		t.Fatalf("status = %q, want fail when binary is group/world writable", check.Status)
	}
}

// TestDoctorReport_SystemModeRunsAllSecurityChecks pins the system-mode
// branch end-to-end: with a fake install.toml + fake systemd dir it
// must contribute every security row (services, sockets, unit
// hardening, fc ownership). Statuses themselves vary because we can't
// easily fake root-owned files in a test, but every check name must
// appear so a future refactor can't silently drop one.
func TestDoctorReport_SystemModeRunsAllSecurityChecks(t *testing.T) {
	d := buildDoctorDaemon(t)

	installDir := t.TempDir()
	installPath := filepath.Join(installDir, "install.toml")
	if err := os.WriteFile(installPath, []byte("owner_user = \"alice\"\nowner_uid = 1000\nowner_gid = 1000\nowner_home = \"/home/alice\"\ninstalled_at = 2026-04-28T00:00:00Z\n"), 0o644); err != nil {
		t.Fatalf("write install.toml: %v", err)
	}
	systemdDir := t.TempDir()
	for _, svc := range []string{"bangerd.service", "bangerd-root.service"} {
		if err := os.WriteFile(filepath.Join(systemdDir, svc), []byte(""), 0o644); err != nil {
			t.Fatalf("write fake unit %s: %v", svc, err)
		}
	}

	report := system.Report{}
	d.addSecurityPostureChecksAt(context.Background(), &report, installPath, systemdDir)

	for _, name := range []string{
		"helper service",
		"owner daemon service",
		"helper socket",
		"daemon socket",
		"helper unit hardening",
		"daemon unit hardening",
		"firecracker binary ownership",
	} {
		if findCheck(report, name) == nil {
			t.Errorf("system-mode security check %q missing from report", name)
		}
	}
	if findCheck(report, "security posture") != nil {
		t.Error("system mode should NOT emit the non-system-mode warn")
	}
}

func TestDoctorReport_StoreErrorSurfacesAsFail(t *testing.T) {
	d := buildDoctorDaemon(t)
	report := d.doctorReport(context.Background(), errors.New("simulated open failure"), false)

	check := findCheck(report, "state store")
	if check == nil {
		t.Fatal("state store check missing from report")
	}
	if check.Status != system.CheckStatusFail {
		t.Fatalf("state store status = %q, want fail (store error should surface)", check.Status)
	}
	joined := strings.Join(check.Details, " ")
	if !strings.Contains(joined, "simulated open failure") {
		t.Fatalf("state store details = %q, want the storeErr message", joined)
	}
}

func TestDoctorReport_StoreMissingSurfacesAsPassForFreshInstall(t *testing.T) {
	d := buildDoctorDaemon(t)
	// Fresh install: the DB file simply doesn't exist yet. doctor must
	// not treat that as a failure — nothing's broken, the first daemon
	// start will create the file. The status message should say so,
	// so a user running `banger doctor` before ever booting a VM
	// doesn't see a scary red check.
	report := d.doctorReport(context.Background(), nil, true)

	check := findCheck(report, "state store")
	if check == nil {
		t.Fatal("state store check missing from report")
	}
	if check.Status != system.CheckStatusPass {
		t.Fatalf("state store status = %q, want pass for a missing DB on fresh install", check.Status)
	}
	joined := strings.Join(check.Details, " ")
	if !strings.Contains(joined, "will be created") {
		t.Fatalf("state store details = %q, want mention of 'will be created' so users know this is expected", joined)
	}
}

func TestDoctorReport_StoreSuccessSurfacesAsPass(t *testing.T) {
	d := buildDoctorDaemon(t)
	report := d.doctorReport(context.Background(), nil, false)

	check := findCheck(report, "state store")
	if check == nil {
		t.Fatal("state store check missing from report")
	}
	if check.Status != system.CheckStatusPass {
		t.Fatalf("state store status = %q, want pass", check.Status)
	}
}

func TestDoctorReport_MissingFirecrackerFailsFirecrackerBinaryCheck(t *testing.T) {
	d := buildDoctorDaemon(t)
	// Point at a nonexistent path. Note: the doctor's PATH lookup
	// looks for the basename, so use an absolute non-existent path
	// (that's the configured-path branch — bare-name lookups would
	// fall through to the test-fixture binDir which DOES contain a
	// fake `firecracker`).
	d.config.FirecrackerBin = filepath.Join(t.TempDir(), "does-not-exist")

	report := d.doctorReport(context.Background(), nil, false)
	check := findCheck(report, "firecracker binary")
	if check == nil {
		t.Fatal("firecracker binary check missing from report")
	}
	if check.Status != system.CheckStatusFail {
		t.Fatalf("firecracker binary status = %q, want fail when binary missing", check.Status)
	}
	joined := strings.Join(check.Details, " ")
	if !strings.Contains(joined, "firecracker-microvm/firecracker/releases") {
		t.Fatalf("missing-binary report should include the upstream URL; got %q", joined)
	}
}

// TestFirecrackerInstallHintDispatchesByDistro pins the per-distro
// install command guess. Pinned IDs are the ones banger is willing to
// suggest a concrete command for; everything else gets only the
// upstream URL.
func TestFirecrackerInstallHintDispatchesByDistro(t *testing.T) {
	t.Parallel()
	for _, tc := range []struct {
		name     string
		release  string
		wantSub  string
		wantNone bool
	}{
		{name: "debian", release: "ID=debian\nVERSION_CODENAME=bookworm\n", wantSub: "apt install firecracker"},
		{name: "ubuntu_id_like_debian", release: "ID=ubuntu\nID_LIKE=debian\n", wantSub: "apt install firecracker"},
		{name: "arch", release: "ID=arch\n", wantSub: "paru -S firecracker"},
		{name: "manjaro_via_id_like", release: "ID=manjaro\nID_LIKE=arch\n", wantSub: "paru -S firecracker"},
		{name: "nixos", release: "ID=nixos\n", wantSub: "nixos.firecracker"},
		{name: "fedora_falls_back_to_url", release: "ID=fedora\n", wantNone: true},
		{name: "missing_file", release: "", wantNone: true},
	} {
		tc := tc
		t.Run(tc.name, func(t *testing.T) {
			t.Parallel()
			osPath := filepath.Join(t.TempDir(), "os-release")
			if tc.release != "" {
				if err := os.WriteFile(osPath, []byte(tc.release), 0o644); err != nil {
					t.Fatalf("write os-release: %v", err)
				}
			}
			hints := firecrackerInstallHint(osPath)
			joined := strings.Join(hints, " ")
			if !strings.Contains(joined, "firecracker-microvm/firecracker/releases") {
				t.Fatalf("hints missing upstream URL; got %q", joined)
			}
			if tc.wantNone {
				// Distro-specific hint must NOT be present — only the URL.
				if len(hints) != 1 {
					t.Fatalf("unrecognised distro got distro-specific hint(s); want only the URL line, got %v", hints)
				}
				return
			}
			if !strings.Contains(joined, tc.wantSub) {
				t.Fatalf("hints %q do not contain expected substring %q", joined, tc.wantSub)
			}
			if len(hints) < 2 {
				t.Fatalf("expected distro hint + URL; got only %v", hints)
			}
		})
	}
}

// firecrackerVersionRunner is a CommandRunner that actually executes
// firecracker --version (via system.Runner) but stubs everything else
// with the permissive default. The doctor uses d.runner for the
// firecracker version query AND for several other checks; this tiny
// dispatcher lets us run a real script for one command without
// rewiring the rest.
type firecrackerVersionRunner struct {
	real    system.Runner
	canned  []byte
	bin     string
}

func (r *firecrackerVersionRunner) Run(ctx context.Context, name string, args ...string) ([]byte, error) {
	if name == r.bin {
		return r.real.Run(ctx, name, args...)
	}
	return r.canned, nil
}

func (r *firecrackerVersionRunner) RunSudo(_ context.Context, _ ...string) ([]byte, error) {
	return r.canned, nil
}

// stubFirecrackerVersion replaces the test daemon's firecracker
// stub with a script that prints the requested version line, then
// swaps d.runner for one that actually executes the script when the
// firecracker path is queried. Returns the resulting daemon ready
// for doctorReport.
func stubFirecrackerVersion(t *testing.T, d *Daemon, version string) {
	t.Helper()
	if err := os.WriteFile(d.config.FirecrackerBin, []byte("#!/bin/sh\necho 'Firecracker v"+version+"'\n"), 0o755); err != nil {
		t.Fatalf("write firecracker stub: %v", err)
	}
	d.runner = &firecrackerVersionRunner{
		real:   system.NewRunner(),
		canned: []byte("default via 10.0.0.1 dev eth0 proto static\n"),
		bin:    d.config.FirecrackerBin,
	}
}

// TestFirecrackerVersionCheckPasses pins the happy path: when the
// configured firecracker reports a tested-range version, doctor
// emits a PASS row.
func TestFirecrackerVersionCheckPasses(t *testing.T) {
	d := buildDoctorDaemon(t)
	stubFirecrackerVersion(t, d, firecracker.KnownTestedVersion)
	report := d.doctorReport(context.Background(), nil, false)
	check := findCheck(report, "firecracker binary")
	if check == nil {
		t.Fatal("firecracker binary check missing from report")
	}
	if check.Status != system.CheckStatusPass {
		t.Fatalf("status = %q, want pass; details=%v", check.Status, check.Details)
	}
}

// TestFirecrackerVersionCheckFailsBelowMin pins the too-old path:
// a binary reporting a version below MinSupportedVersion must FAIL
// with the upgrade hint.
func TestFirecrackerVersionCheckFailsBelowMin(t *testing.T) {
	d := buildDoctorDaemon(t)
	stubFirecrackerVersion(t, d, "0.25.0")
	report := d.doctorReport(context.Background(), nil, false)
	check := findCheck(report, "firecracker binary")
	if check == nil {
		t.Fatal("firecracker binary check missing from report")
	}
	if check.Status != system.CheckStatusFail {
		t.Fatalf("status = %q, want fail for below-min version", check.Status)
	}
}

// TestFirecrackerVersionCheckWarnsAboveTested pins the over-tested
// path: a binary reporting a version newer than KnownTestedVersion
// must WARN — newer firecracker usually works, but it's outside the
// tested window.
func TestFirecrackerVersionCheckWarnsAboveTested(t *testing.T) {
	d := buildDoctorDaemon(t)
	stubFirecrackerVersion(t, d, "99.0.0")
	report := d.doctorReport(context.Background(), nil, false)
	check := findCheck(report, "firecracker binary")
	if check == nil {
		t.Fatal("firecracker binary check missing from report")
	}
	if check.Status != system.CheckStatusWarn {
		t.Fatalf("status = %q, want warn for above-tested version", check.Status)
	}
}

func TestDoctorReport_IncludesEveryDefaultCapability(t *testing.T) {
	d := buildDoctorDaemon(t)
	report := d.doctorReport(context.Background(), nil, false)

	// Every registered capability that implements doctorCapability must
	// contribute a check. Current defaults: work-disk, dns, nat. If a
	// capability is added later it should either extend this list or
	// register its own check name — either way, the assertion makes
	// the contract visible.
	for _, name := range []string{
		"feature /root work disk",
		"feature vm dns",
		"feature nat",
	} {
		if findCheck(report, name) == nil {
			t.Errorf("capability check %q missing from report", name)
		}
	}
}

func TestDoctorReport_EmitsVMDefaultsProvenance(t *testing.T) {
	d := buildDoctorDaemon(t)
	report := d.doctorReport(context.Background(), nil, false)

	check := findCheck(report, "vm defaults")
	if check == nil {
		t.Fatal("vm defaults check missing from report")
	}
	if check.Status != system.CheckStatusPass {
		t.Fatalf("vm defaults status = %q, want pass (this is an always-pass informational check)", check.Status)
	}
	joined := strings.Join(check.Details, "\n")
	for _, needle := range []string{"vcpu:", "memory:", "disk:"} {
		if !strings.Contains(joined, needle) {
			t.Errorf("vm defaults details missing %q; got:\n%s", needle, joined)
		}
	}
}