orca/internal/linux/bootstrap.go

// Package linux implements the SSH-based bootstrap of a generic Linux
// host (Ubuntu/Debian/Alpine) as an orca worker node (REQ-161, P12).
//
// The bootstrap sequence (run via `orca node join --type linux`):
//  1. Generate or load the orca SSH keypair (Ed25519, D-037)
//  2. SSH dial with key auth + TOFU host-key capture (D-035)
//  3. Deploy the orca pubkey to ~orca/.ssh/authorized_keys
//  4. Create the `orca` Linux system user (nologin shell)
//  5. Create the drift-events directory (~orca/drift-events)
//  6. Return the node metadata for the caller to persist
//
// Unlike Proxmox bootstrap, there is NO PVE role, NO sudoers file, and
// NO PVE user — this is a plain Linux worker. Authentication is
// key-based (R-021): the orca SSH key is used for the initial SSH auth
// and pubkey deployment; subsequent orca→worker access uses the same
// key.
//
// All steps are idempotent: re-running the bootstrap on an
// already-configured host is a no-op.
package linux

import (
	"bytes"
	"context"
	"fmt"
	"log/slog"
	"net"
	"os"
	"strings"
	"time"

	"golang.org/x/crypto/ssh"

	"git.cloudinit.dev/coreci/orca/internal/certpaths"
	"git.cloudinit.dev/coreci/orca/internal/proxmox"
	"git.cloudinit.dev/coreci/orca/internal/security"
	"git.cloudinit.dev/coreci/orca/internal/traefik"
)

// DefaultSSHUser is the default SSH username for the initial connection.
const DefaultSSHUser = "root"

// DefaultOrcaUser is the default Linux system user created on the worker.
const DefaultOrcaUser = "orca"

// DefaultSSHPort is the default SSH port.
const DefaultSSHPort = 22

// Options configures a Linux worker bootstrap run.
type Options struct {
	Host               string
	SSHUser            string
	SSHKeyPath         string
	OrcaUser           string
	SSHPort            int
	HostKeyFingerprint string
	Logger             *slog.Logger
}

// Result is the outcome of a successful bootstrap.
type Result struct {
	NodeName           string
	NodeAddress        string
	HostKeyFingerprint string
}

// BootstrapLinux runs the full SSH bootstrap sequence on a remote
// generic Linux host. Returns the node metadata for the caller to
// persist to the registry.
func BootstrapLinux(ctx context.Context, opts Options) (*Result, error) {
	if opts.Host == "" {
		return nil, fmt.Errorf("linux bootstrap: --host is required")
	}
	if opts.SSHKeyPath == "" {
		return nil, fmt.Errorf("linux bootstrap: --ssh-key is required (R-021: no passwords; use --ssh-key or pre-stage the orca key)")
	}
	if opts.SSHUser == "" {
		opts.SSHUser = DefaultSSHUser
	}
	if opts.OrcaUser == "" {
		opts.OrcaUser = DefaultOrcaUser
	}
	if opts.SSHPort == 0 {
		opts.SSHPort = DefaultSSHPort
	}
	if opts.Logger == nil {
		opts.Logger = slog.Default()
	}

	// Step 1: Load the orca SSH keypair.
	privKey, err := os.ReadFile(opts.SSHKeyPath)
	if err != nil {
		return nil, fmt.Errorf("linux bootstrap: read SSH key: %w", err)
	}
	signer, err := ssh.ParsePrivateKey(privKey)
	if err != nil {
		return nil, fmt.Errorf("linux bootstrap: parse SSH key: %w", err)
	}
	pubKey, err := os.ReadFile(certpaths.SSHPubPath())
	if err != nil {
		return nil, fmt.Errorf("linux bootstrap: read orca pubkey: %w", err)
	}
	pubKeyLine := strings.TrimSpace(string(pubKey))

	// Step 2: SSH dial with key auth + TOFU host-key capture.
	sshAddr := net.JoinHostPort(opts.Host, fmt.Sprintf("%d", opts.SSHPort))
	var capturedHostKey ssh.PublicKey
	var hostKeyCallback ssh.HostKeyCallback
	if opts.HostKeyFingerprint != "" {
		hkcb, err := pinnedHostKeyCallback(opts.HostKeyFingerprint, &capturedHostKey)
		if err != nil {
			return nil, fmt.Errorf("linux bootstrap: parse host key fingerprint: %w", err)
		}
		hostKeyCallback = hkcb
	} else {
		// REQ-164 / Phase A3: reuse the tested Proxmox TOFU callback
		// which handles first-connect key capture + known_hosts file
		// creation (create-on-open). The previous inline implementation
		// failed on first connect with a raw KeyError because it never
		// wrote the captured key.
		hkcb, err := proxmox.TOFUHostKeyCallbackPath(certpaths.KnownHostsPath(), sshAddr, &capturedHostKey)
		if err != nil {
			return nil, fmt.Errorf("linux bootstrap: known_hosts: %w", err)
		}
		hostKeyCallback = hkcb
	}

	sshConfig := &ssh.ClientConfig{
		User:            opts.SSHUser,
		Auth:            []ssh.AuthMethod{ssh.PublicKeys(signer)},
		HostKeyCallback: hostKeyCallback,
		Timeout:         30 * time.Second,
	}

	opts.Logger.Info("linux bootstrap: dialing", "addr", sshAddr, "user", opts.SSHUser)
	client, err := ssh.Dial("tcp", sshAddr, sshConfig)
	if err != nil {
		return nil, fmt.Errorf("linux bootstrap: SSH dial %s: %w", sshAddr, err)
	}
	defer client.Close()

	// Step 3: Deploy the orca pubkey to authorized_keys.
	if err := sshExec(client, fmt.Sprintf(
		"mkdir -p ~%s/.ssh && grep -qF '%s' ~%s/.ssh/authorized_keys 2>/dev/null || echo '%s' >> ~%s/.ssh/authorized_keys && chmod 700 ~%s/.ssh && chmod 600 ~%s/.ssh/authorized_keys",
		opts.OrcaUser, pubKeyLine, opts.OrcaUser, pubKeyLine, opts.OrcaUser, opts.OrcaUser, opts.OrcaUser,
	)); err != nil {
		return nil, fmt.Errorf("linux bootstrap: deploy pubkey: %w", err)
	}
	opts.Logger.Info("linux bootstrap: pubkey deployed", "user", opts.OrcaUser)

	// Step 4: Create the orca system user (nologin shell).
	if err := sshExec(client, fmt.Sprintf(
		"id -u %s 2>/dev/null || useradd -r -s /usr/sbin/nologin -d /home/%s -m %s",
		opts.OrcaUser, opts.OrcaUser, opts.OrcaUser,
	)); err != nil {
		return nil, fmt.Errorf("linux bootstrap: create user: %w", err)
	}
	opts.Logger.Info("linux bootstrap: user created", "user", opts.OrcaUser)

	// Step 4d: Ensure orca-traefik podman container on the remote host
	// (REQ-172, R-024). Replaces v0.13 binary+systemd install.
	sshExecFn := func(cmd string) ([]byte, error) {
		session, err := client.NewSession()
		if err != nil {
			return nil, err
		}
		defer session.Close()
		return session.CombinedOutput(cmd)
	}
	ctx, cancelContainer := context.WithTimeout(ctx, 120*time.Second)
	defer cancelContainer()
	if err := traefik.EnsureTraefikContainerRemote(ctx, "", sshExecFn); err != nil {
		opts.Logger.Warn("linux bootstrap: traefik container ensure failed", "err", err)
	}

	// Step 5: Create the drift-events directory.
	if err := sshExec(client, fmt.Sprintf(
		"mkdir -p ~%s/drift-events && chown %s:%s ~%s/drift-events",
		opts.OrcaUser, opts.OrcaUser, opts.OrcaUser, opts.OrcaUser,
	)); err != nil {
		return nil, fmt.Errorf("linux bootstrap: create drift-events dir: %w", err)
	}
	opts.Logger.Info("linux bootstrap: drift-events dir created", "user", opts.OrcaUser)

	// Step 6: Return node metadata.
	hostKeyFP := ""
	if capturedHostKey != nil {
		hostKeyFP = ssh.FingerprintSHA256(capturedHostKey)
	}

	return &Result{
		NodeName:           opts.Host,
		NodeAddress:        fmt.Sprintf("%s:8443", opts.Host),
		HostKeyFingerprint: hostKeyFP,
	}, nil
}

// sshExec runs a command on the remote host and returns an error if
// the exit code is non-zero.
func sshExec(client *ssh.Client, cmd string) error {
	session, err := client.NewSession()
	if err != nil {
		return err
	}
	defer session.Close()
	var stderr bytes.Buffer
	session.Stderr = &stderr
	if err := session.Run(cmd); err != nil {
		return fmt.Errorf("%w: %s", err, strings.TrimSpace(stderr.String()))
	}
	return nil
}

// pinnedHostKeyCallback returns a host key callback that pins to the
// expected fingerprint.
func pinnedHostKeyCallback(expectedSHA256Base64 string, capturedKey *ssh.PublicKey) (ssh.HostKeyCallback, error) {
	if expectedSHA256Base64 == "" {
		return nil, fmt.Errorf("empty fingerprint")
	}
	cb := ssh.HostKeyCallback(func(hostname string, remote net.Addr, key ssh.PublicKey) error {
		got := ssh.FingerprintSHA256(key)
		if got != expectedSHA256Base64 {
			return fmt.Errorf("host key fingerprint mismatch: got %s, want %s", got, expectedSHA256Base64)
		}
		*capturedKey = key
		return nil
	})
	return cb, nil
}

var _ = security.WriteAtomic