tailscale_mcp/

cli.rs

//! Running a local command and reading its failure.
//!
//! The interesting part is the reading. `tailscale` reports everything through
//! a non-zero exit code and a line of English on standard error, so this is
//! where that line becomes one of the error codes an agent can act on. Getting
//! it wrong is not cosmetic: `needs_operator` tells the caller to run one
//! command and try again, while `unsupported_version` tells it to stop asking.

use std::collections::HashMap;
use std::net::IpAddr;

use tailscale_cli::{ExecError, Invocation, Output};

use crate::context::{SelfIdentity, ToolContext};
use crate::error::{ToolError, ToolResult};
use crate::meta::ToolMeta;
use crate::version::{Version, satisfies};

/// Run a command, and turn anything other than a clean exit into a tool error.
pub async fn run(ctx: &ToolContext, meta: &ToolMeta, invocation: Invocation) -> ToolResult<Output> {
    let display = displayed(ctx, &invocation);
    let output = ctx
        .local
        .run(invocation)
        .await
        .map_err(|e| exec_error(ctx, &display, e))?;
    if output.success() {
        return Ok(output);
    }
    Err(command_failure(ctx, meta, &display, &output))
}

/// Run a command whose refusal may be an ordinary answer.
///
/// Several `tailscale` commands report an unremarkable fact about the tailnet
/// through a non-zero exit: `exit-node list` when there are no exit nodes,
/// `routecheck` when no report has been produced yet, `wait` when the timeout
/// passed, `status` when the backend is not running. Reporting those as tool
/// failures would tell a caller that its call went wrong when what it learnt is
/// exactly what it asked for.
///
/// What stays a failure is everything that is not an answer about the tailnet:
/// a binary that is not there, a subcommand it does not know, a refusal to talk
/// to us at all. The caller reads [`Output::success`] for the rest.
pub async fn run_tolerant(
    ctx: &ToolContext,
    meta: &ToolMeta,
    invocation: Invocation,
) -> ToolResult<Output> {
    let display = displayed(ctx, &invocation);
    let output = ctx
        .local
        .run(invocation)
        .await
        .map_err(|e| exec_error(ctx, &display, e))?;
    if output.success() {
        return Ok(output);
    }
    let stderr = ctx.redactor.apply(&output.stderr);
    if is_unrecognised(&stderr) {
        return Err(version_error(ctx, meta));
    }
    if needs_operator(&stderr) {
        return Err(ToolError::needs_operator(&stderr));
    }
    // Deliberately not `is_not_found`: "no exit nodes found" is the answer
    // these commands exist to give.
    Ok(output)
}

/// Run a command and hand back its standard output as text, failures aside.
pub async fn run_text(
    ctx: &ToolContext,
    meta: &ToolMeta,
    invocation: Invocation,
) -> ToolResult<String> {
    let output = run(ctx, meta, invocation).await?;
    Ok(output.stdout_str().into_owned())
}

/// The command line as it may be shown: what ran, with any secret in it gone.
///
/// [`Invocation::display`] is the argument list verbatim, which every typed
/// tool could show as it stands, because the server assembled those arguments
/// itself and knows an auth key only ever reaches the CLI through a file. The
/// passthrough has no such assurance: its arguments are the caller's, so a
/// command line that turns into an error message, a log line or a report goes
/// through here first. It is `pub(crate)` for the last of those: the
/// passthrough puts the command in what it answers with, and a second spelling
/// of this expression is a second place to forget it.
pub(crate) fn displayed(ctx: &ToolContext, invocation: &Invocation) -> String {
    ctx.redactor.apply(&invocation.display()).into_owned()
}

/// Something went wrong before the command produced a result.
///
/// Every message here goes through the session's redactor, not just the
/// shape-based pass [`ToolError::new`] applies for itself. `ExecError::Io`
/// names the command it was talking to, and since the passthrough that command
/// line is the caller's; the rest are redacted too so that the next variant
/// added does not have to be judged for whether it carries one.
fn exec_error(ctx: &ToolContext, display: &str, error: ExecError) -> ToolError {
    let told = |error: &ExecError| ctx.redactor.apply(&error.to_string()).into_owned();
    match error {
        ExecError::BinaryNotFound { .. }
        | ExecError::BinaryNotExecutable { .. }
        | ExecError::Spawn { .. } => {
            ToolError::backend_unavailable("the local surface", &told(&error))
        }
        ExecError::Timeout {
            timeout, printed, ..
        } => ToolError::timeout(display, timeout.as_secs(), &ctx.redactor.apply(&printed)),
        ExecError::Io { .. } | ExecError::SecretFile(_) => {
            ToolError::new(crate::error::ErrorCode::CliFailed, told(&error))
        }
    }
}

/// The command ran and refused. Work out what kind of refusal it was.
pub fn command_failure(
    ctx: &ToolContext,
    meta: &ToolMeta,
    display: &str,
    output: &Output,
) -> ToolError {
    let stderr = ctx.redactor.apply(&output.stderr);
    if is_unrecognised(&stderr) {
        return version_error(ctx, meta);
    }
    if needs_operator(&stderr) {
        return ToolError::needs_operator(&stderr);
    }
    if is_not_found(&stderr) {
        return ToolError::not_found(stderr.trim());
    }
    ToolError::cli_failed(display, output.exit_code, &stderr)
}

/// The binary does not know this subcommand or flag.
///
/// Matched on Go's `flag` package wording and the CLI's own dispatch, both of
/// which are stable across the releases this server supports.
fn is_unrecognised(stderr: &str) -> bool {
    const MARKERS: &[&str] = &[
        "flag provided but not defined",
        "unknown flag",
        "unknown subcommand",
        "unknown command",
        "is not a tailscale command",
        "unrecognized command",
    ];
    let lowered = stderr.to_ascii_lowercase();
    MARKERS.iter().any(|m| lowered.contains(m))
}

/// The command exists but this user may not run it.
fn needs_operator(stderr: &str) -> bool {
    const MARKERS: &[&str] = &[
        "access denied",
        "--operator=",
        "must be run as root",
        "permission denied",
        "you must be root",
    ];
    let lowered = stderr.to_ascii_lowercase();
    MARKERS.iter().any(|m| lowered.contains(m))
}

/// The command ran but its target is not there.
fn is_not_found(stderr: &str) -> bool {
    const MARKERS: &[&str] = &["no such", "not found", "does not exist", "unknown peer"];
    let lowered = stderr.to_ascii_lowercase();
    MARKERS.iter().any(|m| lowered.contains(m))
}

/// Report an unrecognised command as a version problem.
///
/// The minimum comes from the tool's own row when it has one. When it does not
/// — the command predates our floor and should have been there — the floor is
/// the honest answer: something older than anything we model is running.
fn version_error(ctx: &ToolContext, meta: &ToolMeta) -> ToolError {
    let needs = meta
        .min_version
        .map(str::to_owned)
        .unwrap_or_else(|| crate::version::SUPPORTED_FLOOR.to_string());
    let found = ctx
        .cli_version
        .map_or_else(|| "unknown".to_owned(), |v| v.to_string());
    ToolError::unsupported_version(meta.name, &needs, &found)
}

/// Whether a tool's stated minimum is met by the CLI we found.
///
/// Checked before spawning, so that a tool with a known minimum reports the
/// version code with its own number rather than whatever the binary says.
pub fn version_permits(ctx: &ToolContext, meta: &ToolMeta) -> ToolResult<()> {
    if satisfies(ctx.cli_version, meta.min_version) {
        Ok(())
    } else {
        Err(version_error(ctx, meta))
    }
}

/// Read the version out of `tailscale version`.
///
/// A failure here is not an error: the probe runs at startup, and a server that
/// refused to start because it could not read a version string would be worse
/// than one that runs without knowing it.
pub async fn probe_version(backend: &dyn tailscale_cli::LocalBackend) -> Option<Version> {
    let output = backend
        .run(Invocation::read(["version"]))
        .await
        .ok()
        .filter(Output::success)?;
    Version::parse_cli_output(&output.stdout_str())
}

/// Read who this node is from `tailscale status --json`.
///
/// Only the handful of fields that name this node, because the point is to
/// recognise a control-plane operation aimed at ourselves (ticket 21) rather
/// than to model the status document — ticket 08 does that properly.
///
/// A failure gives an identity that matches nothing, which is the safe way
/// round: an operation we cannot prove is aimed at ourselves is treated as an
/// ordinary one, and the operator sees the same confirmation rules as anyone
/// managing another node.
pub async fn probe_identity(backend: &dyn tailscale_cli::LocalBackend) -> SelfIdentity {
    status_document(backend)
        .await
        .as_ref()
        .map(identity_in)
        .unwrap_or_default()
}

/// Who this node is and what it calls its peers, from one reading of status.
///
/// Both at once because both come out of the same document, and running
/// `tailscale status` twice at startup to parse the same JSON two ways is work
/// nobody asked for.
pub async fn probe_node(
    backend: &dyn tailscale_cli::LocalBackend,
) -> (SelfIdentity, HashMap<IpAddr, String>) {
    let Some(document) = status_document(backend).await else {
        return (SelfIdentity::default(), HashMap::new());
    };
    (identity_in(&document), peer_names_in(&document))
}

/// `tailscale status --json`, parsed, or nothing if it could not be had.
pub(crate) async fn status_document(
    backend: &dyn tailscale_cli::LocalBackend,
) -> Option<serde_json::Value> {
    let output = backend
        .run(Invocation::read(["status", "--json"]))
        .await
        .ok()
        .filter(Output::success)?;
    serde_json::from_str(&output.stdout_str()).ok()
}

fn identity_in(document: &serde_json::Value) -> SelfIdentity {
    let node = &document["Self"];
    SelfIdentity {
        node_id: node["ID"].as_str().map(str::to_owned),
        // Status cannot supply this: the numeric id is the control plane's own
        // name for the device and is never sent to the node. Filled in from
        // the control plane by `ToolContext::names_us`.
        numeric_id: None,
        addresses: node["TailscaleIPs"]
            .as_array()
            .map(|ips| {
                ips.iter()
                    .filter_map(|ip| ip.as_str().map(str::to_owned))
                    .collect()
            })
            .unwrap_or_default(),
        dns_name: node["DNSName"].as_str().map(str::to_owned),
    }
}

/// Every node this node can name, by address.
///
/// For the HTTP transport's log line: a request arriving from `100.64.0.2` is
/// more useful in a log as the node it came from, and the local node already
/// knows the mapping. An address this node has never heard of stays an
/// address (ticket 23).
fn peer_names_in(document: &serde_json::Value) -> HashMap<IpAddr, String> {
    let mut named = HashMap::new();
    let peers = document["Peer"]
        .as_object()
        .into_iter()
        .flat_map(|by_key| by_key.values());
    for node in std::iter::once(&document["Self"]).chain(peers) {
        let Some(name) = node["DNSName"].as_str() else {
            continue;
        };
        let name = name.trim_end_matches('.');
        for address in node["TailscaleIPs"].as_array().into_iter().flatten() {
            if let Some(address) = address.as_str().and_then(|a| a.parse().ok()) {
                named.insert(address, name.to_owned());
            }
        }
    }
    named
}

#[cfg(test)]
mod tests {
    use std::sync::Arc;

    use super::*;
    use crate::context::{Identity, PathPolicy};
    use crate::error::{ErrorCode, Redactor};
    use crate::meta::{Tier, ToolMeta, Toolset};
    use crate::testing::StubBackend;

    fn meta(min_version: Option<&'static str>) -> ToolMeta {
        ToolMeta {
            name: "tailscale_service_list",
            toolset: Toolset::LocalStatus,
            tier: Tier::Read,
            summary: "",
            self_severing: false,
            severs_local_node: false,
            requires_confirmation: false,
            idempotent: true,
            varying_tier: false,
            min_version,
            platforms: None,
        }
    }

    fn context(backend: StubBackend, cli_version: Option<Version>) -> ToolContext {
        ToolContext {
            local: Arc::new(backend),
            tailnet: None,
            redactor: Redactor::default(),
            max_result_bytes: 1 << 20,
            identity: Identity::default(),
            cli_version,
            paths: PathPolicy::default(),
            devices: Default::default(),
            max_tier: crate::meta::Tier::Destructive,
        }
    }

    #[tokio::test]
    async fn an_unknown_subcommand_reports_the_minimum_version() {
        let ctx = context(
            StubBackend::failure(1, "tailscale service: unknown subcommand \"list\"\n"),
            Some(Version::new(1, 78, 0)),
        );
        let err = run(
            &ctx,
            &meta(Some("1.94")),
            Invocation::read(["service", "list"]),
        )
        .await
        .expect_err("should fail");
        assert_eq!(err.code, ErrorCode::UnsupportedVersion);
        assert!(err.message.contains("1.94"), "{}", err.message);
        assert!(err.message.contains("1.78.0"), "{}", err.message);
    }

    #[tokio::test]
    async fn an_unknown_flag_reports_the_minimum_version() {
        let ctx = context(
            StubBackend::failure(1, "flag provided but not defined: -report-posture\n"),
            None,
        );
        let err = run(&ctx, &meta(Some("1.58")), Invocation::read(["set"]))
            .await
            .expect_err("should fail");
        assert_eq!(err.code, ErrorCode::UnsupportedVersion);
        assert!(err.message.contains("1.58"), "{}", err.message);
    }

    #[tokio::test]
    async fn a_tool_without_a_minimum_falls_back_to_the_floor() {
        let ctx = context(StubBackend::failure(1, "unknown subcommand\n"), None);
        let err = run(&ctx, &meta(None), Invocation::read(["nonsense"]))
            .await
            .expect_err("should fail");
        assert_eq!(err.code, ErrorCode::UnsupportedVersion);
        assert!(
            err.message
                .contains(&crate::version::SUPPORTED_FLOOR.to_string()),
            "{}",
            err.message
        );
    }

    #[tokio::test]
    async fn a_peer_named_for_an_operator_is_still_not_found() {
        let ctx = context(
            StubBackend::failure(1, "no such peer: operator-laptop\n"),
            None,
        );
        let err = run(&ctx, &meta(None), Invocation::read(["ping"]))
            .await
            .expect_err("should fail");
        assert_eq!(err.code, ErrorCode::NotFound, "{err:?}");
    }

    #[tokio::test]
    async fn a_secret_in_the_error_stream_does_not_reach_the_caller() {
        let ctx = context(
            StubBackend::failure(1, "bad key tskey-auth-example1CNTRL-secretpart\n"),
            None,
        );
        let err = run(&ctx, &meta(None), Invocation::read(["up"]))
            .await
            .expect_err("should fail");
        assert!(
            !err.stderr
                .as_deref()
                .unwrap_or_default()
                .contains("secretpart"),
            "{err:?}"
        );
    }

    #[test]
    fn a_broken_pipe_or_an_unwritable_secret_file_is_a_cli_failure() {
        let ctx = context(StubBackend::ok(""), None);
        for error in [
            ExecError::Io {
                command: "tailscale up".to_owned(),
                source: std::io::Error::other("broken pipe"),
            },
            ExecError::SecretFile(std::io::Error::other("no space left on device")),
        ] {
            let told = error.to_string();
            assert_eq!(
                exec_error(&ctx, "tailscale up", error).code,
                ErrorCode::CliFailed,
                "{told}"
            );
        }
    }

    /// One reading of status, two answers out of it.
    #[tokio::test]
    async fn the_status_probe_names_this_node_and_its_peers() {
        let backend = StubBackend::missing().on(
            ["status", "--json"],
            tailscale_cli::stub::Reply::ok(
                serde_json::json!({
                    "Self": {
                        "ID": "n1111111CNTRL",
                        "DNSName": "workstation.example-tailnet.ts.net.",
                        "TailscaleIPs": ["100.64.0.1", "fd7a:115c:a1e0::1"],
                    },
                    "Peer": {
                        "nodekey:2222": {
                            "DNSName": "laptop.example-tailnet.ts.net.",
                            "TailscaleIPs": ["100.64.0.2"],
                        },
                        // No addresses, so nothing to key it by.
                        "nodekey:3333": {"DNSName": "ghost.example-tailnet.ts.net."},
                    },
                })
                .to_string(),
            ),
        );

        let (identity, peers) = probe_node(&backend).await;
        assert!(identity.matches("n1111111CNTRL"));
        assert!(identity.matches("workstation"));
        assert_eq!(
            peers.get(&"100.64.0.2".parse::<IpAddr>().expect("an address")),
            Some(&"laptop.example-tailnet.ts.net".to_owned()),
            "a peer is named by the address a request would arrive from"
        );
        assert_eq!(
            peers.get(&"100.64.0.1".parse::<IpAddr>().expect("an address")),
            Some(&"workstation.example-tailnet.ts.net".to_owned()),
            "and so is this node, which can reach its own HTTP transport"
        );
        assert_eq!(
            peers.len(),
            3,
            "the trailing dot is dropped, the ghost has no address"
        );

        assert_eq!(
            backend.calls().len(),
            1,
            "one reading of status, not one per answer wanted"
        );
    }
}