Add a basic noble migration check script

Perform a number of checks to ensure the system is ready for the noble
migration. The results are written to a JSON file in /etc/ that other
things like the JI and the upgrade script itself can read from.

The script is run hourly on a systemd timer but can also be run
interactively for administrators who want slightly more details.

Refs #7322.

molecule/testinfra/common/test_release_upgrades.py

@@ -1,3 +1,7 @@
+import json
+import time
+
+import pytest
 import testutils
 
 test_vars = testutils.securedrop_test_vars
@@ -27,3 +31,35 @@ def test_release_manager_upgrade_channel(host):
     _, channel = raw_output.split("=")
 
     assert channel == "never"
+
+
+def test_migration_check(host):
+    """Verify our migration check script works"""
+    if host.system_info.codename != "focal":
+        pytest.skip("only applicable/testable on focal")
+
+    with host.sudo():
+        # remove state file so we can see if it works
+        if host.file("/etc/securedrop-noble-migration.json").exists:
+            host.run("rm /etc/securedrop-noble-migration.json")
+        cmd = host.run("systemctl start securedrop-noble-migration-check")
+        assert cmd.rc == 0
+        while host.service("securedrop-noble-migration-check").is_running:
+            time.sleep(1)
+
+        # JSON state file was created
+        assert host.file("/etc/securedrop-noble-migration.json").exists
+
+        cmd = host.run("cat /etc/securedrop-noble-migration.json")
+        assert cmd.rc == 0
+
+        contents = json.loads(cmd.stdout)
+        print(contents)
+        # The script did not error out
+        assert "error" not in contents
+        # staging CI jobs don't have enough free space, so just check
+        # that it returned a value for it
+        assert "free_space" in contents
+        del contents["free_space"]
+        # All the values should be True
+        assert all(contents.values())

noble-migration/Cargo.toml

@@ -4,3 +4,8 @@ version = "0.1.0"
 edition = "2021"
 
 [dependencies]
+anyhow = "1.0.93"
+rustix = { version = "0.38.40", features = ["process"] }
+serde = { version = "1.0.215", features = ["derive"] }
+serde_json = "1.0.132"
+url = "2.5.3"

noble-migration/src/bin/check.rs

@@ -0,0 +1,313 @@
+//! Check migration of a SecureDrop server from focal to noble
+//!
+//! This script is run as root on both the app and mon servers.
+//!
+//! It is typically run by a systemd service/timer, but we also
+//! support admins running it manually to get more detailed output.
+use anyhow::{bail, Context, Result};
+use rustix::process::geteuid;
+use serde::Serialize;
+use std::{
+    fs,
+    path::Path,
+    process::{self, ExitCode},
+};
+use url::{Host, Url};
+
+/// This file contains the state of the pre-migration checks.
+///
+/// There are four possible states:
+/// * does not exist: check script hasn't run yet
+/// * empty JSON object: script determines it isn't on focal
+/// * {"error": true}: script encountered an error
+/// * JSON object with boolean values for each check (see `State` struct)
+const STATE_PATH: &str = "/etc/securedrop-noble-migration.json";
+
+#[derive(Serialize)]
+struct State {
+    ssh: bool,
+    ufw: bool,
+    free_space: bool,
+    apt: bool,
+    systemd: bool,
+}
+
+impl State {
+    fn is_ready(&self) -> bool {
+        self.ssh && self.ufw && self.free_space && self.apt && self.systemd
+    }
+}
+
+/// Parse the OS codename from /etc/os-release
+fn os_codename() -> Result<String> {
+    let contents = fs::read_to_string("/etc/os-release")
+        .context("reading /etc/os-release failed")?;
+    for line in contents.lines() {
+        if line.starts_with("VERSION_CODENAME=") {
+            // unwrap: Safe because we know the line contains "="
+            let (_, codename) = line.split_once("=").unwrap();
+            return Ok(codename.trim().to_string());
+        }
+    }
+
+    bail!("Could not find VERSION_CODENAME in /etc/os-release")
+}
+
+/// Check that the UNIX "ssh" group has no members
+///
+/// See <https://github.com/freedomofpress/securedrop/issues/7316>.
+fn check_ssh_group() -> Result<bool> {
+    // There are no clean bindings to getgrpname in rustix,
+    // so jut shell out to getent to get group members
+    let output = process::Command::new("getent")
+        .arg("group")
+        .arg("ssh")
+        .output()
+        .context("spawning getent failed")?;
+    if output.status.code() == Some(2) {
+        println!("ssh: group does not exist");
+        return Ok(true);
+    } else if !output.status.success() {
+        bail!(
+            "running getent failed: {}",
+            String::from_utf8_lossy(&output.stderr)
+        );
+    }
+
+    let stdout = String::from_utf8(output.stdout)
+        .context("getent stdout is not utf-8")?;
+    let members = parse_getent_output(&stdout)?;
+    if members.is_empty() {
+        println!("ssh: group is empty");
+        Ok(true)
+    } else {
+        println!("ssh: group is not empty: {members:?}");
+        Ok(false)
+    }
+}
+
+/// Parse the output of `getent group ssh`, return true if empty
+fn parse_getent_output(stdout: &str) -> Result<Vec<&str>> {
+    let stdout = stdout.trim();
+    // The format looks like `ssh:x:123:member1,member2`
+    if !stdout.contains(":") {
+        bail!("unexpected output from getent: '{stdout}'");
+    }
+
+    // unwrap: safe, we know the line contains ":"
+    let (_, members) = stdout.rsplit_once(':').unwrap();
+    if members.is_empty() {
+        Ok(vec![])
+    } else {
+        Ok(members.split(',').collect())
+    }
+}
+
+/// Check that ufw was removed
+///
+/// See <https://github.com/freedomofpress/securedrop/issues/7313>.
+fn check_ufw_removed() -> bool {
+    if Path::new("/usr/sbin/ufw").exists() {
+        println!("ufw: ufw is still installed");
+        false
+    } else {
+        println!("ufw: ufw was removed");
+        true
+    }
+}
+
+const DESIRED_FREE_SPACE: u64 = 10 * 1024 * 1024 * 1024; // 10GB
+
+/// Check that there is enough free space for taking a backup
+/// and downloading all the new packages. 10GB is a bit of an
+/// over-estimate, but it should leave enough headroom for everyone.
+fn check_free_space() -> Result<bool> {
+    // Also no simple bindings to get disk size, so shell out to df
+    let output = process::Command::new("df")
+        .arg("/")
+        .output()
+        .context("spawning df failed")?;
+    if !output.status.success() {
+        bail!(
+            "running df failed: {}",
+            String::from_utf8_lossy(&output.stderr)
+        );
+    }
+
+    let stdout =
+        String::from_utf8(output.stdout).context("df stdout is not utf-8")?;
+    let free_space = parse_df_output(&stdout)?;
+
+    if free_space < DESIRED_FREE_SPACE {
+        println!("free space: not enough free space");
+        Ok(false)
+    } else {
+        println!("free space: enough free space");
+        Ok(true)
+    }
+}
+
+fn parse_df_output(stdout: &str) -> Result<u64> {
+    let line = match stdout.split_once('\n') {
+        Some((_, line)) => line,
+        None => bail!("df output didn't have a newline"),
+    };
+    let parts: Vec<_> = line.split_whitespace().collect();
+
+    if parts.len() < 4 {
+        bail!("df output didn't have enough columns");
+    }
+
+    // vec indexing is safe because we did the bounds check above
+    let free_space = parts[3]
+        .parse::<u64>()
+        .context("parsing free space failed")?;
+
+    Ok(free_space)
+}
+
+const EXPECTED_DOMAINS: [&str; 5] = [
+    // Ubuntu domains
+    "archive.ubuntu.com",
+    "security.ubuntu.com",
+    // SecureDrop domains (including testing ones)
+    "apt.freedom.press",
+    "apt-qa.freedom.press",
+    "apt-test.freedom.press",
+];
+
+/// Verify only expected sources are configured for apt
+fn check_apt() -> Result<bool> {
+    let output = process::Command::new("apt-get")
+        .arg("indextargets")
+        .output()
+        .context("spawning apt-get indextargets failed")?;
+    if !output.status.success() {
+        bail!(
+            "running apt-get indextargets failed: {}",
+            String::from_utf8_lossy(&output.stderr)
+        );
+    }
+
+    let stdout = String::from_utf8(output.stdout)
+        .context("apt-get stdout is not utf-8")?;
+    for line in stdout.lines() {
+        if line.starts_with("URI:") {
+            let uri = line.strip_prefix("URI: ").unwrap();
+            let parsed = Url::parse(uri)?;
+            if let Some(Host::Domain(domain)) = parsed.host() {
+                if !EXPECTED_DOMAINS.contains(&domain) {
+                    println!("apt: unexpected source: {domain}");
+                    return Ok(false);
+                }
+            } else {
+                println!("apt: unexpected source: {uri}");
+                return Ok(false);
+            }
+        }
+    }
+
+    println!("apt: all sources are expected");
+    Ok(true)
+}
+
+/// Check that systemd has no failed units
+fn check_systemd() -> Result<bool> {
+    let output = process::Command::new("systemctl")
+        .arg("is-failed")
+        .output()
+        .context("spawning systemctl failed")?;
+    if output.status.success() {
+        // success means some units are failed
+        println!("systemd: some units are failed");
+        Ok(false)
+    } else {
+        println!("systemd: no failed units");
+        Ok(true)
+    }
+}
+
+fn run() -> Result<()> {
+    let codename = os_codename()?;
+    if codename != "focal" {
+        println!("Unsupported Ubuntu version: {codename}");
+        // nothing to do, write an empty JSON blob
+        fs::write(STATE_PATH, "{}")?;
+        return Ok(());
+    }
+
+    let state = State {
+        ssh: check_ssh_group()?,
+        ufw: check_ufw_removed(),
+        free_space: check_free_space()?,
+        apt: check_apt()?,
+        systemd: check_systemd()?,
+    };
+
+    fs::write(
+        STATE_PATH,
+        serde_json::to_string(&state).context("serializing state failed")?,
+    )
+    .context("writing state file failed")?;
+    if state.is_ready() {
+        println!("All ready for migration!");
+    } else {
+        println!(
+            "Some errors were found that will block migration.
+If you are unsure what to do, please contact the SecureDrop
+support team: <https://docs.securedrop.org/en/stable/getting_support.html>."
+        );
+        // Logically we should exit with a failure here, but we don't
+        // want the systemd unit to fail.
+    }
+    Ok(())
+}
+
+fn main() -> Result<ExitCode> {
+    if !geteuid().is_root() {
+        println!("This script must be run as root");
+        return Ok(ExitCode::FAILURE);
+    }
+
+    match run() {
+        Ok(()) => Ok(ExitCode::SUCCESS),
+        Err(e) => {
+            // Try to log the error in the least complex way possible
+            fs::write(STATE_PATH, "{\"error\": true}")?;
+            eprintln!("Error running migration pre-check: {e}");
+            Ok(ExitCode::FAILURE)
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_parse_getent_output() {
+        // no members
+        assert_eq!(
+            parse_getent_output("ssh:x:123:\n").unwrap(),
+            Vec::<&str>::new()
+        );
+        // one member
+        assert_eq!(
+            parse_getent_output("ssh:x:123:member1\n").unwrap(),
+            vec!["member1"]
+        );
+        // two members
+        assert_eq!(
+            parse_getent_output("ssh:x:123:member1,member2\n").unwrap(),
+            vec!["member1", "member2"]
+        );
+    }
+
+    #[test]
+    fn test_parse_df_output() {
+        // Taken from my Qubes VM, but the output of df is the same on Ubuntu
+        assert_eq!(parse_df_output("Filesystem         1K-blocks     Used Available Use% Mounted on
+/dev/mapper/dmroot  20260052 10727468   8478072  56% /
+").unwrap(), 8478072);
+    }
+}

securedrop/debian/config/lib/systemd/system/securedrop-noble-migration-check.service

@@ -0,0 +1,7 @@
+[Unit]
+Description=Check noble migration readiness
+
+[Service]
+Type=oneshot
+ExecStart=/usr/bin/securedrop-noble-migration-check
+User=root

securedrop/debian/config/lib/systemd/system/securedrop-noble-migration-check.timer

@@ -0,0 +1,10 @@
+[Unit]
+Description=Check noble migration readiness
+
+[Timer]
+OnCalendar=hourly
+Persistent=true
+RandomizedDelaySec=5m
+
+[Install]
+WantedBy=timers.target