core-agent-ide/codex-rs/execpolicy/src/policy.rs

use crate::decision::Decision;
use crate::error::Error;
use crate::error::Result;
use crate::rule::NetworkRule;
use crate::rule::NetworkRuleProtocol;
use crate::rule::PatternToken;
use crate::rule::PrefixPattern;
use crate::rule::PrefixRule;
use crate::rule::RuleMatch;
use crate::rule::RuleRef;
use crate::rule::normalize_network_rule_host;
use multimap::MultiMap;
use serde::Deserialize;
use serde::Serialize;
use std::sync::Arc;

type HeuristicsFallback<'a> = Option<&'a dyn Fn(&[String]) -> Decision>;

#[derive(Clone, Debug)]
pub struct Policy {
    rules_by_program: MultiMap<String, RuleRef>,
    network_rules: Vec<NetworkRule>,
}

impl Policy {
    pub fn new(rules_by_program: MultiMap<String, RuleRef>) -> Self {
        Self::from_parts(rules_by_program, Vec::new())
    }

    pub fn from_parts(
        rules_by_program: MultiMap<String, RuleRef>,
        network_rules: Vec<NetworkRule>,
    ) -> Self {
        Self {
            rules_by_program,
            network_rules,
        }
    }

    pub fn empty() -> Self {
        Self::new(MultiMap::new())
    }

    pub fn rules(&self) -> &MultiMap<String, RuleRef> {
        &self.rules_by_program
    }

    pub fn network_rules(&self) -> &[NetworkRule] {
        &self.network_rules
    }

    pub fn get_allowed_prefixes(&self) -> Vec<Vec<String>> {
        let mut prefixes = Vec::new();

        for (_program, rules) in self.rules_by_program.iter_all() {
            for rule in rules {
                let Some(prefix_rule) = rule.as_any().downcast_ref::<PrefixRule>() else {
                    continue;
                };
                if prefix_rule.decision != Decision::Allow {
                    continue;
                }

                let mut prefix = Vec::with_capacity(prefix_rule.pattern.rest.len() + 1);
                prefix.push(prefix_rule.pattern.first.as_ref().to_string());
                prefix.extend(prefix_rule.pattern.rest.iter().map(render_pattern_token));
                prefixes.push(prefix);
            }
        }

        prefixes.sort();
        prefixes.dedup();
        prefixes
    }

    pub fn add_prefix_rule(&mut self, prefix: &[String], decision: Decision) -> Result<()> {
        let (first_token, rest) = prefix
            .split_first()
            .ok_or_else(|| Error::InvalidPattern("prefix cannot be empty".to_string()))?;

        let rule: RuleRef = Arc::new(PrefixRule {
            pattern: PrefixPattern {
                first: Arc::from(first_token.as_str()),
                rest: rest
                    .iter()
                    .map(|token| PatternToken::Single(token.clone()))
                    .collect::<Vec<_>>()
                    .into(),
            },
            decision,
            justification: None,
        });

        self.rules_by_program.insert(first_token.clone(), rule);
        Ok(())
    }

    pub fn add_network_rule(
        &mut self,
        host: &str,
        protocol: NetworkRuleProtocol,
        decision: Decision,
        justification: Option<String>,
    ) -> Result<()> {
        let host = normalize_network_rule_host(host)?;
        if let Some(raw) = justification.as_deref()
            && raw.trim().is_empty()
        {
            return Err(Error::InvalidRule(
                "justification cannot be empty".to_string(),
            ));
        }
        self.network_rules.push(NetworkRule {
            host,
            protocol,
            decision,
            justification,
        });
        Ok(())
    }

    pub fn compiled_network_domains(&self) -> (Vec<String>, Vec<String>) {
        let mut allowed = Vec::new();
        let mut denied = Vec::new();

        for rule in &self.network_rules {
            match rule.decision {
                Decision::Allow => {
                    denied.retain(|entry| entry != &rule.host);
                    upsert_domain(&mut allowed, &rule.host);
                }
                Decision::Forbidden => {
                    allowed.retain(|entry| entry != &rule.host);
                    upsert_domain(&mut denied, &rule.host);
                }
                Decision::Prompt => {}
            }
        }

        (allowed, denied)
    }

    pub fn check<F>(&self, cmd: &[String], heuristics_fallback: &F) -> Evaluation
    where
        F: Fn(&[String]) -> Decision,
    {
        let matched_rules = self.matches_for_command(cmd, Some(heuristics_fallback));
        Evaluation::from_matches(matched_rules)
    }

    /// Checks multiple commands and aggregates the results.
    pub fn check_multiple<Commands, F>(
        &self,
        commands: Commands,
        heuristics_fallback: &F,
    ) -> Evaluation
    where
        Commands: IntoIterator,
        Commands::Item: AsRef<[String]>,
        F: Fn(&[String]) -> Decision,
    {
        let matched_rules: Vec<RuleMatch> = commands
            .into_iter()
            .flat_map(|command| {
                self.matches_for_command(command.as_ref(), Some(heuristics_fallback))
            })
            .collect();

        Evaluation::from_matches(matched_rules)
    }

    /// Returns matching rules for the given command. If no rules match and
    /// `heuristics_fallback` is provided, returns a single
    /// `HeuristicsRuleMatch` with the decision rendered by
    /// `heuristics_fallback`.
    ///
    /// If `heuristics_fallback.is_some()`, then the returned vector is
    /// guaranteed to be non-empty.
    pub fn matches_for_command(
        &self,
        cmd: &[String],
        heuristics_fallback: HeuristicsFallback<'_>,
    ) -> Vec<RuleMatch> {
        let matched_rules: Vec<RuleMatch> = match cmd.first() {
            Some(first) => self
                .rules_by_program
                .get_vec(first)
                .map(|rules| rules.iter().filter_map(|rule| rule.matches(cmd)).collect())
                .unwrap_or_default(),
            None => Vec::new(),
        };

        if matched_rules.is_empty()
            && let Some(heuristics_fallback) = heuristics_fallback
        {
            vec![RuleMatch::HeuristicsRuleMatch {
                command: cmd.to_vec(),
                decision: heuristics_fallback(cmd),
            }]
        } else {
            matched_rules
        }
    }
}

fn upsert_domain(entries: &mut Vec<String>, host: &str) {
    entries.retain(|entry| entry != host);
    entries.push(host.to_string());
}

fn render_pattern_token(token: &PatternToken) -> String {
    match token {
        PatternToken::Single(value) => value.clone(),
        PatternToken::Alts(alternatives) => format!("[{}]", alternatives.join("|")),
    }
}

#[derive(Clone, Debug, Eq, PartialEq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct Evaluation {
    pub decision: Decision,
    #[serde(rename = "matchedRules")]
    pub matched_rules: Vec<RuleMatch>,
}

impl Evaluation {
    pub fn is_match(&self) -> bool {
        self.matched_rules
            .iter()
            .any(|rule_match| !matches!(rule_match, RuleMatch::HeuristicsRuleMatch { .. }))
    }

    /// Caller is responsible for ensuring that `matched_rules` is non-empty.
    fn from_matches(matched_rules: Vec<RuleMatch>) -> Self {
        let decision = matched_rules.iter().map(RuleMatch::decision).max();
        #[expect(clippy::expect_used)]
        let decision = decision.expect("invariant failed: matched_rules must be non-empty");

        Self {
            decision,
            matched_rules,
        }
    }
}