gitks/bare.rs

//! Copyright (c) 2022-2026 GitDataAi All rights reserved.

use std::path::{Path, PathBuf};

use crate::error::{GitError, GitResult};
use crate::pb::RepositoryHeader;

#[derive(Debug)]
pub struct GitBare {
    pub bare_dir: PathBuf,
}

impl GitBare {
    pub fn new(bare_dir: PathBuf) -> Self {
        Self { bare_dir }
    }

    pub fn gix_repo(&self) -> GitResult<gix::Repository> {
        tracing::debug!(repo = %self.bare_dir.display(), "opening gix repository");
        gix::open(&self.bare_dir).map_err(|e| {
            tracing::error!(repo = %self.bare_dir.display(), error = %e, "failed to open gix repository");
            GitError::Internal(format!("failed to open gix repository: {e}"))
        })
    }

    pub fn from_repository_header(header: &RepositoryHeader) -> GitResult<Self> {
        let storage_path = header.storage_path.trim();
        let relative_path = header.relative_path.trim();
        let storage_name = header.storage_name.trim();
        let _ = storage_name; // reserved for future sharding logic

        // Validate relative_path early to prevent path traversal
        if !relative_path.is_empty() {
            crate::sanitize::validate_relative_path(relative_path)?;
        }

        let base = if !storage_path.is_empty() {
            let p = Path::new(storage_path);
            if !p.is_absolute() {
                return Err(GitError::InvalidArgument(
                    "storage_path must be an absolute path".into(),
                ));
            }
            PathBuf::from(p)
        } else if !relative_path.is_empty() {
            return Err(GitError::InvalidArgument(
                "relative_path requires storage_path to be set".into(),
            ));
        } else {
            return Err(GitError::InvalidArgument("empty repository path".into()));
        };

        let bare_dir = if !relative_path.is_empty() && !storage_path.is_empty() {
            let candidate = base.join(relative_path);
            let base_canon = base.canonicalize().unwrap_or_else(|_| base.clone());

            // Validate that relative_path itself contains no traversal patterns
            // before any filesystem access (mitigates TOCTOU)
            if relative_path.contains("..") {
                return Err(GitError::InvalidArgument(format!(
                    "path traversal detected: relative_path contains '..': {relative_path}"
                )));
            }
            // Reject symlinks in relative_path components
            if relative_path.contains('\0') {
                return Err(GitError::InvalidArgument(
                    "relative_path contains null byte".into(),
                ));
            }

            let canonical = match candidate.canonicalize() {
                Ok(canon) => canon,
                Err(_) => {
                    // Path doesn't exist yet; validate via parent
                    let parent = candidate.parent().unwrap_or(&base);
                    let filename = candidate.file_name().ok_or_else(|| {
                        GitError::InvalidArgument("invalid path: missing filename".into())
                    })?;

                    let parent_canon = parent
                        .canonicalize()
                        .unwrap_or_else(|_| parent.to_path_buf());
                    let constructed = parent_canon.join(filename);

                    let constructed_str = constructed.to_string_lossy();
                    let base_str = base_canon.to_string_lossy();

                    if !constructed_str.starts_with(&*base_str) {
                        tracing::warn!(
                            relative_path = %relative_path,
                            base = %base_canon.display(),
                            "path traversal attempt detected (parent check)"
                        );
                        return Err(GitError::InvalidArgument(format!(
                            "path traversal detected: {relative_path} escapes storage root"
                        )));
                    }

                    constructed
                }
            };

            if !canonical.starts_with(&base_canon) {
                tracing::warn!(
                    relative_path = %relative_path,
                    canonical = %canonical.display(),
                    base = %base_canon.display(),
                    "path traversal attempt detected"
                );
                return Err(GitError::InvalidArgument(format!(
                    "path traversal detected: {relative_path} escapes storage root"
                )));
            }

            // Verify the resolved path has no symlinks in its components
            // by checking that canonicalization is idempotent
            let double_canon = canonical
                .canonicalize()
                .unwrap_or_else(|_| canonical.clone());
            if canonical != double_canon {
                return Err(GitError::InvalidArgument(
                    "path resolved to different target (possible symlink race)".into(),
                ));
            }

            canonical
        } else if !storage_path.is_empty() {
            base.canonicalize().unwrap_or(base)
        } else {
            return Err(GitError::InvalidArgument("empty repository path".into()));
        };

        if !bare_dir.exists() {
            tracing::warn!(path = %bare_dir.display(), "repository not found");
            return Err(GitError::RepoNotFound);
        }
        if !bare_dir.is_dir() {
            return Err(GitError::InvalidArgument(format!(
                "not a directory: {}",
                bare_dir.display()
            )));
        }

        let head_path = bare_dir.join("HEAD");
        if !head_path.exists() {
            let git_dir = bare_dir.join(".git");
            if git_dir.is_dir() && git_dir.join("HEAD").exists() {
                tracing::debug!(path = %git_dir.display(), "resolved non-bare repo via .git subdir");
                return Ok(Self { bare_dir: git_dir });
            }
            return Err(GitError::NotBareRepository);
        }

        Ok(Self { bare_dir })
    }

    pub fn object_format(&self) -> crate::pb::ObjectFormat {
        let repo = self.gix_repo().ok();
        let kind = repo
            .as_ref()
            .map(|r| r.object_hash())
            .unwrap_or(gix::hash::Kind::Sha1);
        match kind {
            gix::hash::Kind::Sha1 => crate::pb::ObjectFormat::Sha1,
            gix::hash::Kind::Sha256 => crate::pb::ObjectFormat::Sha256,
            _ => crate::pb::ObjectFormat::Unspecified,
        }
    }

    /// Convert a hex object id to a protobuf Oid.
    ///
    /// `Oid.value` is the binary hash bytes, while `Oid.hex` keeps the printable
    /// lowercase representation for clients that prefer string IDs.
    pub fn oid_to_pb(&self, hex: impl Into<String>) -> crate::pb::Oid {
        let hex = hex.into().to_lowercase();
        let format = self.object_format();
        crate::pb::Oid {
            value: crate::oid::hex_to_bytes(&hex).unwrap_or_default(),
            hex,
            format: format as i32,
        }
    }
}