Redact Git credentials from pyproject.toml (#6074)

## Summary

We retain them if you use `--raw-sources`, but otherwise they're
removed. We still respect them in the subsequent `uv.lock` via an
in-process store.

Closes #6056.
This commit is contained in:
Charlie Marsh 2024-08-13 21:30:02 -04:00 committed by GitHub
parent 92263108cc
commit 8fac63d4ce
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
10 changed files with 301 additions and 23 deletions

1
Cargo.lock generated
View file

@ -4885,6 +4885,7 @@ dependencies = [
"tokio",
"tracing",
"url",
"uv-auth",
"uv-fs",
]

View file

@ -10,7 +10,7 @@ use std::io::Write;
use url::Url;
#[derive(Clone, Debug, PartialEq)]
pub(crate) struct Credentials {
pub struct Credentials {
/// The name of the user for authentication.
username: Username,
/// The password to use for authentication.
@ -114,7 +114,7 @@ impl Credentials {
/// Parse [`Credentials`] from a URL, if any.
///
/// Returns [`None`] if both [`Url::username`] and [`Url::password`] are not populated.
pub(crate) fn from_url(url: &Url) -> Option<Self> {
pub fn from_url(url: &Url) -> Option<Self> {
if url.username().is_empty() && url.password().is_none() {
return None;
}
@ -203,6 +203,20 @@ impl Credentials {
header
}
/// Apply the credentials to the given URL.
///
/// Any existing credentials will be overridden.
#[must_use]
pub fn apply(&self, mut url: Url) -> Url {
if let Some(username) = self.username() {
let _ = url.set_username(username);
}
if let Some(password) = self.password() {
let _ = url.set_password(Some(password));
}
url
}
/// Attach the credentials to the given request.
///
/// Any existing credentials will be overridden.

View file

@ -1,20 +1,20 @@
use std::sync::{Arc, LazyLock};
use tracing::trace;
use url::Url;
use cache::CredentialsCache;
pub use credentials::Credentials;
pub use keyring::KeyringProvider;
pub use middleware::AuthMiddleware;
use realm::Realm;
mod cache;
mod credentials;
mod keyring;
mod middleware;
mod realm;
use std::sync::{Arc, LazyLock};
use cache::CredentialsCache;
use credentials::Credentials;
pub use keyring::KeyringProvider;
pub use middleware::AuthMiddleware;
use realm::Realm;
use tracing::trace;
use url::Url;
// TODO(zanieb): Consider passing a cache explicitly throughout
/// Global authentication cache for a uv invocation

View file

@ -15,6 +15,7 @@ workspace = true
[dependencies]
cache-key = { workspace = true }
uv-fs = { workspace = true }
uv-auth = { workspace = true }
anyhow = { workspace = true }
cargo-util = { workspace = true }

View file

@ -0,0 +1,21 @@
use std::collections::HashMap;
use std::sync::{Arc, RwLock};
use cache_key::RepositoryUrl;
use uv_auth::Credentials;
/// A store for Git credentials.
#[derive(Debug, Default)]
pub struct GitStore(RwLock<HashMap<RepositoryUrl, Arc<Credentials>>>);
impl GitStore {
/// Insert [`Credentials`] for the given URL into the store.
pub fn insert(&self, url: RepositoryUrl, credentials: Credentials) -> Option<Arc<Credentials>> {
self.0.write().unwrap().insert(url, Arc::new(credentials))
}
/// Get the [`Credentials`] for the given URL, if they exist.
pub fn get(&self, url: &RepositoryUrl) -> Option<Arc<Credentials>> {
self.0.read().unwrap().get(url).cloned()
}
}

View file

@ -1,5 +1,8 @@
use std::sync::LazyLock;
use url::Url;
use crate::credentials::GitStore;
pub use crate::git::GitReference;
pub use crate::resolver::{
GitResolver, GitResolverError, RepositoryReference, ResolvedRepositoryReference,
@ -7,11 +10,17 @@ pub use crate::resolver::{
pub use crate::sha::{GitOid, GitSha, OidParseError};
pub use crate::source::{Fetch, GitSource, Reporter};
mod credentials;
mod git;
mod resolver;
mod sha;
mod source;
/// Global authentication cache for a uv invocation.
///
/// This is used to share Git credentials within a single process.
pub static GIT_STORE: LazyLock<GitStore> = LazyLock::new(GitStore::default);
/// A URL reference to a Git repository.
#[derive(Debug, Clone, PartialEq, PartialOrd, Eq, Hash, Ord)]
pub struct GitUrl {
@ -44,6 +53,7 @@ impl GitUrl {
}
}
/// Set the precise [`GitSha`] to use for this Git URL.
#[must_use]
pub fn with_precise(mut self, precise: GitSha) -> Self {
self.precise = Some(precise);

View file

@ -1,6 +1,8 @@
//! Git support is derived from Cargo's implementation.
//! Cargo is dual-licensed under either Apache 2.0 or MIT, at the user's choice.
//! Source: <https://github.com/rust-lang/cargo/blob/23eb492cf920ce051abfc56bbaf838514dc8365c/src/cargo/sources/git/source.rs>
use std::borrow::Cow;
use std::path::{Path, PathBuf};
use anyhow::Result;
@ -11,7 +13,7 @@ use url::Url;
use cache_key::{cache_digest, RepositoryUrl};
use crate::git::GitRemote;
use crate::{GitOid, GitSha, GitUrl};
use crate::{GitOid, GitSha, GitUrl, GIT_STORE};
/// A remote Git source that can be checked out locally.
pub struct GitSource {
@ -52,11 +54,21 @@ impl GitSource {
/// Fetch the underlying Git repository at the given revision.
#[instrument(skip(self), fields(repository = %self.git.repository, rev = ?self.git.precise))]
pub fn fetch(self) -> Result<Fetch> {
// Compute the canonical URL for the repository.
let canonical = RepositoryUrl::new(&self.git.repository);
// The path to the repo, within the Git database.
let ident = cache_digest(&RepositoryUrl::new(&self.git.repository));
let ident = cache_digest(&canonical);
let db_path = self.cache.join("db").join(&ident);
let remote = GitRemote::new(&self.git.repository);
// Authenticate the URL, if necessary.
let remote = if let Some(credentials) = GIT_STORE.get(&canonical) {
Cow::Owned(credentials.apply(self.git.repository.clone()))
} else {
Cow::Borrowed(&self.git.repository)
};
let remote = GitRemote::new(&remote);
let (db, actual_rev, task) = match (self.git.precise, remote.db_at(&db_path).ok()) {
// If we have a locked revision, and we have a preexisting database
// which has that revision, then no update needs to happen.

View file

@ -2,12 +2,12 @@ use std::collections::hash_map::Entry;
use std::path::PathBuf;
use anyhow::{Context, Result};
use cache_key::RepositoryUrl;
use owo_colors::OwoColorize;
use pep508_rs::{ExtraName, Requirement, VersionOrUrl};
use rustc_hash::{FxBuildHasher, FxHashMap};
use tracing::debug;
use pep508_rs::{ExtraName, Requirement, VersionOrUrl};
use uv_auth::store_credentials_from_url;
use uv_auth::{store_credentials_from_url, Credentials};
use uv_cache::Cache;
use uv_client::{BaseClientBuilder, Connectivity, FlatIndexClient, RegistryClientBuilder};
use uv_configuration::{
@ -16,6 +16,7 @@ use uv_configuration::{
use uv_dispatch::BuildDispatch;
use uv_distribution::DistributionDatabase;
use uv_fs::CWD;
use uv_git::GIT_STORE;
use uv_normalize::PackageName;
use uv_python::{
request_from_version_file, EnvironmentPreference, Interpreter, PythonDownloads,
@ -330,6 +331,37 @@ pub(crate) async fn add(
}
};
// Redact any credentials. By default, we avoid writing sensitive credentials to files that
// will be checked into version control (e.g., `pyproject.toml` and `uv.lock`). Instead,
// we store the credentials in a global store, and reuse them during resolution. The
// expectation is that subsequent resolutions steps will succeed by reading from (e.g.) the
// user's credentials store, rather than by reading from the `pyproject.toml` file.
let source = match source {
Some(Source::Git {
mut git,
subdirectory,
rev,
tag,
branch,
}) => {
let credentials = Credentials::from_url(&git);
if let Some(credentials) = credentials {
debug!("Caching credentials for: {git}");
GIT_STORE.insert(RepositoryUrl::new(&git), credentials);
let _ = git.set_username("");
let _ = git.set_password(None);
};
Some(Source::Git {
git,
subdirectory,
rev,
tag,
branch,
})
}
_ => source,
};
// Update the `pyproject.toml`.
let edit = match dependency_type {
DependencyType::Production => toml.add_dependency(&requirement, source.as_ref())?,

View file

@ -5,7 +5,7 @@ use assert_fs::prelude::*;
use indoc::indoc;
use insta::assert_snapshot;
use crate::common::packse_index_url;
use crate::common::{decode_token, packse_index_url};
use common::{uv_snapshot, TestContext};
mod common;
@ -291,6 +291,196 @@ fn add_git() -> Result<()> {
Ok(())
}
/// Add a Git requirement from a private repository, with credentials. The resolution should
/// succeed, but the `pyproject.toml` should omit the credentials.
#[test]
#[cfg(feature = "git")]
fn add_git_private_source() -> Result<()> {
let context = TestContext::new("3.12");
let token = decode_token(common::READ_ONLY_GITHUB_TOKEN);
let pyproject_toml = context.temp_dir.child("pyproject.toml");
pyproject_toml.write_str(indoc! {r#"
[project]
name = "project"
version = "0.1.0"
requires-python = ">=3.12"
dependencies = []
"#})?;
uv_snapshot!(context.filters(), context.add(&[&format!("uv-private-pypackage @ git+https://{token}@github.com/astral-test/uv-private-pypackage")]).arg("--preview"), @r###"
success: true
exit_code: 0
----- stdout -----
----- stderr -----
Resolved 2 packages in [TIME]
Prepared 2 packages in [TIME]
Installed 2 packages in [TIME]
+ project==0.1.0 (from file://[TEMP_DIR]/)
+ uv-private-pypackage==0.1.0 (from git+https://github.com/astral-test/uv-private-pypackage@d780faf0ac91257d4d5a4f0c5a0e4509608c0071)
"###);
let pyproject_toml = fs_err::read_to_string(context.temp_dir.join("pyproject.toml"))?;
insta::with_settings!({
filters => context.filters(),
}, {
assert_snapshot!(
pyproject_toml, @r###"
[project]
name = "project"
version = "0.1.0"
requires-python = ">=3.12"
dependencies = [
"uv-private-pypackage",
]
[tool.uv.sources]
uv-private-pypackage = { git = "https://github.com/astral-test/uv-private-pypackage" }
"###
);
});
let lock = fs_err::read_to_string(context.temp_dir.join("uv.lock"))?;
insta::with_settings!({
filters => context.filters(),
}, {
assert_snapshot!(
lock, @r###"
version = 1
requires-python = ">=3.12"
[options]
exclude-newer = "2024-03-25 00:00:00 UTC"
[[package]]
name = "project"
version = "0.1.0"
source = { editable = "." }
dependencies = [
{ name = "uv-private-pypackage" },
]
[[package]]
name = "uv-private-pypackage"
version = "0.1.0"
source = { git = "https://github.com/astral-test/uv-private-pypackage#d780faf0ac91257d4d5a4f0c5a0e4509608c0071" }
"###
);
});
// Install from the lockfile.
uv_snapshot!(context.filters(), context.sync().arg("--frozen"), @r###"
success: true
exit_code: 0
----- stdout -----
----- stderr -----
warning: `uv sync` is experimental and may change without warning
Audited 2 packages in [TIME]
"###);
Ok(())
}
/// Add a Git requirement from a private repository, with credentials. Since `--raw-sources` is
/// specified, the `pyproject.toml` should retain the credentials.
#[test]
#[cfg(feature = "git")]
fn add_git_private_raw() -> Result<()> {
let context = TestContext::new("3.12");
let token = decode_token(common::READ_ONLY_GITHUB_TOKEN);
let pyproject_toml = context.temp_dir.child("pyproject.toml");
pyproject_toml.write_str(indoc! {r#"
[project]
name = "project"
version = "0.1.0"
requires-python = ">=3.12"
dependencies = []
"#})?;
uv_snapshot!(context.filters(), context.add(&[&format!("uv-private-pypackage @ git+https://{token}@github.com/astral-test/uv-private-pypackage")]).arg("--raw-sources").arg("--preview"), @r###"
success: true
exit_code: 0
----- stdout -----
----- stderr -----
Resolved 2 packages in [TIME]
Prepared 2 packages in [TIME]
Installed 2 packages in [TIME]
+ project==0.1.0 (from file://[TEMP_DIR]/)
+ uv-private-pypackage==0.1.0 (from git+https://github.com/astral-test/uv-private-pypackage@d780faf0ac91257d4d5a4f0c5a0e4509608c0071)
"###);
let pyproject_toml = fs_err::read_to_string(context.temp_dir.join("pyproject.toml"))?;
let filters: Vec<_> = [(token.as_str(), "***")]
.into_iter()
.chain(context.filters())
.collect();
insta::with_settings!({
filters => filters
}, {
assert_snapshot!(
pyproject_toml, @r###"
[project]
name = "project"
version = "0.1.0"
requires-python = ">=3.12"
dependencies = [
"uv-private-pypackage @ git+https://***@github.com/astral-test/uv-private-pypackage",
]
"###
);
});
let lock = fs_err::read_to_string(context.temp_dir.join("uv.lock"))?;
insta::with_settings!({
filters => context.filters(),
}, {
assert_snapshot!(
lock, @r###"
version = 1
requires-python = ">=3.12"
[options]
exclude-newer = "2024-03-25 00:00:00 UTC"
[[package]]
name = "project"
version = "0.1.0"
source = { editable = "." }
dependencies = [
{ name = "uv-private-pypackage" },
]
[[package]]
name = "uv-private-pypackage"
version = "0.1.0"
source = { git = "https://github.com/astral-test/uv-private-pypackage#d780faf0ac91257d4d5a4f0c5a0e4509608c0071" }
"###
);
});
// Install from the lockfile.
uv_snapshot!(context.filters(), context.sync().arg("--frozen"), @r###"
success: true
exit_code: 0
----- stdout -----
----- stderr -----
warning: `uv sync` is experimental and may change without warning
Audited 2 packages in [TIME]
"###);
Ok(())
}
#[test]
#[cfg(feature = "git")]
fn add_git_error() -> Result<()> {

View file

@ -5271,9 +5271,6 @@ fn lock_redact_git() -> Result<()> {
version = "0.1.0"
requires-python = ">=3.12"
dependencies = ["uv-private-pypackage @ git+https://{token}@github.com/astral-test/uv-private-pypackage"]
[tool.uv]
index-url = "https://public:heron@pypi-proxy.fly.dev/basic-auth/simple"
"#,
token = token,
})?;