OXYGENHUB

Coverage / crates/hub/src/terraform.rs 85.3%

266 of 312 lines run, 39 of 122 functions · run #510 on main · run missed

1
//! Terraform and OpenTofu state, kept by the hub per repository
2
//! (oxygenhub#252): the HTTP backend Terraform and OpenTofu speak, with
3
//! locking, at `/api/v1/repos/OWNER/NAME/terraform/state/STATE`, so a Flow
4
//! job keeps its state without any cloud credential. A job authenticates
5
//! with an identity token of this hub's for the audience `{base}/terraform`
6
//! (its pipeline's `identity` block) or its runner's job token, either for
7
//! its own repository only; people with a token of theirs and their
8
//! repository role. Tokens go as the bearer, or as the password of Basic
9
//! auth, which is what the backend sends.
10
//!
11
//! State is kept outside the repository's history (it often holds secrets),
12
//! sealed at rest with the hub's secrets key, and its last versions are
13
//! kept, so a bad apply can be rolled back: a kept version is restored as a
14
//! new one (#255), so nothing is lost and going forward again is restoring
15
//! a newer one.
16
17
use crate::api::ApiError;
18
use crate::auth::roles::{Access, Role};
19
use crate::{db, Shared};
20
use axum::body::Bytes;
21
use axum::extract::{Path, Query, State};
22
use axum::http::{header, HeaderMap, Method, StatusCode};
23
use axum::response::{IntoResponse, Response};
24
25
/// Versions of a state kept besides the current one.
26
pub const VERSIONS_KEPT: i64 = 20;
27
/// The most a state may weigh.
28
pub const MAX_STATE_BYTES: usize = 32 << 20;
29
30
/// Who may use a repository's states.
31
enum Who {
32
    /// A running job of this repository.
33
    Job { name: String },
34
    /// A person, with their role on the repository.
35
    Person { account: Box<db::accounts::Account>, role: Role },
36
}
37
38
impl Who {
394×
    fn label(&self) -> String {
404×
        match self {
414×
            Who::Job { name } => name.clone(),
420×
            Who::Person { account, .. } => account.username.clone(),
43
        }
444×
    }
45
4624×
    fn may(&self, access: Access) -> bool {
4724×
        match self {
48
            // A job reads and writes its own repository's states.
4913×
            Who::Job { .. } => true,
5011×
            Who::Person { role, .. } => crate::auth::roles::allows(Some(*role), access),
51
        }
5224×
    }
53
}
54
5512×
fn err(status: StatusCode, why: impl Into<String>) -> ApiError {
5612×
    ApiError(status, why.into())
5712×
}
58
59
/// The token a request carries: the bearer, or Basic auth's password.
6029×
fn token_of(headers: &HeaderMap) -> Option<String> {
6129×
    if let Some(t) = oxy_server::bearer_token(headers) {
6215×
        return Some(t.to_string());
6314×
    }
6414×
    let v = headers.get(header::AUTHORIZATION)?.to_str().ok()?;
6513×
    let b64 = v.strip_prefix("Basic ").or_else(|| v.strip_prefix("basic "))?;
6613×
    let raw = String::from_utf8(oxy_core::dsse::base64_decode(b64.trim())?).ok()?;
6713×
    let (_, password) = raw.split_once(':')?;
6813×
    (!password.is_empty()).then(|| password.to_string())
6929×
}
70
71
/// The audience a job's identity token needs here.
725×
pub fn audience(base_url: &str) -> String {
735×
    format!("{}/terraform", base_url.trim_end_matches('/'))
745×
}
75
7630×
fn valid_name(name: &str) -> bool {
7730×
    !name.is_empty()
7830×
        && name.len() <= 100
79119×
        && name.chars().all(|c| c.is_ascii_alphanumeric() || c == '-' || c == '_' || c == '.')
8030×
        && !name.starts_with('.')
8130×
}
82
8329×
async fn who(app: &Shared, headers: &HeaderMap, row: &db::repos::RepoRow) -> Result<Who, ApiError> {
8429×
    let token = token_of(headers).ok_or_else(|| {
851×
        err(StatusCode::UNAUTHORIZED, "a Terraform state needs a token: a job's (TF_HTTP_PASSWORD), or one of yours with repo:write")
861×
    })?;
8728×
    let hash = crate::auth::token_hash(&token);
8828×
    if token.starts_with(crate::flow::JOB_TOKEN_PREFIX) {
8913×
        let job = db::flow::job_by_token(&app.db, &hash, crate::flow::JOB_TOKEN_MAX_AGE_MS)
9013×
            .await
9113×
            .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?
9213×
            .ok_or_else(|| err(StatusCode::UNAUTHORIZED, "this job token is not valid (the job must be running)"))?;
9313×
        let run = db::flow::run_by_id(&app.db, job.run_id)
9413×
            .await
9513×
            .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?
9613×
            .ok_or_else(|| err(StatusCode::UNAUTHORIZED, "this job token is not valid"))?;
9713×
        if run.repo_id != row.id {
981×
            return Err(err(StatusCode::FORBIDDEN, "a job keeps the states of its own repository only"));
9912×
        }
10012×
        return Ok(Who::Job { name: format!("run #{} {}", run.number, job.name) });
10115×
    }
10215×
    let personal = db::accounts::token_by_hash(&app.db, &hash).await.map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
10315×
    let Some(token) = personal.filter(|t| !t.expired(db::now_ms())) else {
104
        // A job's identity token for this hub's Terraform audience.
1054×
        let issuer = app.oidc().await.map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
1064×
        let aud = audience(app.config.base_url());
1074×
        let Some(claims) = issuer.verify(&token, &aud, db::now_ms()) else {
1082×
            return Err(err(StatusCode::UNAUTHORIZED, format!("unknown or expired token (a job's identity token must be for {aud})")));
109
        };
1102×
        let sub = claims["sub"].as_str().unwrap_or("");
1112×
        let repo = claims["repository"]
1122×
            .as_str()
1132×
            .map(String::from)
1142×
            .or_else(|| sub.strip_prefix("repo:").and_then(|r| r.split_once(":bookmark:")).map(|(r, _)| r.to_string()));
1152×
        if repo.as_deref() != Some(row.slug().as_str()) {
1161×
            return Err(err(StatusCode::FORBIDDEN, "a job keeps the states of its own repository only"));
1171×
        }
1181×
        return Ok(Who::Job { name: sub.to_string() });
119
    };
12011×
    if !(token.has_scope("repo:write") || token.has_scope("repo:admin")) {
1210×
        return Err(err(
1220×
            StatusCode::FORBIDDEN,
1230×
            "the token lacks repo:write: a Terraform state, read or written, needs it (states hold secrets)",
1240×
        ));
12511×
    }
12611×
    let slug = row.slug();
12711×
    if !token.covers(&slug) {
1280×
        return Err(err(StatusCode::FORBIDDEN, format!("the token does not cover {slug}")));
12911×
    }
13011×
    let account = db::accounts::account_by_id(&app.db, token.account_id)
13111×
        .await
13211×
        .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?
13311×
        .filter(|a| !a.banned && !a.disabled)
13411×
        .ok_or_else(|| err(StatusCode::UNAUTHORIZED, "unknown token"))?;
13511×
    let role = crate::auth::roles::role_for(&app.db, Some(&account), row)
13611×
        .await
13711×
        .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
13811×
    match role {
13911×
        Some(r) if crate::auth::roles::allows(Some(r), Access::Read) => Ok(Who::Person { account: Box::new(account), role: r }),
1400×
        _ => Err(err(StatusCode::NOT_FOUND, "no such repository")),
141
    }
14229×
}
143
144
#[derive(serde::Deserialize, Default)]
145
pub struct LockQuery {
146
    #[serde(rename = "ID", default)]
147
    id: Option<String>,
148
}
149
15013×
fn json(status: StatusCode, body: String) -> Response {
15113×
    (status, [(header::CONTENT_TYPE, "application/json")], body).into_response()
15213×
}
153
154
/// The lock's ID in a lock-info body (`{"ID": …}`).
1556×
fn lock_id(info: &[u8]) -> Option<String> {
1566×
    serde_json::from_slice::<serde_json::Value>(info).ok()?.get("ID")?.as_str().map(String::from)
1576×
}
158
159
/// Why a restore did not happen.
160
pub enum RestoreError {
161
    /// No such version of that state.
162
    NotFound,
163
    /// Someone holds the state's lock: who.
164
    Locked(String),
165
    /// The kept body does not open or is not JSON.
166
    Broken,
167
    Db(sqlx::Error),
168
}
169
170
impl RestoreError {
1712×
    pub fn message(&self) -> String {
1722×
        match self {
1731×
            RestoreError::NotFound => "no such version of this state".into(),
1741×
            RestoreError::Locked(by) => format!("the state is locked by {by}: restore it once the lock is released"),
1750×
            RestoreError::Broken => "that version does not open: it cannot be restored".into(),
1760×
            RestoreError::Db(e) => e.to_string(),
177
        }
1782×
    }
179
1802×
    fn status(&self) -> StatusCode {
1812×
        match self {
1821×
            RestoreError::NotFound => StatusCode::NOT_FOUND,
1831×
            RestoreError::Locked(_) => StatusCode::LOCKED,
1840×
            RestoreError::Broken | RestoreError::Db(_) => StatusCode::INTERNAL_SERVER_ERROR,
185
        }
1862×
    }
187
}
188
189
/// Restore kept version `id` of a state (#255): written as a new version
190
/// whose serial is above the current one (OpenTofu and Terraform take a
191
/// state only forward), so the state it replaces is kept as a version too.
192
/// Refused while the state is locked. The new serial.
1935×
pub async fn restore(app: &Shared, repo_id: i64, state: &str, id: i64, by: &str) -> Result<i64, RestoreError> {
1945×
    let current = db::terraform::get(&app.db, repo_id, state).await.map_err(RestoreError::Db)?;
1955×
    if let Some(l) = current.as_ref().filter(|s| !s.lock_id.is_empty()) {
1961×
        return Err(RestoreError::Locked(l.locked_by.clone()));
1974×
    }
1983×
    let (body, old_serial) =
1994×
        db::terraform::version_body(&app.db, repo_id, state, id).await.map_err(RestoreError::Db)?.ok_or(RestoreError::NotFound)?;
2003×
    let plain = app.secret_key.open(&body).ok_or(RestoreError::Broken)?;
2013×
    let mut json: serde_json::Value = serde_json::from_str(&plain).map_err(|_| RestoreError::Broken)?;
2023×
    let serial = current.map_or(0, |c| c.serial).max(old_serial) + 1;
2033×
    json["serial"] = serial.into();
2043×
    let text = serde_json::to_string_pretty(&json).map_err(|_| RestoreError::Broken)?;
2053×
    let label = format!("{by}, restoring serial {old_serial}");
2063×
    db::terraform::put(&app.db, repo_id, state, &app.secret_key.seal(&text), serial, &label, VERSIONS_KEPT)
2073×
        .await
2083×
        .map_err(RestoreError::Db)?;
2093×
    Ok(serial)
2105×
}
211
212
/// `GET /api/v1/repos/{owner}/{name}/terraform/state/{state}/versions`:
213
/// the state's kept versions, newest first (#255).
2143×
pub async fn versions(
2153×
    State(app): State<Shared>,
2163×
    headers: HeaderMap,
2173×
    Path((owner, name, state)): Path<(String, String, String)>,
2183×
) -> Result<Response, ApiError> {
2193×
    let (row, _who) = versions_caller(&app, &headers, &owner, &name, &state, Access::Write).await?;
2202×
    let list =
2212×
        db::terraform::version_list(&app.db, row.id, &state).await.map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
2222×
    let out: Vec<serde_json::Value> = list
2232×
        .iter()
2243×
        .map(|v| serde_json::json!({ "id": v.id, "serial": v.serial, "written_by": v.created_by, "written_ms": v.created_ms }))
2252×
        .collect();
2262×
    Ok(json(StatusCode::OK, serde_json::Value::Array(out).to_string()))
2273×
}
228
229
/// `GET …/terraform/state/{state}/versions/{id}`: one kept version's state.
2301×
pub async fn version(
2311×
    State(app): State<Shared>,
2321×
    headers: HeaderMap,
2331×
    Path((owner, name, state, id)): Path<(String, String, String, i64)>,
2341×
) -> Result<Response, ApiError> {
2351×
    let (row, _who) = versions_caller(&app, &headers, &owner, &name, &state, Access::Write).await?;
2361×
    let (body, _) = db::terraform::version_body(&app.db, row.id, &state, id)
2371×
        .await
2381×
        .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?
2391×
        .ok_or_else(|| err(StatusCode::NOT_FOUND, "no such version of this state"))?;
2401×
    let plain = app.secret_key.open(&body).ok_or_else(|| err(StatusCode::INTERNAL_SERVER_ERROR, "the version does not open"))?;
2411×
    Ok(json(StatusCode::OK, plain))
2421×
}
243
244
/// `POST …/terraform/state/{state}/versions/{id}/restore`: restore a kept
245
/// version, by a person with the maintain role (never a job), audited.
2466×
pub async fn restore_version(
2476×
    State(app): State<Shared>,
2486×
    headers: HeaderMap,
2496×
    Path((owner, name, state, id)): Path<(String, String, String, i64)>,
2506×
) -> Result<Response, ApiError> {
2516×
    let (row, who) = versions_caller(&app, &headers, &owner, &name, &state, Access::Maintain).await?;
2525×
    let Who::Person { account, .. } = &who else {
2531×
        return Err(err(StatusCode::FORBIDDEN, "a job does not restore a state: a person with the maintain role does"));
254
    };
2554×
    let serial = restore(&app, row.id, &state, id, &account.username).await.map_err(|e| err(e.status(), e.message()))?;
2562×
    db::accounts::audit(
2572×
        &app.db,
2582×
        Some(account),
2592×
        "terraform.restored",
2602×
        &format!("{}:{state}", row.slug()),
2612×
        &format!("version {id}, now serial {serial}"),
2622×
    )
2632×
    .await
2642×
    .ok();
2652×
    Ok(json(StatusCode::OK, serde_json::json!({ "serial": serial }).to_string()))
2666×
}
267
268
/// The repository and the caller of a versions endpoint, allowed `access`.
26910×
async fn versions_caller(
27010×
    app: &Shared,
27110×
    headers: &HeaderMap,
27210×
    owner: &str,
27310×
    name: &str,
27410×
    state: &str,
27510×
    access: Access,
27610×
) -> Result<(db::repos::RepoRow, Who), ApiError> {
27710×
    if !valid_name(state) {
2780×
        return Err(err(StatusCode::BAD_REQUEST, "a state's name is letters, digits, '-', '_' and '.', up to 100"));
27910×
    }
28010×
    let row = db::repos::get(&app.db, owner, name)
28110×
        .await
28210×
        .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?
28310×
        .ok_or_else(|| err(StatusCode::NOT_FOUND, "no such repository"))?;
28410×
    let who = who(app, headers, &row).await?;
28510×
    if !who.may(access) {
2862×
        return Err(err(
287
            StatusCode::FORBIDDEN,
2882×
            if access == Access::Maintain { "restoring a state needs the maintain role" } else { "a Terraform state needs the write role" },
289
        ));
2908×
    }
2918×
    Ok((row, who))
29210×
}
293
294
/// `GET|POST|DELETE|LOCK|UNLOCK /api/v1/repos/{owner}/{name}/terraform/state/{state}`.
29520×
pub async fn state(
29620×
    State(app): State<Shared>,
29720×
    method: Method,
29820×
    headers: HeaderMap,
29920×
    Path((owner, name, state)): Path<(String, String, String)>,
30020×
    Query(q): Query<LockQuery>,
30120×
    body: Bytes,
30220×
) -> Result<Response, ApiError> {
30320×
    if !valid_name(&state) {
3041×
        return Err(err(StatusCode::BAD_REQUEST, "a state's name is letters, digits, '-', '_' and '.', up to 100"));
30519×
    }
30619×
    let row = db::repos::get(&app.db, &owner, &name)
30719×
        .await
30819×
        .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?
30919×
        .ok_or_else(|| err(StatusCode::NOT_FOUND, "no such repository"))?;
31019×
    let who = who(&app, &headers, &row).await?;
311
    // Reading a state needs the write role too: states hold secrets, and
312
    // every account reads a public or internal repository (#256).
31314×
    if !who.may(Access::Write) {
3141×
        return Err(err(StatusCode::FORBIDDEN, "a Terraform state needs the write role on its repository: states hold secrets"));
31513×
    }
31613×
    let db_err = |e: sqlx::Error| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string());
31713×
    let current = db::terraform::get(&app.db, row.id, &state).await.map_err(db_err)?;
31813×
    let locked = current.as_ref().filter(|s| !s.lock_id.is_empty());
31913×
    match method.as_str() {
32013×
        "GET" => match current.filter(|s| !s.body.is_empty()) {
3213×
            Some(s) => {
3223×
                let plain =
3233×
                    app.secret_key.open(&s.body).ok_or_else(|| err(StatusCode::INTERNAL_SERVER_ERROR, "the state does not open"))?;
3243×
                Ok(json(StatusCode::OK, plain))
325
            }
326
            // No state yet: Terraform starts afresh.
3271×
            None => Ok(StatusCode::NO_CONTENT.into_response()),
328
        },
3299×
        "POST" => {
3303×
            if body.len() > MAX_STATE_BYTES {
3310×
                return Err(err(StatusCode::PAYLOAD_TOO_LARGE, format!("a state may be {} MiB at most", MAX_STATE_BYTES >> 20)));
3323×
            }
3333×
            let text = std::str::from_utf8(&body).map_err(|_| err(StatusCode::BAD_REQUEST, "a state is JSON"))?;
3343×
            let parsed: serde_json::Value = serde_json::from_str(text).map_err(|_| err(StatusCode::BAD_REQUEST, "a state is JSON"))?;
3353×
            if let Some(l) = locked {
3362×
                if q.id.as_deref() != Some(l.lock_id.as_str()) {
3371×
                    return Ok(json(StatusCode::LOCKED, l.lock_info.clone()));
3381×
                }
3391×
            }
3402×
            let serial = parsed.get("serial").and_then(|v| v.as_i64()).unwrap_or(0);
3412×
            let sealed = app.secret_key.seal(text);
3422×
            db::terraform::put(&app.db, row.id, &state, &sealed, serial, &who.label(), VERSIONS_KEPT).await.map_err(db_err)?;
3432×
            Ok(StatusCode::OK.into_response())
344
        }
3456×
        "DELETE" => {
3460×
            if let Some(l) = locked {
3470×
                if q.id.as_deref() != Some(l.lock_id.as_str()) {
3480×
                    return Ok(json(StatusCode::LOCKED, l.lock_info.clone()));
3490×
                }
3500×
            }
3510×
            db::terraform::delete(&app.db, row.id, &state).await.map_err(db_err)?;
3520×
            Ok(StatusCode::OK.into_response())
353
        }
3546×
        "LOCK" => {
3553×
            let info = std::str::from_utf8(&body).unwrap_or("");
3563×
            let Some(id) = lock_id(body.as_ref()) else {
3570×
                return Err(err(StatusCode::BAD_REQUEST, "a lock request is the lock's JSON, with its ID"));
358
            };
3593×
            if let Some(l) = locked {
3601×
                return Ok(json(if l.lock_id == id { StatusCode::OK } else { StatusCode::LOCKED }, l.lock_info.clone()));
3612×
            }
3622×
            if db::terraform::lock(&app.db, row.id, &state, &id, info, &who.label()).await.map_err(db_err)? {
3632×
                Ok(json(StatusCode::OK, info.to_string()))
364
            } else {
365
                // Taken meanwhile.
3660×
                let now = db::terraform::get(&app.db, row.id, &state).await.map_err(db_err)?;
3670×
                Ok(json(StatusCode::LOCKED, now.map(|s| s.lock_info).unwrap_or_default()))
368
            }
369
        }
3703×
        "UNLOCK" => {
3713×
            let Some(l) = locked else { return Ok(StatusCode::OK.into_response()) };
372
            // `terraform force-unlock` sends the ID; a plain unlock its lock info.
3733×
            let id = lock_id(body.as_ref()).or(q.id.clone());
3743×
            if id.as_deref() != Some(l.lock_id.as_str()) {
3751×
                return Ok(json(StatusCode::CONFLICT, l.lock_info.clone()));
3762×
            }
3772×
            db::terraform::unlock(&app.db, row.id, &state).await.map_err(db_err)?;
3782×
            Ok(StatusCode::OK.into_response())
379
        }
3800×
        other => {
3810×
            Err(err(StatusCode::METHOD_NOT_ALLOWED, format!("{other} is not a Terraform backend method (GET, POST, DELETE, LOCK, UNLOCK)")))
382
        }
383
    }
38420×
}