diff options
author | Alex Auvolat <lx@deuxfleurs.fr> | 2025-01-31 18:18:04 +0100 |
---|---|---|
committer | Alex Auvolat <lx@deuxfleurs.fr> | 2025-01-31 18:18:29 +0100 |
commit | 9fa20d45bebab2a3f66b9721c3643dbd607d944d (patch) | |
tree | 4c5cc3dee19f7cbd9e146a90ef5cbddb052716d5 /src/api/common | |
parent | 9330fd79d3466051394f6d419a247d46da8f5151 (diff) | |
download | garage-9fa20d45bebab2a3f66b9721c3643dbd607d944d.tar.gz garage-9fa20d45bebab2a3f66b9721c3643dbd607d944d.zip |
wip: split garage_api into garage_api_{common,s3,k2v,admin}
Diffstat (limited to 'src/api/common')
-rw-r--r-- | src/api/common/Cargo.toml | 73 | ||||
-rw-r--r-- | src/api/common/common_error.rs | 217 | ||||
-rw-r--r-- | src/api/common/encoding.rs | 22 | ||||
-rw-r--r-- | src/api/common/generic_server.rs | 378 | ||||
-rw-r--r-- | src/api/common/helpers.rs | 371 | ||||
-rw-r--r-- | src/api/common/lib.rs | 12 | ||||
-rw-r--r-- | src/api/common/router_macros.rs | 226 | ||||
-rw-r--r-- | src/api/common/signature/error.rs | 32 | ||||
-rw-r--r-- | src/api/common/signature/mod.rs | 78 | ||||
-rw-r--r-- | src/api/common/signature/payload.rs | 562 | ||||
-rw-r--r-- | src/api/common/signature/streaming.rs | 373 |
11 files changed, 2344 insertions, 0 deletions
diff --git a/src/api/common/Cargo.toml b/src/api/common/Cargo.toml new file mode 100644 index 00000000..e5dc57d4 --- /dev/null +++ b/src/api/common/Cargo.toml @@ -0,0 +1,73 @@ +[package] +name = "garage_api_common" +version = "1.0.1" +authors = ["Alex Auvolat <alex@adnab.me>"] +edition = "2018" +license = "AGPL-3.0" +description = "S3 API server crate for the Garage object store" +repository = "https://git.deuxfleurs.fr/Deuxfleurs/garage" +readme = "../../README.md" + +[lib] +path = "lib.rs" + +# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html + +[dependencies] +garage_model.workspace = true +garage_table.workspace = true +garage_block.workspace = true +garage_net.workspace = true +garage_util.workspace = true +garage_rpc.workspace = true + +aes-gcm.workspace = true +argon2.workspace = true +async-compression.workspace = true +async-trait.workspace = true +base64.workspace = true +bytes.workspace = true +chrono.workspace = true +crc32fast.workspace = true +crc32c.workspace = true +crypto-common.workspace = true +err-derive.workspace = true +hex.workspace = true +hmac.workspace = true +idna.workspace = true +tracing.workspace = true +md-5.workspace = true +nom.workspace = true +pin-project.workspace = true +sha1.workspace = true +sha2.workspace = true + +futures.workspace = true +futures-util.workspace = true +tokio.workspace = true +tokio-stream.workspace = true +tokio-util.workspace = true + +form_urlencoded.workspace = true +http.workspace = true +httpdate.workspace = true +http-range.workspace = true +http-body-util.workspace = true +hyper = { workspace = true, default-features = false, features = ["server", "http1"] } +hyper-util.workspace = true +multer.workspace = true +percent-encoding.workspace = true +roxmltree.workspace = true +url.workspace = true + +serde.workspace = true +serde_bytes.workspace = true +serde_json.workspace = true +quick-xml.workspace = true + +opentelemetry.workspace = true +opentelemetry-prometheus = { workspace = true, optional = true } +prometheus = { workspace = true, optional = true } + +[features] +metrics = [ "opentelemetry-prometheus", "prometheus" ] diff --git a/src/api/common/common_error.rs b/src/api/common/common_error.rs new file mode 100644 index 00000000..1e3f9feb --- /dev/null +++ b/src/api/common/common_error.rs @@ -0,0 +1,217 @@ +use std::convert::TryFrom; + +use err_derive::Error; +use hyper::StatusCode; + +use garage_util::error::Error as GarageError; + +use garage_model::helper::error::Error as HelperError; + +/// Errors of this crate +#[derive(Debug, Error)] +pub enum CommonError { + // ---- INTERNAL ERRORS ---- + /// Error related to deeper parts of Garage + #[error(display = "Internal error: {}", _0)] + InternalError(#[error(source)] GarageError), + + /// Error related to Hyper + #[error(display = "Internal error (Hyper error): {}", _0)] + Hyper(#[error(source)] hyper::Error), + + /// Error related to HTTP + #[error(display = "Internal error (HTTP error): {}", _0)] + Http(#[error(source)] http::Error), + + // ---- GENERIC CLIENT ERRORS ---- + /// Proper authentication was not provided + #[error(display = "Forbidden: {}", _0)] + Forbidden(String), + + /// Generic bad request response with custom message + #[error(display = "Bad request: {}", _0)] + BadRequest(String), + + /// The client sent a header with invalid value + #[error(display = "Invalid header value: {}", _0)] + InvalidHeader(#[error(source)] hyper::header::ToStrError), + + // ---- SPECIFIC ERROR CONDITIONS ---- + // These have to be error codes referenced in the S3 spec here: + // https://docs.aws.amazon.com/AmazonS3/latest/API/ErrorResponses.html#ErrorCodeList + /// The bucket requested don't exists + #[error(display = "Bucket not found: {}", _0)] + NoSuchBucket(String), + + /// Tried to create a bucket that already exist + #[error(display = "Bucket already exists")] + BucketAlreadyExists, + + /// Tried to delete a non-empty bucket + #[error(display = "Tried to delete a non-empty bucket")] + BucketNotEmpty, + + // Category: bad request + /// Bucket name is not valid according to AWS S3 specs + #[error(display = "Invalid bucket name: {}", _0)] + InvalidBucketName(String), +} + +impl CommonError { + pub fn http_status_code(&self) -> StatusCode { + match self { + CommonError::InternalError( + GarageError::Timeout | GarageError::RemoteError(_) | GarageError::Quorum(..), + ) => StatusCode::SERVICE_UNAVAILABLE, + CommonError::InternalError(_) | CommonError::Hyper(_) | CommonError::Http(_) => { + StatusCode::INTERNAL_SERVER_ERROR + } + CommonError::BadRequest(_) => StatusCode::BAD_REQUEST, + CommonError::Forbidden(_) => StatusCode::FORBIDDEN, + CommonError::NoSuchBucket(_) => StatusCode::NOT_FOUND, + CommonError::BucketNotEmpty | CommonError::BucketAlreadyExists => StatusCode::CONFLICT, + CommonError::InvalidBucketName(_) | CommonError::InvalidHeader(_) => { + StatusCode::BAD_REQUEST + } + } + } + + pub fn aws_code(&self) -> &'static str { + match self { + CommonError::Forbidden(_) => "AccessDenied", + CommonError::InternalError( + GarageError::Timeout | GarageError::RemoteError(_) | GarageError::Quorum(..), + ) => "ServiceUnavailable", + CommonError::InternalError(_) | CommonError::Hyper(_) | CommonError::Http(_) => { + "InternalError" + } + CommonError::BadRequest(_) => "InvalidRequest", + CommonError::NoSuchBucket(_) => "NoSuchBucket", + CommonError::BucketAlreadyExists => "BucketAlreadyExists", + CommonError::BucketNotEmpty => "BucketNotEmpty", + CommonError::InvalidBucketName(_) => "InvalidBucketName", + CommonError::InvalidHeader(_) => "InvalidHeaderValue", + } + } + + pub fn bad_request<M: ToString>(msg: M) -> Self { + CommonError::BadRequest(msg.to_string()) + } +} + +impl TryFrom<HelperError> for CommonError { + type Error = HelperError; + + fn try_from(err: HelperError) -> Result<Self, HelperError> { + match err { + HelperError::Internal(i) => Ok(Self::InternalError(i)), + HelperError::BadRequest(b) => Ok(Self::BadRequest(b)), + HelperError::InvalidBucketName(n) => Ok(Self::InvalidBucketName(n)), + HelperError::NoSuchBucket(n) => Ok(Self::NoSuchBucket(n)), + e => Err(e), + } + } +} + +/// This function converts HelperErrors into CommonErrors, +/// for variants that exist in CommonError. +/// This is used for helper functions that might return InvalidBucketName +/// or NoSuchBucket for instance, and we want to pass that error +/// up to our caller. +pub fn pass_helper_error(err: HelperError) -> CommonError { + match CommonError::try_from(err) { + Ok(e) => e, + Err(e) => panic!("Helper error `{}` should hot have happenned here", e), + } +} + +pub fn helper_error_as_internal(err: HelperError) -> CommonError { + match err { + HelperError::Internal(e) => CommonError::InternalError(e), + e => CommonError::InternalError(GarageError::Message(e.to_string())), + } +} + +pub trait CommonErrorDerivative: From<CommonError> { + fn internal_error<M: ToString>(msg: M) -> Self { + Self::from(CommonError::InternalError(GarageError::Message( + msg.to_string(), + ))) + } + + fn bad_request<M: ToString>(msg: M) -> Self { + Self::from(CommonError::BadRequest(msg.to_string())) + } + + fn forbidden<M: ToString>(msg: M) -> Self { + Self::from(CommonError::Forbidden(msg.to_string())) + } +} + +/// Trait to map error to the Bad Request error code +pub trait OkOrBadRequest { + type S; + fn ok_or_bad_request<M: AsRef<str>>(self, reason: M) -> Result<Self::S, CommonError>; +} + +impl<T, E> OkOrBadRequest for Result<T, E> +where + E: std::fmt::Display, +{ + type S = T; + fn ok_or_bad_request<M: AsRef<str>>(self, reason: M) -> Result<T, CommonError> { + match self { + Ok(x) => Ok(x), + Err(e) => Err(CommonError::BadRequest(format!( + "{}: {}", + reason.as_ref(), + e + ))), + } + } +} + +impl<T> OkOrBadRequest for Option<T> { + type S = T; + fn ok_or_bad_request<M: AsRef<str>>(self, reason: M) -> Result<T, CommonError> { + match self { + Some(x) => Ok(x), + None => Err(CommonError::BadRequest(reason.as_ref().to_string())), + } + } +} + +/// Trait to map an error to an Internal Error code +pub trait OkOrInternalError { + type S; + fn ok_or_internal_error<M: AsRef<str>>(self, reason: M) -> Result<Self::S, CommonError>; +} + +impl<T, E> OkOrInternalError for Result<T, E> +where + E: std::fmt::Display, +{ + type S = T; + fn ok_or_internal_error<M: AsRef<str>>(self, reason: M) -> Result<T, CommonError> { + match self { + Ok(x) => Ok(x), + Err(e) => Err(CommonError::InternalError(GarageError::Message(format!( + "{}: {}", + reason.as_ref(), + e + )))), + } + } +} + +impl<T> OkOrInternalError for Option<T> { + type S = T; + fn ok_or_internal_error<M: AsRef<str>>(self, reason: M) -> Result<T, CommonError> { + match self { + Some(x) => Ok(x), + None => Err(CommonError::InternalError(GarageError::Message( + reason.as_ref().to_string(), + ))), + } + } +} diff --git a/src/api/common/encoding.rs b/src/api/common/encoding.rs new file mode 100644 index 00000000..e286a784 --- /dev/null +++ b/src/api/common/encoding.rs @@ -0,0 +1,22 @@ +//! Module containing various helpers for encoding + +/// Encode &str for use in a URI +pub fn uri_encode(string: &str, encode_slash: bool) -> String { + let mut result = String::with_capacity(string.len() * 2); + for c in string.chars() { + match c { + 'a'..='z' | 'A'..='Z' | '0'..='9' | '_' | '-' | '~' | '.' => result.push(c), + '/' if encode_slash => result.push_str("%2F"), + '/' if !encode_slash => result.push('/'), + _ => { + result.push_str( + &format!("{}", c) + .bytes() + .map(|b| format!("%{:02X}", b)) + .collect::<String>(), + ); + } + } + } + result +} diff --git a/src/api/common/generic_server.rs b/src/api/common/generic_server.rs new file mode 100644 index 00000000..d92a3465 --- /dev/null +++ b/src/api/common/generic_server.rs @@ -0,0 +1,378 @@ +use std::convert::Infallible; +use std::fs::{self, Permissions}; +use std::os::unix::fs::PermissionsExt; +use std::sync::Arc; +use std::time::Duration; + +use async_trait::async_trait; + +use futures::future::Future; +use futures::stream::{futures_unordered::FuturesUnordered, StreamExt}; + +use http_body_util::BodyExt; +use hyper::header::HeaderValue; +use hyper::server::conn::http1; +use hyper::service::service_fn; +use hyper::{body::Incoming as IncomingBody, Request, Response}; +use hyper::{HeaderMap, StatusCode}; +use hyper_util::rt::TokioIo; + +use tokio::io::{AsyncRead, AsyncWrite}; +use tokio::net::{TcpListener, TcpStream, UnixListener, UnixStream}; +use tokio::sync::watch; +use tokio::time::{sleep_until, Instant}; + +use opentelemetry::{ + global, + metrics::{Counter, ValueRecorder}, + trace::{FutureExt, SpanRef, TraceContextExt, Tracer}, + Context, KeyValue, +}; + +use garage_util::error::Error as GarageError; +use garage_util::forwarded_headers; +use garage_util::metrics::{gen_trace_id, RecordDuration}; +use garage_util::socket_address::UnixOrTCPSocketAddress; + +use crate::helpers::{BoxBody, ErrorBody}; + +pub trait ApiEndpoint: Send + Sync + 'static { + fn name(&self) -> &'static str; + fn add_span_attributes(&self, span: SpanRef<'_>); +} + +pub trait ApiError: std::error::Error + Send + Sync + 'static { + fn http_status_code(&self) -> StatusCode; + fn add_http_headers(&self, header_map: &mut HeaderMap<HeaderValue>); + fn http_body(&self, garage_region: &str, path: &str) -> ErrorBody; +} + +#[async_trait] +pub trait ApiHandler: Send + Sync + 'static { + const API_NAME: &'static str; + const API_NAME_DISPLAY: &'static str; + + type Endpoint: ApiEndpoint; + type Error: ApiError; + + fn parse_endpoint(&self, r: &Request<IncomingBody>) -> Result<Self::Endpoint, Self::Error>; + async fn handle( + &self, + req: Request<IncomingBody>, + endpoint: Self::Endpoint, + ) -> Result<Response<BoxBody<Self::Error>>, Self::Error>; +} + +pub struct ApiServer<A: ApiHandler> { + region: String, + api_handler: A, + + // Metrics + request_counter: Counter<u64>, + error_counter: Counter<u64>, + request_duration: ValueRecorder<f64>, +} + +impl<A: ApiHandler> ApiServer<A> { + pub fn new(region: String, api_handler: A) -> Arc<Self> { + let meter = global::meter("garage/api"); + Arc::new(Self { + region, + api_handler, + request_counter: meter + .u64_counter(format!("api.{}.request_counter", A::API_NAME)) + .with_description(format!( + "Number of API calls to the various {} API endpoints", + A::API_NAME_DISPLAY + )) + .init(), + error_counter: meter + .u64_counter(format!("api.{}.error_counter", A::API_NAME)) + .with_description(format!( + "Number of API calls to the various {} API endpoints that resulted in errors", + A::API_NAME_DISPLAY + )) + .init(), + request_duration: meter + .f64_value_recorder(format!("api.{}.request_duration", A::API_NAME)) + .with_description(format!( + "Duration of API calls to the various {} API endpoints", + A::API_NAME_DISPLAY + )) + .init(), + }) + } + + pub async fn run_server( + self: Arc<Self>, + bind_addr: UnixOrTCPSocketAddress, + unix_bind_addr_mode: Option<u32>, + must_exit: watch::Receiver<bool>, + ) -> Result<(), GarageError> { + let server_name = format!("{} API", A::API_NAME_DISPLAY); + info!("{} server listening on {}", server_name, bind_addr); + + match bind_addr { + UnixOrTCPSocketAddress::TCPSocket(addr) => { + let listener = TcpListener::bind(addr).await?; + + let handler = move |request, socketaddr| self.clone().handler(request, socketaddr); + server_loop(server_name, listener, handler, must_exit).await + } + UnixOrTCPSocketAddress::UnixSocket(ref path) => { + if path.exists() { + fs::remove_file(path)? + } + + let listener = UnixListener::bind(path)?; + let listener = UnixListenerOn(listener, path.display().to_string()); + + fs::set_permissions( + path, + Permissions::from_mode(unix_bind_addr_mode.unwrap_or(0o222)), + )?; + + let handler = move |request, socketaddr| self.clone().handler(request, socketaddr); + server_loop(server_name, listener, handler, must_exit).await + } + } + } + + async fn handler( + self: Arc<Self>, + req: Request<IncomingBody>, + addr: String, + ) -> Result<Response<BoxBody<A::Error>>, http::Error> { + let uri = req.uri().clone(); + + if let Ok(forwarded_for_ip_addr) = + forwarded_headers::handle_forwarded_for_headers(req.headers()) + { + info!( + "{} (via {}) {} {}", + forwarded_for_ip_addr, + addr, + req.method(), + uri + ); + } else { + info!("{} {} {}", addr, req.method(), uri); + } + debug!("{:?}", req); + + let tracer = opentelemetry::global::tracer("garage"); + let span = tracer + .span_builder(format!("{} API call (unknown)", A::API_NAME_DISPLAY)) + .with_trace_id(gen_trace_id()) + .with_attributes(vec![ + KeyValue::new("method", format!("{}", req.method())), + KeyValue::new("uri", req.uri().to_string()), + ]) + .start(&tracer); + + let res = self + .handler_stage2(req) + .with_context(Context::current_with_span(span)) + .await; + + match res { + Ok(x) => { + debug!("{} {:?}", x.status(), x.headers()); + Ok(x) + } + Err(e) => { + let body = e.http_body(&self.region, uri.path()); + let mut http_error_builder = Response::builder().status(e.http_status_code()); + + if let Some(header_map) = http_error_builder.headers_mut() { + e.add_http_headers(header_map) + } + + let http_error = http_error_builder.body(body)?; + + if e.http_status_code().is_server_error() { + warn!("Response: error {}, {}", e.http_status_code(), e); + } else { + info!("Response: error {}, {}", e.http_status_code(), e); + } + Ok(http_error + .map(|body| BoxBody::new(body.map_err(|_: Infallible| unreachable!())))) + } + } + } + + async fn handler_stage2( + &self, + req: Request<IncomingBody>, + ) -> Result<Response<BoxBody<A::Error>>, A::Error> { + let endpoint = self.api_handler.parse_endpoint(&req)?; + debug!("Endpoint: {}", endpoint.name()); + + let current_context = Context::current(); + let current_span = current_context.span(); + current_span.update_name::<String>(format!( + "{} API {}", + A::API_NAME_DISPLAY, + endpoint.name() + )); + current_span.set_attribute(KeyValue::new("endpoint", endpoint.name())); + endpoint.add_span_attributes(current_span); + + let metrics_tags = &[KeyValue::new("api_endpoint", endpoint.name())]; + + let res = self + .api_handler + .handle(req, endpoint) + .record_duration(&self.request_duration, &metrics_tags[..]) + .await; + + self.request_counter.add(1, &metrics_tags[..]); + + let status_code = match &res { + Ok(r) => r.status(), + Err(e) => e.http_status_code(), + }; + if status_code.is_client_error() || status_code.is_server_error() { + self.error_counter.add( + 1, + &[ + metrics_tags[0].clone(), + KeyValue::new("status_code", status_code.as_str().to_string()), + ], + ); + } + + res + } +} + +// ==== helper functions ==== + +#[async_trait] +pub trait Accept: Send + Sync + 'static { + type Stream: AsyncRead + AsyncWrite + Send + Sync + 'static; + async fn accept(&self) -> std::io::Result<(Self::Stream, String)>; +} + +#[async_trait] +impl Accept for TcpListener { + type Stream = TcpStream; + async fn accept(&self) -> std::io::Result<(Self::Stream, String)> { + self.accept() + .await + .map(|(stream, addr)| (stream, addr.to_string())) + } +} + +pub struct UnixListenerOn(pub UnixListener, pub String); + +#[async_trait] +impl Accept for UnixListenerOn { + type Stream = UnixStream; + async fn accept(&self) -> std::io::Result<(Self::Stream, String)> { + self.0 + .accept() + .await + .map(|(stream, _addr)| (stream, self.1.clone())) + } +} + +pub async fn server_loop<A, H, F, E>( + server_name: String, + listener: A, + handler: H, + mut must_exit: watch::Receiver<bool>, +) -> Result<(), GarageError> +where + A: Accept, + H: Fn(Request<IncomingBody>, String) -> F + Send + Sync + Clone + 'static, + F: Future<Output = Result<Response<BoxBody<E>>, http::Error>> + Send + 'static, + E: Send + Sync + std::error::Error + 'static, +{ + let (conn_in, mut conn_out) = tokio::sync::mpsc::unbounded_channel(); + let connection_collector = tokio::spawn({ + let server_name = server_name.clone(); + async move { + let mut connections = FuturesUnordered::<tokio::task::JoinHandle<()>>::new(); + loop { + let collect_next = async { + if connections.is_empty() { + futures::future::pending().await + } else { + connections.next().await + } + }; + tokio::select! { + result = collect_next => { + trace!("{} server: HTTP connection finished: {:?}", server_name, result); + } + new_fut = conn_out.recv() => { + match new_fut { + Some(f) => connections.push(f), + None => break, + } + } + } + } + let deadline = Instant::now() + Duration::from_secs(10); + while !connections.is_empty() { + info!( + "{} server: {} connections still open, deadline in {:.2}s", + server_name, + connections.len(), + (deadline - Instant::now()).as_secs_f32(), + ); + tokio::select! { + conn_res = connections.next() => { + trace!( + "{} server: HTTP connection finished: {:?}", + server_name, + conn_res.unwrap(), + ); + } + _ = sleep_until(deadline) => { + warn!("{} server: exit deadline reached with {} connections still open, killing them now", + server_name, + connections.len()); + for conn in connections.iter() { + conn.abort(); + } + for conn in connections { + assert!(conn.await.unwrap_err().is_cancelled()); + } + break; + } + } + } + } + }); + + while !*must_exit.borrow() { + let (stream, client_addr) = tokio::select! { + acc = listener.accept() => acc?, + _ = must_exit.changed() => continue, + }; + + let io = TokioIo::new(stream); + + let handler = handler.clone(); + let serve = move |req: Request<IncomingBody>| handler(req, client_addr.clone()); + + let fut = tokio::task::spawn(async move { + let io = Box::pin(io); + if let Err(e) = http1::Builder::new() + .serve_connection(io, service_fn(serve)) + .await + { + debug!("Error handling HTTP connection: {}", e); + } + }); + conn_in.send(fut)?; + } + + info!("{} server exiting", server_name); + drop(conn_in); + connection_collector.await?; + + Ok(()) +} diff --git a/src/api/common/helpers.rs b/src/api/common/helpers.rs new file mode 100644 index 00000000..c8586de4 --- /dev/null +++ b/src/api/common/helpers.rs @@ -0,0 +1,371 @@ +use std::convert::Infallible; +use std::sync::Arc; + +use futures::{Stream, StreamExt, TryStreamExt}; + +use http_body_util::{BodyExt, Full as FullBody}; +use hyper::{ + body::{Body, Bytes}, + Request, Response, +}; +use idna::domain_to_unicode; +use serde::{Deserialize, Serialize}; + +use garage_model::bucket_table::BucketParams; +use garage_model::garage::Garage; +use garage_model::key_table::Key; +use garage_util::data::Uuid; +use garage_util::error::Error as GarageError; + +use crate::common_error::{CommonError as Error, *}; + +/// What kind of authorization is required to perform a given action +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Authorization { + /// No authorization is required + None, + /// Having Read permission on bucket + Read, + /// Having Write permission on bucket + Write, + /// Having Owner permission on bucket + Owner, +} + +/// The values which are known for each request related to a bucket +pub struct ReqCtx { + pub garage: Arc<Garage>, + pub bucket_id: Uuid, + pub bucket_name: String, + pub bucket_params: BucketParams, + pub api_key: Key, +} + +/// Host to bucket +/// +/// Convert a host, like "bucket.garage-site.tld" to the corresponding bucket "bucket", +/// considering that ".garage-site.tld" is the "root domain". For domains not matching +/// the provided root domain, no bucket is returned +/// This behavior has been chosen to follow AWS S3 semantic. +pub fn host_to_bucket<'a>(host: &'a str, root: &str) -> Option<&'a str> { + let root = root.trim_start_matches('.'); + let label_root = root.chars().filter(|c| c == &'.').count() + 1; + let root = root.rsplit('.'); + let mut host = host.rsplitn(label_root + 1, '.'); + for root_part in root { + let host_part = host.next()?; + if root_part != host_part { + return None; + } + } + host.next() +} + +/// Extract host from the authority section given by the HTTP host header +/// +/// The HTTP host contains both a host and a port. +/// Extracting the port is more complex than just finding the colon (:) symbol due to IPv6 +/// We do not use the collect pattern as there is no way in std rust to collect over a stack allocated value +/// check here: <https://docs.rs/collect_slice/1.2.0/collect_slice/> +pub fn authority_to_host(authority: &str) -> Result<String, Error> { + let mut iter = authority.chars().enumerate(); + let (_, first_char) = iter + .next() + .ok_or_else(|| Error::bad_request("Authority is empty".to_string()))?; + + let split = match first_char { + '[' => { + let mut iter = iter.skip_while(|(_, c)| c != &']'); + match iter.next() { + Some((_, ']')) => iter.next(), + _ => { + return Err(Error::bad_request(format!( + "Authority {} has an illegal format", + authority + ))) + } + } + } + _ => iter.find(|(_, c)| *c == ':'), + }; + + let authority = match split { + Some((i, ':')) => Ok(&authority[..i]), + None => Ok(authority), + Some((_, _)) => Err(Error::bad_request(format!( + "Authority {} has an illegal format", + authority + ))), + }; + authority.map(|h| domain_to_unicode(h).0) +} + +/// Extract the bucket name and the key name from an HTTP path and possibly a bucket provided in +/// the host header of the request +/// +/// S3 internally manages only buckets and keys. This function splits +/// an HTTP path to get the corresponding bucket name and key. +pub fn parse_bucket_key<'a>( + path: &'a str, + host_bucket: Option<&'a str>, +) -> Result<(&'a str, Option<&'a str>), Error> { + let path = path.trim_start_matches('/'); + + if let Some(bucket) = host_bucket { + if !path.is_empty() { + return Ok((bucket, Some(path))); + } else { + return Ok((bucket, None)); + } + } + + let (bucket, key) = match path.find('/') { + Some(i) => { + let key = &path[i + 1..]; + if !key.is_empty() { + (&path[..i], Some(key)) + } else { + (&path[..i], None) + } + } + None => (path, None), + }; + if bucket.is_empty() { + return Err(Error::bad_request("No bucket specified")); + } + Ok((bucket, key)) +} + +const UTF8_BEFORE_LAST_CHAR: char = '\u{10FFFE}'; + +/// Compute the key after the prefix +pub fn key_after_prefix(pfx: &str) -> Option<String> { + let mut next = pfx.to_string(); + while !next.is_empty() { + let tail = next.pop().unwrap(); + if tail >= char::MAX { + continue; + } + + // Circumvent a limitation of RangeFrom that overflow earlier than needed + // See: https://doc.rust-lang.org/core/ops/struct.RangeFrom.html + let new_tail = if tail == UTF8_BEFORE_LAST_CHAR { + char::MAX + } else { + (tail..).nth(1).unwrap() + }; + + next.push(new_tail); + return Some(next); + } + + None +} + +// =============== body helpers ================= + +pub type EmptyBody = http_body_util::Empty<bytes::Bytes>; +pub type ErrorBody = FullBody<bytes::Bytes>; +pub type BoxBody<E> = http_body_util::combinators::BoxBody<bytes::Bytes, E>; + +pub fn string_body<E>(s: String) -> BoxBody<E> { + bytes_body(bytes::Bytes::from(s.into_bytes())) +} +pub fn bytes_body<E>(b: bytes::Bytes) -> BoxBody<E> { + BoxBody::new(FullBody::new(b).map_err(|_: Infallible| unreachable!())) +} +pub fn empty_body<E>() -> BoxBody<E> { + BoxBody::new(http_body_util::Empty::new().map_err(|_: Infallible| unreachable!())) +} +pub fn error_body(s: String) -> ErrorBody { + ErrorBody::from(bytes::Bytes::from(s.into_bytes())) +} + +pub async fn parse_json_body<T, B, E>(req: Request<B>) -> Result<T, E> +where + T: for<'de> Deserialize<'de>, + B: Body, + E: From<<B as Body>::Error> + From<Error>, +{ + let body = req.into_body().collect().await?.to_bytes(); + let resp: T = serde_json::from_slice(&body).ok_or_bad_request("Invalid JSON")?; + Ok(resp) +} + +pub fn json_ok_response<E, T: Serialize>(res: &T) -> Result<Response<BoxBody<E>>, E> +where + E: From<Error>, +{ + let resp_json = serde_json::to_string_pretty(res) + .map_err(GarageError::from) + .map_err(Error::from)?; + Ok(Response::builder() + .status(hyper::StatusCode::OK) + .header(http::header::CONTENT_TYPE, "application/json") + .body(string_body(resp_json)) + .unwrap()) +} + +pub fn body_stream<B, E>(body: B) -> impl Stream<Item = Result<Bytes, E>> +where + B: Body<Data = Bytes>, + <B as Body>::Error: Into<E>, + E: From<Error>, +{ + let stream = http_body_util::BodyStream::new(body); + let stream = TryStreamExt::map_err(stream, Into::into); + stream.map(|x| { + x.and_then(|f| { + f.into_data() + .map_err(|_| E::from(Error::bad_request("non-data frame"))) + }) + }) +} + +pub fn is_default<T: Default + PartialEq>(v: &T) -> bool { + *v == T::default() +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn parse_bucket_containing_a_key() -> Result<(), Error> { + let (bucket, key) = parse_bucket_key("/my_bucket/a/super/file.jpg", None)?; + assert_eq!(bucket, "my_bucket"); + assert_eq!(key.expect("key must be set"), "a/super/file.jpg"); + Ok(()) + } + + #[test] + fn parse_bucket_containing_no_key() -> Result<(), Error> { + let (bucket, key) = parse_bucket_key("/my_bucket/", None)?; + assert_eq!(bucket, "my_bucket"); + assert!(key.is_none()); + let (bucket, key) = parse_bucket_key("/my_bucket", None)?; + assert_eq!(bucket, "my_bucket"); + assert!(key.is_none()); + Ok(()) + } + + #[test] + fn parse_bucket_containing_no_bucket() { + let parsed = parse_bucket_key("", None); + assert!(parsed.is_err()); + let parsed = parse_bucket_key("/", None); + assert!(parsed.is_err()); + let parsed = parse_bucket_key("////", None); + assert!(parsed.is_err()); + } + + #[test] + fn parse_bucket_with_vhost_and_key() -> Result<(), Error> { + let (bucket, key) = parse_bucket_key("/a/super/file.jpg", Some("my-bucket"))?; + assert_eq!(bucket, "my-bucket"); + assert_eq!(key.expect("key must be set"), "a/super/file.jpg"); + Ok(()) + } + + #[test] + fn parse_bucket_with_vhost_no_key() -> Result<(), Error> { + let (bucket, key) = parse_bucket_key("", Some("my-bucket"))?; + assert_eq!(bucket, "my-bucket"); + assert!(key.is_none()); + let (bucket, key) = parse_bucket_key("/", Some("my-bucket"))?; + assert_eq!(bucket, "my-bucket"); + assert!(key.is_none()); + Ok(()) + } + + #[test] + fn authority_to_host_with_port() -> Result<(), Error> { + let domain = authority_to_host("[::1]:3902")?; + assert_eq!(domain, "[::1]"); + let domain2 = authority_to_host("garage.tld:65200")?; + assert_eq!(domain2, "garage.tld"); + let domain3 = authority_to_host("127.0.0.1:80")?; + assert_eq!(domain3, "127.0.0.1"); + Ok(()) + } + + #[test] + fn authority_to_host_without_port() -> Result<(), Error> { + let domain = authority_to_host("[::1]")?; + assert_eq!(domain, "[::1]"); + let domain2 = authority_to_host("garage.tld")?; + assert_eq!(domain2, "garage.tld"); + let domain3 = authority_to_host("127.0.0.1")?; + assert_eq!(domain3, "127.0.0.1"); + assert!(authority_to_host("[").is_err()); + assert!(authority_to_host("[hello").is_err()); + Ok(()) + } + + #[test] + fn host_to_bucket_test() { + assert_eq!( + host_to_bucket("john.doe.garage.tld", ".garage.tld").unwrap(), + "john.doe" + ); + + assert_eq!( + host_to_bucket("john.doe.garage.tld", "garage.tld").unwrap(), + "john.doe" + ); + + assert_eq!(host_to_bucket("john.doe.com", "garage.tld"), None); + + assert_eq!(host_to_bucket("john.doe.com", ".garage.tld"), None); + + assert_eq!(host_to_bucket("garage.tld", "garage.tld"), None); + + assert_eq!(host_to_bucket("garage.tld", ".garage.tld"), None); + + assert_eq!(host_to_bucket("not-garage.tld", "garage.tld"), None); + assert_eq!(host_to_bucket("not-garage.tld", ".garage.tld"), None); + } + + #[test] + fn test_key_after_prefix() { + use std::iter::FromIterator; + + assert_eq!(UTF8_BEFORE_LAST_CHAR as u32, (char::MAX as u32) - 1); + assert_eq!(key_after_prefix("a/b/").unwrap().as_str(), "a/b0"); + assert_eq!(key_after_prefix("€").unwrap().as_str(), "₭"); + assert_eq!( + key_after_prefix("").unwrap().as_str(), + String::from(char::from_u32(0x10FFFE).unwrap()) + ); + + // When the last character is the biggest UTF8 char + let a = String::from_iter(['a', char::MAX].iter()); + assert_eq!(key_after_prefix(a.as_str()).unwrap().as_str(), "b"); + + // When all characters are the biggest UTF8 char + let b = String::from_iter([char::MAX; 3].iter()); + assert!(key_after_prefix(b.as_str()).is_none()); + + // Check utf8 surrogates + let c = String::from('\u{D7FF}'); + assert_eq!( + key_after_prefix(c.as_str()).unwrap().as_str(), + String::from('\u{E000}') + ); + + // Check the character before the biggest one + let d = String::from('\u{10FFFE}'); + assert_eq!( + key_after_prefix(d.as_str()).unwrap().as_str(), + String::from(char::MAX) + ); + } +} + +#[derive(Serialize)] +pub struct CustomApiErrorBody { + pub code: String, + pub message: String, + pub region: String, + pub path: String, +} diff --git a/src/api/common/lib.rs b/src/api/common/lib.rs new file mode 100644 index 00000000..49d463d7 --- /dev/null +++ b/src/api/common/lib.rs @@ -0,0 +1,12 @@ +//! Crate for serving a S3 compatible API +#[macro_use] +extern crate tracing; + +pub mod common_error; + +pub mod encoding; +pub mod generic_server; +pub mod helpers; +pub mod router_macros; +/// This mode is public only to help testing. Don't expect stability here +pub mod signature; diff --git a/src/api/common/router_macros.rs b/src/api/common/router_macros.rs new file mode 100644 index 00000000..d9fe86db --- /dev/null +++ b/src/api/common/router_macros.rs @@ -0,0 +1,226 @@ +/// This macro is used to generate very repetitive match {} blocks in this module +/// It is _not_ made to be used anywhere else +#[macro_export] +macro_rules! router_match { + (@match $enum:expr , [ $($endpoint:ident,)* ]) => {{ + // usage: router_match {@match my_enum, [ VariantWithField1, VariantWithField2 ..] } + // returns true if the variant was one of the listed variants, false otherwise. + match $enum { + $( + Endpoint::$endpoint { .. } => true, + )* + _ => false + } + }}; + (@extract $enum:expr , $param:ident, [ $($endpoint:ident,)* ]) => {{ + // usage: router_match {@extract my_enum, field_name, [ VariantWithField1, VariantWithField2 ..] } + // returns Some(field_value), or None if the variant was not one of the listed variants. + match $enum { + $( + Endpoint::$endpoint {$param, ..} => Some($param), + )* + _ => None + } + }}; + (@gen_path_parser ($method:expr, $reqpath:expr, $query:expr) + [ + $($meth:ident $path:pat $(if $required:ident)? => $api:ident $(($($conv:ident :: $param:ident),*))?,)* + ]) => {{ + { + #[allow(unused_parens)] + match ($method, $reqpath) { + $( + (&Method::$meth, $path) if true $(&& $query.$required.is_some())? => Endpoint::$api { + $($( + $param: router_match!(@@parse_param $query, $conv, $param), + )*)? + }, + )* + (m, p) => { + return Err(Error::bad_request(format!( + "Unknown API endpoint: {} {}", + m, p + ))) + } + } + } + }}; + (@gen_parser ($keyword:expr, $key:ident, $query:expr, $header:expr), + key: [$($kw_k:ident $(if $required_k:ident)? $(header $header_k:expr)? => $api_k:ident $(($($conv_k:ident :: $param_k:ident),*))?,)*], + no_key: [$($kw_nk:ident $(if $required_nk:ident)? $(if_header $header_nk:expr)? => $api_nk:ident $(($($conv_nk:ident :: $param_nk:ident),*))?,)*]) => {{ + // usage: router_match {@gen_parser (keyword, key, query, header), + // key: [ + // SOME_KEYWORD => VariantWithKey, + // ... + // ], + // no_key: [ + // SOME_KEYWORD => VariantWithoutKey, + // ... + // ] + // } + // See in from_{method} for more detailed usage. + match ($keyword, !$key.is_empty()){ + $( + (Keyword::$kw_k, true) if true $(&& $query.$required_k.is_some())? $(&& $header.contains_key($header_k))? => Ok(Endpoint::$api_k { + $key, + $($( + $param_k: router_match!(@@parse_param $query, $conv_k, $param_k), + )*)? + }), + )* + $( + (Keyword::$kw_nk, false) $(if $query.$required_nk.is_some())? $(if $header.contains($header_nk))? => Ok(Endpoint::$api_nk { + $($( + $param_nk: router_match!(@@parse_param $query, $conv_nk, $param_nk), + )*)? + }), + )* + (kw, _) => Err(Error::bad_request(format!("Invalid endpoint: {}", kw))) + } + }}; + + (@@parse_param $query:expr, query_opt, $param:ident) => {{ + // extract optional query parameter + $query.$param.take().map(|param| param.into_owned()) + }}; + (@@parse_param $query:expr, query, $param:ident) => {{ + // extract mendatory query parameter + $query.$param.take().ok_or_bad_request("Missing argument for endpoint")?.into_owned() + }}; + (@@parse_param $query:expr, opt_parse, $param:ident) => {{ + // extract and parse optional query parameter + // missing parameter is file, however parse error is reported as an error + $query.$param + .take() + .map(|param| param.parse()) + .transpose() + .map_err(|_| Error::bad_request("Failed to parse query parameter"))? + }}; + (@@parse_param $query:expr, parse, $param:ident) => {{ + // extract and parse mandatory query parameter + // both missing and un-parseable parameters are reported as errors + $query.$param.take().ok_or_bad_request("Missing argument for endpoint")? + .parse() + .map_err(|_| Error::bad_request("Failed to parse query parameter"))? + }}; + (@func + $(#[$doc:meta])* + pub enum Endpoint { + $( + $(#[$outer:meta])* + $variant:ident $({ + $($name:ident: $ty:ty,)* + })?, + )* + }) => { + $(#[$doc])* + pub enum Endpoint { + $( + $(#[$outer])* + $variant $({ + $($name: $ty, )* + })?, + )* + } + impl Endpoint { + pub fn name(&self) -> &'static str { + match self { + $(Endpoint::$variant $({ $($name: _,)* .. })? => stringify!($variant),)* + } + } + } + }; +} + +/// This macro is used to generate part of the code in this module. It must be called only one, and +/// is useless outside of this module. +#[macro_export] +macro_rules! generateQueryParameters { + ( + keywords: [ $($kw_param:expr => $kw_name: ident),* ], + fields: [ $($f_param:expr => $f_name:ident),* ] + ) => { + #[derive(Debug)] + #[allow(non_camel_case_types)] + #[allow(clippy::upper_case_acronyms)] + enum Keyword { + EMPTY, + $( $kw_name, )* + } + + impl std::fmt::Display for Keyword { + fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result { + match self { + Keyword::EMPTY => write!(f, "``"), + $( Keyword::$kw_name => write!(f, "`{}`", $kw_param), )* + } + } + } + + impl Default for Keyword { + fn default() -> Self { + Keyword::EMPTY + } + } + + /// Struct containing all query parameters used in endpoints. Think of it as an HashMap, + /// but with keys statically known. + #[derive(Debug, Default)] + struct QueryParameters<'a> { + keyword: Option<Keyword>, + $( + $f_name: Option<Cow<'a, str>>, + )* + } + + impl<'a> QueryParameters<'a> { + /// Build this struct from the query part of an URI. + fn from_query(query: &'a str) -> Result<Self, Error> { + let mut res: Self = Default::default(); + for (k, v) in url::form_urlencoded::parse(query.as_bytes()) { + match k.as_ref() { + $( + $kw_param => if let Some(prev_kw) = res.keyword.replace(Keyword::$kw_name) { + return Err(Error::bad_request(format!( + "Multiple keywords: '{}' and '{}'", prev_kw, $kw_param + ))); + }, + )* + $( + $f_param => if !v.is_empty() { + if res.$f_name.replace(v).is_some() { + return Err(Error::bad_request(format!( + "Query parameter repeated: '{}'", k + ))); + } + }, + )* + _ => { + if !(k.starts_with("response-") || k.starts_with("X-Amz-")) { + debug!("Received an unknown query parameter: '{}'", k); + } + } + }; + } + Ok(res) + } + + /// Get an error message in case not all parameters where used when extracting them to + /// build an Endpoint variant + fn nonempty_message(&self) -> Option<&str> { + if self.keyword.is_some() { + Some("Keyword not used") + } $( + else if self.$f_name.is_some() { + Some(concat!("'", $f_param, "'")) + } + )* else { + None + } + } + } + } +} + +pub use generateQueryParameters; +pub use router_match; diff --git a/src/api/common/signature/error.rs b/src/api/common/signature/error.rs new file mode 100644 index 00000000..2d92a072 --- /dev/null +++ b/src/api/common/signature/error.rs @@ -0,0 +1,32 @@ +use err_derive::Error; + +use crate::common_error::CommonError; +pub use crate::common_error::{CommonErrorDerivative, OkOrBadRequest, OkOrInternalError}; + +/// Errors of this crate +#[derive(Debug, Error)] +pub enum Error { + #[error(display = "{}", _0)] + /// Error from common error + Common(CommonError), + + /// Authorization Header Malformed + #[error(display = "Authorization header malformed, unexpected scope: {}", _0)] + AuthorizationHeaderMalformed(String), + + // Category: bad request + /// The request contained an invalid UTF-8 sequence in its path or in other parameters + #[error(display = "Invalid UTF-8: {}", _0)] + InvalidUtf8Str(#[error(source)] std::str::Utf8Error), +} + +impl<T> From<T> for Error +where + CommonError: From<T>, +{ + fn from(err: T) -> Self { + Error::Common(CommonError::from(err)) + } +} + +impl CommonErrorDerivative for Error {} diff --git a/src/api/common/signature/mod.rs b/src/api/common/signature/mod.rs new file mode 100644 index 00000000..6514da43 --- /dev/null +++ b/src/api/common/signature/mod.rs @@ -0,0 +1,78 @@ +use chrono::{DateTime, Utc}; +use hmac::{Hmac, Mac}; +use sha2::Sha256; + +use hyper::{body::Incoming as IncomingBody, Request}; + +use garage_model::garage::Garage; +use garage_model::key_table::Key; +use garage_util::data::{sha256sum, Hash}; + +use error::*; + +pub mod error; +pub mod payload; +pub mod streaming; + +pub const SHORT_DATE: &str = "%Y%m%d"; +pub const LONG_DATETIME: &str = "%Y%m%dT%H%M%SZ"; + +type HmacSha256 = Hmac<Sha256>; + +pub async fn verify_request( + garage: &Garage, + mut req: Request<IncomingBody>, + service: &'static str, +) -> Result<(Request<streaming::ReqBody>, Key, Option<Hash>), Error> { + let (api_key, mut content_sha256) = + payload::check_payload_signature(&garage, &mut req, service).await?; + let api_key = + api_key.ok_or_else(|| Error::forbidden("Garage does not support anonymous access yet"))?; + + let req = streaming::parse_streaming_body( + &api_key, + req, + &mut content_sha256, + &garage.config.s3_api.s3_region, + service, + )?; + + Ok((req, api_key, content_sha256)) +} + +pub fn verify_signed_content(expected_sha256: Hash, body: &[u8]) -> Result<(), Error> { + if expected_sha256 != sha256sum(body) { + return Err(Error::bad_request( + "Request content hash does not match signed hash".to_string(), + )); + } + Ok(()) +} + +pub fn signing_hmac( + datetime: &DateTime<Utc>, + secret_key: &str, + region: &str, + service: &str, +) -> Result<HmacSha256, crypto_common::InvalidLength> { + let secret = String::from("AWS4") + secret_key; + let mut date_hmac = HmacSha256::new_from_slice(secret.as_bytes())?; + date_hmac.update(datetime.format(SHORT_DATE).to_string().as_bytes()); + let mut region_hmac = HmacSha256::new_from_slice(&date_hmac.finalize().into_bytes())?; + region_hmac.update(region.as_bytes()); + let mut service_hmac = HmacSha256::new_from_slice(®ion_hmac.finalize().into_bytes())?; + service_hmac.update(service.as_bytes()); + let mut signing_hmac = HmacSha256::new_from_slice(&service_hmac.finalize().into_bytes())?; + signing_hmac.update(b"aws4_request"); + let hmac = HmacSha256::new_from_slice(&signing_hmac.finalize().into_bytes())?; + Ok(hmac) +} + +pub fn compute_scope(datetime: &DateTime<Utc>, region: &str, service: &str) -> String { + format!( + "{}/{}/{}/aws4_request", + datetime.format(SHORT_DATE), + region, + service + ) +} diff --git a/src/api/common/signature/payload.rs b/src/api/common/signature/payload.rs new file mode 100644 index 00000000..81541e4a --- /dev/null +++ b/src/api/common/signature/payload.rs @@ -0,0 +1,562 @@ +use std::collections::HashMap; +use std::convert::TryFrom; + +use chrono::{DateTime, Duration, NaiveDateTime, TimeZone, Utc}; +use hmac::Mac; +use hyper::header::{HeaderMap, HeaderName, HeaderValue, AUTHORIZATION, HOST}; +use hyper::{body::Incoming as IncomingBody, Method, Request}; +use sha2::{Digest, Sha256}; + +use garage_table::*; +use garage_util::data::Hash; + +use garage_model::garage::Garage; +use garage_model::key_table::*; + +use super::LONG_DATETIME; +use super::{compute_scope, signing_hmac}; + +use crate::encoding::uri_encode; +use crate::signature::error::*; + +pub const X_AMZ_ALGORITHM: HeaderName = HeaderName::from_static("x-amz-algorithm"); +pub const X_AMZ_CREDENTIAL: HeaderName = HeaderName::from_static("x-amz-credential"); +pub const X_AMZ_DATE: HeaderName = HeaderName::from_static("x-amz-date"); +pub const X_AMZ_EXPIRES: HeaderName = HeaderName::from_static("x-amz-expires"); +pub const X_AMZ_SIGNEDHEADERS: HeaderName = HeaderName::from_static("x-amz-signedheaders"); +pub const X_AMZ_SIGNATURE: HeaderName = HeaderName::from_static("x-amz-signature"); +pub const X_AMZ_CONTENT_SH256: HeaderName = HeaderName::from_static("x-amz-content-sha256"); + +pub const AWS4_HMAC_SHA256: &str = "AWS4-HMAC-SHA256"; +pub const UNSIGNED_PAYLOAD: &str = "UNSIGNED-PAYLOAD"; +pub const STREAMING_AWS4_HMAC_SHA256_PAYLOAD: &str = "STREAMING-AWS4-HMAC-SHA256-PAYLOAD"; + +pub type QueryMap = HeaderMap<QueryValue>; +pub struct QueryValue { + /// Original key with potential uppercase characters, + /// for use in signature calculation + key: String, + value: String, +} + +pub async fn check_payload_signature( + garage: &Garage, + request: &mut Request<IncomingBody>, + service: &'static str, +) -> Result<(Option<Key>, Option<Hash>), Error> { + let query = parse_query_map(request.uri())?; + + if query.contains_key(&X_AMZ_ALGORITHM) { + // We check for presigned-URL-style authentication first, because + // the browser or something else could inject an Authorization header + // that is totally unrelated to AWS signatures. + check_presigned_signature(garage, service, request, query).await + } else if request.headers().contains_key(AUTHORIZATION) { + check_standard_signature(garage, service, request, query).await + } else { + // Unsigned (anonymous) request + let content_sha256 = request + .headers() + .get("x-amz-content-sha256") + .filter(|c| c.as_bytes() != UNSIGNED_PAYLOAD.as_bytes()); + if let Some(content_sha256) = content_sha256 { + let sha256 = hex::decode(content_sha256) + .ok() + .and_then(|bytes| Hash::try_from(&bytes)) + .ok_or_bad_request("Invalid content sha256 hash")?; + Ok((None, Some(sha256))) + } else { + Ok((None, None)) + } + } +} + +async fn check_standard_signature( + garage: &Garage, + service: &'static str, + request: &Request<IncomingBody>, + query: QueryMap, +) -> Result<(Option<Key>, Option<Hash>), Error> { + let authorization = Authorization::parse_header(request.headers())?; + + // Verify that all necessary request headers are included in signed_headers + // The following must be included for all signatures: + // - the Host header (mandatory) + // - all x-amz-* headers used in the request + // AWS also indicates that the Content-Type header should be signed if + // it is used, but Minio client doesn't sign it so we don't check it for compatibility. + let signed_headers = split_signed_headers(&authorization)?; + verify_signed_headers(request.headers(), &signed_headers)?; + + let canonical_request = canonical_request( + service, + request.method(), + request.uri().path(), + &query, + request.headers(), + &signed_headers, + &authorization.content_sha256, + )?; + let string_to_sign = string_to_sign( + &authorization.date, + &authorization.scope, + &canonical_request, + ); + + trace!("canonical request:\n{}", canonical_request); + trace!("string to sign:\n{}", string_to_sign); + + let key = verify_v4(garage, service, &authorization, string_to_sign.as_bytes()).await?; + + let content_sha256 = if authorization.content_sha256 == UNSIGNED_PAYLOAD { + None + } else if authorization.content_sha256 == STREAMING_AWS4_HMAC_SHA256_PAYLOAD { + let bytes = hex::decode(authorization.signature).ok_or_bad_request("Invalid signature")?; + Some(Hash::try_from(&bytes).ok_or_bad_request("Invalid signature")?) + } else { + let bytes = hex::decode(authorization.content_sha256) + .ok_or_bad_request("Invalid content sha256 hash")?; + Some(Hash::try_from(&bytes).ok_or_bad_request("Invalid content sha256 hash")?) + }; + + Ok((Some(key), content_sha256)) +} + +async fn check_presigned_signature( + garage: &Garage, + service: &'static str, + request: &mut Request<IncomingBody>, + mut query: QueryMap, +) -> Result<(Option<Key>, Option<Hash>), Error> { + let algorithm = query.get(&X_AMZ_ALGORITHM).unwrap(); + let authorization = Authorization::parse_presigned(&algorithm.value, &query)?; + + // Verify that all necessary request headers are included in signed_headers + // For AWSv4 pre-signed URLs, the following must be included: + // - the Host header (mandatory) + // - all x-amz-* headers used in the request + let signed_headers = split_signed_headers(&authorization)?; + verify_signed_headers(request.headers(), &signed_headers)?; + + // The X-Amz-Signature value is passed as a query parameter, + // but the signature cannot be computed from a string that contains itself. + // AWS specifies that all query params except X-Amz-Signature are included + // in the canonical request. + query.remove(&X_AMZ_SIGNATURE); + let canonical_request = canonical_request( + service, + request.method(), + request.uri().path(), + &query, + request.headers(), + &signed_headers, + &authorization.content_sha256, + )?; + let string_to_sign = string_to_sign( + &authorization.date, + &authorization.scope, + &canonical_request, + ); + + trace!("canonical request (presigned url):\n{}", canonical_request); + trace!("string to sign (presigned url):\n{}", string_to_sign); + + let key = verify_v4(garage, service, &authorization, string_to_sign.as_bytes()).await?; + + // In the page on presigned URLs, AWS specifies that if a signed query + // parameter and a signed header of the same name have different values, + // then an InvalidRequest error is raised. + let headers_mut = request.headers_mut(); + for (name, value) in query.iter() { + if let Some(existing) = headers_mut.get(name) { + if signed_headers.contains(&name) && existing.as_bytes() != value.value.as_bytes() { + return Err(Error::bad_request(format!( + "Conflicting values for `{}` in query parameters and request headers", + name + ))); + } + } + if name.as_str().starts_with("x-amz-") { + // Query parameters that start by x-amz- are actually intended to stand in for + // headers that can't be added at the time the request is made. + // What we do is just add them to the Request object as regular headers, + // that will be handled downstream as if they were included like in a normal request. + // (Here we allow such query parameters to override headers with the same name + // that are not signed, however there is not much reason that this would happen) + headers_mut.insert( + name, + HeaderValue::from_bytes(value.value.as_bytes()) + .ok_or_bad_request("invalid query parameter value")?, + ); + } + } + + // Presigned URLs always use UNSIGNED-PAYLOAD, + // so there is no sha256 hash to return. + Ok((Some(key), None)) +} + +pub fn parse_query_map(uri: &http::uri::Uri) -> Result<QueryMap, Error> { + let mut query = QueryMap::with_capacity(0); + if let Some(query_str) = uri.query() { + let query_pairs = url::form_urlencoded::parse(query_str.as_bytes()); + for (key, val) in query_pairs { + let name = + HeaderName::from_bytes(key.as_bytes()).ok_or_bad_request("Invalid header name")?; + + let value = QueryValue { + key: key.to_string(), + value: val.into_owned(), + }; + + if query.insert(name, value).is_some() { + return Err(Error::bad_request(format!( + "duplicate query parameter: `{}`", + key + ))); + } + } + } + Ok(query) +} + +fn parse_credential(cred: &str) -> Result<(String, String), Error> { + let first_slash = cred + .find('/') + .ok_or_bad_request("Credentials does not contain '/' in authorization field")?; + let (key_id, scope) = cred.split_at(first_slash); + Ok(( + key_id.to_string(), + scope.trim_start_matches('/').to_string(), + )) +} + +fn split_signed_headers(authorization: &Authorization) -> Result<Vec<HeaderName>, Error> { + let mut signed_headers = authorization + .signed_headers + .split(';') + .map(HeaderName::try_from) + .collect::<Result<Vec<HeaderName>, _>>() + .ok_or_bad_request("invalid header name")?; + signed_headers.sort_by(|h1, h2| h1.as_str().cmp(h2.as_str())); + Ok(signed_headers) +} + +fn verify_signed_headers(headers: &HeaderMap, signed_headers: &[HeaderName]) -> Result<(), Error> { + if !signed_headers.contains(&HOST) { + return Err(Error::bad_request("Header `Host` should be signed")); + } + for (name, _) in headers.iter() { + if name.as_str().starts_with("x-amz-") { + if !signed_headers.contains(name) { + return Err(Error::bad_request(format!( + "Header `{}` should be signed", + name + ))); + } + } + } + Ok(()) +} + +pub fn string_to_sign(datetime: &DateTime<Utc>, scope_string: &str, canonical_req: &str) -> String { + let mut hasher = Sha256::default(); + hasher.update(canonical_req.as_bytes()); + [ + AWS4_HMAC_SHA256, + &datetime.format(LONG_DATETIME).to_string(), + scope_string, + &hex::encode(hasher.finalize().as_slice()), + ] + .join("\n") +} + +pub fn canonical_request( + service: &'static str, + method: &Method, + canonical_uri: &str, + query: &QueryMap, + headers: &HeaderMap, + signed_headers: &[HeaderName], + content_sha256: &str, +) -> Result<String, Error> { + // There seems to be evidence that in AWSv4 signatures, the path component is url-encoded + // a second time when building the canonical request, as specified in this documentation page: + // -> https://docs.aws.amazon.com/rolesanywhere/latest/userguide/authentication-sign-process.html + // However this documentation page is for a specific service ("roles anywhere"), and + // in the S3 service we know for a fact that there is no double-urlencoding, because all of + // the tests we made with external software work without it. + // + // The theory is that double-urlencoding occurs for all services except S3, + // which is what is implemented in rusoto_signature: + // -> https://docs.rs/rusoto_signature/latest/src/rusoto_signature/signature.rs.html#464 + // + // Digging into the code of the official AWS Rust SDK, we learn that double-URI-encoding can + // be set or unset on a per-request basis (the signature crates, aws-sigv4 and aws-sig-auth, + // are agnostic to this). Grepping the codebase confirms that S3 is the only API for which + // double_uri_encode is set to false, meaning it is true (its default value) for all other + // AWS services. We will therefore implement this behavior in Garage as well. + // + // Note that this documentation page, which is touted as the "authoritative reference" on + // AWSv4 signatures, makes no mention of either single- or double-urlencoding: + // -> https://docs.aws.amazon.com/IAM/latest/UserGuide/create-signed-request.html + // This page of the S3 documentation does also not mention anything specific: + // -> https://docs.aws.amazon.com/AmazonS3/latest/API/sig-v4-header-based-auth.html + // + // Note that there is also the issue of path normalization, which I hope is unrelated to the + // one of URI-encoding. At least in aws-sigv4 both parameters can be set independently, + // and rusoto_signature does not seem to do any effective path normalization, even though + // it mentions it in the comments (same link to the source code as above). + // We make the explicit choice of NOT normalizing paths in the K2V API because doing so + // would make non-normalized paths invalid K2V partition keys, and we don't want that. + let canonical_uri: std::borrow::Cow<str> = if service != "s3" { + uri_encode(canonical_uri, false).into() + } else { + canonical_uri.into() + }; + + // Canonical query string from passed HeaderMap + let canonical_query_string = { + let mut items = Vec::with_capacity(query.len()); + for (_, QueryValue { key, value }) in query.iter() { + items.push(uri_encode(&key, true) + "=" + &uri_encode(&value, true)); + } + items.sort(); + items.join("&") + }; + + // Canonical header string calculated from signed headers + let canonical_header_string = signed_headers + .iter() + .map(|name| { + let value = headers + .get(name) + .ok_or_bad_request(format!("signed header `{}` is not present", name))?; + let value = std::str::from_utf8(value.as_bytes())?; + Ok(format!("{}:{}", name.as_str(), value.trim())) + }) + .collect::<Result<Vec<String>, Error>>()? + .join("\n"); + let signed_headers = signed_headers.join(";"); + + let list = [ + method.as_str(), + &canonical_uri, + &canonical_query_string, + &canonical_header_string, + "", + &signed_headers, + content_sha256, + ]; + Ok(list.join("\n")) +} + +pub fn parse_date(date: &str) -> Result<DateTime<Utc>, Error> { + let date: NaiveDateTime = + NaiveDateTime::parse_from_str(date, LONG_DATETIME).ok_or_bad_request("Invalid date")?; + Ok(Utc.from_utc_datetime(&date)) +} + +pub async fn verify_v4( + garage: &Garage, + service: &str, + auth: &Authorization, + payload: &[u8], +) -> Result<Key, Error> { + let scope_expected = compute_scope(&auth.date, &garage.config.s3_api.s3_region, service); + if auth.scope != scope_expected { + return Err(Error::AuthorizationHeaderMalformed(auth.scope.to_string())); + } + + let key = garage + .key_table + .get(&EmptyKey, &auth.key_id) + .await? + .filter(|k| !k.state.is_deleted()) + .ok_or_else(|| Error::forbidden(format!("No such key: {}", &auth.key_id)))?; + let key_p = key.params().unwrap(); + + let mut hmac = signing_hmac( + &auth.date, + &key_p.secret_key, + &garage.config.s3_api.s3_region, + service, + ) + .ok_or_internal_error("Unable to build signing HMAC")?; + hmac.update(payload); + let signature = + hex::decode(&auth.signature).map_err(|_| Error::forbidden("Invalid signature"))?; + if hmac.verify_slice(&signature).is_err() { + return Err(Error::forbidden("Invalid signature")); + } + + Ok(key) +} + +// ============ Authorization header, or X-Amz-* query params ========= + +pub struct Authorization { + key_id: String, + scope: String, + signed_headers: String, + signature: String, + content_sha256: String, + date: DateTime<Utc>, +} + +impl Authorization { + fn parse_header(headers: &HeaderMap) -> Result<Self, Error> { + let authorization = headers + .get(AUTHORIZATION) + .ok_or_bad_request("Missing authorization header")? + .to_str()?; + + let (auth_kind, rest) = authorization + .split_once(' ') + .ok_or_bad_request("Authorization field to short")?; + + if auth_kind != AWS4_HMAC_SHA256 { + return Err(Error::bad_request("Unsupported authorization method")); + } + + let mut auth_params = HashMap::new(); + for auth_part in rest.split(',') { + let auth_part = auth_part.trim(); + let eq = auth_part + .find('=') + .ok_or_bad_request("Field without value in authorization header")?; + let (key, value) = auth_part.split_at(eq); + auth_params.insert(key.to_string(), value.trim_start_matches('=').to_string()); + } + + let cred = auth_params + .get("Credential") + .ok_or_bad_request("Could not find Credential in Authorization field")?; + let signed_headers = auth_params + .get("SignedHeaders") + .ok_or_bad_request("Could not find SignedHeaders in Authorization field")? + .to_string(); + let signature = auth_params + .get("Signature") + .ok_or_bad_request("Could not find Signature in Authorization field")? + .to_string(); + + let content_sha256 = headers + .get(X_AMZ_CONTENT_SH256) + .ok_or_bad_request("Missing X-Amz-Content-Sha256 field")?; + + let date = headers + .get(X_AMZ_DATE) + .ok_or_bad_request("Missing X-Amz-Date field") + .map_err(Error::from)? + .to_str()?; + let date = parse_date(date)?; + + if Utc::now() - date > Duration::hours(24) { + return Err(Error::bad_request("Date is too old".to_string())); + } + + let (key_id, scope) = parse_credential(cred)?; + let auth = Authorization { + key_id, + scope, + signed_headers, + signature, + content_sha256: content_sha256.to_str()?.to_string(), + date, + }; + Ok(auth) + } + + fn parse_presigned(algorithm: &str, query: &QueryMap) -> Result<Self, Error> { + if algorithm != AWS4_HMAC_SHA256 { + return Err(Error::bad_request( + "Unsupported authorization method".to_string(), + )); + } + + let cred = query + .get(&X_AMZ_CREDENTIAL) + .ok_or_bad_request("X-Amz-Credential not found in query parameters")?; + let signed_headers = query + .get(&X_AMZ_SIGNEDHEADERS) + .ok_or_bad_request("X-Amz-SignedHeaders not found in query parameters")?; + let signature = query + .get(&X_AMZ_SIGNATURE) + .ok_or_bad_request("X-Amz-Signature not found in query parameters")?; + + let duration = query + .get(&X_AMZ_EXPIRES) + .ok_or_bad_request("X-Amz-Expires not found in query parameters")? + .value + .parse() + .map_err(|_| Error::bad_request("X-Amz-Expires is not a number".to_string()))?; + + if duration > 7 * 24 * 3600 { + return Err(Error::bad_request( + "X-Amz-Expires may not exceed a week".to_string(), + )); + } + + let date = query + .get(&X_AMZ_DATE) + .ok_or_bad_request("Missing X-Amz-Date field")?; + let date = parse_date(&date.value)?; + + if Utc::now() - date > Duration::seconds(duration) { + return Err(Error::bad_request("Date is too old".to_string())); + } + + let (key_id, scope) = parse_credential(&cred.value)?; + Ok(Authorization { + key_id, + scope, + signed_headers: signed_headers.value.clone(), + signature: signature.value.clone(), + content_sha256: UNSIGNED_PAYLOAD.to_string(), + date, + }) + } + + pub fn parse_form(params: &HeaderMap) -> Result<Self, Error> { + let algorithm = params + .get(X_AMZ_ALGORITHM) + .ok_or_bad_request("Missing X-Amz-Algorithm header")? + .to_str()?; + if algorithm != AWS4_HMAC_SHA256 { + return Err(Error::bad_request( + "Unsupported authorization method".to_string(), + )); + } + + let credential = params + .get(X_AMZ_CREDENTIAL) + .ok_or_else(|| Error::forbidden("Garage does not support anonymous access yet"))? + .to_str()?; + let signature = params + .get(X_AMZ_SIGNATURE) + .ok_or_bad_request("No signature was provided")? + .to_str()? + .to_string(); + let date = params + .get(X_AMZ_DATE) + .ok_or_bad_request("No date was provided")? + .to_str()?; + let date = parse_date(date)?; + + if Utc::now() - date > Duration::hours(24) { + return Err(Error::bad_request("Date is too old".to_string())); + } + + let (key_id, scope) = parse_credential(credential)?; + let auth = Authorization { + key_id, + scope, + signed_headers: "".to_string(), + signature, + content_sha256: UNSIGNED_PAYLOAD.to_string(), + date, + }; + Ok(auth) + } +} diff --git a/src/api/common/signature/streaming.rs b/src/api/common/signature/streaming.rs new file mode 100644 index 00000000..e223d1b1 --- /dev/null +++ b/src/api/common/signature/streaming.rs @@ -0,0 +1,373 @@ +use std::pin::Pin; + +use chrono::{DateTime, NaiveDateTime, TimeZone, Utc}; +use futures::prelude::*; +use futures::task; +use garage_model::key_table::Key; +use hmac::Mac; +use http_body_util::StreamBody; +use hyper::body::{Bytes, Incoming as IncomingBody}; +use hyper::Request; + +use garage_util::data::Hash; + +use super::{compute_scope, sha256sum, HmacSha256, LONG_DATETIME}; + +use crate::helpers::*; +use crate::signature::error::*; +use crate::signature::payload::{ + STREAMING_AWS4_HMAC_SHA256_PAYLOAD, X_AMZ_CONTENT_SH256, X_AMZ_DATE, +}; + +pub const AWS4_HMAC_SHA256_PAYLOAD: &str = "AWS4-HMAC-SHA256-PAYLOAD"; + +pub type ReqBody = BoxBody<Error>; + +pub fn parse_streaming_body( + api_key: &Key, + req: Request<IncomingBody>, + content_sha256: &mut Option<Hash>, + region: &str, + service: &str, +) -> Result<Request<ReqBody>, Error> { + match req.headers().get(X_AMZ_CONTENT_SH256) { + Some(header) if header == STREAMING_AWS4_HMAC_SHA256_PAYLOAD => { + let signature = content_sha256 + .take() + .ok_or_bad_request("No signature provided")?; + + let secret_key = &api_key + .state + .as_option() + .ok_or_internal_error("Deleted key state")? + .secret_key; + + let date = req + .headers() + .get(X_AMZ_DATE) + .ok_or_bad_request("Missing X-Amz-Date field")? + .to_str()?; + let date: NaiveDateTime = NaiveDateTime::parse_from_str(date, LONG_DATETIME) + .ok_or_bad_request("Invalid date")?; + let date: DateTime<Utc> = Utc.from_utc_datetime(&date); + + let scope = compute_scope(&date, region, service); + let signing_hmac = crate::signature::signing_hmac(&date, secret_key, region, service) + .ok_or_internal_error("Unable to build signing HMAC")?; + + Ok(req.map(move |body| { + let stream = body_stream::<_, Error>(body); + let signed_payload_stream = + SignedPayloadStream::new(stream, signing_hmac, date, &scope, signature) + .map(|x| x.map(hyper::body::Frame::data)) + .map_err(Error::from); + ReqBody::new(StreamBody::new(signed_payload_stream)) + })) + } + _ => Ok(req.map(|body| ReqBody::new(http_body_util::BodyExt::map_err(body, Error::from)))), + } +} + +/// Result of `sha256("")` +const EMPTY_STRING_HEX_DIGEST: &str = + "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"; + +fn compute_streaming_payload_signature( + signing_hmac: &HmacSha256, + date: DateTime<Utc>, + scope: &str, + previous_signature: Hash, + content_sha256: Hash, +) -> Result<Hash, Error> { + let string_to_sign = [ + AWS4_HMAC_SHA256_PAYLOAD, + &date.format(LONG_DATETIME).to_string(), + scope, + &hex::encode(previous_signature), + EMPTY_STRING_HEX_DIGEST, + &hex::encode(content_sha256), + ] + .join("\n"); + + let mut hmac = signing_hmac.clone(); + hmac.update(string_to_sign.as_bytes()); + + Ok(Hash::try_from(&hmac.finalize().into_bytes()).ok_or_internal_error("Invalid signature")?) +} + +mod payload { + use garage_util::data::Hash; + + pub enum Error<I> { + Parser(nom::error::Error<I>), + BadSignature, + } + + impl<I> Error<I> { + pub fn description(&self) -> &str { + match *self { + Error::Parser(ref e) => e.code.description(), + Error::BadSignature => "Bad signature", + } + } + } + + #[derive(Debug, Clone)] + pub struct Header { + pub size: usize, + pub signature: Hash, + } + + impl Header { + pub fn parse(input: &[u8]) -> nom::IResult<&[u8], Self, Error<&[u8]>> { + use nom::bytes::streaming::tag; + use nom::character::streaming::hex_digit1; + use nom::combinator::map_res; + use nom::number::streaming::hex_u32; + + macro_rules! try_parse { + ($expr:expr) => { + $expr.map_err(|e| e.map(Error::Parser))? + }; + } + + let (input, size) = try_parse!(hex_u32(input)); + let (input, _) = try_parse!(tag(";")(input)); + + let (input, _) = try_parse!(tag("chunk-signature=")(input)); + let (input, data) = try_parse!(map_res(hex_digit1, hex::decode)(input)); + let signature = Hash::try_from(&data).ok_or(nom::Err::Failure(Error::BadSignature))?; + + let (input, _) = try_parse!(tag("\r\n")(input)); + + let header = Header { + size: size as usize, + signature, + }; + + Ok((input, header)) + } + } +} + +#[derive(Debug)] +pub enum SignedPayloadStreamError { + Stream(Error), + InvalidSignature, + Message(String), +} + +impl SignedPayloadStreamError { + fn message(msg: &str) -> Self { + SignedPayloadStreamError::Message(msg.into()) + } +} + +impl From<SignedPayloadStreamError> for Error { + fn from(err: SignedPayloadStreamError) -> Self { + match err { + SignedPayloadStreamError::Stream(e) => e, + SignedPayloadStreamError::InvalidSignature => { + Error::bad_request("Invalid payload signature") + } + SignedPayloadStreamError::Message(e) => { + Error::bad_request(format!("Chunk format error: {}", e)) + } + } + } +} + +impl<I> From<payload::Error<I>> for SignedPayloadStreamError { + fn from(err: payload::Error<I>) -> Self { + Self::message(err.description()) + } +} + +impl<I> From<nom::error::Error<I>> for SignedPayloadStreamError { + fn from(err: nom::error::Error<I>) -> Self { + Self::message(err.code.description()) + } +} + +struct SignedPayload { + header: payload::Header, + data: Bytes, +} + +#[pin_project::pin_project] +pub struct SignedPayloadStream<S> +where + S: Stream<Item = Result<Bytes, Error>>, +{ + #[pin] + stream: S, + buf: bytes::BytesMut, + datetime: DateTime<Utc>, + scope: String, + signing_hmac: HmacSha256, + previous_signature: Hash, +} + +impl<S> SignedPayloadStream<S> +where + S: Stream<Item = Result<Bytes, Error>>, +{ + pub fn new( + stream: S, + signing_hmac: HmacSha256, + datetime: DateTime<Utc>, + scope: &str, + seed_signature: Hash, + ) -> Self { + Self { + stream, + buf: bytes::BytesMut::new(), + datetime, + scope: scope.into(), + signing_hmac, + previous_signature: seed_signature, + } + } + + fn parse_next(input: &[u8]) -> nom::IResult<&[u8], SignedPayload, SignedPayloadStreamError> { + use nom::bytes::streaming::{tag, take}; + + macro_rules! try_parse { + ($expr:expr) => { + $expr.map_err(nom::Err::convert)? + }; + } + + let (input, header) = try_parse!(payload::Header::parse(input)); + + // 0-sized chunk is the last + if header.size == 0 { + return Ok(( + input, + SignedPayload { + header, + data: Bytes::new(), + }, + )); + } + + let (input, data) = try_parse!(take::<_, _, nom::error::Error<_>>(header.size)(input)); + let (input, _) = try_parse!(tag::<_, _, nom::error::Error<_>>("\r\n")(input)); + + let data = Bytes::from(data.to_vec()); + + Ok((input, SignedPayload { header, data })) + } +} + +impl<S> Stream for SignedPayloadStream<S> +where + S: Stream<Item = Result<Bytes, Error>> + Unpin, +{ + type Item = Result<Bytes, SignedPayloadStreamError>; + + fn poll_next( + self: Pin<&mut Self>, + cx: &mut task::Context<'_>, + ) -> task::Poll<Option<Self::Item>> { + use std::task::Poll; + + let mut this = self.project(); + + loop { + let (input, payload) = match Self::parse_next(this.buf) { + Ok(res) => res, + Err(nom::Err::Incomplete(_)) => { + match futures::ready!(this.stream.as_mut().poll_next(cx)) { + Some(Ok(bytes)) => { + this.buf.extend(bytes); + continue; + } + Some(Err(e)) => { + return Poll::Ready(Some(Err(SignedPayloadStreamError::Stream(e)))) + } + None => { + return Poll::Ready(Some(Err(SignedPayloadStreamError::message( + "Unexpected EOF", + )))); + } + } + } + Err(nom::Err::Error(e)) | Err(nom::Err::Failure(e)) => { + return Poll::Ready(Some(Err(e))) + } + }; + + // 0-sized chunk is the last + if payload.data.is_empty() { + return Poll::Ready(None); + } + + let data_sha256sum = sha256sum(&payload.data); + + let expected_signature = compute_streaming_payload_signature( + this.signing_hmac, + *this.datetime, + this.scope, + *this.previous_signature, + data_sha256sum, + ) + .map_err(|e| { + SignedPayloadStreamError::Message(format!("Could not build signature: {}", e)) + })?; + + if payload.header.signature != expected_signature { + return Poll::Ready(Some(Err(SignedPayloadStreamError::InvalidSignature))); + } + + *this.buf = input.into(); + *this.previous_signature = payload.header.signature; + + return Poll::Ready(Some(Ok(payload.data))); + } + } + + fn size_hint(&self) -> (usize, Option<usize>) { + self.stream.size_hint() + } +} + +#[cfg(test)] +mod tests { + use futures::prelude::*; + + use super::{SignedPayloadStream, SignedPayloadStreamError}; + + #[tokio::test] + async fn test_interrupted_signed_payload_stream() { + use chrono::{DateTime, Utc}; + + use garage_util::data::Hash; + + let datetime = DateTime::parse_from_rfc3339("2021-12-13T13:12:42+01:00") // TODO UNIX 0 + .unwrap() + .with_timezone(&Utc); + let secret_key = "test"; + let region = "test"; + let scope = crate::signature::compute_scope(&datetime, region, "s3"); + let signing_hmac = + crate::signature::signing_hmac(&datetime, secret_key, region, "s3").unwrap(); + + let data: &[&[u8]] = &[b"1"]; + let body = futures::stream::iter(data.iter().map(|block| Ok(block.to_vec().into()))); + + let seed_signature = Hash::default(); + + let mut stream = + SignedPayloadStream::new(body, signing_hmac, datetime, &scope, seed_signature); + + assert!(stream.try_next().await.is_err()); + match stream.try_next().await { + Err(SignedPayloadStreamError::Message(msg)) if msg == "Unexpected EOF" => {} + item => panic!( + "Unexpected result, expected early EOF error, got {:?}", + item + ), + } + } +} |