garage/src/web/web_server.rs

206 lines
5.6 KiB
Rust

use std::borrow::Cow;
use std::net::SocketAddr;
use std::sync::Arc;
use futures::future::Future;
use hyper::header::HOST;
use hyper::server::conn::AddrStream;
use hyper::service::{make_service_fn, service_fn};
use hyper::{Body, Request, Response, Server};
use garage_model::garage::Garage;
use garage_util::error::Error;
pub async fn run_web_server(
garage: Arc<Garage>,
shutdown_signal: impl Future<Output = ()>,
) -> Result<(), Error> {
let addr = &garage.config.s3_web.bind_addr;
let service = make_service_fn(|conn: &AddrStream| {
let garage = garage.clone();
let client_addr = conn.remote_addr();
async move {
Ok::<_, Error>(service_fn(move |req: Request<Body>| {
let garage = garage.clone();
handler(garage, req, client_addr)
}))
}
});
let server = Server::bind(&addr).serve(service);
let graceful = server.with_graceful_shutdown(shutdown_signal);
info!("Web server listening on http://{}", addr);
graceful.await?;
Ok(())
}
async fn handler(
garage: Arc<Garage>,
req: Request<Body>,
addr: SocketAddr,
) -> Result<Response<Body>, Error> {
// Get http authority string (eg. [::1]:3902 or garage.tld:80)
let authority = req
.headers()
.get(HOST)
.ok_or(Error::BadRequest(format!("HOST header required")))?
.to_str()?;
// Get bucket
let host = authority_to_host(authority)?;
let root = &garage.config.s3_web.root_domain;
let bucket = host_to_bucket(&host, root);
// Get path
let path = req.uri().path().to_string();
let index = &garage.config.s3_web.index;
let key = path_to_key(&path, &index)?;
info!("Selected bucket: \"{}\", selected key: \"{}\"", bucket, key);
// Get bucket descriptor
let object = garage
.object_table
.get(&bucket.to_string(), &key.to_string())
.await?
.ok_or(Error::NotFound)?;
Ok(Response::new(Body::from("hello world\n")))
}
/// 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/
fn authority_to_host(authority: &str) -> Result<&str, Error> {
let mut iter = authority.chars().enumerate();
let (_, first_char) = iter
.next()
.ok_or(Error::BadRequest(format!("Authority is empty")))?;
let split = match first_char {
'[' => {
let mut iter = iter.skip_while(|(_, c)| c != &']');
iter.next().expect("Authority parsing logic error");
iter.next()
}
_ => iter.skip_while(|(_, c)| c != &':').next(),
};
match split {
Some((i, ':')) => Ok(&authority[..i]),
None => Ok(authority),
Some((_, _)) => Err(Error::BadRequest(format!(
"Authority {} has an illegal format",
authority
))),
}
}
/// Host to bucket
///
/// Convert a host, like "bucket.garage-site.tld" or "john.doe.com"
/// to the corresponding bucket, resp. "bucket" and "john.doe.com"
/// considering that ".garage-site.tld" is the "root domain".
/// This behavior has been chosen to follow AWS S3 semantic.
fn host_to_bucket<'a>(host: &'a str, root: &str) -> &'a str {
if root.len() >= host.len() || !host.ends_with(root) {
return host;
}
let len_diff = host.len() - root.len();
let missing_starting_dot = root.chars().next() != Some('.');
let cursor = if missing_starting_dot {
len_diff - 1
} else {
len_diff
};
&host[..cursor]
}
/// Path to key
///
/// Convert the provided path to the internal key
/// When a path ends with "/", we append the index name to match traditional web server behavior
/// which is also AWS S3 behavior.
fn path_to_key<'a>(path: &'a str, index: &str) -> Result<Cow<'a, str>, Error> {
let path_utf8 = percent_encoding::percent_decode_str(&path).decode_utf8()?;
match path_utf8.chars().last() {
None => Err(Error::BadRequest(format!(
"Path must have at least a character"
))),
Some('/') => {
let mut key = String::with_capacity(path_utf8.len() + index.len());
key.push_str(&path_utf8);
key.push_str(index);
Ok(key.into())
}
Some(_) => Ok(path_utf8.into()),
}
}
#[cfg(test)]
mod tests {
use super::*;
#[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");
Ok(())
}
#[test]
fn host_to_bucket_test() {
assert_eq!(
host_to_bucket("john.doe.garage.tld", ".garage.tld"),
"john.doe"
);
assert_eq!(
host_to_bucket("john.doe.garage.tld", "garage.tld"),
"john.doe"
);
assert_eq!(host_to_bucket("john.doe.com", "garage.tld"), "john.doe.com");
assert_eq!(
host_to_bucket("john.doe.com", ".garage.tld"),
"john.doe.com"
);
assert_eq!(host_to_bucket("garage.tld", "garage.tld"), "garage.tld");
assert_eq!(host_to_bucket("garage.tld", ".garage.tld"), "garage.tld");
}
#[test]
fn path_to_key_test() -> Result<(), Error> {
assert_eq!(path_to_key("/file%20.jpg", "index.html")?, "/file .jpg");
assert_eq!(path_to_key("/%20t/", "index.html")?, "/ t/index.html");
assert_eq!(path_to_key("/", "index.html")?, "/index.html");
assert!(path_to_key("", "index.html").is_err());
Ok(())
}
}