1
0
Fork 0
mirror of https://github.com/denoland/deno.git synced 2024-11-01 09:24:20 -04:00
denoland-deno/ext/fetch/lib.rs

686 lines
19 KiB
Rust
Raw Normal View History

// Copyright 2018-2023 the Deno authors. All rights reserved. MIT license.
2020-09-18 09:20:55 -04:00
mod byte_stream;
mod fs_fetch_handler;
use data_url::DataUrl;
2020-09-18 09:20:55 -04:00
use deno_core::error::type_error;
use deno_core::error::AnyError;
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
use deno_core::futures::stream::Peekable;
use deno_core::futures::Future;
use deno_core::futures::Stream;
use deno_core::futures::StreamExt;
use deno_core::include_js_files;
use deno_core::op;
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
use deno_core::BufView;
use deno_core::WriteOutcome;
2020-09-18 09:20:55 -04:00
use deno_core::url::Url;
use deno_core::AsyncRefCell;
use deno_core::AsyncResult;
use deno_core::ByteString;
use deno_core::CancelFuture;
use deno_core::CancelHandle;
use deno_core::CancelTryFuture;
use deno_core::Canceled;
use deno_core::Extension;
2020-09-18 09:20:55 -04:00
use deno_core::OpState;
use deno_core::RcRef;
use deno_core::Resource;
use deno_core::ResourceId;
2020-09-18 09:20:55 -04:00
use deno_core::ZeroCopyBuf;
use deno_tls::rustls::RootCertStore;
use deno_tls::Proxy;
use http::header::CONTENT_LENGTH;
use http::Uri;
use reqwest::header::HeaderMap;
2020-09-18 09:20:55 -04:00
use reqwest::header::HeaderName;
use reqwest::header::HeaderValue;
use reqwest::header::ACCEPT_ENCODING;
use reqwest::header::HOST;
use reqwest::header::RANGE;
use reqwest::header::USER_AGENT;
use reqwest::redirect::Policy;
use reqwest::Body;
2020-09-18 09:20:55 -04:00
use reqwest::Client;
use reqwest::Method;
use reqwest::RequestBuilder;
2020-09-18 09:20:55 -04:00
use reqwest::Response;
use serde::Deserialize;
use serde::Serialize;
use std::borrow::Cow;
2020-09-18 09:20:55 -04:00
use std::cell::RefCell;
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
use std::cmp::min;
2020-09-18 09:20:55 -04:00
use std::convert::From;
use std::path::Path;
use std::path::PathBuf;
use std::pin::Pin;
2020-09-18 09:20:55 -04:00
use std::rc::Rc;
use tokio::sync::mpsc;
2020-09-18 09:20:55 -04:00
// Re-export reqwest and data_url
pub use data_url;
pub use reqwest;
2020-09-18 12:31:30 -04:00
pub use fs_fetch_handler::FsFetchHandler;
use crate::byte_stream::MpscByteStream;
#[derive(Clone)]
pub struct Options {
pub user_agent: String,
pub root_cert_store: Option<RootCertStore>,
pub proxy: Option<Proxy>,
pub request_builder_hook: Option<fn(RequestBuilder) -> RequestBuilder>,
pub unsafely_ignore_certificate_errors: Option<Vec<String>>,
pub client_cert_chain_and_key: Option<(String, String)>,
pub file_fetch_handler: Rc<dyn FetchHandler>,
}
impl Default for Options {
fn default() -> Self {
Self {
user_agent: "".to_string(),
root_cert_store: None,
proxy: None,
request_builder_hook: None,
unsafely_ignore_certificate_errors: None,
client_cert_chain_and_key: None,
file_fetch_handler: Rc::new(DefaultFileFetchHandler),
}
}
}
pub fn init<FP>(options: Options) -> Extension
where
FP: FetchPermissions + 'static,
{
Extension::builder(env!("CARGO_PKG_NAME"))
.dependencies(vec!["deno_webidl", "deno_web", "deno_url", "deno_console"])
.esm(include_js_files!(
"20_headers.js",
"21_formdata.js",
"22_body.js",
"22_http_client.js",
"23_request.js",
"23_response.js",
"26_fetch.js",
))
.ops(vec![
op_fetch::decl::<FP>(),
op_fetch_send::decl(),
op_fetch_custom_client::decl::<FP>(),
])
.state(move |state| {
state.put::<Options>(options.clone());
state.put::<reqwest::Client>({
create_http_client(
options.user_agent.clone(),
options.root_cert_store.clone(),
vec![],
options.proxy.clone(),
options.unsafely_ignore_certificate_errors.clone(),
options.client_cert_chain_and_key.clone(),
)
.unwrap()
});
Ok(())
})
.build()
2020-09-18 09:20:55 -04:00
}
pub type CancelableResponseFuture =
Pin<Box<dyn Future<Output = CancelableResponseResult>>>;
pub trait FetchHandler: dyn_clone::DynClone {
// Return the result of the fetch request consisting of a tuple of the
// cancelable response result, the optional fetch body resource and the
// optional cancel handle.
fn fetch_file(
&self,
state: &mut OpState,
url: Url,
) -> (
CancelableResponseFuture,
Option<FetchRequestBodyResource>,
Option<Rc<CancelHandle>>,
);
}
dyn_clone::clone_trait_object!(FetchHandler);
/// A default implementation which will error for every request.
#[derive(Clone)]
pub struct DefaultFileFetchHandler;
impl FetchHandler for DefaultFileFetchHandler {
fn fetch_file(
&self,
_state: &mut OpState,
_url: Url,
) -> (
CancelableResponseFuture,
Option<FetchRequestBodyResource>,
Option<Rc<CancelHandle>>,
) {
let fut = async move {
Ok(Err(type_error(
"NetworkError when attempting to fetch resource.",
)))
};
(Box::pin(fut), None, None)
}
}
2020-09-18 09:20:55 -04:00
pub trait FetchPermissions {
fn check_net_url(
&mut self,
_url: &Url,
api_name: &str,
) -> Result<(), AnyError>;
fn check_read(&mut self, _p: &Path, api_name: &str) -> Result<(), AnyError>;
}
pub fn get_declaration() -> PathBuf {
PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("lib.deno_fetch.d.ts")
}
#[derive(Serialize)]
#[serde(rename_all = "camelCase")]
pub struct FetchReturn {
request_rid: ResourceId,
request_body_rid: Option<ResourceId>,
cancel_handle_rid: Option<ResourceId>,
}
#[op]
pub fn op_fetch<FP>(
state: &mut OpState,
method: ByteString,
url: String,
headers: Vec<(ByteString, ByteString)>,
client_rid: Option<u32>,
has_body: bool,
body_length: Option<u64>,
data: Option<ZeroCopyBuf>,
) -> Result<FetchReturn, AnyError>
2020-09-18 09:20:55 -04:00
where
FP: FetchPermissions + 'static,
{
let client = if let Some(rid) = client_rid {
let r = state.resource_table.get::<HttpClientResource>(rid)?;
2020-09-18 09:20:55 -04:00
r.client.clone()
} else {
let client = state.borrow::<reqwest::Client>();
2020-09-18 09:20:55 -04:00
client.clone()
};
let method = Method::from_bytes(&method)?;
let url = Url::parse(&url)?;
2020-09-18 09:20:55 -04:00
// Check scheme before asking for net permission
let scheme = url.scheme();
let (request_rid, request_body_rid, cancel_handle_rid) = match scheme {
"file" => {
let path = url.to_file_path().map_err(|_| {
type_error("NetworkError when attempting to fetch resource.")
})?;
let permissions = state.borrow_mut::<FP>();
permissions.check_read(&path, "fetch()")?;
if method != Method::GET {
return Err(type_error(format!(
"Fetching files only supports the GET method. Received {method}."
)));
}
let Options {
file_fetch_handler, ..
} = state.borrow_mut::<Options>();
let file_fetch_handler = file_fetch_handler.clone();
let (request, maybe_request_body, maybe_cancel_handle) =
file_fetch_handler.fetch_file(state, url);
let request_rid = state.resource_table.add(FetchRequestResource(request));
let maybe_request_body_rid =
maybe_request_body.map(|r| state.resource_table.add(r));
let maybe_cancel_handle_rid = maybe_cancel_handle
.map(|ch| state.resource_table.add(FetchCancelHandle(ch)));
(request_rid, maybe_request_body_rid, maybe_cancel_handle_rid)
}
"http" | "https" => {
let permissions = state.borrow_mut::<FP>();
permissions.check_net_url(&url, "fetch()")?;
2020-09-18 09:20:55 -04:00
// Make sure that we have a valid URI early, as reqwest's `RequestBuilder::send`
// internally uses `expect_uri`, which panics instead of returning a usable `Result`.
if url.as_str().parse::<Uri>().is_err() {
return Err(type_error("Invalid URL"));
}
let mut request = client.request(method.clone(), url);
2020-09-18 09:20:55 -04:00
let request_body_rid = if has_body {
match data {
None => {
// If no body is passed, we return a writer for streaming the body.
let (stream, tx) = MpscByteStream::new();
// If the size of the body is known, we include a content-length
// header explicitly.
if let Some(body_size) = body_length {
request =
request.header(CONTENT_LENGTH, HeaderValue::from(body_size))
}
request = request.body(Body::wrap_stream(stream));
let request_body_rid =
state.resource_table.add(FetchRequestBodyResource {
body: AsyncRefCell::new(tx),
cancel: CancelHandle::default(),
});
Some(request_body_rid)
}
Some(data) => {
// If a body is passed, we use it, and don't return a body for streaming.
request = request.body(Vec::from(&*data));
None
}
}
} else {
// POST and PUT requests should always have a 0 length content-length,
// if there is no body. https://fetch.spec.whatwg.org/#http-network-or-cache-fetch
if matches!(method, Method::POST | Method::PUT) {
request = request.header(CONTENT_LENGTH, HeaderValue::from(0));
}
None
};
let mut header_map = HeaderMap::new();
for (key, value) in headers {
let name = HeaderName::from_bytes(&key)
.map_err(|err| type_error(err.to_string()))?;
let v = HeaderValue::from_bytes(&value)
.map_err(|err| type_error(err.to_string()))?;
if !matches!(name, HOST | CONTENT_LENGTH) {
header_map.append(name, v);
}
}
if header_map.contains_key(RANGE) {
// https://fetch.spec.whatwg.org/#http-network-or-cache-fetch step 18
// If httpRequests header list contains `Range`, then append (`Accept-Encoding`, `identity`)
header_map
.insert(ACCEPT_ENCODING, HeaderValue::from_static("identity"));
}
request = request.headers(header_map);
let options = state.borrow::<Options>();
if let Some(request_builder_hook) = options.request_builder_hook {
request = request_builder_hook(request);
}
let cancel_handle = CancelHandle::new_rc();
let cancel_handle_ = cancel_handle.clone();
let fut = async move {
request
.send()
.or_cancel(cancel_handle_)
.await
.map(|res| res.map_err(|err| type_error(err.to_string())))
};
let request_rid = state
.resource_table
.add(FetchRequestResource(Box::pin(fut)));
let cancel_handle_rid =
state.resource_table.add(FetchCancelHandle(cancel_handle));
(request_rid, request_body_rid, Some(cancel_handle_rid))
}
"data" => {
let data_url = DataUrl::process(url.as_str())
.map_err(|e| type_error(format!("{e:?}")))?;
2020-09-18 09:20:55 -04:00
let (body, _) = data_url
.decode_to_vec()
.map_err(|e| type_error(format!("{e:?}")))?;
2020-09-18 09:20:55 -04:00
let response = http::Response::builder()
.status(http::StatusCode::OK)
.header(http::header::CONTENT_TYPE, data_url.mime_type().to_string())
.body(reqwest::Body::from(body))?;
let fut = async move { Ok(Ok(Response::from(response))) };
let request_rid = state
.resource_table
.add(FetchRequestResource(Box::pin(fut)));
(request_rid, None, None)
}
"blob" => {
// Blob URL resolution happens in the JS side of fetch. If we got here is
// because the URL isn't an object URL.
return Err(type_error("Blob for the given URL not found."));
}
_ => return Err(type_error(format!("scheme '{scheme}' not supported"))),
};
Ok(FetchReturn {
request_rid,
request_body_rid,
cancel_handle_rid,
})
}
#[derive(Serialize)]
2021-03-17 17:33:29 -04:00
#[serde(rename_all = "camelCase")]
pub struct FetchResponse {
status: u16,
status_text: String,
headers: Vec<(ByteString, ByteString)>,
url: String,
response_rid: ResourceId,
content_length: Option<u64>,
2021-03-17 17:33:29 -04:00
}
#[op]
pub async fn op_fetch_send(
state: Rc<RefCell<OpState>>,
rid: ResourceId,
) -> Result<FetchResponse, AnyError> {
let request = state
.borrow_mut()
.resource_table
.take::<FetchRequestResource>(rid)?;
let request = Rc::try_unwrap(request)
.ok()
.expect("multiple op_fetch_send ongoing");
let res = match request.0.await {
Ok(Ok(res)) => res,
Ok(Err(err)) => return Err(type_error(err.to_string())),
Err(_) => return Err(type_error("request was cancelled")),
};
2020-09-18 09:20:55 -04:00
//debug!("Fetch response {}", url);
let status = res.status();
let url = res.url().to_string();
2020-09-18 09:20:55 -04:00
let mut res_headers = Vec::new();
for (key, val) in res.headers().iter() {
res_headers.push((key.as_str().into(), val.as_bytes().into()));
2020-09-18 09:20:55 -04:00
}
let content_length = res.content_length();
let stream: BytesStream = Box::pin(res.bytes_stream().map(|r| {
r.map_err(|err| std::io::Error::new(std::io::ErrorKind::Other, err))
}));
let rid = state
.borrow_mut()
.resource_table
.add(FetchResponseBodyResource {
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
reader: AsyncRefCell::new(stream.peekable()),
cancel: CancelHandle::default(),
size: content_length,
});
2020-09-18 09:20:55 -04:00
Ok(FetchResponse {
status: status.as_u16(),
status_text: status.canonical_reason().unwrap_or("").to_string(),
headers: res_headers,
url,
response_rid: rid,
content_length,
})
2021-03-17 17:33:29 -04:00
}
type CancelableResponseResult = Result<Result<Response, AnyError>, Canceled>;
struct FetchRequestResource(
Pin<Box<dyn Future<Output = CancelableResponseResult>>>,
);
impl Resource for FetchRequestResource {
fn name(&self) -> Cow<str> {
"fetchRequest".into()
}
}
2020-09-18 09:20:55 -04:00
struct FetchCancelHandle(Rc<CancelHandle>);
impl Resource for FetchCancelHandle {
fn name(&self) -> Cow<str> {
"fetchCancelHandle".into()
}
fn close(self: Rc<Self>) {
self.0.cancel()
}
}
pub struct FetchRequestBodyResource {
body: AsyncRefCell<mpsc::Sender<Option<bytes::Bytes>>>,
cancel: CancelHandle,
}
impl Resource for FetchRequestBodyResource {
fn name(&self) -> Cow<str> {
"fetchRequestBody".into()
}
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
fn write(self: Rc<Self>, buf: BufView) -> AsyncResult<WriteOutcome> {
Box::pin(async move {
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
let bytes: bytes::Bytes = buf.into();
let nwritten = bytes.len();
let body = RcRef::map(&self, |r| &r.body).borrow_mut().await;
let cancel = RcRef::map(self, |r| &r.cancel);
body
.send(Some(bytes))
.or_cancel(cancel)
.await?
.map_err(|_| {
type_error("request body receiver not connected (request closed)")
})?;
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
Ok(WriteOutcome::Full { nwritten })
})
}
fn shutdown(self: Rc<Self>) -> AsyncResult<()> {
Box::pin(async move {
let body = RcRef::map(&self, |r| &r.body).borrow_mut().await;
let cancel = RcRef::map(self, |r| &r.cancel);
// There is a case where hyper knows the size of the response body up
// front (through content-length header on the resp), where it will drop
// the body once that content length has been reached, regardless of if
// the stream is complete or not. This is expected behaviour, but it means
// that if you stream a body with an up front known size (eg a Blob),
// explicit shutdown can never succeed because the body (and by extension
// the receiver) will have dropped by the time we try to shutdown. As such
// we ignore if the receiver is closed, because we know that the request
// is complete in good health in that case.
body.send(None).or_cancel(cancel).await?.ok();
Ok(())
})
}
fn close(self: Rc<Self>) {
self.cancel.cancel()
}
}
type BytesStream =
Pin<Box<dyn Stream<Item = Result<bytes::Bytes, std::io::Error>> + Unpin>>;
struct FetchResponseBodyResource {
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
reader: AsyncRefCell<Peekable<BytesStream>>,
cancel: CancelHandle,
size: Option<u64>,
}
impl Resource for FetchResponseBodyResource {
fn name(&self) -> Cow<str> {
"fetchResponseBody".into()
}
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
fn read(self: Rc<Self>, limit: usize) -> AsyncResult<BufView> {
Box::pin(async move {
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
let reader = RcRef::map(&self, |r| &r.reader).borrow_mut().await;
let fut = async move {
let mut reader = Pin::new(reader);
loop {
match reader.as_mut().peek_mut().await {
Some(Ok(chunk)) if !chunk.is_empty() => {
let len = min(limit, chunk.len());
let chunk = chunk.split_to(len);
break Ok(chunk.into());
}
// This unwrap is safe because `peek_mut()` returned `Some`, and thus
// currently has a peeked value that can be synchronously returned
// from `next()`.
//
// The future returned from `next()` is always ready, so we can
// safely call `await` on it without creating a race condition.
Some(_) => match reader.as_mut().next().await.unwrap() {
Ok(chunk) => assert!(chunk.is_empty()),
Err(err) => break Err(type_error(err.to_string())),
feat(core): improve resource read & write traits (#16115) This commit introduces two new buffer wrapper types to `deno_core`. The main benefit of these new wrappers is that they can wrap a number of different underlying buffer types. This allows for a more flexible read and write API on resources that will require less copying of data between different buffer representations. - `BufView` is a read-only view onto a buffer. It can be backed by `ZeroCopyBuf`, `Vec<u8>`, and `bytes::Bytes`. - `BufViewMut` is a read-write view onto a buffer. It can be cheaply converted into a `BufView`. It can be backed by `ZeroCopyBuf` or `Vec<u8>`. Both new buffer views have a cursor. This means that the start point of the view can be constrained to write / read from just a slice of the view. Only the start point of the slice can be adjusted. The end point is fixed. To adjust the end point, the underlying buffer needs to be truncated. Readable resources have been changed to better cater to resources that do not support BYOB reads. The basic `read` method now returns a `BufView` instead of taking a `ZeroCopyBuf` to fill. This allows the operation to return buffers that the resource has already allocated, instead of forcing the caller to allocate the buffer. BYOB reads are still very useful for resources that support them, so a new `read_byob` method has been added that takes a `BufViewMut` to fill. `op_read` attempts to use `read_byob` if the resource supports it, which falls back to `read` and performs an additional copy if it does not. For Rust->JS reads this change should have no impact, but for Rust->Rust reads, this allows the caller to avoid an additional copy in many scenarios. This combined with the support for `BufView` to be backed by `bytes::Bytes` allows us to avoid one data copy when piping from a `fetch` response into an `ext/http` response. Writable resources have been changed to take a `BufView` instead of a `ZeroCopyBuf` as an argument. This allows for less copying of data in certain scenarios, as described above. Additionally a new `Resource::write_all` method has been added that takes a `BufView` and continually attempts to write the resource until the entire buffer has been written. Certain resources like files can override this method to provide a more efficient `write_all` implementation.
2022-10-09 10:49:25 -04:00
},
None => break Ok(BufView::empty()),
}
}
};
let cancel_handle = RcRef::map(self, |r| &r.cancel);
fut.try_or_cancel(cancel_handle).await
})
}
fn size_hint(&self) -> (u64, Option<u64>) {
(self.size.unwrap_or(0), self.size)
}
fn close(self: Rc<Self>) {
self.cancel.cancel()
}
2020-09-18 09:20:55 -04:00
}
struct HttpClientResource {
client: Client,
}
impl Resource for HttpClientResource {
fn name(&self) -> Cow<str> {
"httpClient".into()
}
}
2020-09-18 09:20:55 -04:00
impl HttpClientResource {
fn new(client: Client) -> Self {
Self { client }
}
}
#[derive(Deserialize, Debug)]
2021-03-17 17:33:29 -04:00
#[serde(rename_all = "camelCase")]
pub struct CreateHttpClientOptions {
ca_certs: Vec<String>,
proxy: Option<Proxy>,
cert_chain: Option<String>,
private_key: Option<String>,
}
#[op]
pub fn op_fetch_custom_client<FP>(
2020-09-18 09:20:55 -04:00
state: &mut OpState,
2021-03-17 17:33:29 -04:00
args: CreateHttpClientOptions,
) -> Result<ResourceId, AnyError>
2020-09-18 09:20:55 -04:00
where
FP: FetchPermissions + 'static,
{
if let Some(proxy) = args.proxy.clone() {
let permissions = state.borrow_mut::<FP>();
let url = Url::parse(&proxy.url)?;
permissions.check_net_url(&url, "Deno.createHttpClient()")?;
}
let client_cert_chain_and_key = {
if args.cert_chain.is_some() || args.private_key.is_some() {
let cert_chain = args
.cert_chain
.ok_or_else(|| type_error("No certificate chain provided"))?;
let private_key = args
.private_key
.ok_or_else(|| type_error("No private key provided"))?;
Some((cert_chain, private_key))
} else {
None
}
};
let options = state.borrow::<Options>();
let ca_certs = args
.ca_certs
.into_iter()
.map(|cert| cert.into_bytes())
.collect::<Vec<_>>();
let client = create_http_client(
options.user_agent.clone(),
options.root_cert_store.clone(),
ca_certs,
args.proxy,
options.unsafely_ignore_certificate_errors.clone(),
client_cert_chain_and_key,
)?;
2020-09-18 09:20:55 -04:00
let rid = state.resource_table.add(HttpClientResource::new(client));
Ok(rid)
2020-09-18 09:20:55 -04:00
}
/// Create new instance of async reqwest::Client. This client supports
/// proxies and doesn't follow redirects.
pub fn create_http_client(
user_agent: String,
root_cert_store: Option<RootCertStore>,
ca_certs: Vec<Vec<u8>>,
proxy: Option<Proxy>,
unsafely_ignore_certificate_errors: Option<Vec<String>>,
client_cert_chain_and_key: Option<(String, String)>,
) -> Result<Client, AnyError> {
let mut tls_config = deno_tls::create_client_config(
root_cert_store,
ca_certs,
unsafely_ignore_certificate_errors,
client_cert_chain_and_key,
)?;
tls_config.alpn_protocols = vec!["h2".into(), "http/1.1".into()];
let mut headers = HeaderMap::new();
headers.insert(USER_AGENT, user_agent.parse().unwrap());
let mut builder = Client::builder()
.redirect(Policy::none())
.default_headers(headers)
.use_preconfigured_tls(tls_config);
if let Some(proxy) = proxy {
let mut reqwest_proxy = reqwest::Proxy::all(&proxy.url)?;
if let Some(basic_auth) = &proxy.basic_auth {
reqwest_proxy =
reqwest_proxy.basic_auth(&basic_auth.username, &basic_auth.password);
}
builder = builder.proxy(reqwest_proxy);
}
// unwrap here because it can only fail when native TLS is used.
Ok(builder.build().unwrap())
}