Source code
Revision control
Copy as Markdown
Other Tools
use super::*;
use crate::codec::UserError;
use crate::frame::{self, PushPromiseHeaderError, Reason, DEFAULT_INITIAL_WINDOW_SIZE};
use crate::proto::{self, Error};
use http::{HeaderMap, Request, Response};
use std::cmp::Ordering;
use std::io;
use std::task::{Context, Poll, Waker};
use std::time::{Duration, Instant};
#[derive(Debug)]
pub(super) struct Recv {
/// Initial window size of remote initiated streams
init_window_sz: WindowSize,
/// Connection level flow control governing received data
flow: FlowControl,
/// Amount of connection window capacity currently used by outstanding streams.
in_flight_data: WindowSize,
/// The lowest stream ID that is still idle
next_stream_id: Result<StreamId, StreamIdOverflow>,
/// The stream ID of the last processed stream
last_processed_id: StreamId,
/// Any streams with a higher ID are ignored.
///
/// This starts as MAX, but is lowered when a GOAWAY is received.
///
/// > After sending a GOAWAY frame, the sender can discard frames for
/// > streams initiated by the receiver with identifiers higher than
/// > the identified last stream.
max_stream_id: StreamId,
/// Streams that have pending window updates
pending_window_updates: store::Queue<stream::NextWindowUpdate>,
/// New streams to be accepted
pending_accept: store::Queue<stream::NextAccept>,
/// Locally reset streams that should be reaped when they expire
pending_reset_expired: store::Queue<stream::NextResetExpire>,
/// How long locally reset streams should ignore received frames
reset_duration: Duration,
/// Holds frames that are waiting to be read
buffer: Buffer<Event>,
/// Refused StreamId, this represents a frame that must be sent out.
refused: Option<StreamId>,
/// If push promises are allowed to be received.
is_push_enabled: bool,
/// If extended connect protocol is enabled.
is_extended_connect_protocol_enabled: bool,
}
#[derive(Debug)]
pub(super) enum Event {
Headers(peer::PollMessage),
Data(Bytes),
Trailers(HeaderMap),
}
#[derive(Debug)]
pub(super) enum RecvHeaderBlockError<T> {
Oversize(T),
State(Error),
}
#[derive(Debug)]
pub(crate) enum Open {
PushPromise,
Headers,
}
impl Recv {
pub fn new(peer: peer::Dyn, config: &Config) -> Self {
let next_stream_id = if peer.is_server() { 1 } else { 2 };
let mut flow = FlowControl::new();
// connections always have the default window size, regardless of
// settings
flow.inc_window(DEFAULT_INITIAL_WINDOW_SIZE)
.expect("invalid initial remote window size");
flow.assign_capacity(DEFAULT_INITIAL_WINDOW_SIZE).unwrap();
Recv {
init_window_sz: config.local_init_window_sz,
flow,
in_flight_data: 0 as WindowSize,
next_stream_id: Ok(next_stream_id.into()),
pending_window_updates: store::Queue::new(),
last_processed_id: StreamId::ZERO,
max_stream_id: StreamId::MAX,
pending_accept: store::Queue::new(),
pending_reset_expired: store::Queue::new(),
reset_duration: config.local_reset_duration,
buffer: Buffer::new(),
refused: None,
is_push_enabled: config.local_push_enabled,
is_extended_connect_protocol_enabled: config.extended_connect_protocol_enabled,
}
}
/// Returns the initial receive window size
pub fn init_window_sz(&self) -> WindowSize {
self.init_window_sz
}
/// Returns the ID of the last processed stream
pub fn last_processed_id(&self) -> StreamId {
self.last_processed_id
}
/// Update state reflecting a new, remotely opened stream
///
/// Returns the stream state if successful. `None` if refused
pub fn open(
&mut self,
id: StreamId,
mode: Open,
counts: &mut Counts,
) -> Result<Option<StreamId>, Error> {
assert!(self.refused.is_none());
counts.peer().ensure_can_open(id, mode)?;
let next_id = self.next_stream_id()?;
if id < next_id {
proto_err!(conn: "id ({:?}) < next_id ({:?})", id, next_id);
return Err(Error::library_go_away(Reason::PROTOCOL_ERROR));
}
self.next_stream_id = id.next_id();
if !counts.can_inc_num_recv_streams() {
self.refused = Some(id);
return Ok(None);
}
Ok(Some(id))
}
/// Transition the stream state based on receiving headers
///
/// The caller ensures that the frame represents headers and not trailers.
pub fn recv_headers(
&mut self,
frame: frame::Headers,
stream: &mut store::Ptr,
counts: &mut Counts,
) -> Result<(), RecvHeaderBlockError<Option<frame::Headers>>> {
tracing::trace!("opening stream; init_window={}", self.init_window_sz);
let is_initial = stream.state.recv_open(&frame)?;
if is_initial {
// TODO: be smarter about this logic
if frame.stream_id() > self.last_processed_id {
self.last_processed_id = frame.stream_id();
}
// Increment the number of concurrent streams
counts.inc_num_recv_streams(stream);
}
if !stream.content_length.is_head() {
use super::stream::ContentLength;
use http::header;
if let Some(content_length) = frame.fields().get(header::CONTENT_LENGTH) {
let content_length = match frame::parse_u64(content_length.as_bytes()) {
Ok(v) => v,
Err(_) => {
proto_err!(stream: "could not parse content-length; stream={:?}", stream.id);
return Err(Error::library_reset(stream.id, Reason::PROTOCOL_ERROR).into());
}
};
stream.content_length = ContentLength::Remaining(content_length);
}
}
if frame.is_over_size() {
// A frame is over size if the decoded header block was bigger than
// SETTINGS_MAX_HEADER_LIST_SIZE.
//
// > A server that receives a larger header block than it is willing
// > to handle can send an HTTP 431 (Request Header Fields Too
// > Large) status code [RFC6585]. A client can discard responses
// > that it cannot process.
//
// So, if peer is a server, we'll send a 431. In either case,
// an error is recorded, which will send a REFUSED_STREAM,
// since we don't want any of the data frames either.
tracing::debug!(
"stream error REQUEST_HEADER_FIELDS_TOO_LARGE -- \
recv_headers: frame is over size; stream={:?}",
stream.id
);
return if counts.peer().is_server() && is_initial {
let mut res = frame::Headers::new(
stream.id,
frame::Pseudo::response(::http::StatusCode::REQUEST_HEADER_FIELDS_TOO_LARGE),
HeaderMap::new(),
);
res.set_end_stream();
Err(RecvHeaderBlockError::Oversize(Some(res)))
} else {
Err(RecvHeaderBlockError::Oversize(None))
};
}
let stream_id = frame.stream_id();
let (pseudo, fields) = frame.into_parts();
if pseudo.protocol.is_some()
&& counts.peer().is_server()
&& !self.is_extended_connect_protocol_enabled
{
proto_err!(stream: "cannot use :protocol if extended connect protocol is disabled; stream={:?}", stream.id);
return Err(Error::library_reset(stream.id, Reason::PROTOCOL_ERROR).into());
}
if pseudo.status.is_some() && counts.peer().is_server() {
proto_err!(stream: "cannot use :status header for requests; stream={:?}", stream.id);
return Err(Error::library_reset(stream.id, Reason::PROTOCOL_ERROR).into());
}
if !pseudo.is_informational() {
let message = counts
.peer()
.convert_poll_message(pseudo, fields, stream_id)?;
// Push the frame onto the stream's recv buffer
stream
.pending_recv
.push_back(&mut self.buffer, Event::Headers(message));
stream.notify_recv();
// Only servers can receive a headers frame that initiates the stream.
// This is verified in `Streams` before calling this function.
if counts.peer().is_server() {
// Correctness: never push a stream to `pending_accept` without having the
// corresponding headers frame pushed to `stream.pending_recv`.
self.pending_accept.push(stream);
}
}
Ok(())
}
/// Called by the server to get the request
///
/// # Panics
///
/// Panics if `stream.pending_recv` has no `Event::Headers` queued.
///
pub fn take_request(&mut self, stream: &mut store::Ptr) -> Request<()> {
use super::peer::PollMessage::*;
match stream.pending_recv.pop_front(&mut self.buffer) {
Some(Event::Headers(Server(request))) => request,
_ => unreachable!("server stream queue must start with Headers"),
}
}
/// Called by the client to get pushed response
pub fn poll_pushed(
&mut self,
cx: &Context,
stream: &mut store::Ptr,
) -> Poll<Option<Result<(Request<()>, store::Key), proto::Error>>> {
use super::peer::PollMessage::*;
let mut ppp = stream.pending_push_promises.take();
let pushed = ppp.pop(stream.store_mut()).map(|mut pushed| {
match pushed.pending_recv.pop_front(&mut self.buffer) {
Some(Event::Headers(Server(headers))) => (headers, pushed.key()),
// When frames are pushed into the queue, it is verified that
// the first frame is a HEADERS frame.
_ => panic!("Headers not set on pushed stream"),
}
});
stream.pending_push_promises = ppp;
if let Some(p) = pushed {
Poll::Ready(Some(Ok(p)))
} else {
let is_open = stream.state.ensure_recv_open()?;
if is_open {
stream.recv_task = Some(cx.waker().clone());
Poll::Pending
} else {
Poll::Ready(None)
}
}
}
/// Called by the client to get the response
pub fn poll_response(
&mut self,
cx: &Context,
stream: &mut store::Ptr,
) -> Poll<Result<Response<()>, proto::Error>> {
use super::peer::PollMessage::*;
// If the buffer is not empty, then the first frame must be a HEADERS
// frame or the user violated the contract.
match stream.pending_recv.pop_front(&mut self.buffer) {
Some(Event::Headers(Client(response))) => Poll::Ready(Ok(response)),
Some(_) => panic!("poll_response called after response returned"),
None => {
if !stream.state.ensure_recv_open()? {
proto_err!(stream: "poll_response: stream={:?} is not opened;", stream.id);
return Poll::Ready(Err(Error::library_reset(
stream.id,
Reason::PROTOCOL_ERROR,
)));
}
stream.recv_task = Some(cx.waker().clone());
Poll::Pending
}
}
}
/// Transition the stream based on receiving trailers
pub fn recv_trailers(
&mut self,
frame: frame::Headers,
stream: &mut store::Ptr,
) -> Result<(), Error> {
// Transition the state
stream.state.recv_close()?;
if stream.ensure_content_length_zero().is_err() {
proto_err!(stream: "recv_trailers: content-length is not zero; stream={:?};", stream.id);
return Err(Error::library_reset(stream.id, Reason::PROTOCOL_ERROR));
}
let trailers = frame.into_fields();
// Push the frame onto the stream's recv buffer
stream
.pending_recv
.push_back(&mut self.buffer, Event::Trailers(trailers));
stream.notify_recv();
Ok(())
}
/// Releases capacity of the connection
pub fn release_connection_capacity(&mut self, capacity: WindowSize, task: &mut Option<Waker>) {
tracing::trace!(
"release_connection_capacity; size={}, connection in_flight_data={}",
capacity,
self.in_flight_data,
);
// Decrement in-flight data
self.in_flight_data -= capacity;
// Assign capacity to connection
// TODO: proper error handling
let _res = self.flow.assign_capacity(capacity);
debug_assert!(_res.is_ok());
if self.flow.unclaimed_capacity().is_some() {
if let Some(task) = task.take() {
task.wake();
}
}
}
/// Releases capacity back to the connection & stream
pub fn release_capacity(
&mut self,
capacity: WindowSize,
stream: &mut store::Ptr,
task: &mut Option<Waker>,
) -> Result<(), UserError> {
tracing::trace!("release_capacity; size={}", capacity);
if capacity > stream.in_flight_recv_data {
return Err(UserError::ReleaseCapacityTooBig);
}
self.release_connection_capacity(capacity, task);
// Decrement in-flight data
stream.in_flight_recv_data -= capacity;
// Assign capacity to stream
// TODO: proper error handling
let _res = stream.recv_flow.assign_capacity(capacity);
debug_assert!(_res.is_ok());
if stream.recv_flow.unclaimed_capacity().is_some() {
// Queue the stream for sending the WINDOW_UPDATE frame.
self.pending_window_updates.push(stream);
if let Some(task) = task.take() {
task.wake();
}
}
Ok(())
}
/// Release any unclaimed capacity for a closed stream.
pub fn release_closed_capacity(&mut self, stream: &mut store::Ptr, task: &mut Option<Waker>) {
debug_assert_eq!(stream.ref_count, 0);
if stream.in_flight_recv_data == 0 {
return;
}
tracing::trace!(
"auto-release closed stream ({:?}) capacity: {:?}",
stream.id,
stream.in_flight_recv_data,
);
self.release_connection_capacity(stream.in_flight_recv_data, task);
stream.in_flight_recv_data = 0;
self.clear_recv_buffer(stream);
}
/// Set the "target" connection window size.
///
/// By default, all new connections start with 64kb of window size. As
/// streams used and release capacity, we will send WINDOW_UPDATEs for the
/// connection to bring it back up to the initial "target".
///
/// Setting a target means that we will try to tell the peer about
/// WINDOW_UPDATEs so the peer knows it has about `target` window to use
/// for the whole connection.
///
/// The `task` is an optional parked task for the `Connection` that might
/// be blocked on needing more window capacity.
pub fn set_target_connection_window(
&mut self,
target: WindowSize,
task: &mut Option<Waker>,
) -> Result<(), Reason> {
tracing::trace!(
"set_target_connection_window; target={}; available={}, reserved={}",
target,
self.flow.available(),
self.in_flight_data,
);
// The current target connection window is our `available` plus any
// in-flight data reserved by streams.
//
// Update the flow controller with the difference between the new
// target and the current target.
let current = self
.flow
.available()
.add(self.in_flight_data)?
.checked_size();
if target > current {
self.flow.assign_capacity(target - current)?;
} else {
self.flow.claim_capacity(current - target)?;
}
// If changing the target capacity means we gained a bunch of capacity,
// enough that we went over the update threshold, then schedule sending
// a connection WINDOW_UPDATE.
if self.flow.unclaimed_capacity().is_some() {
if let Some(task) = task.take() {
task.wake();
}
}
Ok(())
}
pub(crate) fn apply_local_settings(
&mut self,
settings: &frame::Settings,
store: &mut Store,
) -> Result<(), proto::Error> {
if let Some(val) = settings.is_extended_connect_protocol_enabled() {
self.is_extended_connect_protocol_enabled = val;
}
if let Some(target) = settings.initial_window_size() {
let old_sz = self.init_window_sz;
self.init_window_sz = target;
tracing::trace!("update_initial_window_size; new={}; old={}", target, old_sz,);
// Per RFC 7540 §6.9.2:
//
// In addition to changing the flow-control window for streams that are
// not yet active, a SETTINGS frame can alter the initial flow-control
// window size for streams with active flow-control windows (that is,
// streams in the "open" or "half-closed (remote)" state). When the
// value of SETTINGS_INITIAL_WINDOW_SIZE changes, a receiver MUST adjust
// the size of all stream flow-control windows that it maintains by the
// difference between the new value and the old value.
//
// A change to `SETTINGS_INITIAL_WINDOW_SIZE` can cause the available
// space in a flow-control window to become negative. A sender MUST
// track the negative flow-control window and MUST NOT send new
// flow-controlled frames until it receives WINDOW_UPDATE frames that
// cause the flow-control window to become positive.
match target.cmp(&old_sz) {
Ordering::Less => {
// We must decrease the (local) window on every open stream.
let dec = old_sz - target;
tracing::trace!("decrementing all windows; dec={}", dec);
store.try_for_each(|mut stream| {
stream
.recv_flow
.dec_recv_window(dec)
.map_err(proto::Error::library_go_away)?;
Ok::<_, proto::Error>(())
})?;
}
Ordering::Greater => {
// We must increase the (local) window on every open stream.
let inc = target - old_sz;
tracing::trace!("incrementing all windows; inc={}", inc);
store.try_for_each(|mut stream| {
// XXX: Shouldn't the peer have already noticed our
// overflow and sent us a GOAWAY?
stream
.recv_flow
.inc_window(inc)
.map_err(proto::Error::library_go_away)?;
stream
.recv_flow
.assign_capacity(inc)
.map_err(proto::Error::library_go_away)?;
Ok::<_, proto::Error>(())
})?;
}
Ordering::Equal => (),
}
}
Ok(())
}
pub fn is_end_stream(&self, stream: &store::Ptr) -> bool {
if !stream.state.is_recv_closed() {
return false;
}
stream.pending_recv.is_empty()
}
pub fn recv_data(&mut self, frame: frame::Data, stream: &mut store::Ptr) -> Result<(), Error> {
let sz = frame.payload().len();
// This should have been enforced at the codec::FramedRead layer, so
// this is just a sanity check.
assert!(sz <= MAX_WINDOW_SIZE as usize);
let sz = sz as WindowSize;
let is_ignoring_frame = stream.state.is_local_error();
if !is_ignoring_frame && !stream.state.is_recv_streaming() {
// TODO: There are cases where this can be a stream error of
// STREAM_CLOSED instead...
// Receiving a DATA frame when not expecting one is a protocol
// error.
proto_err!(conn: "unexpected DATA frame; stream={:?}", stream.id);
return Err(Error::library_go_away(Reason::PROTOCOL_ERROR));
}
tracing::trace!(
"recv_data; size={}; connection={}; stream={}",
sz,
self.flow.window_size(),
stream.recv_flow.window_size()
);
if is_ignoring_frame {
tracing::trace!(
"recv_data; frame ignored on locally reset {:?} for some time",
stream.id,
);
return self.ignore_data(sz);
}
// Ensure that there is enough capacity on the connection before acting
// on the stream.
self.consume_connection_window(sz)?;
if stream.recv_flow.window_size() < sz {
// > A receiver MAY respond with a stream error (Section 5.4.2) or
// > connection error (Section 5.4.1) of type FLOW_CONTROL_ERROR if
// > it is unable to accept a frame.
//
// So, for violating the **stream** window, we can send either a
// stream or connection error. We've opted to send a stream
// error.
return Err(Error::library_reset(stream.id, Reason::FLOW_CONTROL_ERROR));
}
if stream.dec_content_length(frame.payload().len()).is_err() {
proto_err!(stream:
"recv_data: content-length overflow; stream={:?}; len={:?}",
stream.id,
frame.payload().len(),
);
return Err(Error::library_reset(stream.id, Reason::PROTOCOL_ERROR));
}
if frame.is_end_stream() {
if stream.ensure_content_length_zero().is_err() {
proto_err!(stream:
"recv_data: content-length underflow; stream={:?}; len={:?}",
stream.id,
frame.payload().len(),
);
return Err(Error::library_reset(stream.id, Reason::PROTOCOL_ERROR));
}
if stream.state.recv_close().is_err() {
proto_err!(conn: "recv_data: failed to transition to closed state; stream={:?}", stream.id);
return Err(Error::library_go_away(Reason::PROTOCOL_ERROR));
}
}
// Received a frame, but no one cared about it. fix issue#648
if !stream.is_recv {
tracing::trace!(
"recv_data; frame ignored on stream release {:?} for some time",
stream.id,
);
self.release_connection_capacity(sz, &mut None);
return Ok(());
}
// Update stream level flow control
stream
.recv_flow
.send_data(sz)
.map_err(proto::Error::library_go_away)?;
// Track the data as in-flight
stream.in_flight_recv_data += sz;
let event = Event::Data(frame.into_payload());
// Push the frame onto the recv buffer
stream.pending_recv.push_back(&mut self.buffer, event);
stream.notify_recv();
Ok(())
}
pub fn ignore_data(&mut self, sz: WindowSize) -> Result<(), Error> {
// Ensure that there is enough capacity on the connection...
self.consume_connection_window(sz)?;
// Since we are ignoring this frame,
// we aren't returning the frame to the user. That means they
// have no way to release the capacity back to the connection. So
// we have to release it automatically.
//
// This call doesn't send a WINDOW_UPDATE immediately, just marks
// the capacity as available to be reclaimed. When the available
// capacity meets a threshold, a WINDOW_UPDATE is then sent.
self.release_connection_capacity(sz, &mut None);
Ok(())
}
pub fn consume_connection_window(&mut self, sz: WindowSize) -> Result<(), Error> {
if self.flow.window_size() < sz {
tracing::debug!(
"connection error FLOW_CONTROL_ERROR -- window_size ({:?}) < sz ({:?});",
self.flow.window_size(),
sz,
);
return Err(Error::library_go_away(Reason::FLOW_CONTROL_ERROR));
}
// Update connection level flow control
self.flow.send_data(sz).map_err(Error::library_go_away)?;
// Track the data as in-flight
self.in_flight_data += sz;
Ok(())
}
pub fn recv_push_promise(
&mut self,
frame: frame::PushPromise,
stream: &mut store::Ptr,
) -> Result<(), Error> {
stream.state.reserve_remote()?;
if frame.is_over_size() {
// A frame is over size if the decoded header block was bigger than
// SETTINGS_MAX_HEADER_LIST_SIZE.
//
// > A server that receives a larger header block than it is willing
// > to handle can send an HTTP 431 (Request Header Fields Too
// > Large) status code [RFC6585]. A client can discard responses
// > that it cannot process.
//
// So, if peer is a server, we'll send a 431. In either case,
// an error is recorded, which will send a REFUSED_STREAM,
// since we don't want any of the data frames either.
tracing::debug!(
"stream error REFUSED_STREAM -- recv_push_promise: \
headers frame is over size; promised_id={:?};",
frame.promised_id(),
);
return Err(Error::library_reset(
frame.promised_id(),
Reason::REFUSED_STREAM,
));
}
let promised_id = frame.promised_id();
let (pseudo, fields) = frame.into_parts();
let req = crate::server::Peer::convert_poll_message(pseudo, fields, promised_id)?;
if let Err(e) = frame::PushPromise::validate_request(&req) {
use PushPromiseHeaderError::*;
match e {
NotSafeAndCacheable => proto_err!(
stream:
"recv_push_promise: method {} is not safe and cacheable; promised_id={:?}",
req.method(),
promised_id,
),
InvalidContentLength(e) => proto_err!(
stream:
"recv_push_promise; promised request has invalid content-length {:?}; promised_id={:?}",
e,
promised_id,
),
}
return Err(Error::library_reset(promised_id, Reason::PROTOCOL_ERROR));
}
use super::peer::PollMessage::*;
stream
.pending_recv
.push_back(&mut self.buffer, Event::Headers(Server(req)));
stream.notify_recv();
Ok(())
}
/// Ensures that `id` is not in the `Idle` state.
pub fn ensure_not_idle(&self, id: StreamId) -> Result<(), Reason> {
if let Ok(next) = self.next_stream_id {
if id >= next {
tracing::debug!(
"stream ID implicitly closed, PROTOCOL_ERROR; stream={:?}",
id
);
return Err(Reason::PROTOCOL_ERROR);
}
}
// if next_stream_id is overflowed, that's ok.
Ok(())
}
/// Handle remote sending an explicit RST_STREAM.
pub fn recv_reset(
&mut self,
frame: frame::Reset,
stream: &mut Stream,
counts: &mut Counts,
) -> Result<(), Error> {
// Reseting a stream that the user hasn't accepted is possible,
// but should be done with care. These streams will continue
// to take up memory in the accept queue, but will no longer be
// counted as "concurrent" streams.
//
// So, we have a separate limit for these.
//
if stream.is_pending_accept {
if counts.can_inc_num_remote_reset_streams() {
counts.inc_num_remote_reset_streams();
} else {
tracing::warn!(
"recv_reset; remotely-reset pending-accept streams reached limit ({:?})",
counts.max_remote_reset_streams(),
);
return Err(Error::library_go_away_data(
Reason::ENHANCE_YOUR_CALM,
"too_many_resets",
));
}
}
// Notify the stream
stream.state.recv_reset(frame, stream.is_pending_send);
stream.notify_send();
stream.notify_recv();
Ok(())
}
/// Handle a connection-level error
pub fn handle_error(&mut self, err: &proto::Error, stream: &mut Stream) {
// Receive an error
stream.state.handle_error(err);
// If a receiver is waiting, notify it
stream.notify_send();
stream.notify_recv();
}
pub fn go_away(&mut self, last_processed_id: StreamId) {
assert!(self.max_stream_id >= last_processed_id);
self.max_stream_id = last_processed_id;
}
pub fn recv_eof(&mut self, stream: &mut Stream) {
stream.state.recv_eof();
stream.notify_send();
stream.notify_recv();
}
pub(super) fn clear_recv_buffer(&mut self, stream: &mut Stream) {
while stream.pending_recv.pop_front(&mut self.buffer).is_some() {
// drop it
}
}
/// Get the max ID of streams we can receive.
///
/// This gets lowered if we send a GOAWAY frame.
pub fn max_stream_id(&self) -> StreamId {
self.max_stream_id
}
pub fn next_stream_id(&self) -> Result<StreamId, Error> {
if let Ok(id) = self.next_stream_id {
Ok(id)
} else {
Err(Error::library_go_away(Reason::PROTOCOL_ERROR))
}
}
pub fn may_have_created_stream(&self, id: StreamId) -> bool {
if let Ok(next_id) = self.next_stream_id {
// Peer::is_local_init should have been called beforehand
debug_assert_eq!(id.is_server_initiated(), next_id.is_server_initiated(),);
id < next_id
} else {
true
}
}
pub(super) fn maybe_reset_next_stream_id(&mut self, id: StreamId) {
if let Ok(next_id) = self.next_stream_id {
// !Peer::is_local_init should have been called beforehand
debug_assert_eq!(id.is_server_initiated(), next_id.is_server_initiated());
if id >= next_id {
self.next_stream_id = id.next_id();
}
}
}
/// Returns true if the remote peer can reserve a stream with the given ID.
pub fn ensure_can_reserve(&self) -> Result<(), Error> {
if !self.is_push_enabled {
proto_err!(conn: "recv_push_promise: push is disabled");
return Err(Error::library_go_away(Reason::PROTOCOL_ERROR));
}
Ok(())
}
/// Add a locally reset stream to queue to be eventually reaped.
pub fn enqueue_reset_expiration(&mut self, stream: &mut store::Ptr, counts: &mut Counts) {
if !stream.state.is_local_error() || stream.is_pending_reset_expiration() {
return;
}
tracing::trace!("enqueue_reset_expiration; {:?}", stream.id);
if counts.can_inc_num_reset_streams() {
counts.inc_num_reset_streams();
self.pending_reset_expired.push(stream);
}
}
/// Send any pending refusals.
pub fn send_pending_refusal<T, B>(
&mut self,
cx: &mut Context,
dst: &mut Codec<T, Prioritized<B>>,
) -> Poll<io::Result<()>>
where
T: AsyncWrite + Unpin,
B: Buf,
{
if let Some(stream_id) = self.refused {
ready!(dst.poll_ready(cx))?;
// Create the RST_STREAM frame
let frame = frame::Reset::new(stream_id, Reason::REFUSED_STREAM);
// Buffer the frame
dst.buffer(frame.into()).expect("invalid RST_STREAM frame");
}
self.refused = None;
Poll::Ready(Ok(()))
}
pub fn clear_expired_reset_streams(&mut self, store: &mut Store, counts: &mut Counts) {
if !self.pending_reset_expired.is_empty() {
let now = Instant::now();
let reset_duration = self.reset_duration;
while let Some(stream) = self.pending_reset_expired.pop_if(store, |stream| {
let reset_at = stream.reset_at.expect("reset_at must be set if in queue");
// rust-lang/rust#86470 tracks a bug in the standard library where `Instant`
// subtraction can panic (because, on some platforms, `Instant` isn't actually
// monotonic). We use a saturating operation to avoid this panic here.
now.saturating_duration_since(reset_at) > reset_duration
}) {
counts.transition_after(stream, true);
}
}
}
pub fn clear_queues(
&mut self,
clear_pending_accept: bool,
store: &mut Store,
counts: &mut Counts,
) {
self.clear_stream_window_update_queue(store, counts);
self.clear_all_reset_streams(store, counts);
if clear_pending_accept {
self.clear_all_pending_accept(store, counts);
}
}
fn clear_stream_window_update_queue(&mut self, store: &mut Store, counts: &mut Counts) {
while let Some(stream) = self.pending_window_updates.pop(store) {
counts.transition(stream, |_, stream| {
tracing::trace!("clear_stream_window_update_queue; stream={:?}", stream.id);
})
}
}
/// Called on EOF
fn clear_all_reset_streams(&mut self, store: &mut Store, counts: &mut Counts) {
while let Some(stream) = self.pending_reset_expired.pop(store) {
counts.transition_after(stream, true);
}
}
fn clear_all_pending_accept(&mut self, store: &mut Store, counts: &mut Counts) {
while let Some(stream) = self.pending_accept.pop(store) {
counts.transition_after(stream, false);
}
}
pub fn poll_complete<T, B>(
&mut self,
cx: &mut Context,
store: &mut Store,
counts: &mut Counts,
dst: &mut Codec<T, Prioritized<B>>,
) -> Poll<io::Result<()>>
where
T: AsyncWrite + Unpin,
B: Buf,
{
// Send any pending connection level window updates
ready!(self.send_connection_window_update(cx, dst))?;
// Send any pending stream level window updates
ready!(self.send_stream_window_updates(cx, store, counts, dst))?;
Poll::Ready(Ok(()))
}
/// Send connection level window update
fn send_connection_window_update<T, B>(
&mut self,
cx: &mut Context,
dst: &mut Codec<T, Prioritized<B>>,
) -> Poll<io::Result<()>>
where
T: AsyncWrite + Unpin,
B: Buf,
{
if let Some(incr) = self.flow.unclaimed_capacity() {
let frame = frame::WindowUpdate::new(StreamId::zero(), incr);
// Ensure the codec has capacity
ready!(dst.poll_ready(cx))?;
// Buffer the WINDOW_UPDATE frame
dst.buffer(frame.into())
.expect("invalid WINDOW_UPDATE frame");
// Update flow control
self.flow
.inc_window(incr)
.expect("unexpected flow control state");
}
Poll::Ready(Ok(()))
}
/// Send stream level window update
pub fn send_stream_window_updates<T, B>(
&mut self,
cx: &mut Context,
store: &mut Store,
counts: &mut Counts,
dst: &mut Codec<T, Prioritized<B>>,
) -> Poll<io::Result<()>>
where
T: AsyncWrite + Unpin,
B: Buf,
{
loop {
// Ensure the codec has capacity
ready!(dst.poll_ready(cx))?;
// Get the next stream
let stream = match self.pending_window_updates.pop(store) {
Some(stream) => stream,
None => return Poll::Ready(Ok(())),
};
counts.transition(stream, |_, stream| {
tracing::trace!("pending_window_updates -- pop; stream={:?}", stream.id);
debug_assert!(!stream.is_pending_window_update);
if !stream.state.is_recv_streaming() {
// No need to send window updates on the stream if the stream is
// no longer receiving data.
//
// TODO: is this correct? We could possibly send a window
// update on a ReservedRemote stream if we already know
// we want to stream the data faster...
return;
}
// TODO: de-dup
if let Some(incr) = stream.recv_flow.unclaimed_capacity() {
// Create the WINDOW_UPDATE frame
let frame = frame::WindowUpdate::new(stream.id, incr);
// Buffer it
dst.buffer(frame.into())
.expect("invalid WINDOW_UPDATE frame");
// Update flow control
stream
.recv_flow
.inc_window(incr)
.expect("unexpected flow control state");
}
})
}
}
pub fn next_incoming(&mut self, store: &mut Store) -> Option<store::Key> {
self.pending_accept.pop(store).map(|ptr| ptr.key())
}
pub fn poll_data(
&mut self,
cx: &Context,
stream: &mut Stream,
) -> Poll<Option<Result<Bytes, proto::Error>>> {
match stream.pending_recv.pop_front(&mut self.buffer) {
Some(Event::Data(payload)) => Poll::Ready(Some(Ok(payload))),
Some(event) => {
// Frame is trailer
stream.pending_recv.push_front(&mut self.buffer, event);
// Notify the recv task. This is done just in case
// `poll_trailers` was called.
//
// It is very likely that `notify_recv` will just be a no-op (as
// the task will be None), so this isn't really much of a
// performance concern. It also means we don't have to track
// state to see if `poll_trailers` was called before `poll_data`
// returned `None`.
stream.notify_recv();
// No more data frames
Poll::Ready(None)
}
None => self.schedule_recv(cx, stream),
}
}
pub fn poll_trailers(
&mut self,
cx: &Context,
stream: &mut Stream,
) -> Poll<Option<Result<HeaderMap, proto::Error>>> {
match stream.pending_recv.pop_front(&mut self.buffer) {
Some(Event::Trailers(trailers)) => Poll::Ready(Some(Ok(trailers))),
Some(event) => {
// Frame is not trailers.. not ready to poll trailers yet.
stream.pending_recv.push_front(&mut self.buffer, event);
Poll::Pending
}
None => self.schedule_recv(cx, stream),
}
}
fn schedule_recv<T>(
&mut self,
cx: &Context,
stream: &mut Stream,
) -> Poll<Option<Result<T, proto::Error>>> {
if stream.state.ensure_recv_open()? {
// Request to get notified once more frames arrive
stream.recv_task = Some(cx.waker().clone());
Poll::Pending
} else {
// No more frames will be received
Poll::Ready(None)
}
}
}
// ===== impl Open =====
impl Open {
pub fn is_push_promise(&self) -> bool {
matches!(*self, Self::PushPromise)
}
}
// ===== impl RecvHeaderBlockError =====
impl<T> From<Error> for RecvHeaderBlockError<T> {
fn from(err: Error) -> Self {
RecvHeaderBlockError::State(err)
}
}