blob: 4ccc2e52782b9f0cfb1584a089f15bbcf16983d8 [file] [log] [blame]
/*
* Copyright (c) 2021 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "net/dcsctp/rx/traditional_reassembly_streams.h"
#include <stddef.h>
#include <cstdint>
#include <functional>
#include <iterator>
#include <map>
#include <numeric>
#include <utility>
#include <vector>
#include "absl/algorithm/container.h"
#include "absl/types/optional.h"
#include "api/array_view.h"
#include "net/dcsctp/common/sequence_numbers.h"
#include "net/dcsctp/packet/chunk/forward_tsn_common.h"
#include "net/dcsctp/packet/data.h"
#include "net/dcsctp/public/dcsctp_message.h"
#include "rtc_base/logging.h"
namespace dcsctp {
namespace {
// Given a map (`chunks`) and an iterator to within that map (`iter`), this
// function will return an iterator to the first chunk in that message, which
// has the `is_beginning` flag set. If there are any gaps, or if the beginning
// can't be found, `absl::nullopt` is returned.
absl::optional<std::map<UnwrappedTSN, Data>::iterator> FindBeginning(
const std::map<UnwrappedTSN, Data>& chunks,
std::map<UnwrappedTSN, Data>::iterator iter) {
UnwrappedTSN prev_tsn = iter->first;
for (;;) {
if (iter->second.is_beginning) {
return iter;
}
if (iter == chunks.begin()) {
return absl::nullopt;
}
--iter;
if (iter->first.next_value() != prev_tsn) {
return absl::nullopt;
}
prev_tsn = iter->first;
}
}
// Given a map (`chunks`) and an iterator to within that map (`iter`), this
// function will return an iterator to the chunk after the last chunk in that
// message, which has the `is_end` flag set. If there are any gaps, or if the
// end can't be found, `absl::nullopt` is returned.
absl::optional<std::map<UnwrappedTSN, Data>::iterator> FindEnd(
std::map<UnwrappedTSN, Data>& chunks,
std::map<UnwrappedTSN, Data>::iterator iter) {
UnwrappedTSN prev_tsn = iter->first;
for (;;) {
if (iter->second.is_end) {
return ++iter;
}
++iter;
if (iter == chunks.end()) {
return absl::nullopt;
}
if (iter->first != prev_tsn.next_value()) {
return absl::nullopt;
}
prev_tsn = iter->first;
}
}
} // namespace
TraditionalReassemblyStreams::TraditionalReassemblyStreams(
absl::string_view log_prefix,
OnAssembledMessage on_assembled_message,
const DcSctpSocketHandoverState* handover_state)
: log_prefix_(log_prefix),
on_assembled_message_(std::move(on_assembled_message)) {
if (handover_state) {
for (const DcSctpSocketHandoverState::OrderedStream& state_stream :
handover_state->rx.ordered_streams) {
ordered_streams_.emplace(
std::piecewise_construct,
std::forward_as_tuple(StreamID(state_stream.id)),
std::forward_as_tuple(this, SSN(state_stream.next_ssn)));
}
for (const DcSctpSocketHandoverState::UnorderedStream& state_stream :
handover_state->rx.unordered_streams) {
unordered_streams_.emplace(
std::piecewise_construct,
std::forward_as_tuple(StreamID(state_stream.id)),
std::forward_as_tuple(this));
}
}
}
int TraditionalReassemblyStreams::UnorderedStream::Add(UnwrappedTSN tsn,
Data data) {
int queued_bytes = data.size();
auto [it, inserted] = chunks_.emplace(tsn, std::move(data));
if (!inserted) {
return 0;
}
queued_bytes -= TryToAssembleMessage(it);
return queued_bytes;
}
size_t TraditionalReassemblyStreams::UnorderedStream::TryToAssembleMessage(
ChunkMap::iterator iter) {
// TODO(boivie): This method is O(N) with the number of fragments in a
// message, which can be inefficient for very large values of N. This could be
// optimized by e.g. only trying to assemble a message once _any_ beginning
// and _any_ end has been found.
absl::optional<ChunkMap::iterator> start = FindBeginning(chunks_, iter);
if (!start.has_value()) {
return 0;
}
absl::optional<ChunkMap::iterator> end = FindEnd(chunks_, iter);
if (!end.has_value()) {
return 0;
}
size_t bytes_assembled = AssembleMessage(*start, *end);
chunks_.erase(*start, *end);
return bytes_assembled;
}
size_t TraditionalReassemblyStreams::StreamBase::AssembleMessage(
const ChunkMap::iterator start,
const ChunkMap::iterator end) {
size_t count = std::distance(start, end);
if (count == 1) {
// Fast path - zero-copy
const Data& data = start->second;
size_t payload_size = start->second.size();
UnwrappedTSN tsns[1] = {start->first};
DcSctpMessage message(data.stream_id, data.ppid, std::move(data.payload));
parent_.on_assembled_message_(tsns, std::move(message));
return payload_size;
}
// Slow path - will need to concatenate the payload.
std::vector<UnwrappedTSN> tsns;
std::vector<uint8_t> payload;
size_t payload_size = std::accumulate(
start, end, 0,
[](size_t v, const auto& p) { return v + p.second.size(); });
tsns.reserve(count);
payload.reserve(payload_size);
for (auto it = start; it != end; ++it) {
const Data& data = it->second;
tsns.push_back(it->first);
payload.insert(payload.end(), data.payload.begin(), data.payload.end());
}
DcSctpMessage message(start->second.stream_id, start->second.ppid,
std::move(payload));
parent_.on_assembled_message_(tsns, std::move(message));
return payload_size;
}
size_t TraditionalReassemblyStreams::UnorderedStream::EraseTo(
UnwrappedTSN tsn) {
auto end_iter = chunks_.upper_bound(tsn);
size_t removed_bytes = std::accumulate(
chunks_.begin(), end_iter, 0,
[](size_t r, const auto& p) { return r + p.second.size(); });
chunks_.erase(chunks_.begin(), end_iter);
return removed_bytes;
}
size_t TraditionalReassemblyStreams::OrderedStream::TryToAssembleMessage() {
if (chunks_by_ssn_.empty() || chunks_by_ssn_.begin()->first != next_ssn_) {
return 0;
}
ChunkMap& chunks = chunks_by_ssn_.begin()->second;
if (!chunks.begin()->second.is_beginning || !chunks.rbegin()->second.is_end) {
return 0;
}
uint32_t tsn_diff =
UnwrappedTSN::Difference(chunks.rbegin()->first, chunks.begin()->first);
if (tsn_diff != chunks.size() - 1) {
return 0;
}
size_t assembled_bytes = AssembleMessage(chunks.begin(), chunks.end());
chunks_by_ssn_.erase(chunks_by_ssn_.begin());
next_ssn_.Increment();
return assembled_bytes;
}
size_t TraditionalReassemblyStreams::OrderedStream::TryToAssembleMessages() {
size_t assembled_bytes = 0;
for (;;) {
size_t assembled_bytes_this_iter = TryToAssembleMessage();
if (assembled_bytes_this_iter == 0) {
break;
}
assembled_bytes += assembled_bytes_this_iter;
}
return assembled_bytes;
}
int TraditionalReassemblyStreams::OrderedStream::Add(UnwrappedTSN tsn,
Data data) {
int queued_bytes = data.size();
UnwrappedSSN ssn = ssn_unwrapper_.Unwrap(data.ssn);
auto [unused, inserted] = chunks_by_ssn_[ssn].emplace(tsn, std::move(data));
if (!inserted) {
return 0;
}
if (ssn == next_ssn_) {
queued_bytes -= TryToAssembleMessages();
}
return queued_bytes;
}
size_t TraditionalReassemblyStreams::OrderedStream::EraseTo(SSN ssn) {
UnwrappedSSN unwrapped_ssn = ssn_unwrapper_.Unwrap(ssn);
auto end_iter = chunks_by_ssn_.upper_bound(unwrapped_ssn);
size_t removed_bytes = std::accumulate(
chunks_by_ssn_.begin(), end_iter, 0, [](size_t r1, const auto& p) {
return r1 +
absl::c_accumulate(p.second, 0, [](size_t r2, const auto& q) {
return r2 + q.second.size();
});
});
chunks_by_ssn_.erase(chunks_by_ssn_.begin(), end_iter);
if (unwrapped_ssn >= next_ssn_) {
unwrapped_ssn.Increment();
next_ssn_ = unwrapped_ssn;
}
removed_bytes += TryToAssembleMessages();
return removed_bytes;
}
int TraditionalReassemblyStreams::Add(UnwrappedTSN tsn, Data data) {
if (data.is_unordered) {
auto it = unordered_streams_.emplace(data.stream_id, this).first;
return it->second.Add(tsn, std::move(data));
}
auto it = ordered_streams_.emplace(data.stream_id, this).first;
return it->second.Add(tsn, std::move(data));
}
size_t TraditionalReassemblyStreams::HandleForwardTsn(
UnwrappedTSN new_cumulative_ack_tsn,
rtc::ArrayView<const AnyForwardTsnChunk::SkippedStream> skipped_streams) {
size_t bytes_removed = 0;
// The `skipped_streams` only cover ordered messages - need to
// iterate all unordered streams manually to remove those chunks.
for (auto& [unused, stream] : unordered_streams_) {
bytes_removed += stream.EraseTo(new_cumulative_ack_tsn);
}
for (const auto& skipped_stream : skipped_streams) {
auto it = ordered_streams_.find(skipped_stream.stream_id);
if (it != ordered_streams_.end()) {
bytes_removed += it->second.EraseTo(skipped_stream.ssn);
}
}
return bytes_removed;
}
void TraditionalReassemblyStreams::ResetStreams(
rtc::ArrayView<const StreamID> stream_ids) {
if (stream_ids.empty()) {
for (auto& [stream_id, stream] : ordered_streams_) {
RTC_DLOG(LS_VERBOSE) << log_prefix_
<< "Resetting implicit stream_id=" << *stream_id;
stream.Reset();
}
} else {
for (StreamID stream_id : stream_ids) {
auto it = ordered_streams_.find(stream_id);
if (it != ordered_streams_.end()) {
RTC_DLOG(LS_VERBOSE)
<< log_prefix_ << "Resetting explicit stream_id=" << *stream_id;
it->second.Reset();
}
}
}
}
HandoverReadinessStatus TraditionalReassemblyStreams::GetHandoverReadiness()
const {
HandoverReadinessStatus status;
for (const auto& [unused, stream] : ordered_streams_) {
if (stream.has_unassembled_chunks()) {
status.Add(HandoverUnreadinessReason::kOrderedStreamHasUnassembledChunks);
break;
}
}
for (const auto& [unused, stream] : unordered_streams_) {
if (stream.has_unassembled_chunks()) {
status.Add(
HandoverUnreadinessReason::kUnorderedStreamHasUnassembledChunks);
break;
}
}
return status;
}
void TraditionalReassemblyStreams::AddHandoverState(
DcSctpSocketHandoverState& state) {
for (const auto& [stream_id, stream] : ordered_streams_) {
DcSctpSocketHandoverState::OrderedStream state_stream;
state_stream.id = stream_id.value();
state_stream.next_ssn = stream.next_ssn().value();
state.rx.ordered_streams.push_back(std::move(state_stream));
}
for (const auto& [stream_id, unused] : unordered_streams_) {
DcSctpSocketHandoverState::UnorderedStream state_stream;
state_stream.id = stream_id.value();
state.rx.unordered_streams.push_back(std::move(state_stream));
}
}
} // namespace dcsctp