tdlight/test/http.cpp

565 lines
16 KiB
C++
Raw Normal View History

//
2024-01-01 01:07:21 +01:00
// Copyright Aliaksei Levin (levlam@telegram.org), Arseny Smirnov (arseny30@gmail.com) 2014-2024
//
// Distributed under the Boost Software License, Version 1.0. (See accompanying
// file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
//
#include "data.h"
2021-09-24 12:41:22 +02:00
#if TD_DARWIN_WATCH_OS
2021-03-26 16:52:45 +01:00
#include "td/net/DarwinHttp.h"
#endif
#include "td/net/HttpChunkedByteFlow.h"
#include "td/net/HttpHeaderCreator.h"
#include "td/net/HttpQuery.h"
#include "td/net/HttpReader.h"
#include "td/utils/AesCtrByteFlow.h"
#include "td/utils/algorithm.h"
#include "td/utils/base64.h"
#include "td/utils/buffer.h"
#include "td/utils/BufferedFd.h"
#include "td/utils/ByteFlow.h"
#include "td/utils/common.h"
#include "td/utils/crypto.h"
#include "td/utils/format.h"
#include "td/utils/Gzip.h"
#include "td/utils/GzipByteFlow.h"
#include "td/utils/logging.h"
#include "td/utils/misc.h"
#include "td/utils/port/detail/PollableFd.h"
#include "td/utils/port/FileFd.h"
#include "td/utils/port/path.h"
#include "td/utils/port/PollFlags.h"
#include "td/utils/port/thread_local.h"
#include "td/utils/Random.h"
#include "td/utils/Slice.h"
#include "td/utils/SliceBuilder.h"
#include "td/utils/Status.h"
#include "td/utils/tests.h"
#include "td/utils/UInt.h"
#include <algorithm>
2021-03-26 16:04:01 +01:00
#include <condition_variable>
#include <limits>
2021-03-27 03:19:22 +01:00
#include <mutex>
2021-03-26 16:04:01 +01:00
static td::string make_chunked(const td::string &str) {
auto v = td::rand_split(str);
td::string res;
for (auto &s : v) {
res += PSTRING() << td::format::as_hex_dump(static_cast<td::int32>(s.size()));
res += "\r\n";
res += s;
res += "\r\n";
}
res += "0\r\n\r\n";
return res;
}
static td::string gen_http_content() {
int t = td::Random::fast(0, 2);
int len;
if (t == 0) {
len = td::Random::fast(1, 10);
} else if (t == 1) {
len = td::Random::fast(100, 200);
} else {
len = td::Random::fast(1000, 20000);
}
return td::rand_string(std::numeric_limits<char>::min(), std::numeric_limits<char>::max(), len);
}
static td::string make_http_query(td::string content, td::string content_type, bool is_chunked, bool is_gzip,
double gzip_k = 5, td::string zip_override = td::string()) {
td::HttpHeaderCreator hc;
hc.init_post("/");
hc.add_header("jfkdlsahhjk", td::rand_string('a', 'z', td::Random::fast(1, 2000)));
if (!content_type.empty()) {
hc.add_header("content-type", content_type);
2021-03-19 16:01:25 +01:00
}
if (is_gzip) {
td::BufferSlice zip;
if (zip_override.empty()) {
zip = td::gzencode(content, gzip_k);
} else {
zip = td::BufferSlice(zip_override);
}
if (!zip.empty()) {
hc.add_header("content-encoding", "gzip");
content = zip.as_slice().str();
}
}
if (is_chunked) {
hc.add_header("transfer-encoding", "chunked");
content = make_chunked(content);
} else {
hc.set_content_size(content.size());
}
auto r_header = hc.finish();
CHECK(r_header.is_ok());
return PSTRING() << r_header.ok() << content;
}
static td::string rand_http_query(td::string content) {
bool is_chunked = td::Random::fast_bool();
bool is_gzip = td::Random::fast_bool();
return make_http_query(std::move(content), td::string(), is_chunked, is_gzip);
}
static td::string join(const td::vector<td::string> &v) {
td::string res;
for (auto &s : v) {
res += s;
}
return res;
}
TEST(Http, stack_overflow) {
td::ChainBufferWriter writer;
td::BufferSlice slice(td::string(256, 'A'));
for (int i = 0; i < 1000000; i++) {
td::ChainBufferWriter tmp_writer;
writer.append(slice.clone());
}
{
auto reader = writer.extract_reader();
reader.sync_with_writer();
}
}
TEST(Http, reader) {
#if TD_ANDROID || TD_TIZEN
return;
#endif
td::clear_thread_locals();
auto start_mem = td::BufferAllocator::get_buffer_mem();
auto start_size = td::BufferAllocator::get_buffer_slice_size();
{
td::BufferSlice a("test test");
td::BufferSlice b = std::move(a);
#if TD_CLANG
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wunknown-pragmas"
#pragma clang diagnostic ignored "-Wunknown-warning-option"
#pragma clang diagnostic ignored "-Wself-move"
2024-03-24 22:01:32 +01:00
#endif
#if TD_GCC
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wconversion"
#endif
a = std::move(a);
b = std::move(b);
2024-03-24 22:01:32 +01:00
#if TD_GCC
#pragma GCC diagnostic pop
#endif
#if TD_CLANG
#pragma clang diagnostic pop
#endif
a = std::move(b);
td::BufferSlice c = a.from_slice(a);
CHECK(c.size() == a.size());
}
td::clear_thread_locals();
ASSERT_EQ(start_mem, td::BufferAllocator::get_buffer_mem());
ASSERT_EQ(start_size, td::BufferAllocator::get_buffer_slice_size());
for (int i = 0; i < 20; i++) {
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::HttpReader reader;
int max_post_size = 10000;
reader.init(&input, max_post_size, 0);
td::vector<td::string> contents(100);
std::generate(contents.begin(), contents.end(), gen_http_content);
auto v = td::transform(contents, rand_http_query);
auto vec_str = td::rand_split(join(v));
td::HttpQuery q;
td::vector<td::string> res;
for (auto &str : vec_str) {
input_writer.append(str);
input.sync_with_writer();
while (true) {
auto r_state = reader.read_next(&q);
LOG_IF(ERROR, r_state.is_error()) << r_state.error() << td::tag("ok", res.size());
ASSERT_TRUE(r_state.is_ok());
auto state = r_state.ok();
if (state == 0) {
if (q.files_.empty()) {
ASSERT_TRUE(td::narrow_cast<int>(q.content_.size()) <= max_post_size);
auto expected = contents[res.size()];
ASSERT_EQ(expected, q.content_.str());
res.push_back(q.content_.str());
} else {
auto r_fd = td::FileFd::open(q.files_[0].temp_file_name, td::FileFd::Read);
ASSERT_TRUE(r_fd.is_ok());
auto fd = r_fd.move_as_ok();
td::string content(td::narrow_cast<std::size_t>(q.files_[0].size), '\0');
auto r_size = fd.read(td::MutableSlice(content));
ASSERT_TRUE(r_size.is_ok());
ASSERT_TRUE(r_size.ok() == content.size());
ASSERT_TRUE(td::narrow_cast<int>(content.size()) > max_post_size);
ASSERT_EQ(contents[res.size()], content);
res.push_back(content);
fd.close();
}
} else {
break;
}
}
}
ASSERT_EQ(contents.size(), res.size());
ASSERT_EQ(contents, res);
}
td::clear_thread_locals();
ASSERT_EQ(start_mem, td::BufferAllocator::get_buffer_mem());
ASSERT_EQ(start_size, td::BufferAllocator::get_buffer_slice_size());
}
TEST(Http, gzip_bomb) {
#if TD_ANDROID || TD_TIZEN || TD_EMSCRIPTEN // the test must be disabled on low-memory systems
return;
#endif
auto gzip_bomb_str =
td::gzdecode(td::gzdecode(td::base64url_decode(td::Slice(gzip_bomb, gzip_bomb_size)).ok()).as_slice())
.as_slice()
.str();
auto query = make_http_query(td::string(), td::string(), false, true, 0.01, gzip_bomb_str);
auto parts = td::rand_split(query);
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::HttpReader reader;
td::HttpQuery q;
reader.init(&input, 100000000, 0);
for (auto &part : parts) {
input_writer.append(part);
input.sync_with_writer();
auto r_state = reader.read_next(&q);
if (r_state.is_error()) {
LOG(INFO) << r_state.error();
return;
}
ASSERT_TRUE(r_state.ok() != 0);
}
}
2021-03-19 16:01:25 +01:00
TEST(Http, gzip) {
auto gzip_str = td::gzdecode(td::base64url_decode(td::Slice(gzip, gzip_size)).ok()).as_slice().str();
2021-03-19 16:01:25 +01:00
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::HttpReader reader;
2021-03-19 16:01:25 +01:00
reader.init(&input, 0, 0);
auto query = make_http_query(td::string(), "application/json", false, true, 0.01, gzip_str);
2021-03-19 16:01:25 +01:00
input_writer.append(query);
input.sync_with_writer();
td::HttpQuery q;
2021-03-19 16:01:25 +01:00
auto r_state = reader.read_next(&q);
ASSERT_TRUE(r_state.is_error());
ASSERT_EQ(413, r_state.error().code());
}
TEST(Http, aes_ctr_encode_decode_flow) {
auto str = td::rand_string('a', 'z', 1000000);
auto parts = td::rand_split(str);
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::ByteFlowSource source(&input);
td::UInt256 key;
td::UInt128 iv;
td::Random::secure_bytes(key.raw, sizeof(key));
td::Random::secure_bytes(iv.raw, sizeof(iv));
td::AesCtrByteFlow aes_encode;
aes_encode.init(key, iv);
td::AesCtrByteFlow aes_decode;
aes_decode.init(key, iv);
td::ByteFlowSink sink;
source >> aes_encode >> aes_decode >> sink;
ASSERT_TRUE(!sink.is_ready());
for (auto &part : parts) {
input_writer.append(part);
source.wakeup();
}
ASSERT_TRUE(!sink.is_ready());
source.close_input(td::Status::OK());
ASSERT_TRUE(sink.is_ready());
LOG_IF(ERROR, sink.status().is_error()) << sink.status();
ASSERT_TRUE(sink.status().is_ok());
ASSERT_EQ(str, sink.result()->move_as_buffer_slice().as_slice().str());
}
TEST(Http, aes_file_encryption) {
auto str = td::rand_string('a', 'z', 1000000);
td::CSlice name = "test_encryption";
td::unlink(name).ignore();
td::UInt256 key;
td::UInt128 iv;
td::Random::secure_bytes(key.raw, sizeof(key));
td::Random::secure_bytes(iv.raw, sizeof(iv));
{
td::BufferedFdBase<td::FileFd> fd(td::FileFd::open(name, td::FileFd::Write | td::FileFd::Create).move_as_ok());
auto parts = td::rand_split(str);
td::ChainBufferWriter output_writer;
auto output_reader = output_writer.extract_reader();
td::ByteFlowSource source(&output_reader);
td::AesCtrByteFlow aes_encode;
aes_encode.init(key, iv);
td::ByteFlowSink sink;
source >> aes_encode >> sink;
fd.set_output_reader(sink.get_output());
for (auto &part : parts) {
output_writer.append(part);
source.wakeup();
fd.flush_write().ensure();
}
fd.close();
}
{
td::BufferedFdBase<td::FileFd> fd(td::FileFd::open(name, td::FileFd::Read).move_as_ok());
td::ChainBufferWriter input_writer;
auto input_reader = input_writer.extract_reader();
td::ByteFlowSource source(&input_reader);
td::AesCtrByteFlow aes_encode;
aes_encode.init(key, iv);
td::ByteFlowSink sink;
source >> aes_encode >> sink;
fd.set_input_writer(&input_writer);
fd.get_poll_info().add_flags(td::PollFlags::Read());
while (can_read_local(fd)) {
fd.flush_read(4096).ensure();
source.wakeup();
}
fd.close();
source.close_input(td::Status::OK());
ASSERT_TRUE(sink.is_ready());
LOG_IF(ERROR, sink.status().is_error()) << sink.status();
ASSERT_TRUE(sink.status().is_ok());
auto result = sink.result()->move_as_buffer_slice().as_slice().str();
ASSERT_EQ(str, result);
}
td::unlink(name).ignore();
}
TEST(Http, chunked_flow) {
auto str = td::rand_string('a', 'z', 100);
auto parts = td::rand_split(make_chunked(str));
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::ByteFlowSource source(&input);
td::HttpChunkedByteFlow chunked_flow;
td::ByteFlowSink sink;
source >> chunked_flow >> sink;
for (auto &part : parts) {
input_writer.append(part);
source.wakeup();
}
source.close_input(td::Status::OK());
ASSERT_TRUE(sink.is_ready());
LOG_IF(ERROR, sink.status().is_error()) << sink.status();
ASSERT_TRUE(sink.status().is_ok());
auto res = sink.result()->move_as_buffer_slice().as_slice().str();
ASSERT_EQ(str.size(), res.size());
ASSERT_EQ(str, res);
}
TEST(Http, chunked_flow_error) {
auto str = td::rand_string('a', 'z', 100000);
for (int d = 1; d < 100; d += 10) {
auto new_str = make_chunked(str);
new_str.resize(str.size() - d);
auto parts = td::rand_split(new_str);
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::ByteFlowSource source(&input);
td::HttpChunkedByteFlow chunked_flow;
td::ByteFlowSink sink;
source >> chunked_flow >> sink;
for (auto &part : parts) {
input_writer.append(part);
source.wakeup();
}
ASSERT_TRUE(!sink.is_ready());
source.close_input(td::Status::OK());
ASSERT_TRUE(sink.is_ready());
ASSERT_TRUE(!sink.status().is_ok());
}
}
TEST(Http, gzip_chunked_flow) {
auto str = td::rand_string('a', 'z', 1000000);
auto parts = td::rand_split(make_chunked(td::gzencode(str, 2.0).as_slice().str()));
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::ByteFlowSource source(&input);
td::HttpChunkedByteFlow chunked_flow;
td::GzipByteFlow gzip_flow(td::Gzip::Mode::Decode);
td::ByteFlowSink sink;
source >> chunked_flow >> gzip_flow >> sink;
for (auto &part : parts) {
input_writer.append(part);
source.wakeup();
}
source.close_input(td::Status::OK());
ASSERT_TRUE(sink.is_ready());
LOG_IF(ERROR, sink.status().is_error()) << sink.status();
ASSERT_TRUE(sink.status().is_ok());
ASSERT_EQ(str, sink.result()->move_as_buffer_slice().as_slice().str());
}
TEST(Http, gzip_bomb_with_limit) {
td::string gzip_bomb_str;
{
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::GzipByteFlow gzip_flow(td::Gzip::Mode::Encode);
td::ByteFlowSource source(&input);
td::ByteFlowSink sink;
source >> gzip_flow >> sink;
td::string s(1 << 16, 'a');
for (int i = 0; i < 1000; i++) {
input_writer.append(s);
source.wakeup();
}
source.close_input(td::Status::OK());
ASSERT_TRUE(sink.is_ready());
LOG_IF(ERROR, sink.status().is_error()) << sink.status();
ASSERT_TRUE(sink.status().is_ok());
gzip_bomb_str = sink.result()->move_as_buffer_slice().as_slice().str();
}
auto query = make_http_query(td::string(), td::string(), false, true, 0.01, gzip_bomb_str);
auto parts = td::rand_split(query);
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::HttpReader reader;
td::HttpQuery q;
reader.init(&input, 1000000);
bool ok = false;
for (auto &part : parts) {
input_writer.append(part);
input.sync_with_writer();
auto r_state = reader.read_next(&q);
if (r_state.is_error()) {
LOG(FATAL) << r_state.error();
return;
} else if (r_state.ok() == 0) {
ok = true;
}
}
ASSERT_TRUE(ok);
}
2021-03-26 16:04:01 +01:00
2023-11-25 10:35:21 +01:00
TEST(Http, partial_form_data) {
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::HttpReader reader;
reader.init(&input, 0, 0);
auto query =
make_http_query("------abcd\r\nCo", "Content-Type: multipart/form-data; boundary=----abcd", false, false);
input_writer.append(query);
input.sync_with_writer();
td::HttpQuery q;
auto r_state = reader.read_next(&q);
ASSERT_TRUE(r_state.is_error());
ASSERT_EQ(400, r_state.error().code());
}
TEST(Http, form_data) {
td::ChainBufferWriter input_writer;
auto input = input_writer.extract_reader();
td::HttpReader reader;
reader.init(&input, 0, 1);
auto query = make_http_query(
"------abcd\r\n"
"Content-Disposition: form-data; name=\"text\"\r\n"
"\r\n"
"some text\r\n"
"------abcd\r\n"
"Content-Disposition: form-data; name=\"text2\"\r\n"
"\r\n"
"some text\r\n"
"more text\r\n"
"------abcd\r\n"
"Content-Disposition: form-data; name=\"file\"; filename=\"file.txt\"\r\n"
"Content-Type: text/plain\r\n"
"\r\n"
"File content\r\n"
"------abcd--",
"Content-Type: multipart/form-data; boundary=----abcd", false, false);
input_writer.append(query);
input.sync_with_writer();
td::HttpQuery q;
auto r_state = reader.read_next(&q);
ASSERT_TRUE(r_state.is_ok());
ASSERT_EQ(2u, q.args_.size());
ASSERT_EQ("text", q.args_[0].first);
ASSERT_EQ("some text", q.args_[0].second);
ASSERT_EQ("text2", q.args_[1].first);
ASSERT_EQ("some text\r\nmore text", q.args_[1].second);
ASSERT_EQ(1u, q.files_.size());
ASSERT_EQ("file.txt", q.files_[0].name);
ASSERT_EQ("file", q.files_[0].field_name);
ASSERT_EQ("text/plain", q.files_[0].content_type);
ASSERT_EQ(12, q.files_[0].size);
ASSERT_TRUE(!q.files_[0].temp_file_name.empty());
}
2021-09-24 12:41:22 +02:00
#if TD_DARWIN_WATCH_OS
2021-03-26 16:04:01 +01:00
struct Baton {
std::mutex mutex;
std::condition_variable cond;
bool is_ready{false};
void wait() {
std::unique_lock<std::mutex> lock(mutex);
cond.wait(lock, [&] { return is_ready; });
}
void post() {
{
std::unique_lock<std::mutex> lock(mutex);
is_ready = true;
}
cond.notify_all();
}
void reset() {
is_ready = false;
}
};
TEST(Http, Darwin) {
Baton baton;
td::DarwinHttp::get("http://example.com", [&](td::BufferSlice data) { baton.post(); });
2021-03-26 16:04:01 +01:00
baton.wait();
}
2021-03-26 16:52:45 +01:00
#endif