resolv/DnsTlsSocket.cpp

/*
 * Copyright (C) 2018 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#define LOG_TAG "DnsTlsSocket"
//#define LOG_NDEBUG 0

#include "netd_resolv/DnsTlsSocket.h"

#include <arpa/inet.h>
#include <arpa/nameser.h>
#include <errno.h>
#include <linux/tcp.h>
#include <openssl/err.h>
#include <openssl/sha.h>
#include <sys/eventfd.h>
#include <sys/poll.h>
#include <algorithm>

#include "netd_resolv/DnsTlsSessionCache.h"
#include "netd_resolv/IDnsTlsSocketObserver.h"

#include "log/log.h"
#include "netdutils/SocketOption.h"

namespace android {

using netdutils::enableSockopt;
using netdutils::enableTcpKeepAlives;
using netdutils::isOk;
using netdutils::Status;

namespace net {
namespace {

constexpr const char kCaCertDir[] = "/system/etc/security/cacerts";
constexpr size_t SHA256_SIZE = SHA256_DIGEST_LENGTH;

int waitForReading(int fd) {
    struct pollfd fds = { .fd = fd, .events = POLLIN };
    const int ret = TEMP_FAILURE_RETRY(poll(&fds, 1, -1));
    return ret;
}

int waitForWriting(int fd) {
    struct pollfd fds = { .fd = fd, .events = POLLOUT };
    const int ret = TEMP_FAILURE_RETRY(poll(&fds, 1, -1));
    return ret;
}

}  // namespace

Status DnsTlsSocket::tcpConnect() {
    ALOGV("%u connecting TCP socket", mMark);
    int type = SOCK_NONBLOCK | SOCK_CLOEXEC;
    switch (mServer.protocol) {
        case IPPROTO_TCP:
            type |= SOCK_STREAM;
            break;
        default:
            return Status(EPROTONOSUPPORT);
    }

    mSslFd.reset(socket(mServer.ss.ss_family, type, mServer.protocol));
    if (mSslFd.get() == -1) {
        ALOGE("Failed to create socket");
        return Status(errno);
    }

    const socklen_t len = sizeof(mMark);
    if (setsockopt(mSslFd.get(), SOL_SOCKET, SO_MARK, &mMark, len) == -1) {
        ALOGE("Failed to set socket mark");
        mSslFd.reset();
        return Status(errno);
    }

    const Status tfo = enableSockopt(mSslFd.get(), SOL_TCP, TCP_FASTOPEN_CONNECT);
    if (!isOk(tfo) && tfo.code() != ENOPROTOOPT) {
        ALOGI("Failed to enable TFO: %s", tfo.msg().c_str());
    }

    // Send 5 keepalives, 3 seconds apart, after 15 seconds of inactivity.
    enableTcpKeepAlives(mSslFd.get(), 15U, 5U, 3U).ignoreError();

    if (connect(mSslFd.get(), reinterpret_cast<const struct sockaddr *>(&mServer.ss),
                sizeof(mServer.ss)) != 0 &&
            errno != EINPROGRESS) {
        ALOGV("Socket failed to connect");
        mSslFd.reset();
        return Status(errno);
    }

    return netdutils::status::ok;
}

bool getSPKIDigest(const X509* cert, std::vector<uint8_t>* out) {
    int spki_len = i2d_X509_PUBKEY(X509_get_X509_PUBKEY(cert), nullptr);
    unsigned char spki[spki_len];
    unsigned char* temp = spki;
    if (spki_len != i2d_X509_PUBKEY(X509_get_X509_PUBKEY(cert), &temp)) {
        ALOGW("SPKI length mismatch");
        return false;
    }
    out->resize(SHA256_SIZE);
    unsigned int digest_len = 0;
    int ret = EVP_Digest(spki, spki_len, out->data(), &digest_len, EVP_sha256(), nullptr);
    if (ret != 1) {
        ALOGW("Server cert digest extraction failed");
        return false;
    }
    if (digest_len != out->size()) {
        ALOGW("Wrong digest length: %d", digest_len);
        return false;
    }
    return true;
}

bool DnsTlsSocket::initialize() {
    // This method should only be called once, at the beginning, so locking should be
    // unnecessary.  This lock only serves to help catch bugs in code that calls this method.
    std::lock_guard guard(mLock);
    if (mSslCtx) {
        // This is a bug in the caller.
        return false;
    }
    mSslCtx.reset(SSL_CTX_new(TLS_method()));
    if (!mSslCtx) {
        return false;
    }

    // Load system CA certs for hostname verification.
    //
    // For discussion of alternative, sustainable approaches see b/71909242.
    if (SSL_CTX_load_verify_locations(mSslCtx.get(), nullptr, kCaCertDir) != 1) {
        ALOGE("Failed to load CA cert dir: %s", kCaCertDir);
        return false;
    }

    // Enable TLS false start
    SSL_CTX_set_false_start_allowed_without_alpn(mSslCtx.get(), 1);
    SSL_CTX_set_mode(mSslCtx.get(), SSL_MODE_ENABLE_FALSE_START);

    // Enable session cache
    mCache->prepareSslContext(mSslCtx.get());

    // Connect
    Status status = tcpConnect();
    if (!status.ok()) {
        return false;
    }
    mSsl = sslConnect(mSslFd.get());
    if (!mSsl) {
        return false;
    }

    mEventFd.reset(eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC));

    // Start the I/O loop.
    mLoopThread.reset(new std::thread(&DnsTlsSocket::loop, this));

    return true;
}

bssl::UniquePtr<SSL> DnsTlsSocket::sslConnect(int fd) {
    if (!mSslCtx) {
        ALOGE("Internal error: context is null in sslConnect");
        return nullptr;
    }
    if (!SSL_CTX_set_min_proto_version(mSslCtx.get(), TLS1_2_VERSION)) {
        ALOGE("Failed to set minimum TLS version");
        return nullptr;
    }

    bssl::UniquePtr<SSL> ssl(SSL_new(mSslCtx.get()));
    // This file descriptor is owned by mSslFd, so don't let libssl close it.
    bssl::UniquePtr<BIO> bio(BIO_new_socket(fd, BIO_NOCLOSE));
    SSL_set_bio(ssl.get(), bio.get(), bio.get());
    bio.release();

    if (!mCache->prepareSsl(ssl.get())) {
        return nullptr;
    }

    if (!mServer.name.empty()) {
        if (SSL_set_tlsext_host_name(ssl.get(), mServer.name.c_str()) != 1) {
            ALOGE("Failed to set SNI to %s", mServer.name.c_str());
            return nullptr;
        }
        X509_VERIFY_PARAM* param = SSL_get0_param(ssl.get());
        if (X509_VERIFY_PARAM_set1_host(param, mServer.name.data(), mServer.name.size()) != 1) {
            ALOGE("Failed to set verify host param to %s", mServer.name.c_str());
            return nullptr;
        }
        // This will cause the handshake to fail if certificate verification fails.
        SSL_set_verify(ssl.get(), SSL_VERIFY_PEER, nullptr);
    }

    bssl::UniquePtr<SSL_SESSION> session = mCache->getSession();
    if (session) {
        ALOGV("Setting session");
        SSL_set_session(ssl.get(), session.get());
    } else {
        ALOGV("No session available");
    }

    for (;;) {
        ALOGV("%u Calling SSL_connect", mMark);
        int ret = SSL_connect(ssl.get());
        ALOGV("%u SSL_connect returned %d", mMark, ret);
        if (ret == 1) break;  // SSL handshake complete;

        const int ssl_err = SSL_get_error(ssl.get(), ret);
        switch (ssl_err) {
            case SSL_ERROR_WANT_READ:
                if (waitForReading(fd) != 1) {
                    ALOGW("SSL_connect read error: %d", errno);
                    return nullptr;
                }
                break;
            case SSL_ERROR_WANT_WRITE:
                if (waitForWriting(fd) != 1) {
                    ALOGW("SSL_connect write error");
                    return nullptr;
                }
                break;
            default:
                ALOGW("SSL_connect error %d, errno=%d", ssl_err, errno);
                return nullptr;
        }
    }

    // TODO: Call SSL_shutdown before discarding the session if validation fails.
    if (!mServer.fingerprints.empty()) {
        ALOGV("Checking DNS over TLS fingerprint");

        // We only care that the chain is internally self-consistent, not that
        // it chains to a trusted root, so we can ignore some kinds of errors.
        // TODO: Add a CA root verification mode that respects these errors.
        int verify_result = SSL_get_verify_result(ssl.get());
        switch (verify_result) {
            case X509_V_OK:
            case X509_V_ERR_DEPTH_ZERO_SELF_SIGNED_CERT:
            case X509_V_ERR_SELF_SIGNED_CERT_IN_CHAIN:
            case X509_V_ERR_CERT_UNTRUSTED:
                break;
            default:
                ALOGW("Invalid certificate chain, error %d", verify_result);
                return nullptr;
        }

        STACK_OF(X509) *chain = SSL_get_peer_cert_chain(ssl.get());
        if (!chain) {
            ALOGW("Server has null certificate");
            return nullptr;
        }
        // Chain and its contents are owned by ssl, so we don't need to free explicitly.
        bool matched = false;
        for (size_t i = 0; i < sk_X509_num(chain); ++i) {
            // This appears to be O(N^2), but there doesn't seem to be a straightforward
            // way to walk a STACK_OF nondestructively in linear time.
            X509* cert = sk_X509_value(chain, i);
            std::vector<uint8_t> digest;
            if (!getSPKIDigest(cert, &digest)) {
                ALOGE("Digest computation failed");
                return nullptr;
            }

            if (mServer.fingerprints.count(digest) > 0) {
                matched = true;
                break;
            }
        }

        if (!matched) {
            ALOGW("No matching fingerprint");
            return nullptr;
        }

        ALOGV("DNS over TLS fingerprint is correct");
    }

    ALOGV("%u handshake complete", mMark);

    return ssl;
}

void DnsTlsSocket::sslDisconnect() {
    if (mSsl) {
        SSL_shutdown(mSsl.get());
        mSsl.reset();
    }
    mSslFd.reset();
}

bool DnsTlsSocket::sslWrite(const Slice buffer) {
    ALOGV("%u Writing %zu bytes", mMark, buffer.size());
    for (;;) {
        int ret = SSL_write(mSsl.get(), buffer.base(), buffer.size());
        if (ret == int(buffer.size())) break;  // SSL write complete;

        if (ret < 1) {
            const int ssl_err = SSL_get_error(mSsl.get(), ret);
            switch (ssl_err) {
                case SSL_ERROR_WANT_WRITE:
                    if (waitForWriting(mSslFd.get()) != 1) {
                        ALOGV("SSL_write error");
                        return false;
                    }
                    continue;
                case 0:
                    break;  // SSL write complete;
                default:
                    ALOGV("SSL_write error %d", ssl_err);
                    return false;
            }
        }
    }
    ALOGV("%u Wrote %zu bytes", mMark, buffer.size());
    return true;
}

void DnsTlsSocket::loop() {
    std::lock_guard guard(mLock);
    std::deque<std::vector<uint8_t>> q;

    const int timeout_msecs = DnsTlsSocket::kIdleTimeout.count() * 1000;
    while (true) {
        // poll() ignores negative fds
        struct pollfd fds[2] = { { .fd = -1 }, { .fd = -1 } };
        enum { SSLFD = 0, EVENTFD = 1 };

        // Always listen for a response from server.
        fds[SSLFD].fd = mSslFd.get();
        fds[SSLFD].events = POLLIN;

        // If we have pending queries, wait for space to write one.
        // Otherwise, listen for new queries.
        if (!q.empty()) {
            fds[SSLFD].events |= POLLOUT;
        } else {
            fds[EVENTFD].fd = mEventFd.get();
            fds[EVENTFD].events = POLLIN;
        }

        const int s = TEMP_FAILURE_RETRY(poll(fds, std::size(fds), timeout_msecs));
        if (s == 0) {
            ALOGV("Idle timeout");
            break;
        }
        if (s < 0) {
            ALOGV("Poll failed: %d", errno);
            break;
        }
        if (fds[SSLFD].revents & (POLLIN | POLLERR)) {
            if (!readResponse()) {
                ALOGV("SSL remote close or read error.");
                break;
            }
        }
        if (fds[EVENTFD].revents & (POLLIN | POLLERR)) {
            int64_t num_queries;
            ssize_t res = read(mEventFd.get(), &num_queries, sizeof(num_queries));
            if (res < 0) {
                ALOGW("Error during eventfd read");
                break;
            } else if (res == 0) {
                ALOGV("eventfd closed; disconnecting");
                break;
            } else if (res != sizeof(num_queries)) {
                ALOGE("Int size mismatch: %zd != %zu", res, sizeof(num_queries));
                break;
            } else if (num_queries <= 0) {
                ALOGE("eventfd reads should always be positive");
                break;
            }
            // Take ownership of all pending queries.  (q is always empty here.)
            mQueue.swap(q);
            // The writing thread writes to mQueue and then increments mEventFd, so
            // there should be at least num_queries entries in mQueue.
            if (q.size() < (uint64_t) num_queries) {
                ALOGE("Synchronization error");
                break;
            }
        } else if (fds[SSLFD].revents & POLLOUT) {
            // q cannot be empty here.
            // Sending the entire queue here would risk a TCP flow control deadlock, so
            // we only send a single query on each cycle of this loop.
            // TODO: Coalesce multiple pending queries if there is enough space in the
            // write buffer.
            if (!sendQuery(q.front())) {
                break;
            }
            q.pop_front();
        }
    }
    ALOGV("Closing event FD");
    mEventFd.reset();
    ALOGV("Disconnecting");
    sslDisconnect();
    ALOGV("Calling onClosed");
    mObserver->onClosed();
    ALOGV("Ending loop");
}

DnsTlsSocket::~DnsTlsSocket() {
    ALOGV("Destructor");
    // This will trigger an orderly shutdown in loop().
    // In principle there is a data race here: If there is an I/O error in the network thread
    // simultaneous with a call to the destructor in a different thread, both threads could
    // attempt to call mEventFd.reset() at the same time.  However, the implementation of
    // UniqueFd::reset appears to be thread-safe, and neither thread reads or writes mEventFd
    // after this point, so we don't expect an issue in practice.
    mEventFd.reset();
    {
        // Wait for the orderly shutdown to complete.
        std::lock_guard guard(mLock);
        if (mLoopThread && std::this_thread::get_id() == mLoopThread->get_id()) {
            ALOGE("Violation of re-entrance precondition");
            return;
        }
    }
    if (mLoopThread) {
        ALOGV("Waiting for loop thread to terminate");
        mLoopThread->join();
        mLoopThread.reset();
    }
    ALOGV("Destructor completed");
}

bool DnsTlsSocket::query(uint16_t id, const Slice query) {
    if (!mEventFd) {
        return false;
    }

    // Compose the entire message in a single buffer, so that it can be
    // sent as a single TLS record.
    std::vector<uint8_t> buf(query.size() + 4);
    // Write 2-byte length
    uint16_t len = query.size() + 2;  // + 2 for the ID.
    buf[0] = len >> 8;
    buf[1] = len;
    // Write 2-byte ID
    buf[2] = id >> 8;
    buf[3] = id;
    // Copy body
    std::memcpy(buf.data() + 4, query.base(), query.size());

    mQueue.push(std::move(buf));
    // Increment the mEventFd counter by 1.
    constexpr int64_t num_queries = 1;
    int written = write(mEventFd.get(), &num_queries, sizeof(num_queries));
    return written == sizeof(num_queries);
}

// Read exactly len bytes into buffer or fail with an SSL error code
int DnsTlsSocket::sslRead(const Slice buffer, bool wait) {
    size_t remaining = buffer.size();
    while (remaining > 0) {
        int ret = SSL_read(mSsl.get(), buffer.limit() - remaining, remaining);
        if (ret == 0) {
            ALOGW_IF(remaining < buffer.size(), "SSL closed with %zu of %zu bytes remaining",
                     remaining, buffer.size());
            return SSL_ERROR_ZERO_RETURN;
        }

        if (ret < 0) {
            const int ssl_err = SSL_get_error(mSsl.get(), ret);
            if (wait && ssl_err == SSL_ERROR_WANT_READ) {
                if (waitForReading(mSslFd.get()) != 1) {
                    ALOGV("Poll failed in sslRead: %d", errno);
                    return SSL_ERROR_SYSCALL;
                }
                continue;
            } else {
                ALOGV("SSL_read error %d", ssl_err);
                return ssl_err;
            }
        }

        remaining -= ret;
        wait = true;  // Once a read is started, try to finish.
    }
    return SSL_ERROR_NONE;
}

bool DnsTlsSocket::sendQuery(const std::vector<uint8_t>& buf) {
    if (!sslWrite(netdutils::makeSlice(buf))) {
        return false;
    }
    ALOGV("%u SSL_write complete", mMark);
    return true;
}

bool DnsTlsSocket::readResponse() {
    ALOGV("reading response");
    uint8_t responseHeader[2];
    int err = sslRead(Slice(responseHeader, 2), false);
    if (err == SSL_ERROR_WANT_READ) {
        ALOGV("Ignoring spurious wakeup from server");
        return true;
    }
    if (err != SSL_ERROR_NONE) {
        return false;
    }
    // Truncate responses larger than MAX_SIZE.  This is safe because a DNS packet is
    // always invalid when truncated, so the response will be treated as an error.
    constexpr uint16_t MAX_SIZE = 8192;
    const uint16_t responseSize = (responseHeader[0] << 8) | responseHeader[1];
    ALOGV("%u Expecting response of size %i", mMark, responseSize);
    std::vector<uint8_t> response(std::min(responseSize, MAX_SIZE));
    if (sslRead(netdutils::makeSlice(response), true) != SSL_ERROR_NONE) {
        ALOGV("%u Failed to read %zu bytes", mMark, response.size());
        return false;
    }
    uint16_t remainingBytes = responseSize - response.size();
    while (remainingBytes > 0) {
        constexpr uint16_t CHUNK_SIZE = 2048;
        std::vector<uint8_t> discard(std::min(remainingBytes, CHUNK_SIZE));
        if (sslRead(netdutils::makeSlice(discard), true) != SSL_ERROR_NONE) {
            ALOGV("%u Failed to discard %zu bytes", mMark, discard.size());
            return false;
        }
        remainingBytes -= discard.size();
    }
    ALOGV("%u SSL_read complete", mMark);

    mObserver->onResponse(std::move(response));
    return true;
}

}  // end of namespace net
}  // end of namespace android