X-Git-Url: https://git.sesse.net/?p=cubemap;a=blobdiff_plain;f=httpinput.cpp;h=62ed663611402cef5d4517cef00d8033300391bc;hp=72a8f3aaf021e3c96afe345c6c7635b506ebb251;hb=0229f4414b3b63934c057030deb88cbf6926bb1f;hpb=9abaf929b9da3158e9b7f799775ac56e88d28740 diff --git a/httpinput.cpp b/httpinput.cpp index 72a8f3a..62ed663 100644 --- a/httpinput.cpp +++ b/httpinput.cpp @@ -1,15 +1,22 @@ #include #include +#include +#include #include #include #include #include -#include #include #include #include #include +#include +#include +#include +#include #include +#include +#include #include #include #include @@ -17,249 +24,473 @@ #include "httpinput.h" #include "log.h" -#include "metacube.h" +#include "metacube2.h" #include "parse.h" #include "serverpool.h" #include "state.pb.h" +#include "stream.h" +#include "timespec.h" +#include "util.h" #include "version.h" using namespace std; +namespace { + +string host_header(const string &host, const string &port) +{ + if (port == "http" || atoi(port.c_str()) == 80) { + return host; + } else { + return host + ":" + port; + } +} + +} // namespace + extern ServerPool *servers; - -HTTPInput::HTTPInput(const string &stream_id, const string &url) + +HTTPInput::HTTPInput(const string &url, Input::Encoding encoding) : state(NOT_CONNECTED), - stream_id(stream_id), url(url), - has_metacube_header(false), - sock(-1) + encoding(encoding) { + stats.url = url; + stats.bytes_received = 0; + stats.data_bytes_received = 0; + stats.metadata_bytes_received = 0; + stats.connect_time = -1; + stats.latency_sec = HUGE_VAL; } HTTPInput::HTTPInput(const InputProto &serialized) : state(State(serialized.state())), - stream_id(serialized.stream_id()), url(serialized.url()), + encoding(serialized.is_metacube_encoded() ? + Input::INPUT_ENCODING_METACUBE : + Input::INPUT_ENCODING_RAW), request(serialized.request()), request_bytes_sent(serialized.request_bytes_sent()), response(serialized.response()), http_header(serialized.http_header()), + stream_header(serialized.stream_header()), has_metacube_header(serialized.has_metacube_header()), - sock(serialized.sock()) + sock(serialized.sock()), + child_pid(serialized.child_pid()) { + // Set back the close-on-exec flag for the socket. + // (This can't leak into a child, since we haven't been started yet.) + fcntl(sock, F_SETFD, O_CLOEXEC); + pending_data.resize(serialized.pending_data().size()); memcpy(&pending_data[0], serialized.pending_data().data(), serialized.pending_data().size()); - string protocol; - parse_url(url, &protocol, &host, &port, &path); // Don't care if it fails. + string protocol, user; + parse_url(url, &protocol, &user, &host, &port, &path); // Don't care if it fails. + + stats.url = url; + stats.bytes_received = serialized.bytes_received(); + stats.data_bytes_received = serialized.data_bytes_received(); + stats.metadata_bytes_received = serialized.metadata_bytes_received(); + if (serialized.has_connect_time()) { + stats.connect_time = serialized.connect_time(); + } else { + stats.connect_time = time(nullptr); + } + if (serialized.has_latency_sec()) { + stats.latency_sec = serialized.latency_sec(); + } else { + stats.latency_sec = HUGE_VAL; + } + + last_verbose_connection.tv_sec = -3600; + last_verbose_connection.tv_nsec = 0; } void HTTPInput::close_socket() { - int ret; - do { - ret = close(sock); - } while (ret == -1 && errno == EINTR); - - if (ret == -1) { - log_perror("close()"); + if (sock != -1) { + safe_close(sock); + sock = -1; + } + if (child_pid != -1) { + // Kill the child process group, forcibly. + // TODO: Consider using a pidfd on newer kernels, so that we're guaranteed + // never to kill the wrong process. + kill(-child_pid, SIGKILL); } + child_pid = -1; + + lock_guard lock(stats_mutex); + stats.connect_time = -1; } InputProto HTTPInput::serialize() const { + // Unset the close-on-exec flag for the socket. + // (This can't leak into a child, since there's only one thread left.) + fcntl(sock, F_SETFD, 0); + InputProto serialized; serialized.set_state(state); - serialized.set_stream_id(stream_id); serialized.set_url(url); serialized.set_request(request); serialized.set_request_bytes_sent(request_bytes_sent); serialized.set_response(response); serialized.set_http_header(http_header); + serialized.set_stream_header(stream_header); serialized.set_pending_data(string(pending_data.begin(), pending_data.end())); serialized.set_has_metacube_header(has_metacube_header); serialized.set_sock(sock); + serialized.set_child_pid(child_pid); + serialized.set_bytes_received(stats.bytes_received); + serialized.set_data_bytes_received(stats.data_bytes_received); + if (isfinite(stats.latency_sec)) { + serialized.set_latency_sec(stats.latency_sec); + } + serialized.set_connect_time(stats.connect_time); + if (encoding == Input::INPUT_ENCODING_METACUBE) { + serialized.set_is_metacube_encoded(true); + } else { + assert(encoding == Input::INPUT_ENCODING_RAW); + serialized.set_is_metacube_encoded(false); + } return serialized; } int HTTPInput::lookup_and_connect(const string &host, const string &port) { addrinfo *ai; - int err = getaddrinfo(host.c_str(), port.c_str(), NULL, &ai); - if (err == -1) { - log(WARNING, "[%s] Lookup of '%s' failed (%s).", - stream_id.c_str(), host.c_str(), gai_strerror(err)); - freeaddrinfo(ai); + int err = getaddrinfo(host.c_str(), port.c_str(), nullptr, &ai); + if (err != 0) { + if (!suppress_logging) { + log(WARNING, "[%s] Lookup of '%s' failed (%s).", + url.c_str(), host.c_str(), gai_strerror(err)); + } return -1; } addrinfo *base_ai = ai; // Connect to everything in turn until we have a socket. - while (ai && !should_stop) { - int sock = socket(ai->ai_family, SOCK_STREAM, IPPROTO_TCP); + for ( ; ai && !should_stop(); ai = ai->ai_next) { + // Now do a non-blocking connect. This is important because we want to be able to be + // woken up, even though it's rather cumbersome. + int sock = socket(ai->ai_family, SOCK_STREAM | SOCK_NONBLOCK | SOCK_CLOEXEC, IPPROTO_TCP); if (sock == -1) { // Could be e.g. EPROTONOSUPPORT. The show must go on. continue; } + // Do a non-blocking connect. do { err = connect(sock, ai->ai_addr, ai->ai_addrlen); } while (err == -1 && errno == EINTR); - if (err != -1) { - freeaddrinfo(base_ai); - return sock; + if (err == -1 && errno != EINPROGRESS) { + log_perror("connect"); + safe_close(sock); + continue; } - do { - err = close(sock); - } while (err == -1 && errno == EINTR); + // Wait for the connect to complete, or an error to happen. + for ( ;; ) { + bool complete = wait_for_activity(sock, POLLIN | POLLOUT, nullptr); + if (should_stop()) { + safe_close(sock); + return -1; + } + if (complete) { + break; + } + } - if (err == -1) { - log_perror("close"); - // Can still continue. + // Check whether it ended in an error or not. + socklen_t err_size = sizeof(err); + if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &err_size) == -1) { + log_perror("getsockopt"); + safe_close(sock); + continue; + } + + errno = err; + + if (err == 0) { + // Successful connect. + freeaddrinfo(base_ai); + return sock; } - ai = ai->ai_next; + safe_close(sock); } // Give the last one as error. - log(WARNING, "[%s] Connect to '%s' failed (%s)", - stream_id.c_str(), host.c_str(), strerror(errno)); + if (!suppress_logging) { + log(WARNING, "[%s] Connect to '%s' failed (%s)", + url.c_str(), host.c_str(), strerror(errno)); + } freeaddrinfo(base_ai); return -1; } + +int HTTPInput::open_child_process(const string &cmdline) +{ + int devnullfd = open("/dev/null", O_RDONLY | O_CLOEXEC); + if (devnullfd == -1) { + log_perror("/dev/null"); + return -1; + } + + int pipefd[2]; + if (pipe2(pipefd, O_CLOEXEC) == -1) { + log_perror("pipe2()"); + close(devnullfd); + return -1; + } + + // Point stdout to us, stdin to /dev/null, and stderr remains where it is + // (probably the systemd log). All other file descriptors should be marked + // as close-on-exec, and should thus not leak into the child. + posix_spawn_file_actions_t actions; + posix_spawn_file_actions_init(&actions); + posix_spawn_file_actions_adddup2(&actions, devnullfd, 0); + posix_spawn_file_actions_adddup2(&actions, pipefd[1], 1); + + // Make the process a leader of its own process group, so that we can easily + // kill it and any of its child processes (unless it's started new process + // groups itself, of course). + posix_spawnattr_t attr; + posix_spawnattr_init(&attr); + posix_spawnattr_setflags(&attr, POSIX_SPAWN_SETPGROUP); + posix_spawnattr_setpgroup(&attr, 0); + + char * const argv[] = { + strdup("/bin/sh"), + strdup("-c"), + strdup(path.c_str()), + nullptr + }; + int err = posix_spawn(&child_pid, "/bin/sh", &actions, &attr, argv, /*envp=*/nullptr); + posix_spawn_file_actions_destroy(&actions); + posix_spawnattr_destroy(&attr); + free(argv[0]); + free(argv[1]); + free(argv[2]); + close(devnullfd); + close(pipefd[1]); + + if (err == 0) { + return pipefd[0]; + } else { + child_pid = -1; + log_perror(cmdline.c_str()); + close(pipefd[0]); + return -1; + } +} -bool HTTPInput::parse_response(const std::string &request) +bool HTTPInput::parse_response(const string &request) { vector lines = split_lines(response); if (lines.empty()) { - log(WARNING, "[%s] Empty HTTP response from input.", stream_id.c_str()); + if (!suppress_logging) { + log(WARNING, "[%s] Empty HTTP response from input.", url.c_str()); + } return false; } vector first_line_tokens = split_tokens(lines[0]); if (first_line_tokens.size() < 2) { - log(WARNING, "[%s] Malformed response line '%s' from input.", - stream_id.c_str(), lines[0].c_str()); + if (!suppress_logging) { + log(WARNING, "[%s] Malformed response line '%s' from input.", + url.c_str(), lines[0].c_str()); + } return false; } int response = atoi(first_line_tokens[1].c_str()); if (response != 200) { - log(WARNING, "[%s] Non-200 response '%s' from input.", - stream_id.c_str(), lines[0].c_str()); + if (!suppress_logging) { + log(WARNING, "[%s] Non-200 response '%s' from input.", + url.c_str(), lines[0].c_str()); + } return false; } - multimap parameters; - for (size_t i = 1; i < lines.size(); ++i) { - size_t split = lines[i].find(":"); - if (split == string::npos) { - log(WARNING, "[%s] Ignoring malformed HTTP response line '%s'", - stream_id.c_str(), lines[i].c_str()); - continue; - } - - string key(lines[i].begin(), lines[i].begin() + split); - - // Skip any spaces after the colon. - do { - ++split; - } while (split < lines[i].size() && lines[i][split] == ' '); + HTTPHeaderMultimap parameters = extract_headers(lines, url); - string value(lines[i].begin() + split, lines[i].end()); - - // Remove “Content-encoding: metacube”. - // TODO: Make case-insensitive. - if (key == "Content-encoding" && value == "metacube") { - continue; - } - - parameters.insert(make_pair(key, value)); + // Remove “Content-encoding: metacube”. + const auto encoding_it = parameters.find("Content-Encoding"); + if (encoding_it != parameters.end() && encoding_it->second == "metacube") { + parameters.erase(encoding_it); } // Change “Server: foo” to “Server: metacube/0.1 (reflecting: foo)” - // TODO: Make case-insensitive. // XXX: Use a Via: instead? if (parameters.count("Server") == 0) { parameters.insert(make_pair("Server", SERVER_IDENTIFICATION)); } else { - for (multimap::iterator it = parameters.begin(); - it != parameters.end(); - ++it) { - if (it->first != "Server") { + for (auto &key_and_value : parameters) { + if (key_and_value.first != "Server") { continue; } - it->second = SERVER_IDENTIFICATION " (reflecting: " + it->second + ")"; + key_and_value.second = SERVER_IDENTIFICATION " (reflecting: " + key_and_value.second + ")"; } } + // Erase “Connection: close”; we'll set it on the sending side if needed. + parameters.erase("Connection"); + // Construct the new HTTP header. http_header = "HTTP/1.0 200 OK\r\n"; - for (multimap::iterator it = parameters.begin(); - it != parameters.end(); - ++it) { - http_header.append(it->first + ": " + it->second + "\r\n"); + for (const auto &key_and_value : parameters) { + http_header.append(key_and_value.first + ": " + key_and_value.second + "\r\n"); + } + + for (int stream_index : stream_indices) { + servers->set_header(stream_index, http_header, stream_header); } - http_header.append("\r\n"); - servers->set_header(stream_id, http_header); return true; } void HTTPInput::do_work() { - while (!should_stop) { + timespec last_activity; + + // TODO: Make the timeout persist across restarts. + if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) { + int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity); + assert(err != -1); + } + + while (!should_stop()) { if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) { - // Since we are non-blocking, we need to wait for the right state first. - // Wait up to 50 ms, then check should_stop. - pollfd pfd; - pfd.fd = sock; - pfd.events = (state == SENDING_REQUEST) ? POLLOUT : POLLIN; - pfd.events |= POLLRDHUP; - - int nfds = poll(&pfd, 1, 50); - if (nfds == 0 || (nfds == -1 && errno == EINTR)) { + // Give the socket 30 seconds since last activity before we time out. + static const int timeout_secs = 30; + + timespec now; + int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now); + assert(err != -1); + + timespec elapsed = clock_diff(last_activity, now); + if (elapsed.tv_sec >= timeout_secs) { + // Timeout! + if (!suppress_logging) { + log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec); + } + state = CLOSING_SOCKET; continue; } - if (nfds == -1) { - log_perror("poll"); - state = CLOSING_SOCKET; + + // Basically calculate (30 - (now - last_activity)) = (30 + (last_activity - now)). + // Add a second of slack to account for differences between clocks. + timespec timeout = clock_diff(now, last_activity); + timeout.tv_sec += timeout_secs + 1; + assert(timeout.tv_sec > 0 || (timeout.tv_sec >= 0 && timeout.tv_nsec > 0)); + + bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout); + if (activity) { + err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity); + assert(err != -1); + } else { + // OK. Most likely, should_stop was set, or we have timed out. + continue; } } switch (state) { - case NOT_CONNECTED: + case NOT_CONNECTED: { + // Reap any exited children. + int wstatus, err; + do { + err = waitpid(-1, &wstatus, WNOHANG); + if (err == -1) { + if (errno == EINTR) { + continue; + } + if (errno == ECHILD) { + break; + } + log_perror("waitpid"); + break; + } + } while (err != 0); + child_pid = -1; + request.clear(); request_bytes_sent = 0; response.clear(); pending_data.clear(); - servers->set_header(stream_id, ""); + has_metacube_header = false; + for (int stream_index : stream_indices) { + // Don't zero out the header; it might still be of use to HLS clients. + servers->set_unavailable(stream_index); + } + string protocol; { - string protocol; // Thrown away. - if (!parse_url(url, &protocol, &host, &port, &path)) { - log(WARNING, "[%s] Failed to parse URL '%s'", stream_id.c_str(), url.c_str()); + string user; // Thrown away. + if (!parse_url(url, &protocol, &user, &host, &port, &path)) { + if (!suppress_logging) { + log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str()); + } break; } + + // Remove the brackets around IPv6 address literals. + // TODO: See if we can join this with the code in parse_ip_address(), + // or maybe even more it into parse_url(). + if (!host.empty() && host[0] == '[' && host[host.size() - 1] == ']') { + host = host.substr(1, host.size() - 2); + } } - sock = lookup_and_connect(host, port); - if (sock != -1) { - // Yay, successful connect. Try to set it as nonblocking. - int one = 1; - if (ioctl(sock, FIONBIO, &one) == -1) { - log_perror("ioctl(FIONBIO)"); - state = CLOSING_SOCKET; - } else { + if (suppress_logging) { + // See if there's more than one minute since last time we made a connection + // with logging enabled. If so, turn it on again. + timespec now; + int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now); + assert(err != -1); + + double elapsed = now.tv_sec - last_verbose_connection.tv_sec + + 1e-9 * (now.tv_nsec - last_verbose_connection.tv_nsec); + if (elapsed > 60.0) { + suppress_logging = false; + } + } + if (!suppress_logging) { + int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_verbose_connection); + assert(err != -1); + } + ++num_connection_attempts; + if (protocol == "pipe") { + sock = open_child_process(path.c_str()); + + if (sock != -1) { + // Construct a minimal HTTP header. + http_header = "HTTP/1.0 200 OK\r\n"; + for (int stream_index : stream_indices) { + servers->set_header(stream_index, http_header, stream_header); + } + state = RECEIVING_DATA; + } + } else { + sock = lookup_and_connect(host, port); + if (sock != -1) { + // Yay, successful connect. state = SENDING_REQUEST; - request = "GET " + path + " HTTP/1.0\r\nUser-Agent: cubemap\r\n\r\n"; + request = "GET " + path + " HTTP/1.0\r\nHost: " + host_header(host, port) + "\r\nUser-Agent: cubemap\r\n\r\n"; request_bytes_sent = 0; } } + if (sock != -1) { + lock_guard lock(stats_mutex); + stats.connect_time = time(nullptr); + clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity); + } break; + } case SENDING_REQUEST: { size_t to_send = request.size() - request_bytes_sent; int ret; @@ -298,8 +529,10 @@ void HTTPInput::do_work() if (ret == 0) { // This really shouldn't happen... - log(ERROR, "[%s] Socket unexpectedly closed while reading header", - stream_id.c_str()); + if (!suppress_logging) { + log(ERROR, "[%s] Socket unexpectedly closed while reading header", + url.c_str()); + } state = CLOSING_SOCKET; continue; } @@ -307,7 +540,9 @@ void HTTPInput::do_work() RequestParseStatus status = wait_for_double_newline(&response, buf, ret); if (status == RP_OUT_OF_SPACE) { - log(WARNING, "[%s] Sever sent overlong HTTP response!", stream_id.c_str()); + if (!suppress_logging) { + log(WARNING, "[%s] Server sent overlong HTTP response!", url.c_str()); + } state = CLOSING_SOCKET; continue; } else if (status == RP_NOT_FINISHED_YET) { @@ -320,8 +555,8 @@ void HTTPInput::do_work() if (status == RP_EXTRA_DATA) { char *ptr = static_cast( memmem(response.data(), response.size(), "\r\n\r\n", 4)); - assert(ptr != NULL); - extra_data = string(ptr, &response[0] + response.size()); + assert(ptr != nullptr); + extra_data = string(ptr + 4, &response[0] + response.size()); response.resize(ptr - response.data()); } @@ -334,8 +569,16 @@ void HTTPInput::do_work() process_data(&extra_data[0], extra_data.size()); } - log(INFO, "[%s] Connected to '%s', receiving data.", - stream_id.c_str(), url.c_str()); + if (!suppress_logging) { + if (encoding == Input::INPUT_ENCODING_RAW) { + log(INFO, "[%s] Connected to '%s', receiving raw data.", + url.c_str(), url.c_str()); + } else { + assert(encoding == Input::INPUT_ENCODING_METACUBE); + log(INFO, "[%s] Connected to '%s', receiving data.", + url.c_str(), url.c_str()); + } + } state = RECEIVING_DATA; break; } @@ -355,25 +598,33 @@ void HTTPInput::do_work() if (ret == 0) { // This really shouldn't happen... - log(ERROR, "[%s] Socket unexpectedly closed while reading header", - stream_id.c_str()); + if (!suppress_logging) { + log(ERROR, "[%s] Socket unexpectedly closed while reading data", + url.c_str()); + } state = CLOSING_SOCKET; continue; } + num_connection_attempts = 0; // Reset, since we have a successful read. + if (suppress_logging) { + // This was suppressed earlier, so print it out now. + if (encoding == Input::INPUT_ENCODING_RAW) { + log(INFO, "[%s] Connected to '%s', receiving raw data.", + url.c_str(), url.c_str()); + } else { + assert(encoding == Input::INPUT_ENCODING_METACUBE); + log(INFO, "[%s] Connected to '%s', receiving data.", + url.c_str(), url.c_str()); + } + suppress_logging = false; + } + process_data(buf, ret); break; } case CLOSING_SOCKET: { - int err; - do { - err = close(sock); - } while (err == -1 && errno == EINTR); - - if (err == -1) { - log_perror("close"); - } - + close_socket(); state = NOT_CONNECTED; break; } @@ -384,20 +635,44 @@ void HTTPInput::do_work() // If we are still in NOT_CONNECTED, either something went wrong, // or the connection just got closed. // The earlier steps have already given the error message, if any. - if (state == NOT_CONNECTED && !should_stop) { - log(INFO, "[%s] Waiting 0.2 second and restarting...", stream_id.c_str()); - usleep(200000); + if (state == NOT_CONNECTED && !should_stop()) { + if (!suppress_logging) { + log(INFO, "[%s] Waiting 0.2 seconds and restarting...", url.c_str()); + } + + if (num_connection_attempts >= 3 && !suppress_logging) { + log(INFO, "[%s] %d failed connection attempts, suppressing logging for one minute.", + url.c_str(), num_connection_attempts); + suppress_logging = true; + } + timespec timeout_ts; + timeout_ts.tv_sec = 0; + timeout_ts.tv_nsec = 200000000; + wait_for_wakeup(&timeout_ts); } } } void HTTPInput::process_data(char *ptr, size_t bytes) { + { + lock_guard lock(stats_mutex); + stats.bytes_received += bytes; + } + + if (encoding == Input::INPUT_ENCODING_RAW) { + for (int stream_index : stream_indices) { + servers->add_data(stream_index, ptr, bytes, /*metacube_flags=*/0, /*pts=*/RationalPTS()); + } + return; + } + + assert(encoding == Input::INPUT_ENCODING_METACUBE); pending_data.insert(pending_data.end(), ptr, ptr + bytes); for ( ;; ) { // If we don't have enough data (yet) for even the Metacube header, just return. - if (pending_data.size() < sizeof(metacube_block_header)) { + if (pending_data.size() < sizeof(metacube2_block_header)) { return; } @@ -406,13 +681,13 @@ void HTTPInput::process_data(char *ptr, size_t bytes) if (!has_metacube_header) { char *ptr = static_cast( memmem(pending_data.data(), pending_data.size(), - METACUBE_SYNC, strlen(METACUBE_SYNC))); - if (ptr == NULL) { + METACUBE2_SYNC, strlen(METACUBE2_SYNC))); + if (ptr == nullptr) { // OK, so we didn't find the sync marker. We know then that // we do not have the _full_ marker in the buffer, but we // could have N-1 bytes. Drop everything before that, // and then give up. - drop_pending_data(pending_data.size() - (strlen(METACUBE_SYNC) - 1)); + drop_pending_data(pending_data.size() - (strlen(METACUBE2_SYNC) - 1)); return; } else { // Yay, we found the header. Drop everything (if anything) before it. @@ -420,36 +695,73 @@ void HTTPInput::process_data(char *ptr, size_t bytes) has_metacube_header = true; // Re-check that we have the entire header; we could have dropped data. - if (pending_data.size() < sizeof(metacube_block_header)) { + if (pending_data.size() < sizeof(metacube2_block_header)) { return; } } } // Now it's safe to read the header. - metacube_block_header *hdr = reinterpret_cast(pending_data.data()); - assert(memcmp(hdr->sync, METACUBE_SYNC, sizeof(hdr->sync)) == 0); - uint32_t size = ntohl(hdr->size); - uint32_t flags = ntohl(hdr->flags); + metacube2_block_header hdr; + memcpy(&hdr, pending_data.data(), sizeof(hdr)); + assert(memcmp(hdr.sync, METACUBE2_SYNC, sizeof(hdr.sync)) == 0); + uint32_t size = ntohl(hdr.size); + uint16_t flags = ntohs(hdr.flags); + uint16_t expected_csum = metacube2_compute_crc(&hdr); + + if (expected_csum != ntohs(hdr.csum)) { + log(WARNING, "[%s] Metacube checksum failed (expected 0x%x, got 0x%x), " + "not reading block claiming to be %d bytes (flags=%x).", + url.c_str(), expected_csum, ntohs(hdr.csum), + size, flags); + + // Drop only the first byte, and let the rest of the code handle resync. + pending_data.erase(pending_data.begin(), pending_data.begin() + 1); + has_metacube_header = false; + continue; + } + if (size > 10485760) { + log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header??", + url.c_str(), size, flags); + } // See if we have the entire block. If not, wait for more data. - if (pending_data.size() < sizeof(metacube_block_header) + size) { + if (pending_data.size() < sizeof(metacube2_block_header) + size) { return; } - // Send this block on to the data. - char *inner_data = pending_data.data() + sizeof(metacube_block_header); - if (flags & METACUBE_FLAGS_HEADER) { - string header(inner_data, inner_data + size); - servers->set_header(stream_id, http_header + header); - } else { - servers->add_data(stream_id, inner_data, size); + // See if this is a metadata block. If so, we don't want to send it on, + // but rather process it ourselves. + // TODO: Keep metadata when sending on to other Metacube users. + if (flags & METACUBE_FLAGS_METADATA) { + { + lock_guard lock(stats_mutex); + stats.metadata_bytes_received += size; + } + process_metacube_metadata_block(hdr, pending_data.data() + sizeof(hdr), size); + } else { + // Send this block on to the servers. + { + lock_guard lock(stats_mutex); + stats.data_bytes_received += size; + } + char *inner_data = pending_data.data() + sizeof(metacube2_block_header); + if (flags & METACUBE_FLAGS_HEADER) { + stream_header = string(inner_data, inner_data + size); + for (int stream_index : stream_indices) { + servers->set_header(stream_index, http_header, stream_header); + } + } + for (int stream_index : stream_indices) { + servers->add_data(stream_index, inner_data, size, flags, next_block_pts); + } + next_block_pts = RationalPTS(); } // Consume the block. This isn't the most efficient way of dealing with things // should we have many blocks, but these routines don't need to be too efficient // anyway. - pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube_block_header) + size); + pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube2_block_header) + size); has_metacube_header = false; } } @@ -459,8 +771,63 @@ void HTTPInput::drop_pending_data(size_t num_bytes) if (num_bytes == 0) { return; } - log(WARNING, "[%s] Dropping %lld junk bytes from stream, maybe it is not a Metacube stream?", - stream_id.c_str(), (long long)num_bytes); + log(WARNING, "[%s] Dropping %lld junk bytes; not a Metacube2 stream, or data was dropped from the middle of the stream.", + url.c_str(), (long long)num_bytes); + assert(pending_data.size() >= num_bytes); pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes); } +void HTTPInput::add_destination(int stream_index) +{ + stream_indices.push_back(stream_index); + servers->set_header(stream_index, http_header, stream_header); +} + +InputStats HTTPInput::get_stats() const +{ + lock_guard lock(stats_mutex); + return stats; +} + +void HTTPInput::process_metacube_metadata_block(const metacube2_block_header &hdr, const char *payload, uint32_t payload_size) +{ + if (payload_size < sizeof(uint64_t)) { + log(WARNING, "[%s] Undersized Metacube metadata block (%d bytes); corrupted header?", + url.c_str(), payload_size); + return; + } + + uint64_t type = be64toh(*(const uint64_t *)payload); + if (type == METACUBE_METADATA_TYPE_ENCODER_TIMESTAMP) { + timespec now; + clock_gettime(CLOCK_REALTIME, &now); + + const metacube2_timestamp_packet *pkt = (const metacube2_timestamp_packet *)payload; + if (payload_size != sizeof(*pkt)) { + log(WARNING, "[%s] Metacube timestamp block of wrong size (%d bytes); ignoring.", + url.c_str(), payload_size); + return; + } + + double elapsed = now.tv_sec - be64toh(pkt->tv_sec) + + 1e-9 * (now.tv_nsec - long(be64toh(pkt->tv_nsec))); + { + lock_guard lock(stats_mutex); + stats.latency_sec = elapsed; + } + } else if (type == METACUBE_METADATA_TYPE_NEXT_BLOCK_PTS) { + const metacube2_pts_packet *pkt = (const metacube2_pts_packet *)payload; + if (payload_size != sizeof(*pkt)) { + log(WARNING, "[%s] Metacube pts block of wrong size (%d bytes); ignoring.", + url.c_str(), payload_size); + return; + } + next_block_pts.pts = be64toh(pkt->pts); + next_block_pts.timebase_num = be64toh(pkt->timebase_num); + next_block_pts.timebase_den = be64toh(pkt->timebase_den); + } else { + // Unknown metadata block, ignore + log(INFO, "[%s] Metadata block %llu\n", url.c_str(), type); + return; + } +}