6 #include <netinet/in.h>
11 #include <sys/ioctl.h>
12 #include <sys/socket.h>
14 #include <sys/types.h>
25 #include "httpinput.h"
27 #include "metacube2.h"
29 #include "serverpool.h"
40 string host_header(const string &host, const string &port)
42 if (port == "http" || atoi(port.c_str()) == 80) {
45 return host + ":" + port;
51 extern ServerPool *servers;
53 HTTPInput::HTTPInput(const string &url, Input::Encoding encoding)
54 : state(NOT_CONNECTED),
59 stats.bytes_received = 0;
60 stats.data_bytes_received = 0;
61 stats.metadata_bytes_received = 0;
62 stats.connect_time = -1;
63 stats.latency_sec = HUGE_VAL;
66 HTTPInput::HTTPInput(const InputProto &serialized)
67 : state(State(serialized.state())),
68 url(serialized.url()),
69 encoding(serialized.is_metacube_encoded() ?
70 Input::INPUT_ENCODING_METACUBE :
71 Input::INPUT_ENCODING_RAW),
72 request(serialized.request()),
73 request_bytes_sent(serialized.request_bytes_sent()),
74 response(serialized.response()),
75 http_header(serialized.http_header()),
76 stream_header(serialized.stream_header()),
77 has_metacube_header(serialized.has_metacube_header()),
78 sock(serialized.sock()),
79 child_pid(serialized.child_pid())
81 // Set back the close-on-exec flag for the socket.
82 // (This can't leak into a child, since we haven't been started yet.)
83 fcntl(sock, F_SETFD, 1);
85 pending_data.resize(serialized.pending_data().size());
86 memcpy(&pending_data[0], serialized.pending_data().data(), serialized.pending_data().size());
88 string protocol, user;
89 parse_url(url, &protocol, &user, &host, &port, &path); // Don't care if it fails.
92 stats.bytes_received = serialized.bytes_received();
93 stats.data_bytes_received = serialized.data_bytes_received();
94 stats.metadata_bytes_received = serialized.metadata_bytes_received();
95 if (serialized.has_connect_time()) {
96 stats.connect_time = serialized.connect_time();
98 stats.connect_time = time(nullptr);
100 if (serialized.has_latency_sec()) {
101 stats.latency_sec = serialized.latency_sec();
103 stats.latency_sec = HUGE_VAL;
106 last_verbose_connection.tv_sec = -3600;
107 last_verbose_connection.tv_nsec = 0;
110 void HTTPInput::close_socket()
116 if (child_pid != -1) {
117 // Kill the child process group, forcibly.
118 // TODO: Consider using a pidfd on newer kernels, so that we're guaranteed
119 // never to kill the wrong process.
120 kill(-child_pid, SIGKILL);
124 lock_guard<mutex> lock(stats_mutex);
125 stats.connect_time = -1;
128 InputProto HTTPInput::serialize() const
130 // Unset the close-on-exec flag for the socket.
131 // (This can't leak into a child, since there's only one thread left.)
132 fcntl(sock, F_SETFD, 0);
134 InputProto serialized;
135 serialized.set_state(state);
136 serialized.set_url(url);
137 serialized.set_request(request);
138 serialized.set_request_bytes_sent(request_bytes_sent);
139 serialized.set_response(response);
140 serialized.set_http_header(http_header);
141 serialized.set_stream_header(stream_header);
142 serialized.set_pending_data(string(pending_data.begin(), pending_data.end()));
143 serialized.set_has_metacube_header(has_metacube_header);
144 serialized.set_sock(sock);
145 serialized.set_child_pid(child_pid);
146 serialized.set_bytes_received(stats.bytes_received);
147 serialized.set_data_bytes_received(stats.data_bytes_received);
148 if (isfinite(stats.latency_sec)) {
149 serialized.set_latency_sec(stats.latency_sec);
151 serialized.set_connect_time(stats.connect_time);
152 if (encoding == Input::INPUT_ENCODING_METACUBE) {
153 serialized.set_is_metacube_encoded(true);
155 assert(encoding == Input::INPUT_ENCODING_RAW);
156 serialized.set_is_metacube_encoded(false);
161 int HTTPInput::lookup_and_connect(const string &host, const string &port)
164 int err = getaddrinfo(host.c_str(), port.c_str(), nullptr, &ai);
166 if (!suppress_logging) {
167 log(WARNING, "[%s] Lookup of '%s' failed (%s).",
168 url.c_str(), host.c_str(), gai_strerror(err));
173 addrinfo *base_ai = ai;
175 // Connect to everything in turn until we have a socket.
176 for ( ; ai && !should_stop(); ai = ai->ai_next) {
177 // Now do a non-blocking connect. This is important because we want to be able to be
178 // woken up, even though it's rather cumbersome.
179 int sock = socket(ai->ai_family, SOCK_STREAM | SOCK_NONBLOCK | SOCK_CLOEXEC, IPPROTO_TCP);
181 // Could be e.g. EPROTONOSUPPORT. The show must go on.
185 // Do a non-blocking connect.
187 err = connect(sock, ai->ai_addr, ai->ai_addrlen);
188 } while (err == -1 && errno == EINTR);
190 if (err == -1 && errno != EINPROGRESS) {
191 log_perror("connect");
196 // Wait for the connect to complete, or an error to happen.
198 bool complete = wait_for_activity(sock, POLLIN | POLLOUT, nullptr);
208 // Check whether it ended in an error or not.
209 socklen_t err_size = sizeof(err);
210 if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &err_size) == -1) {
211 log_perror("getsockopt");
219 // Successful connect.
220 freeaddrinfo(base_ai);
227 // Give the last one as error.
228 if (!suppress_logging) {
229 log(WARNING, "[%s] Connect to '%s' failed (%s)",
230 url.c_str(), host.c_str(), strerror(errno));
232 freeaddrinfo(base_ai);
236 int HTTPInput::open_child_process(const string &cmdline)
238 int devnullfd = open("/dev/null", O_RDONLY | O_CLOEXEC);
239 if (devnullfd == -1) {
240 log_perror("/dev/null");
245 if (pipe2(pipefd, O_CLOEXEC) == -1) {
246 log_perror("pipe2()");
251 // Point stdout to us, stdin to /dev/null, and stderr remains where it is
252 // (probably the systemd log). All other file descriptors should be marked
253 // as close-on-exec, and should thus not leak into the child.
254 posix_spawn_file_actions_t actions;
255 posix_spawn_file_actions_init(&actions);
256 posix_spawn_file_actions_adddup2(&actions, devnullfd, 0);
257 posix_spawn_file_actions_adddup2(&actions, pipefd[1], 1);
259 // Make the process a leader of its own process group, so that we can easily
260 // kill it and any of its child processes (unless it's started new process
261 // groups itself, of course).
262 posix_spawnattr_t attr;
263 posix_spawnattr_init(&attr);
264 posix_spawnattr_setflags(&attr, POSIX_SPAWN_SETPGROUP);
265 posix_spawnattr_setpgroup(&attr, 0);
267 char * const argv[] = {
270 strdup(path.c_str()),
273 int err = posix_spawn(&child_pid, "/bin/sh", &actions, &attr, argv, /*envp=*/nullptr);
274 posix_spawn_file_actions_destroy(&actions);
275 posix_spawnattr_destroy(&attr);
285 log_perror(cmdline.c_str());
291 bool HTTPInput::parse_response(const string &request)
293 vector<string> lines = split_lines(response);
295 if (!suppress_logging) {
296 log(WARNING, "[%s] Empty HTTP response from input.", url.c_str());
301 vector<string> first_line_tokens = split_tokens(lines[0]);
302 if (first_line_tokens.size() < 2) {
303 if (!suppress_logging) {
304 log(WARNING, "[%s] Malformed response line '%s' from input.",
305 url.c_str(), lines[0].c_str());
310 int response = atoi(first_line_tokens[1].c_str());
311 if (response != 200) {
312 if (!suppress_logging) {
313 log(WARNING, "[%s] Non-200 response '%s' from input.",
314 url.c_str(), lines[0].c_str());
319 HTTPHeaderMultimap parameters = extract_headers(lines, url);
321 // Remove “Content-encoding: metacube”.
322 const auto encoding_it = parameters.find("Content-Encoding");
323 if (encoding_it != parameters.end() && encoding_it->second == "metacube") {
324 parameters.erase(encoding_it);
327 // Change “Server: foo” to “Server: metacube/0.1 (reflecting: foo)”
328 // XXX: Use a Via: instead?
329 if (parameters.count("Server") == 0) {
330 parameters.insert(make_pair("Server", SERVER_IDENTIFICATION));
332 for (auto &key_and_value : parameters) {
333 if (key_and_value.first != "Server") {
336 key_and_value.second = SERVER_IDENTIFICATION " (reflecting: " + key_and_value.second + ")";
340 // Erase “Connection: close”; we'll set it on the sending side if needed.
341 parameters.erase("Connection");
343 // Construct the new HTTP header.
344 http_header = "HTTP/1.0 200 OK\r\n";
345 for (const auto &key_and_value : parameters) {
346 http_header.append(key_and_value.first + ": " + key_and_value.second + "\r\n");
349 for (int stream_index : stream_indices) {
350 servers->set_header(stream_index, http_header, stream_header);
356 void HTTPInput::do_work()
358 timespec last_activity;
360 // TODO: Make the timeout persist across restarts.
361 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
362 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
366 while (!should_stop()) {
367 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
368 // Give the socket 30 seconds since last activity before we time out.
369 static const int timeout_secs = 30;
372 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
375 timespec elapsed = clock_diff(last_activity, now);
376 if (elapsed.tv_sec >= timeout_secs) {
378 if (!suppress_logging) {
379 log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec);
381 state = CLOSING_SOCKET;
385 // Basically calculate (30 - (now - last_activity)) = (30 + (last_activity - now)).
386 // Add a second of slack to account for differences between clocks.
387 timespec timeout = clock_diff(now, last_activity);
388 timeout.tv_sec += timeout_secs + 1;
389 assert(timeout.tv_sec > 0 || (timeout.tv_sec >= 0 && timeout.tv_nsec > 0));
391 bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout);
393 err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
396 // OK. Most likely, should_stop was set, or we have timed out.
402 case NOT_CONNECTED: {
403 // Reap any exited children.
406 err = waitpid(-1, &wstatus, WNOHANG);
408 if (errno == EINTR) {
411 if (errno == ECHILD) {
414 log_perror("waitpid");
421 request_bytes_sent = 0;
423 pending_data.clear();
424 has_metacube_header = false;
425 for (int stream_index : stream_indices) {
426 // Don't zero out the header; it might still be of use to HLS clients.
427 servers->set_unavailable(stream_index);
432 string user; // Thrown away.
433 if (!parse_url(url, &protocol, &user, &host, &port, &path)) {
434 if (!suppress_logging) {
435 log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str());
440 // Remove the brackets around IPv6 address literals.
441 // TODO: See if we can join this with the code in parse_ip_address(),
442 // or maybe even more it into parse_url().
443 if (!host.empty() && host[0] == '[' && host[host.size() - 1] == ']') {
444 host = host.substr(1, host.size() - 2);
448 if (suppress_logging) {
449 // See if there's more than one minute since last time we made a connection
450 // with logging enabled. If so, turn it on again.
452 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
455 double elapsed = now.tv_sec - last_verbose_connection.tv_sec +
456 1e-9 * (now.tv_nsec - last_verbose_connection.tv_nsec);
457 if (elapsed > 60.0) {
458 suppress_logging = false;
461 if (!suppress_logging) {
462 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_verbose_connection);
465 ++num_connection_attempts;
466 if (protocol == "pipe") {
467 sock = open_child_process(path.c_str());
470 // Construct a minimal HTTP header.
471 http_header = "HTTP/1.0 200 OK\r\n";
472 for (int stream_index : stream_indices) {
473 servers->set_header(stream_index, http_header, stream_header);
475 state = RECEIVING_DATA;
478 sock = lookup_and_connect(host, port);
480 // Yay, successful connect.
481 state = SENDING_REQUEST;
482 request = "GET " + path + " HTTP/1.0\r\nHost: " + host_header(host, port) + "\r\nUser-Agent: cubemap\r\n\r\n";
483 request_bytes_sent = 0;
487 lock_guard<mutex> lock(stats_mutex);
488 stats.connect_time = time(nullptr);
489 clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
493 case SENDING_REQUEST: {
494 size_t to_send = request.size() - request_bytes_sent;
498 ret = write(sock, request.data() + request_bytes_sent, to_send);
499 } while (ret == -1 && errno == EINTR);
503 state = CLOSING_SOCKET;
508 request_bytes_sent += ret;
510 if (request_bytes_sent == request.size()) {
511 state = RECEIVING_HEADER;
515 case RECEIVING_HEADER: {
520 ret = read(sock, buf, sizeof(buf));
521 } while (ret == -1 && errno == EINTR);
525 state = CLOSING_SOCKET;
530 // This really shouldn't happen...
531 if (!suppress_logging) {
532 log(ERROR, "[%s] Socket unexpectedly closed while reading header",
535 state = CLOSING_SOCKET;
539 RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
541 if (status == RP_OUT_OF_SPACE) {
542 if (!suppress_logging) {
543 log(WARNING, "[%s] Server sent overlong HTTP response!", url.c_str());
545 state = CLOSING_SOCKET;
547 } else if (status == RP_NOT_FINISHED_YET) {
551 // OK, so we're fine, but there might be some of the actual data after the response.
552 // We'll need to deal with that separately.
554 if (status == RP_EXTRA_DATA) {
555 char *ptr = static_cast<char *>(
556 memmem(response.data(), response.size(), "\r\n\r\n", 4));
557 assert(ptr != nullptr);
558 extra_data = string(ptr + 4, &response[0] + response.size());
559 response.resize(ptr - response.data());
562 if (!parse_response(response)) {
563 state = CLOSING_SOCKET;
567 if (!extra_data.empty()) {
568 process_data(&extra_data[0], extra_data.size());
571 if (!suppress_logging) {
572 if (encoding == Input::INPUT_ENCODING_RAW) {
573 log(INFO, "[%s] Connected to '%s', receiving raw data.",
574 url.c_str(), url.c_str());
576 assert(encoding == Input::INPUT_ENCODING_METACUBE);
577 log(INFO, "[%s] Connected to '%s', receiving data.",
578 url.c_str(), url.c_str());
581 state = RECEIVING_DATA;
584 case RECEIVING_DATA: {
589 ret = read(sock, buf, sizeof(buf));
590 } while (ret == -1 && errno == EINTR);
594 state = CLOSING_SOCKET;
599 // This really shouldn't happen...
600 if (!suppress_logging) {
601 log(ERROR, "[%s] Socket unexpectedly closed while reading data",
604 state = CLOSING_SOCKET;
608 num_connection_attempts = 0; // Reset, since we have a successful read.
609 if (suppress_logging) {
610 // This was suppressed earlier, so print it out now.
611 if (encoding == Input::INPUT_ENCODING_RAW) {
612 log(INFO, "[%s] Connected to '%s', receiving raw data.",
613 url.c_str(), url.c_str());
615 assert(encoding == Input::INPUT_ENCODING_METACUBE);
616 log(INFO, "[%s] Connected to '%s', receiving data.",
617 url.c_str(), url.c_str());
619 suppress_logging = false;
622 process_data(buf, ret);
625 case CLOSING_SOCKET: {
627 state = NOT_CONNECTED;
634 // If we are still in NOT_CONNECTED, either something went wrong,
635 // or the connection just got closed.
636 // The earlier steps have already given the error message, if any.
637 if (state == NOT_CONNECTED && !should_stop()) {
638 if (!suppress_logging) {
639 log(INFO, "[%s] Waiting 0.2 seconds and restarting...", url.c_str());
642 if (num_connection_attempts >= 3 && !suppress_logging) {
643 log(INFO, "[%s] %d failed connection attempts, suppressing logging for one minute.",
644 url.c_str(), num_connection_attempts);
645 suppress_logging = true;
648 timeout_ts.tv_sec = 0;
649 timeout_ts.tv_nsec = 200000000;
650 wait_for_wakeup(&timeout_ts);
655 void HTTPInput::process_data(char *ptr, size_t bytes)
658 lock_guard<mutex> lock(stats_mutex);
659 stats.bytes_received += bytes;
662 if (encoding == Input::INPUT_ENCODING_RAW) {
663 for (int stream_index : stream_indices) {
664 servers->add_data(stream_index, ptr, bytes, /*metacube_flags=*/0, /*pts=*/RationalPTS());
669 assert(encoding == Input::INPUT_ENCODING_METACUBE);
670 pending_data.insert(pending_data.end(), ptr, ptr + bytes);
673 // If we don't have enough data (yet) for even the Metacube header, just return.
674 if (pending_data.size() < sizeof(metacube2_block_header)) {
678 // Make sure we have the Metacube sync header at the start.
679 // We may need to skip over junk data (it _should_ not happen, though).
680 if (!has_metacube_header) {
681 char *ptr = static_cast<char *>(
682 memmem(pending_data.data(), pending_data.size(),
683 METACUBE2_SYNC, strlen(METACUBE2_SYNC)));
684 if (ptr == nullptr) {
685 // OK, so we didn't find the sync marker. We know then that
686 // we do not have the _full_ marker in the buffer, but we
687 // could have N-1 bytes. Drop everything before that,
689 drop_pending_data(pending_data.size() - (strlen(METACUBE2_SYNC) - 1));
692 // Yay, we found the header. Drop everything (if anything) before it.
693 drop_pending_data(ptr - pending_data.data());
694 has_metacube_header = true;
696 // Re-check that we have the entire header; we could have dropped data.
697 if (pending_data.size() < sizeof(metacube2_block_header)) {
703 // Now it's safe to read the header.
704 metacube2_block_header hdr;
705 memcpy(&hdr, pending_data.data(), sizeof(hdr));
706 assert(memcmp(hdr.sync, METACUBE2_SYNC, sizeof(hdr.sync)) == 0);
707 uint32_t size = ntohl(hdr.size);
708 uint16_t flags = ntohs(hdr.flags);
709 uint16_t expected_csum = metacube2_compute_crc(&hdr);
711 if (expected_csum != ntohs(hdr.csum)) {
712 log(WARNING, "[%s] Metacube checksum failed (expected 0x%x, got 0x%x), "
713 "not reading block claiming to be %d bytes (flags=%x).",
714 url.c_str(), expected_csum, ntohs(hdr.csum),
717 // Drop only the first byte, and let the rest of the code handle resync.
718 pending_data.erase(pending_data.begin(), pending_data.begin() + 1);
719 has_metacube_header = false;
722 if (size > 10485760) {
723 log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header??",
724 url.c_str(), size, flags);
727 // See if we have the entire block. If not, wait for more data.
728 if (pending_data.size() < sizeof(metacube2_block_header) + size) {
732 // See if this is a metadata block. If so, we don't want to send it on,
733 // but rather process it ourselves.
734 // TODO: Keep metadata when sending on to other Metacube users.
735 if (flags & METACUBE_FLAGS_METADATA) {
737 lock_guard<mutex> lock(stats_mutex);
738 stats.metadata_bytes_received += size;
740 process_metacube_metadata_block(hdr, pending_data.data() + sizeof(hdr), size);
742 // Send this block on to the servers.
744 lock_guard<mutex> lock(stats_mutex);
745 stats.data_bytes_received += size;
747 char *inner_data = pending_data.data() + sizeof(metacube2_block_header);
748 if (flags & METACUBE_FLAGS_HEADER) {
749 stream_header = string(inner_data, inner_data + size);
750 for (int stream_index : stream_indices) {
751 servers->set_header(stream_index, http_header, stream_header);
754 for (int stream_index : stream_indices) {
755 servers->add_data(stream_index, inner_data, size, flags, next_block_pts);
757 next_block_pts = RationalPTS();
760 // Consume the block. This isn't the most efficient way of dealing with things
761 // should we have many blocks, but these routines don't need to be too efficient
763 pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube2_block_header) + size);
764 has_metacube_header = false;
768 void HTTPInput::drop_pending_data(size_t num_bytes)
770 if (num_bytes == 0) {
773 log(WARNING, "[%s] Dropping %lld junk bytes; not a Metacube2 stream, or data was dropped from the middle of the stream.",
774 url.c_str(), (long long)num_bytes);
775 assert(pending_data.size() >= num_bytes);
776 pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
779 void HTTPInput::add_destination(int stream_index)
781 stream_indices.push_back(stream_index);
782 servers->set_header(stream_index, http_header, stream_header);
785 InputStats HTTPInput::get_stats() const
787 lock_guard<mutex> lock(stats_mutex);
791 void HTTPInput::process_metacube_metadata_block(const metacube2_block_header &hdr, const char *payload, uint32_t payload_size)
793 if (payload_size < sizeof(uint64_t)) {
794 log(WARNING, "[%s] Undersized Metacube metadata block (%d bytes); corrupted header?",
795 url.c_str(), payload_size);
799 uint64_t type = be64toh(*(const uint64_t *)payload);
800 if (type == METACUBE_METADATA_TYPE_ENCODER_TIMESTAMP) {
802 clock_gettime(CLOCK_REALTIME, &now);
804 const metacube2_timestamp_packet *pkt = (const metacube2_timestamp_packet *)payload;
805 if (payload_size != sizeof(*pkt)) {
806 log(WARNING, "[%s] Metacube timestamp block of wrong size (%d bytes); ignoring.",
807 url.c_str(), payload_size);
811 double elapsed = now.tv_sec - be64toh(pkt->tv_sec) +
812 1e-9 * (now.tv_nsec - long(be64toh(pkt->tv_nsec)));
814 lock_guard<mutex> lock(stats_mutex);
815 stats.latency_sec = elapsed;
817 } else if (type == METACUBE_METADATA_TYPE_NEXT_BLOCK_PTS) {
818 const metacube2_pts_packet *pkt = (const metacube2_pts_packet *)payload;
819 if (payload_size != sizeof(*pkt)) {
820 log(WARNING, "[%s] Metacube pts block of wrong size (%d bytes); ignoring.",
821 url.c_str(), payload_size);
824 next_block_pts.pts = be64toh(pkt->pts);
825 next_block_pts.timebase_num = be64toh(pkt->timebase_num);
826 next_block_pts.timebase_den = be64toh(pkt->timebase_den);
828 // Unknown metadata block, ignore
829 log(INFO, "[%s] Metadata block %llu\n", url.c_str(), type);