4 #include <netinet/in.h>
10 #include <sys/socket.h>
19 #include "httpinput.h"
21 #include "metacube2.h"
22 #include "mutexlock.h"
24 #include "serverpool.h"
35 string host_header(const string &host, const string &port)
37 if (port == "http" || atoi(port.c_str()) == 80) {
40 return host + ":" + port;
46 extern ServerPool *servers;
48 HTTPInput::HTTPInput(const string &url, Input::Encoding encoding)
49 : state(NOT_CONNECTED),
52 has_metacube_header(false),
54 num_connection_attempts(0),
55 suppress_logging(false)
57 pthread_mutex_init(&stats_mutex, NULL);
59 stats.bytes_received = 0;
60 stats.data_bytes_received = 0;
61 stats.connect_time = -1;
63 last_verbose_connection.tv_sec = -3600;
64 last_verbose_connection.tv_nsec = 0;
67 HTTPInput::HTTPInput(const InputProto &serialized)
68 : state(State(serialized.state())),
69 url(serialized.url()),
70 encoding(serialized.is_metacube_encoded() ?
71 Input::INPUT_ENCODING_METACUBE :
72 Input::INPUT_ENCODING_RAW),
73 request(serialized.request()),
74 request_bytes_sent(serialized.request_bytes_sent()),
75 response(serialized.response()),
76 http_header(serialized.http_header()),
77 stream_header(serialized.stream_header()),
78 has_metacube_header(serialized.has_metacube_header()),
79 sock(serialized.sock())
81 pending_data.resize(serialized.pending_data().size());
82 memcpy(&pending_data[0], serialized.pending_data().data(), serialized.pending_data().size());
84 string protocol, user;
85 parse_url(url, &protocol, &user, &host, &port, &path); // Don't care if it fails.
87 pthread_mutex_init(&stats_mutex, NULL);
89 stats.bytes_received = serialized.bytes_received();
90 stats.data_bytes_received = serialized.data_bytes_received();
91 if (serialized.has_connect_time()) {
92 stats.connect_time = serialized.connect_time();
94 stats.connect_time = time(NULL);
98 void HTTPInput::close_socket()
105 MutexLock lock(&stats_mutex);
106 stats.connect_time = -1;
109 InputProto HTTPInput::serialize() const
111 InputProto serialized;
112 serialized.set_state(state);
113 serialized.set_url(url);
114 serialized.set_request(request);
115 serialized.set_request_bytes_sent(request_bytes_sent);
116 serialized.set_response(response);
117 serialized.set_http_header(http_header);
118 serialized.set_stream_header(stream_header);
119 serialized.set_pending_data(string(pending_data.begin(), pending_data.end()));
120 serialized.set_has_metacube_header(has_metacube_header);
121 serialized.set_sock(sock);
122 serialized.set_bytes_received(stats.bytes_received);
123 serialized.set_data_bytes_received(stats.data_bytes_received);
124 serialized.set_connect_time(stats.connect_time);
125 if (encoding == Input::INPUT_ENCODING_METACUBE) {
126 serialized.set_is_metacube_encoded(true);
128 assert(encoding == Input::INPUT_ENCODING_RAW);
129 serialized.set_is_metacube_encoded(false);
134 int HTTPInput::lookup_and_connect(const string &host, const string &port)
137 int err = getaddrinfo(host.c_str(), port.c_str(), NULL, &ai);
139 log(WARNING, "[%s] Lookup of '%s' failed (%s).",
140 url.c_str(), host.c_str(), gai_strerror(err));
144 addrinfo *base_ai = ai;
146 // Connect to everything in turn until we have a socket.
147 for ( ; ai && !should_stop(); ai = ai->ai_next) {
148 int sock = socket(ai->ai_family, SOCK_STREAM, IPPROTO_TCP);
150 // Could be e.g. EPROTONOSUPPORT. The show must go on.
154 // Now do a non-blocking connect. This is important because we want to be able to be
155 // woken up, even though it's rather cumbersome.
157 // Set the socket as nonblocking.
159 if (ioctl(sock, FIONBIO, &one) == -1) {
160 log_perror("ioctl(FIONBIO)");
165 // Do a non-blocking connect.
167 err = connect(sock, ai->ai_addr, ai->ai_addrlen);
168 } while (err == -1 && errno == EINTR);
170 if (err == -1 && errno != EINPROGRESS) {
171 log_perror("connect");
176 // Wait for the connect to complete, or an error to happen.
178 bool complete = wait_for_activity(sock, POLLIN | POLLOUT, NULL);
188 // Check whether it ended in an error or not.
189 socklen_t err_size = sizeof(err);
190 if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &err_size) == -1) {
191 log_perror("getsockopt");
199 // Successful connect.
200 freeaddrinfo(base_ai);
207 // Give the last one as error.
208 log(WARNING, "[%s] Connect to '%s' failed (%s)",
209 url.c_str(), host.c_str(), strerror(errno));
210 freeaddrinfo(base_ai);
214 bool HTTPInput::parse_response(const string &request)
216 vector<string> lines = split_lines(response);
218 if (!suppress_logging) {
219 log(WARNING, "[%s] Empty HTTP response from input.", url.c_str());
224 vector<string> first_line_tokens = split_tokens(lines[0]);
225 if (first_line_tokens.size() < 2) {
226 if (!suppress_logging) {
227 log(WARNING, "[%s] Malformed response line '%s' from input.",
228 url.c_str(), lines[0].c_str());
233 int response = atoi(first_line_tokens[1].c_str());
234 if (response != 200) {
235 if (!suppress_logging) {
236 log(WARNING, "[%s] Non-200 response '%s' from input.",
237 url.c_str(), lines[0].c_str());
242 multimap<string, string> parameters = extract_headers(lines, url);
244 // Remove “Content-encoding: metacube”.
245 // TODO: Make case-insensitive.
246 multimap<string, string>::iterator encoding_it =
247 parameters.find("Content-encoding");
248 if (encoding_it != parameters.end() && encoding_it->second == "metacube") {
249 parameters.erase(encoding_it);
252 // Change “Server: foo” to “Server: metacube/0.1 (reflecting: foo)”
253 // TODO: Make case-insensitive.
254 // XXX: Use a Via: instead?
255 if (parameters.count("Server") == 0) {
256 parameters.insert(make_pair("Server", SERVER_IDENTIFICATION));
258 for (multimap<string, string>::iterator it = parameters.begin();
259 it != parameters.end();
261 if (it->first != "Server") {
264 it->second = SERVER_IDENTIFICATION " (reflecting: " + it->second + ")";
268 // Set “Connection: close”.
269 // TODO: Make case-insensitive.
270 parameters.erase("Connection");
271 parameters.insert(make_pair("Connection", "close"));
273 // Construct the new HTTP header.
274 http_header = "HTTP/1.0 200 OK\r\n";
275 for (multimap<string, string>::iterator it = parameters.begin();
276 it != parameters.end();
278 http_header.append(it->first + ": " + it->second + "\r\n");
281 for (size_t i = 0; i < stream_indices.size(); ++i) {
282 servers->set_header(stream_indices[i], http_header, stream_header);
288 void HTTPInput::do_work()
290 timespec last_activity;
292 // TODO: Make the timeout persist across restarts.
293 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
294 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
298 while (!should_stop()) {
299 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
300 // Give the socket 30 seconds since last activity before we time out.
301 static const int timeout_secs = 30;
304 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
307 timespec elapsed = clock_diff(last_activity, now);
308 if (elapsed.tv_sec >= timeout_secs) {
310 if (!suppress_logging) {
311 log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec);
313 state = CLOSING_SOCKET;
317 // Basically calculate (30 - (now - last_activity)) = (30 + (last_activity - now)).
318 // Add a second of slack to account for differences between clocks.
319 timespec timeout = clock_diff(now, last_activity);
320 timeout.tv_sec += timeout_secs + 1;
321 assert(timeout.tv_sec > 0 || (timeout.tv_sec >= 0 && timeout.tv_nsec > 0));
323 bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout);
325 err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
328 // OK. Most likely, should_stop was set, or we have timed out.
336 request_bytes_sent = 0;
338 pending_data.clear();
339 has_metacube_header = false;
340 for (size_t i = 0; i < stream_indices.size(); ++i) {
341 servers->set_header(stream_indices[i], "", "");
345 string protocol, user; // Thrown away.
346 if (!parse_url(url, &protocol, &user, &host, &port, &path)) {
347 if (!suppress_logging) {
348 log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str());
353 // Remove the brackets around IPv6 address literals.
354 // TODO: See if we can join this with the code in parse_ip_address(),
355 // or maybe even more it into parse_url().
356 if (!host.empty() && host[0] == '[' && host[host.size() - 1] == ']') {
357 host = host.substr(1, host.size() - 2);
361 if (suppress_logging) {
362 // See if there's more than one minute since last time we made a connection
363 // with logging enabled. If so, turn it on again.
365 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
368 double elapsed = now.tv_sec - last_verbose_connection.tv_sec +
369 1e-9 * (now.tv_nsec - last_verbose_connection.tv_nsec);
370 if (elapsed > 60.0) {
371 suppress_logging = false;
374 if (!suppress_logging) {
375 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_verbose_connection);
378 ++num_connection_attempts;
379 sock = lookup_and_connect(host, port);
381 // Yay, successful connect. Try to set it as nonblocking.
383 if (ioctl(sock, FIONBIO, &one) == -1) {
384 log_perror("ioctl(FIONBIO)");
385 state = CLOSING_SOCKET;
387 state = SENDING_REQUEST;
388 request = "GET " + path + " HTTP/1.0\r\nHost: " + host_header(host, port) + "\r\nUser-Agent: cubemap\r\n\r\n";
389 request_bytes_sent = 0;
392 MutexLock lock(&stats_mutex);
393 stats.connect_time = time(NULL);
394 clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
397 case SENDING_REQUEST: {
398 size_t to_send = request.size() - request_bytes_sent;
402 ret = write(sock, request.data() + request_bytes_sent, to_send);
403 } while (ret == -1 && errno == EINTR);
407 state = CLOSING_SOCKET;
412 request_bytes_sent += ret;
414 if (request_bytes_sent == request.size()) {
415 state = RECEIVING_HEADER;
419 case RECEIVING_HEADER: {
424 ret = read(sock, buf, sizeof(buf));
425 } while (ret == -1 && errno == EINTR);
429 state = CLOSING_SOCKET;
434 // This really shouldn't happen...
435 if (!suppress_logging) {
436 log(ERROR, "[%s] Socket unexpectedly closed while reading header",
439 state = CLOSING_SOCKET;
443 RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
445 if (status == RP_OUT_OF_SPACE) {
446 if (!suppress_logging) {
447 log(WARNING, "[%s] Server sent overlong HTTP response!", url.c_str());
449 state = CLOSING_SOCKET;
451 } else if (status == RP_NOT_FINISHED_YET) {
455 // OK, so we're fine, but there might be some of the actual data after the response.
456 // We'll need to deal with that separately.
458 if (status == RP_EXTRA_DATA) {
459 char *ptr = static_cast<char *>(
460 memmem(response.data(), response.size(), "\r\n\r\n", 4));
462 extra_data = string(ptr + 4, &response[0] + response.size());
463 response.resize(ptr - response.data());
466 if (!parse_response(response)) {
467 state = CLOSING_SOCKET;
471 if (!extra_data.empty()) {
472 process_data(&extra_data[0], extra_data.size());
475 if (!suppress_logging) {
476 if (encoding == Input::INPUT_ENCODING_RAW) {
477 log(INFO, "[%s] Connected to '%s', receiving raw data.",
478 url.c_str(), url.c_str());
480 assert(encoding == Input::INPUT_ENCODING_METACUBE);
481 log(INFO, "[%s] Connected to '%s', receiving data.",
482 url.c_str(), url.c_str());
485 state = RECEIVING_DATA;
488 case RECEIVING_DATA: {
493 ret = read(sock, buf, sizeof(buf));
494 } while (ret == -1 && errno == EINTR);
498 state = CLOSING_SOCKET;
503 // This really shouldn't happen...
504 if (!suppress_logging) {
505 log(ERROR, "[%s] Socket unexpectedly closed while reading data",
508 state = CLOSING_SOCKET;
512 num_connection_attempts = 0; // Reset, since we have a successful read.
513 if (suppress_logging) {
514 // This was suppressed earlier, so print it out now.
515 if (encoding == Input::INPUT_ENCODING_RAW) {
516 log(INFO, "[%s] Connected to '%s', receiving raw data.",
517 url.c_str(), url.c_str());
519 assert(encoding == Input::INPUT_ENCODING_METACUBE);
520 log(INFO, "[%s] Connected to '%s', receiving data.",
521 url.c_str(), url.c_str());
523 suppress_logging = false;
526 process_data(buf, ret);
529 case CLOSING_SOCKET: {
531 state = NOT_CONNECTED;
538 // If we are still in NOT_CONNECTED, either something went wrong,
539 // or the connection just got closed.
540 // The earlier steps have already given the error message, if any.
541 if (state == NOT_CONNECTED && !should_stop()) {
542 if (!suppress_logging) {
543 log(INFO, "[%s] Waiting 0.2 seconds and restarting...", url.c_str());
546 if (num_connection_attempts >= 3 && !suppress_logging) {
547 log(INFO, "[%s] %d failed connection attempts, suppressing logging for one minute.",
548 url.c_str(), num_connection_attempts);
549 suppress_logging = true;
552 timeout_ts.tv_sec = 0;
553 timeout_ts.tv_nsec = 200000000;
554 wait_for_wakeup(&timeout_ts);
559 void HTTPInput::process_data(char *ptr, size_t bytes)
562 MutexLock mutex(&stats_mutex);
563 stats.bytes_received += bytes;
566 if (encoding == Input::INPUT_ENCODING_RAW) {
567 for (size_t i = 0; i < stream_indices.size(); ++i) {
568 servers->add_data(stream_indices[i], ptr, bytes, /*metacube_flags=*/0);
573 assert(encoding == Input::INPUT_ENCODING_METACUBE);
574 pending_data.insert(pending_data.end(), ptr, ptr + bytes);
577 // If we don't have enough data (yet) for even the Metacube header, just return.
578 if (pending_data.size() < sizeof(metacube2_block_header)) {
582 // Make sure we have the Metacube sync header at the start.
583 // We may need to skip over junk data (it _should_ not happen, though).
584 if (!has_metacube_header) {
585 char *ptr = static_cast<char *>(
586 memmem(pending_data.data(), pending_data.size(),
587 METACUBE2_SYNC, strlen(METACUBE2_SYNC)));
589 // OK, so we didn't find the sync marker. We know then that
590 // we do not have the _full_ marker in the buffer, but we
591 // could have N-1 bytes. Drop everything before that,
593 drop_pending_data(pending_data.size() - (strlen(METACUBE2_SYNC) - 1));
596 // Yay, we found the header. Drop everything (if anything) before it.
597 drop_pending_data(ptr - pending_data.data());
598 has_metacube_header = true;
600 // Re-check that we have the entire header; we could have dropped data.
601 if (pending_data.size() < sizeof(metacube2_block_header)) {
607 // Now it's safe to read the header.
608 metacube2_block_header hdr;
609 memcpy(&hdr, pending_data.data(), sizeof(hdr));
610 assert(memcmp(hdr.sync, METACUBE2_SYNC, sizeof(hdr.sync)) == 0);
611 uint32_t size = ntohl(hdr.size);
612 uint16_t flags = ntohs(hdr.flags);
613 uint16_t expected_csum = metacube2_compute_crc(&hdr);
615 if (expected_csum != ntohs(hdr.csum)) {
616 log(WARNING, "[%s] Metacube checksum failed (expected 0x%x, got 0x%x), "
617 "not reading block claiming to be %d bytes (flags=%x).",
618 url.c_str(), expected_csum, ntohs(hdr.csum),
621 // Drop only the first byte, and let the rest of the code handle resync.
622 pending_data.erase(pending_data.begin(), pending_data.begin() + 1);
623 has_metacube_header = false;
626 if (size > 10485760) {
627 log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header??",
628 url.c_str(), size, flags);
631 // See if we have the entire block. If not, wait for more data.
632 if (pending_data.size() < sizeof(metacube2_block_header) + size) {
636 // Send this block on to the servers.
638 MutexLock lock(&stats_mutex);
639 stats.data_bytes_received += size;
641 char *inner_data = pending_data.data() + sizeof(metacube2_block_header);
642 if (flags & METACUBE_FLAGS_HEADER) {
643 stream_header = string(inner_data, inner_data + size);
644 for (size_t i = 0; i < stream_indices.size(); ++i) {
645 servers->set_header(stream_indices[i], http_header, stream_header);
648 for (size_t i = 0; i < stream_indices.size(); ++i) {
649 servers->add_data(stream_indices[i], inner_data, size, flags);
652 // Consume the block. This isn't the most efficient way of dealing with things
653 // should we have many blocks, but these routines don't need to be too efficient
655 pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube2_block_header) + size);
656 has_metacube_header = false;
660 void HTTPInput::drop_pending_data(size_t num_bytes)
662 if (num_bytes == 0) {
665 log(WARNING, "[%s] Dropping %lld junk bytes; not a Metacube2 stream, or data was dropped from the middle of the stream.",
666 url.c_str(), (long long)num_bytes);
667 assert(pending_data.size() >= num_bytes);
668 pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
671 void HTTPInput::add_destination(int stream_index)
673 stream_indices.push_back(stream_index);
674 servers->set_header(stream_index, http_header, stream_header);
677 InputStats HTTPInput::get_stats() const
679 MutexLock lock(&stats_mutex);