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()),
80 num_connection_attempts(0),
81 suppress_logging(false)
83 pending_data.resize(serialized.pending_data().size());
84 memcpy(&pending_data[0], serialized.pending_data().data(), serialized.pending_data().size());
86 string protocol, user;
87 parse_url(url, &protocol, &user, &host, &port, &path); // Don't care if it fails.
89 pthread_mutex_init(&stats_mutex, NULL);
91 stats.bytes_received = serialized.bytes_received();
92 stats.data_bytes_received = serialized.data_bytes_received();
93 if (serialized.has_connect_time()) {
94 stats.connect_time = serialized.connect_time();
96 stats.connect_time = time(NULL);
99 last_verbose_connection.tv_sec = -3600;
100 last_verbose_connection.tv_nsec = 0;
103 void HTTPInput::close_socket()
110 MutexLock lock(&stats_mutex);
111 stats.connect_time = -1;
114 InputProto HTTPInput::serialize() const
116 InputProto serialized;
117 serialized.set_state(state);
118 serialized.set_url(url);
119 serialized.set_request(request);
120 serialized.set_request_bytes_sent(request_bytes_sent);
121 serialized.set_response(response);
122 serialized.set_http_header(http_header);
123 serialized.set_stream_header(stream_header);
124 serialized.set_pending_data(string(pending_data.begin(), pending_data.end()));
125 serialized.set_has_metacube_header(has_metacube_header);
126 serialized.set_sock(sock);
127 serialized.set_bytes_received(stats.bytes_received);
128 serialized.set_data_bytes_received(stats.data_bytes_received);
129 serialized.set_connect_time(stats.connect_time);
130 if (encoding == Input::INPUT_ENCODING_METACUBE) {
131 serialized.set_is_metacube_encoded(true);
133 assert(encoding == Input::INPUT_ENCODING_RAW);
134 serialized.set_is_metacube_encoded(false);
139 int HTTPInput::lookup_and_connect(const string &host, const string &port)
142 int err = getaddrinfo(host.c_str(), port.c_str(), NULL, &ai);
144 if (!suppress_logging) {
145 log(WARNING, "[%s] Lookup of '%s' failed (%s).",
146 url.c_str(), host.c_str(), gai_strerror(err));
151 addrinfo *base_ai = ai;
153 // Connect to everything in turn until we have a socket.
154 for ( ; ai && !should_stop(); ai = ai->ai_next) {
155 int sock = socket(ai->ai_family, SOCK_STREAM, IPPROTO_TCP);
157 // Could be e.g. EPROTONOSUPPORT. The show must go on.
161 // Now do a non-blocking connect. This is important because we want to be able to be
162 // woken up, even though it's rather cumbersome.
164 // Set the socket as nonblocking.
166 if (ioctl(sock, FIONBIO, &one) == -1) {
167 log_perror("ioctl(FIONBIO)");
172 // Do a non-blocking connect.
174 err = connect(sock, ai->ai_addr, ai->ai_addrlen);
175 } while (err == -1 && errno == EINTR);
177 if (err == -1 && errno != EINPROGRESS) {
178 log_perror("connect");
183 // Wait for the connect to complete, or an error to happen.
185 bool complete = wait_for_activity(sock, POLLIN | POLLOUT, NULL);
195 // Check whether it ended in an error or not.
196 socklen_t err_size = sizeof(err);
197 if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &err_size) == -1) {
198 log_perror("getsockopt");
206 // Successful connect.
207 freeaddrinfo(base_ai);
214 // Give the last one as error.
215 if (!suppress_logging) {
216 log(WARNING, "[%s] Connect to '%s' failed (%s)",
217 url.c_str(), host.c_str(), strerror(errno));
219 freeaddrinfo(base_ai);
223 bool HTTPInput::parse_response(const string &request)
225 vector<string> lines = split_lines(response);
227 if (!suppress_logging) {
228 log(WARNING, "[%s] Empty HTTP response from input.", url.c_str());
233 vector<string> first_line_tokens = split_tokens(lines[0]);
234 if (first_line_tokens.size() < 2) {
235 if (!suppress_logging) {
236 log(WARNING, "[%s] Malformed response line '%s' from input.",
237 url.c_str(), lines[0].c_str());
242 int response = atoi(first_line_tokens[1].c_str());
243 if (response != 200) {
244 if (!suppress_logging) {
245 log(WARNING, "[%s] Non-200 response '%s' from input.",
246 url.c_str(), lines[0].c_str());
251 multimap<string, string> parameters = extract_headers(lines, url);
253 // Remove “Content-encoding: metacube”.
254 // TODO: Make case-insensitive.
255 multimap<string, string>::iterator encoding_it =
256 parameters.find("Content-encoding");
257 if (encoding_it != parameters.end() && encoding_it->second == "metacube") {
258 parameters.erase(encoding_it);
261 // Change “Server: foo” to “Server: metacube/0.1 (reflecting: foo)”
262 // TODO: Make case-insensitive.
263 // XXX: Use a Via: instead?
264 if (parameters.count("Server") == 0) {
265 parameters.insert(make_pair("Server", SERVER_IDENTIFICATION));
267 for (multimap<string, string>::iterator it = parameters.begin();
268 it != parameters.end();
270 if (it->first != "Server") {
273 it->second = SERVER_IDENTIFICATION " (reflecting: " + it->second + ")";
277 // Set “Connection: close”.
278 // TODO: Make case-insensitive.
279 parameters.erase("Connection");
280 parameters.insert(make_pair("Connection", "close"));
282 // Construct the new HTTP header.
283 http_header = "HTTP/1.0 200 OK\r\n";
284 for (multimap<string, string>::iterator it = parameters.begin();
285 it != parameters.end();
287 http_header.append(it->first + ": " + it->second + "\r\n");
290 for (size_t i = 0; i < stream_indices.size(); ++i) {
291 servers->set_header(stream_indices[i], http_header, stream_header);
297 void HTTPInput::do_work()
299 timespec last_activity;
301 // TODO: Make the timeout persist across restarts.
302 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
303 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
307 while (!should_stop()) {
308 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
309 // Give the socket 30 seconds since last activity before we time out.
310 static const int timeout_secs = 30;
313 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
316 timespec elapsed = clock_diff(last_activity, now);
317 if (elapsed.tv_sec >= timeout_secs) {
319 if (!suppress_logging) {
320 log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec);
322 state = CLOSING_SOCKET;
326 // Basically calculate (30 - (now - last_activity)) = (30 + (last_activity - now)).
327 // Add a second of slack to account for differences between clocks.
328 timespec timeout = clock_diff(now, last_activity);
329 timeout.tv_sec += timeout_secs + 1;
330 assert(timeout.tv_sec > 0 || (timeout.tv_sec >= 0 && timeout.tv_nsec > 0));
332 bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout);
334 err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
337 // OK. Most likely, should_stop was set, or we have timed out.
345 request_bytes_sent = 0;
347 pending_data.clear();
348 has_metacube_header = false;
349 for (size_t i = 0; i < stream_indices.size(); ++i) {
350 servers->set_header(stream_indices[i], "", "");
354 string protocol, user; // Thrown away.
355 if (!parse_url(url, &protocol, &user, &host, &port, &path)) {
356 if (!suppress_logging) {
357 log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str());
362 // Remove the brackets around IPv6 address literals.
363 // TODO: See if we can join this with the code in parse_ip_address(),
364 // or maybe even more it into parse_url().
365 if (!host.empty() && host[0] == '[' && host[host.size() - 1] == ']') {
366 host = host.substr(1, host.size() - 2);
370 if (suppress_logging) {
371 // See if there's more than one minute since last time we made a connection
372 // with logging enabled. If so, turn it on again.
374 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
377 double elapsed = now.tv_sec - last_verbose_connection.tv_sec +
378 1e-9 * (now.tv_nsec - last_verbose_connection.tv_nsec);
379 if (elapsed > 60.0) {
380 suppress_logging = false;
383 if (!suppress_logging) {
384 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_verbose_connection);
387 ++num_connection_attempts;
388 sock = lookup_and_connect(host, port);
390 // Yay, successful connect. Try to set it as nonblocking.
392 if (ioctl(sock, FIONBIO, &one) == -1) {
393 log_perror("ioctl(FIONBIO)");
394 state = CLOSING_SOCKET;
396 state = SENDING_REQUEST;
397 request = "GET " + path + " HTTP/1.0\r\nHost: " + host_header(host, port) + "\r\nUser-Agent: cubemap\r\n\r\n";
398 request_bytes_sent = 0;
401 MutexLock lock(&stats_mutex);
402 stats.connect_time = time(NULL);
403 clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
406 case SENDING_REQUEST: {
407 size_t to_send = request.size() - request_bytes_sent;
411 ret = write(sock, request.data() + request_bytes_sent, to_send);
412 } while (ret == -1 && errno == EINTR);
416 state = CLOSING_SOCKET;
421 request_bytes_sent += ret;
423 if (request_bytes_sent == request.size()) {
424 state = RECEIVING_HEADER;
428 case RECEIVING_HEADER: {
433 ret = read(sock, buf, sizeof(buf));
434 } while (ret == -1 && errno == EINTR);
438 state = CLOSING_SOCKET;
443 // This really shouldn't happen...
444 if (!suppress_logging) {
445 log(ERROR, "[%s] Socket unexpectedly closed while reading header",
448 state = CLOSING_SOCKET;
452 RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
454 if (status == RP_OUT_OF_SPACE) {
455 if (!suppress_logging) {
456 log(WARNING, "[%s] Server sent overlong HTTP response!", url.c_str());
458 state = CLOSING_SOCKET;
460 } else if (status == RP_NOT_FINISHED_YET) {
464 // OK, so we're fine, but there might be some of the actual data after the response.
465 // We'll need to deal with that separately.
467 if (status == RP_EXTRA_DATA) {
468 char *ptr = static_cast<char *>(
469 memmem(response.data(), response.size(), "\r\n\r\n", 4));
471 extra_data = string(ptr + 4, &response[0] + response.size());
472 response.resize(ptr - response.data());
475 if (!parse_response(response)) {
476 state = CLOSING_SOCKET;
480 if (!extra_data.empty()) {
481 process_data(&extra_data[0], extra_data.size());
484 if (!suppress_logging) {
485 if (encoding == Input::INPUT_ENCODING_RAW) {
486 log(INFO, "[%s] Connected to '%s', receiving raw data.",
487 url.c_str(), url.c_str());
489 assert(encoding == Input::INPUT_ENCODING_METACUBE);
490 log(INFO, "[%s] Connected to '%s', receiving data.",
491 url.c_str(), url.c_str());
494 state = RECEIVING_DATA;
497 case RECEIVING_DATA: {
502 ret = read(sock, buf, sizeof(buf));
503 } while (ret == -1 && errno == EINTR);
507 state = CLOSING_SOCKET;
512 // This really shouldn't happen...
513 if (!suppress_logging) {
514 log(ERROR, "[%s] Socket unexpectedly closed while reading data",
517 state = CLOSING_SOCKET;
521 num_connection_attempts = 0; // Reset, since we have a successful read.
522 if (suppress_logging) {
523 // This was suppressed earlier, so print it out now.
524 if (encoding == Input::INPUT_ENCODING_RAW) {
525 log(INFO, "[%s] Connected to '%s', receiving raw data.",
526 url.c_str(), url.c_str());
528 assert(encoding == Input::INPUT_ENCODING_METACUBE);
529 log(INFO, "[%s] Connected to '%s', receiving data.",
530 url.c_str(), url.c_str());
532 suppress_logging = false;
535 process_data(buf, ret);
538 case CLOSING_SOCKET: {
540 state = NOT_CONNECTED;
547 // If we are still in NOT_CONNECTED, either something went wrong,
548 // or the connection just got closed.
549 // The earlier steps have already given the error message, if any.
550 if (state == NOT_CONNECTED && !should_stop()) {
551 if (!suppress_logging) {
552 log(INFO, "[%s] Waiting 0.2 seconds and restarting...", url.c_str());
555 if (num_connection_attempts >= 3 && !suppress_logging) {
556 log(INFO, "[%s] %d failed connection attempts, suppressing logging for one minute.",
557 url.c_str(), num_connection_attempts);
558 suppress_logging = true;
561 timeout_ts.tv_sec = 0;
562 timeout_ts.tv_nsec = 200000000;
563 wait_for_wakeup(&timeout_ts);
568 void HTTPInput::process_data(char *ptr, size_t bytes)
571 MutexLock mutex(&stats_mutex);
572 stats.bytes_received += bytes;
575 if (encoding == Input::INPUT_ENCODING_RAW) {
576 for (size_t i = 0; i < stream_indices.size(); ++i) {
577 servers->add_data(stream_indices[i], ptr, bytes, /*metacube_flags=*/0);
582 assert(encoding == Input::INPUT_ENCODING_METACUBE);
583 pending_data.insert(pending_data.end(), ptr, ptr + bytes);
586 // If we don't have enough data (yet) for even the Metacube header, just return.
587 if (pending_data.size() < sizeof(metacube2_block_header)) {
591 // Make sure we have the Metacube sync header at the start.
592 // We may need to skip over junk data (it _should_ not happen, though).
593 if (!has_metacube_header) {
594 char *ptr = static_cast<char *>(
595 memmem(pending_data.data(), pending_data.size(),
596 METACUBE2_SYNC, strlen(METACUBE2_SYNC)));
598 // OK, so we didn't find the sync marker. We know then that
599 // we do not have the _full_ marker in the buffer, but we
600 // could have N-1 bytes. Drop everything before that,
602 drop_pending_data(pending_data.size() - (strlen(METACUBE2_SYNC) - 1));
605 // Yay, we found the header. Drop everything (if anything) before it.
606 drop_pending_data(ptr - pending_data.data());
607 has_metacube_header = true;
609 // Re-check that we have the entire header; we could have dropped data.
610 if (pending_data.size() < sizeof(metacube2_block_header)) {
616 // Now it's safe to read the header.
617 metacube2_block_header hdr;
618 memcpy(&hdr, pending_data.data(), sizeof(hdr));
619 assert(memcmp(hdr.sync, METACUBE2_SYNC, sizeof(hdr.sync)) == 0);
620 uint32_t size = ntohl(hdr.size);
621 uint16_t flags = ntohs(hdr.flags);
622 uint16_t expected_csum = metacube2_compute_crc(&hdr);
624 if (expected_csum != ntohs(hdr.csum)) {
625 log(WARNING, "[%s] Metacube checksum failed (expected 0x%x, got 0x%x), "
626 "not reading block claiming to be %d bytes (flags=%x).",
627 url.c_str(), expected_csum, ntohs(hdr.csum),
630 // Drop only the first byte, and let the rest of the code handle resync.
631 pending_data.erase(pending_data.begin(), pending_data.begin() + 1);
632 has_metacube_header = false;
635 if (size > 10485760) {
636 log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header??",
637 url.c_str(), size, flags);
640 // See if we have the entire block. If not, wait for more data.
641 if (pending_data.size() < sizeof(metacube2_block_header) + size) {
645 // Send this block on to the servers.
647 MutexLock lock(&stats_mutex);
648 stats.data_bytes_received += size;
650 char *inner_data = pending_data.data() + sizeof(metacube2_block_header);
651 if (flags & METACUBE_FLAGS_HEADER) {
652 stream_header = string(inner_data, inner_data + size);
653 for (size_t i = 0; i < stream_indices.size(); ++i) {
654 servers->set_header(stream_indices[i], http_header, stream_header);
657 for (size_t i = 0; i < stream_indices.size(); ++i) {
658 servers->add_data(stream_indices[i], inner_data, size, flags);
661 // Consume the block. This isn't the most efficient way of dealing with things
662 // should we have many blocks, but these routines don't need to be too efficient
664 pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube2_block_header) + size);
665 has_metacube_header = false;
669 void HTTPInput::drop_pending_data(size_t num_bytes)
671 if (num_bytes == 0) {
674 log(WARNING, "[%s] Dropping %lld junk bytes; not a Metacube2 stream, or data was dropped from the middle of the stream.",
675 url.c_str(), (long long)num_bytes);
676 assert(pending_data.size() >= num_bytes);
677 pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
680 void HTTPInput::add_destination(int stream_index)
682 stream_indices.push_back(stream_index);
683 servers->set_header(stream_index, http_header, stream_header);
686 InputStats HTTPInput::get_stats() const
688 MutexLock lock(&stats_mutex);