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),
55 pthread_mutex_init(&stats_mutex, NULL);
57 stats.bytes_received = 0;
58 stats.data_bytes_received = 0;
59 stats.connect_time = -1;
62 HTTPInput::HTTPInput(const InputProto &serialized)
63 : state(State(serialized.state())),
64 url(serialized.url()),
65 encoding(serialized.is_metacube_encoded() ?
66 Input::INPUT_ENCODING_METACUBE :
67 Input::INPUT_ENCODING_RAW),
68 request(serialized.request()),
69 request_bytes_sent(serialized.request_bytes_sent()),
70 response(serialized.response()),
71 http_header(serialized.http_header()),
72 stream_header(serialized.stream_header()),
73 has_metacube_header(serialized.has_metacube_header()),
74 sock(serialized.sock())
76 pending_data.resize(serialized.pending_data().size());
77 memcpy(&pending_data[0], serialized.pending_data().data(), serialized.pending_data().size());
79 string protocol, user;
80 parse_url(url, &protocol, &user, &host, &port, &path); // Don't care if it fails.
82 pthread_mutex_init(&stats_mutex, NULL);
84 stats.bytes_received = serialized.bytes_received();
85 stats.data_bytes_received = serialized.data_bytes_received();
86 if (serialized.has_connect_time()) {
87 stats.connect_time = serialized.connect_time();
89 stats.connect_time = time(NULL);
93 void HTTPInput::close_socket()
100 MutexLock lock(&stats_mutex);
101 stats.connect_time = -1;
104 InputProto HTTPInput::serialize() const
106 InputProto serialized;
107 serialized.set_state(state);
108 serialized.set_url(url);
109 serialized.set_request(request);
110 serialized.set_request_bytes_sent(request_bytes_sent);
111 serialized.set_response(response);
112 serialized.set_http_header(http_header);
113 serialized.set_stream_header(stream_header);
114 serialized.set_pending_data(string(pending_data.begin(), pending_data.end()));
115 serialized.set_has_metacube_header(has_metacube_header);
116 serialized.set_sock(sock);
117 serialized.set_bytes_received(stats.bytes_received);
118 serialized.set_data_bytes_received(stats.data_bytes_received);
119 serialized.set_connect_time(stats.connect_time);
120 if (encoding == Input::INPUT_ENCODING_METACUBE) {
121 serialized.set_is_metacube_encoded(true);
123 assert(encoding == Input::INPUT_ENCODING_RAW);
124 serialized.set_is_metacube_encoded(false);
129 int HTTPInput::lookup_and_connect(const string &host, const string &port)
132 int err = getaddrinfo(host.c_str(), port.c_str(), NULL, &ai);
134 log(WARNING, "[%s] Lookup of '%s' failed (%s).",
135 url.c_str(), host.c_str(), gai_strerror(err));
139 addrinfo *base_ai = ai;
141 // Connect to everything in turn until we have a socket.
142 for ( ; ai && !should_stop(); ai = ai->ai_next) {
143 int sock = socket(ai->ai_family, SOCK_STREAM, IPPROTO_TCP);
145 // Could be e.g. EPROTONOSUPPORT. The show must go on.
149 // Now do a non-blocking connect. This is important because we want to be able to be
150 // woken up, even though it's rather cumbersome.
152 // Set the socket as nonblocking.
154 if (ioctl(sock, FIONBIO, &one) == -1) {
155 log_perror("ioctl(FIONBIO)");
160 // Do a non-blocking connect.
162 err = connect(sock, ai->ai_addr, ai->ai_addrlen);
163 } while (err == -1 && errno == EINTR);
165 if (err == -1 && errno != EINPROGRESS) {
166 log_perror("connect");
171 // Wait for the connect to complete, or an error to happen.
173 bool complete = wait_for_activity(sock, POLLIN | POLLOUT, NULL);
183 // Check whether it ended in an error or not.
184 socklen_t err_size = sizeof(err);
185 if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &err_size) == -1) {
186 log_perror("getsockopt");
194 // Successful connect.
195 freeaddrinfo(base_ai);
202 // Give the last one as error.
203 log(WARNING, "[%s] Connect to '%s' failed (%s)",
204 url.c_str(), host.c_str(), strerror(errno));
205 freeaddrinfo(base_ai);
209 bool HTTPInput::parse_response(const string &request)
211 vector<string> lines = split_lines(response);
213 log(WARNING, "[%s] Empty HTTP response from input.", url.c_str());
217 vector<string> first_line_tokens = split_tokens(lines[0]);
218 if (first_line_tokens.size() < 2) {
219 log(WARNING, "[%s] Malformed response line '%s' from input.",
220 url.c_str(), lines[0].c_str());
224 int response = atoi(first_line_tokens[1].c_str());
225 if (response != 200) {
226 log(WARNING, "[%s] Non-200 response '%s' from input.",
227 url.c_str(), lines[0].c_str());
231 multimap<string, string> parameters = extract_headers(lines, url);
233 // Remove “Content-encoding: metacube”.
234 // TODO: Make case-insensitive.
235 multimap<string, string>::iterator encoding_it =
236 parameters.find("Content-encoding");
237 if (encoding_it != parameters.end() && encoding_it->second == "metacube") {
238 parameters.erase(encoding_it);
241 // Change “Server: foo” to “Server: metacube/0.1 (reflecting: foo)”
242 // TODO: Make case-insensitive.
243 // XXX: Use a Via: instead?
244 if (parameters.count("Server") == 0) {
245 parameters.insert(make_pair("Server", SERVER_IDENTIFICATION));
247 for (multimap<string, string>::iterator it = parameters.begin();
248 it != parameters.end();
250 if (it->first != "Server") {
253 it->second = SERVER_IDENTIFICATION " (reflecting: " + it->second + ")";
257 // Set “Connection: close”.
258 // TODO: Make case-insensitive.
259 parameters.erase("Connection");
260 parameters.insert(make_pair("Connection", "close"));
262 // Construct the new HTTP header.
263 http_header = "HTTP/1.0 200 OK\r\n";
264 for (multimap<string, string>::iterator it = parameters.begin();
265 it != parameters.end();
267 http_header.append(it->first + ": " + it->second + "\r\n");
270 for (size_t i = 0; i < stream_indices.size(); ++i) {
271 servers->set_header(stream_indices[i], http_header, stream_header);
277 void HTTPInput::do_work()
279 timespec last_activity;
281 // TODO: Make the timeout persist across restarts.
282 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
283 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
287 while (!should_stop()) {
288 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
289 // Give the socket 30 seconds since last activity before we time out.
290 static const int timeout_secs = 30;
293 int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
296 timespec elapsed = clock_diff(last_activity, now);
297 if (elapsed.tv_sec >= timeout_secs) {
299 log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec);
300 state = CLOSING_SOCKET;
304 // Basically calculate (30 - (now - last_activity)) = (30 + (last_activity - now)).
305 // Add a second of slack to account for differences between clocks.
306 timespec timeout = clock_diff(now, last_activity);
307 timeout.tv_sec += timeout_secs + 1;
308 assert(timeout.tv_sec > 0 || (timeout.tv_sec >= 0 && timeout.tv_nsec > 0));
310 bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout);
312 err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
315 // OK. Most likely, should_stop was set, or we have timed out.
323 request_bytes_sent = 0;
325 pending_data.clear();
326 has_metacube_header = false;
327 for (size_t i = 0; i < stream_indices.size(); ++i) {
328 servers->set_header(stream_indices[i], "", "");
332 string protocol, user; // Thrown away.
333 if (!parse_url(url, &protocol, &user, &host, &port, &path)) {
334 log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str());
338 // Remove the brackets around IPv6 address literals.
339 // TODO: See if we can join this with the code in parse_ip_address(),
340 // or maybe even more it into parse_url().
341 if (!host.empty() && host[0] == '[' && host[host.size() - 1] == ']') {
342 host = host.substr(1, host.size() - 2);
346 sock = lookup_and_connect(host, port);
348 // Yay, successful connect. Try to set it as nonblocking.
350 if (ioctl(sock, FIONBIO, &one) == -1) {
351 log_perror("ioctl(FIONBIO)");
352 state = CLOSING_SOCKET;
354 state = SENDING_REQUEST;
355 request = "GET " + path + " HTTP/1.0\r\nHost: " + host_header(host, port) + "\r\nUser-Agent: cubemap\r\n\r\n";
356 request_bytes_sent = 0;
359 MutexLock lock(&stats_mutex);
360 stats.connect_time = time(NULL);
361 clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity);
364 case SENDING_REQUEST: {
365 size_t to_send = request.size() - request_bytes_sent;
369 ret = write(sock, request.data() + request_bytes_sent, to_send);
370 } while (ret == -1 && errno == EINTR);
374 state = CLOSING_SOCKET;
379 request_bytes_sent += ret;
381 if (request_bytes_sent == request.size()) {
382 state = RECEIVING_HEADER;
386 case RECEIVING_HEADER: {
391 ret = read(sock, buf, sizeof(buf));
392 } while (ret == -1 && errno == EINTR);
396 state = CLOSING_SOCKET;
401 // This really shouldn't happen...
402 log(ERROR, "[%s] Socket unexpectedly closed while reading header",
404 state = CLOSING_SOCKET;
408 RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
410 if (status == RP_OUT_OF_SPACE) {
411 log(WARNING, "[%s] Server sent overlong HTTP response!", url.c_str());
412 state = CLOSING_SOCKET;
414 } else if (status == RP_NOT_FINISHED_YET) {
418 // OK, so we're fine, but there might be some of the actual data after the response.
419 // We'll need to deal with that separately.
421 if (status == RP_EXTRA_DATA) {
422 char *ptr = static_cast<char *>(
423 memmem(response.data(), response.size(), "\r\n\r\n", 4));
425 extra_data = string(ptr + 4, &response[0] + response.size());
426 response.resize(ptr - response.data());
429 if (!parse_response(response)) {
430 state = CLOSING_SOCKET;
434 if (!extra_data.empty()) {
435 process_data(&extra_data[0], extra_data.size());
438 if (encoding == Input::INPUT_ENCODING_RAW) {
439 log(INFO, "[%s] Connected to '%s', receiving raw data.",
440 url.c_str(), url.c_str());
442 assert(encoding == Input::INPUT_ENCODING_METACUBE);
443 log(INFO, "[%s] Connected to '%s', receiving data.",
444 url.c_str(), url.c_str());
446 state = RECEIVING_DATA;
449 case RECEIVING_DATA: {
454 ret = read(sock, buf, sizeof(buf));
455 } while (ret == -1 && errno == EINTR);
459 state = CLOSING_SOCKET;
464 // This really shouldn't happen...
465 log(ERROR, "[%s] Socket unexpectedly closed while reading data",
467 state = CLOSING_SOCKET;
471 process_data(buf, ret);
474 case CLOSING_SOCKET: {
476 state = NOT_CONNECTED;
483 // If we are still in NOT_CONNECTED, either something went wrong,
484 // or the connection just got closed.
485 // The earlier steps have already given the error message, if any.
486 if (state == NOT_CONNECTED && !should_stop()) {
487 log(INFO, "[%s] Waiting 0.2 second and restarting...", url.c_str());
489 timeout_ts.tv_sec = 0;
490 timeout_ts.tv_nsec = 200000000;
491 wait_for_wakeup(&timeout_ts);
496 void HTTPInput::process_data(char *ptr, size_t bytes)
498 pending_data.insert(pending_data.end(), ptr, ptr + bytes);
500 MutexLock mutex(&stats_mutex);
501 stats.bytes_received += bytes;
504 if (encoding == Input::INPUT_ENCODING_RAW) {
505 for (size_t i = 0; i < stream_indices.size(); ++i) {
506 servers->add_data(stream_indices[i], ptr, bytes, /*metacube_flags=*/0);
511 assert(encoding == Input::INPUT_ENCODING_METACUBE);
514 // If we don't have enough data (yet) for even the Metacube header, just return.
515 if (pending_data.size() < sizeof(metacube2_block_header)) {
519 // Make sure we have the Metacube sync header at the start.
520 // We may need to skip over junk data (it _should_ not happen, though).
521 if (!has_metacube_header) {
522 char *ptr = static_cast<char *>(
523 memmem(pending_data.data(), pending_data.size(),
524 METACUBE2_SYNC, strlen(METACUBE2_SYNC)));
526 // OK, so we didn't find the sync marker. We know then that
527 // we do not have the _full_ marker in the buffer, but we
528 // could have N-1 bytes. Drop everything before that,
530 drop_pending_data(pending_data.size() - (strlen(METACUBE2_SYNC) - 1));
533 // Yay, we found the header. Drop everything (if anything) before it.
534 drop_pending_data(ptr - pending_data.data());
535 has_metacube_header = true;
537 // Re-check that we have the entire header; we could have dropped data.
538 if (pending_data.size() < sizeof(metacube2_block_header)) {
544 // Now it's safe to read the header.
545 metacube2_block_header hdr;
546 memcpy(&hdr, pending_data.data(), sizeof(hdr));
547 assert(memcmp(hdr.sync, METACUBE2_SYNC, sizeof(hdr.sync)) == 0);
548 uint32_t size = ntohl(hdr.size);
549 uint16_t flags = ntohs(hdr.flags);
550 uint16_t expected_csum = metacube2_compute_crc(&hdr);
552 if (expected_csum != ntohs(hdr.csum)) {
553 log(WARNING, "[%s] Metacube checksum failed (expected 0x%x, got 0x%x), "
554 "not reading block claiming to be %d bytes (flags=%x).",
555 url.c_str(), expected_csum, ntohs(hdr.csum),
558 // Drop only the first byte, and let the rest of the code handle resync.
559 pending_data.erase(pending_data.begin(), pending_data.begin() + 1);
560 has_metacube_header = false;
563 if (size > 10485760) {
564 log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header??",
565 url.c_str(), size, flags);
568 // See if we have the entire block. If not, wait for more data.
569 if (pending_data.size() < sizeof(metacube2_block_header) + size) {
573 // Send this block on to the servers.
575 MutexLock lock(&stats_mutex);
576 stats.data_bytes_received += size;
578 char *inner_data = pending_data.data() + sizeof(metacube2_block_header);
579 if (flags & METACUBE_FLAGS_HEADER) {
580 stream_header = string(inner_data, inner_data + size);
581 for (size_t i = 0; i < stream_indices.size(); ++i) {
582 servers->set_header(stream_indices[i], http_header, stream_header);
585 for (size_t i = 0; i < stream_indices.size(); ++i) {
586 servers->add_data(stream_indices[i], inner_data, size, flags);
589 // Consume the block. This isn't the most efficient way of dealing with things
590 // should we have many blocks, but these routines don't need to be too efficient
592 pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube2_block_header) + size);
593 has_metacube_header = false;
597 void HTTPInput::drop_pending_data(size_t num_bytes)
599 if (num_bytes == 0) {
602 log(WARNING, "[%s] Dropping %lld junk bytes; not a Metacube2 stream, or data was dropped from the middle of the stream.",
603 url.c_str(), (long long)num_bytes);
604 assert(pending_data.size() >= num_bytes);
605 pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
608 void HTTPInput::add_destination(int stream_index)
610 stream_indices.push_back(stream_index);
611 servers->set_header(stream_index, http_header, stream_header);
614 InputStats HTTPInput::get_stats() const
616 MutexLock lock(&stats_mutex);