Support joining multicast addresses (both ASM and SSM).
[cubemap] / httpinput.cpp
1 #include <assert.h>
2 #include <errno.h>
3 #include <netdb.h>
4 #include <netinet/in.h>
5 #include <poll.h>
6 #include <stdint.h>
7 #include <stdlib.h>
8 #include <string.h>
9 #include <sys/ioctl.h>
10 #include <sys/socket.h>
11 #include <sys/time.h>
12 #include <time.h>
13 #include <unistd.h>
14 #include <map>
15 #include <string>
16 #include <utility>
17 #include <vector>
18
19 #include "httpinput.h"
20 #include "log.h"
21 #include "metacube2.h"
22 #include "mutexlock.h"
23 #include "parse.h"
24 #include "serverpool.h"
25 #include "state.pb.h"
26 #include "stream.h"
27 #include "util.h"
28 #include "version.h"
29
30 using namespace std;
31
32 extern ServerPool *servers;
33
34 namespace {
35
36 // Compute b-a.
37 timespec clock_diff(const timespec &a, const timespec &b)
38 {
39         timespec ret;
40         ret.tv_sec = b.tv_sec - a.tv_sec;
41         ret.tv_nsec = b.tv_nsec - a.tv_nsec;
42         if (ret.tv_nsec < 0) {
43                 ret.tv_sec--;
44                 ret.tv_nsec += 1000000000;
45         }
46         assert(ret.tv_nsec >= 0);
47         return ret;
48 }
49
50 }  // namespace
51
52 HTTPInput::HTTPInput(const string &url)
53         : state(NOT_CONNECTED),
54           url(url),
55           has_metacube_header(false),
56           sock(-1)
57 {
58         pthread_mutex_init(&stats_mutex, NULL);
59         stats.url = url;
60         stats.bytes_received = 0;
61         stats.data_bytes_received = 0;
62         stats.connect_time = -1;
63 }
64
65 HTTPInput::HTTPInput(const InputProto &serialized)
66         : state(State(serialized.state())),
67           url(serialized.url()),
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())
75 {
76         pending_data.resize(serialized.pending_data().size());
77         memcpy(&pending_data[0], serialized.pending_data().data(), serialized.pending_data().size());
78
79         string protocol, user;
80         parse_url(url, &protocol, &user, &host, &port, &path);  // Don't care if it fails.
81
82         pthread_mutex_init(&stats_mutex, NULL);
83         stats.url = url;
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();
88         } else {
89                 stats.connect_time = time(NULL);
90         }
91 }
92
93 void HTTPInput::close_socket()
94 {
95         if (sock != -1) {
96                 safe_close(sock);
97                 sock = -1;
98         }
99
100         MutexLock lock(&stats_mutex);
101         stats.connect_time = -1;
102 }
103
104 InputProto HTTPInput::serialize() const
105 {
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         return serialized;
121 }
122
123 int HTTPInput::lookup_and_connect(const string &host, const string &port)
124 {
125         addrinfo *ai;
126         int err = getaddrinfo(host.c_str(), port.c_str(), NULL, &ai);
127         if (err != 0) {
128                 log(WARNING, "[%s] Lookup of '%s' failed (%s).",
129                         url.c_str(), host.c_str(), gai_strerror(err));
130                 return -1;
131         }
132
133         addrinfo *base_ai = ai;
134
135         // Connect to everything in turn until we have a socket.
136         for ( ; ai && !should_stop(); ai = ai->ai_next) {
137                 int sock = socket(ai->ai_family, SOCK_STREAM, IPPROTO_TCP);
138                 if (sock == -1) {
139                         // Could be e.g. EPROTONOSUPPORT. The show must go on.
140                         continue;
141                 }
142
143                 // Now do a non-blocking connect. This is important because we want to be able to be
144                 // woken up, even though it's rather cumbersome.
145
146                 // Set the socket as nonblocking.
147                 int one = 1;
148                 if (ioctl(sock, FIONBIO, &one) == -1) {
149                         log_perror("ioctl(FIONBIO)");
150                         safe_close(sock);
151                         return -1;                      
152                 }
153
154                 // Do a non-blocking connect.
155                 do {
156                         err = connect(sock, ai->ai_addr, ai->ai_addrlen);
157                 } while (err == -1 && errno == EINTR);
158
159                 if (err == -1 && errno != EINPROGRESS) {
160                         log_perror("connect");
161                         safe_close(sock);
162                         continue;
163                 }
164
165                 // Wait for the connect to complete, or an error to happen.
166                 for ( ;; ) {
167                         bool complete = wait_for_activity(sock, POLLIN | POLLOUT, NULL);
168                         if (should_stop()) {
169                                 safe_close(sock);
170                                 return -1;
171                         }
172                         if (complete) {
173                                 break;
174                         }
175                 }
176
177                 // Check whether it ended in an error or not.
178                 socklen_t err_size = sizeof(err);
179                 if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &err, &err_size) == -1) {
180                         log_perror("getsockopt");
181                         safe_close(sock);
182                         continue;
183                 }
184
185                 errno = err;
186
187                 if (err == 0) {
188                         // Successful connect.
189                         freeaddrinfo(base_ai);
190                         return sock;
191                 }
192
193                 safe_close(sock);
194         }
195
196         // Give the last one as error.
197         log(WARNING, "[%s] Connect to '%s' failed (%s)",
198                 url.c_str(), host.c_str(), strerror(errno));
199         freeaddrinfo(base_ai);
200         return -1;
201 }
202         
203 bool HTTPInput::parse_response(const std::string &request)
204 {
205         vector<string> lines = split_lines(response);
206         if (lines.empty()) {
207                 log(WARNING, "[%s] Empty HTTP response from input.", url.c_str());
208                 return false;
209         }
210
211         vector<string> first_line_tokens = split_tokens(lines[0]);
212         if (first_line_tokens.size() < 2) {
213                 log(WARNING, "[%s] Malformed response line '%s' from input.",
214                         url.c_str(), lines[0].c_str());
215                 return false;
216         }
217
218         int response = atoi(first_line_tokens[1].c_str());
219         if (response != 200) {
220                 log(WARNING, "[%s] Non-200 response '%s' from input.",
221                         url.c_str(), lines[0].c_str());
222                 return false;
223         }
224
225         multimap<string, string> parameters;
226         for (size_t i = 1; i < lines.size(); ++i) {
227                 size_t split = lines[i].find(":");
228                 if (split == string::npos) {
229                         log(WARNING, "[%s] Ignoring malformed HTTP response line '%s'",
230                                 url.c_str(), lines[i].c_str());
231                         continue;
232                 }
233
234                 string key(lines[i].begin(), lines[i].begin() + split);
235
236                 // Skip any spaces after the colon.
237                 do {
238                         ++split;
239                 } while (split < lines[i].size() && lines[i][split] == ' ');
240
241                 string value(lines[i].begin() + split, lines[i].end());
242
243                 // Remove “Content-encoding: metacube”.
244                 // TODO: Make case-insensitive.
245                 if (key == "Content-encoding" && value == "metacube") {
246                         continue;
247                 }
248
249                 parameters.insert(make_pair(key, value));
250         }
251
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));
257         } else {
258                 for (multimap<string, string>::iterator it = parameters.begin();
259                      it != parameters.end();
260                      ++it) {
261                         if (it->first != "Server") {
262                                 continue;
263                         }
264                         it->second = SERVER_IDENTIFICATION " (reflecting: " + it->second + ")";
265                 }
266         }
267
268         // Set “Connection: close”.
269         // TODO: Make case-insensitive.
270         parameters.erase("Connection");
271         parameters.insert(make_pair("Connection", "close"));
272
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();
277              ++it) {
278                 http_header.append(it->first + ": " + it->second + "\r\n");
279         }
280
281         for (size_t i = 0; i < stream_indices.size(); ++i) {
282                 servers->set_header(stream_indices[i], http_header, stream_header);
283         }
284
285         return true;
286 }
287
288 void HTTPInput::do_work()
289 {
290         timespec last_activity;
291
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, &last_activity);
295                 assert(err != -1);
296         }
297
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;
302
303                         timespec now;
304                         int err = clock_gettime(CLOCK_MONOTONIC, &now);
305                         assert(err != -1);
306
307                         timespec elapsed = clock_diff(last_activity, now);
308                         if (elapsed.tv_sec >= timeout_secs) {
309                                 // Timeout!
310                                 log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec);
311                                 state = CLOSING_SOCKET;
312                                 continue;
313                         }
314
315                         // Basically calculate (30 - (now - last_activity)) = (30 + (last_activity - now)).
316                         // Add a second of slack to account for differences between clocks.
317                         timespec timeout = clock_diff(now, last_activity);
318                         timeout.tv_sec += timeout_secs + 1;
319                         assert(timeout.tv_sec > 0 || (timeout.tv_sec >= 0 && timeout.tv_nsec > 0));
320
321                         bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout);
322                         if (activity) {
323                                 err = clock_gettime(CLOCK_MONOTONIC, &last_activity);
324                                 assert(err != -1);
325                         } else {
326                                 // OK. Most likely, should_stop was set, or we have timed out.
327                                 continue;
328                         }
329                 }
330
331                 switch (state) {
332                 case NOT_CONNECTED:
333                         request.clear();
334                         request_bytes_sent = 0;
335                         response.clear();
336                         pending_data.clear();
337                         has_metacube_header = false;
338                         for (size_t i = 0; i < stream_indices.size(); ++i) {
339                                 servers->set_header(stream_indices[i], "", "");
340                         }
341
342                         {
343                                 string protocol, user;  // Thrown away.
344                                 if (!parse_url(url, &protocol, &user, &host, &port, &path)) {
345                                         log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str());
346                                         break;
347                                 }
348                         }
349
350                         sock = lookup_and_connect(host, port);
351                         if (sock != -1) {
352                                 // Yay, successful connect. Try to set it as nonblocking.
353                                 int one = 1;
354                                 if (ioctl(sock, FIONBIO, &one) == -1) {
355                                         log_perror("ioctl(FIONBIO)");
356                                         state = CLOSING_SOCKET;
357                                 } else {
358                                         state = SENDING_REQUEST;
359                                         request = "GET " + path + " HTTP/1.0\r\nUser-Agent: cubemap\r\n\r\n";
360                                         request_bytes_sent = 0;
361                                 }
362
363                                 MutexLock lock(&stats_mutex);
364                                 stats.connect_time = time(NULL);
365                                 clock_gettime(CLOCK_MONOTONIC, &last_activity);
366                         }
367                         break;
368                 case SENDING_REQUEST: {
369                         size_t to_send = request.size() - request_bytes_sent;
370                         int ret;
371
372                         do {
373                                 ret = write(sock, request.data() + request_bytes_sent, to_send);
374                         } while (ret == -1 && errno == EINTR);
375
376                         if (ret == -1) {
377                                 log_perror("write");
378                                 state = CLOSING_SOCKET;
379                                 continue;
380                         }
381
382                         assert(ret >= 0);
383                         request_bytes_sent += ret;
384
385                         if (request_bytes_sent == request.size()) {
386                                 state = RECEIVING_HEADER;
387                         }
388                         break;
389                 }
390                 case RECEIVING_HEADER: {
391                         char buf[4096];
392                         int ret;
393
394                         do {
395                                 ret = read(sock, buf, sizeof(buf));
396                         } while (ret == -1 && errno == EINTR);
397
398                         if (ret == -1) {
399                                 log_perror("read");
400                                 state = CLOSING_SOCKET;
401                                 continue;
402                         }
403
404                         if (ret == 0) {
405                                 // This really shouldn't happen...
406                                 log(ERROR, "[%s] Socket unexpectedly closed while reading header",
407                                            url.c_str());
408                                 state = CLOSING_SOCKET;
409                                 continue;
410                         }
411                         
412                         RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
413                         
414                         if (status == RP_OUT_OF_SPACE) {
415                                 log(WARNING, "[%s] Sever sent overlong HTTP response!", url.c_str());
416                                 state = CLOSING_SOCKET;
417                                 continue;
418                         } else if (status == RP_NOT_FINISHED_YET) {
419                                 continue;
420                         }
421         
422                         // OK, so we're fine, but there might be some of the actual data after the response.
423                         // We'll need to deal with that separately.
424                         string extra_data;
425                         if (status == RP_EXTRA_DATA) {
426                                 char *ptr = static_cast<char *>(
427                                         memmem(response.data(), response.size(), "\r\n\r\n", 4));
428                                 assert(ptr != NULL);
429                                 extra_data = string(ptr + 4, &response[0] + response.size());
430                                 response.resize(ptr - response.data());
431                         }
432
433                         if (!parse_response(response)) {
434                                 state = CLOSING_SOCKET;
435                                 continue;
436                         }
437
438                         if (!extra_data.empty()) {
439                                 process_data(&extra_data[0], extra_data.size());
440                         }
441
442                         log(INFO, "[%s] Connected to '%s', receiving data.",
443                                    url.c_str(), url.c_str());
444                         state = RECEIVING_DATA;
445                         break;
446                 }
447                 case RECEIVING_DATA: {
448                         char buf[4096];
449                         int ret;
450
451                         do {
452                                 ret = read(sock, buf, sizeof(buf));
453                         } while (ret == -1 && errno == EINTR);
454
455                         if (ret == -1) {
456                                 log_perror("read");
457                                 state = CLOSING_SOCKET;
458                                 continue;
459                         }
460
461                         if (ret == 0) {
462                                 // This really shouldn't happen...
463                                 log(ERROR, "[%s] Socket unexpectedly closed while reading data",
464                                            url.c_str());
465                                 state = CLOSING_SOCKET;
466                                 continue;
467                         }
468
469                         process_data(buf, ret);
470                         break;
471                 }
472                 case CLOSING_SOCKET: {
473                         close_socket();
474                         state = NOT_CONNECTED;
475                         break;
476                 }
477                 default:
478                         assert(false);
479                 }
480
481                 // If we are still in NOT_CONNECTED, either something went wrong,
482                 // or the connection just got closed.
483                 // The earlier steps have already given the error message, if any.
484                 if (state == NOT_CONNECTED && !should_stop()) {
485                         log(INFO, "[%s] Waiting 0.2 second and restarting...", url.c_str());
486                         timespec timeout_ts;
487                         timeout_ts.tv_sec = 0;
488                         timeout_ts.tv_nsec = 200000000;
489                         wait_for_wakeup(&timeout_ts);
490                 }
491         }
492 }
493
494 void HTTPInput::process_data(char *ptr, size_t bytes)
495 {
496         pending_data.insert(pending_data.end(), ptr, ptr + bytes);
497         {
498                 MutexLock mutex(&stats_mutex);
499                 stats.bytes_received += bytes;
500         }
501
502         for ( ;; ) {
503                 // If we don't have enough data (yet) for even the Metacube header, just return.
504                 if (pending_data.size() < sizeof(metacube2_block_header)) {
505                         return;
506                 }
507
508                 // Make sure we have the Metacube sync header at the start.
509                 // We may need to skip over junk data (it _should_ not happen, though).
510                 if (!has_metacube_header) {
511                         char *ptr = static_cast<char *>(
512                                 memmem(pending_data.data(), pending_data.size(),
513                                        METACUBE2_SYNC, strlen(METACUBE2_SYNC)));
514                         if (ptr == NULL) {
515                                 // OK, so we didn't find the sync marker. We know then that
516                                 // we do not have the _full_ marker in the buffer, but we
517                                 // could have N-1 bytes. Drop everything before that,
518                                 // and then give up.
519                                 drop_pending_data(pending_data.size() - (strlen(METACUBE2_SYNC) - 1));
520                                 return;
521                         } else {
522                                 // Yay, we found the header. Drop everything (if anything) before it.
523                                 drop_pending_data(ptr - pending_data.data());
524                                 has_metacube_header = true;
525
526                                 // Re-check that we have the entire header; we could have dropped data.
527                                 if (pending_data.size() < sizeof(metacube2_block_header)) {
528                                         return;
529                                 }
530                         }
531                 }
532
533                 // Now it's safe to read the header.
534                 metacube2_block_header hdr;
535                 memcpy(&hdr, pending_data.data(), sizeof(hdr));
536                 assert(memcmp(hdr.sync, METACUBE2_SYNC, sizeof(hdr.sync)) == 0);
537                 uint32_t size = ntohl(hdr.size);
538                 uint16_t flags = ntohs(hdr.flags);
539                 uint16_t expected_csum = metacube2_compute_crc(&hdr);
540
541                 if (expected_csum != ntohs(hdr.csum)) {
542                         log(WARNING, "[%s] Metacube checksum failed (expected 0x%x, got 0x%x), "
543                                 "not reading block claiming to be %d bytes (flags=%x).",
544                                 url.c_str(), expected_csum, ntohs(hdr.csum),
545                                 size, flags);
546
547                         // Drop only the first byte, and let the rest of the code handle resync.
548                         pending_data.erase(pending_data.begin(), pending_data.begin() + 1);
549                         has_metacube_header = false;
550                         continue;
551                 }
552                 if (size > 262144) {
553                         log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header?",
554                                 url.c_str(), size, flags);
555                 }
556
557                 // See if we have the entire block. If not, wait for more data.
558                 if (pending_data.size() < sizeof(metacube2_block_header) + size) {
559                         return;
560                 }
561
562                 // Send this block on to the servers.
563                 {
564                         MutexLock lock(&stats_mutex);
565                         stats.data_bytes_received += size;
566                 }
567                 char *inner_data = pending_data.data() + sizeof(metacube2_block_header);
568                 if (flags & METACUBE_FLAGS_HEADER) {
569                         stream_header = string(inner_data, inner_data + size);
570                         for (size_t i = 0; i < stream_indices.size(); ++i) {
571                                 servers->set_header(stream_indices[i], http_header, stream_header);
572                         }
573                 } else {
574                         StreamStartSuitability suitable_for_stream_start;
575                         if (flags & METACUBE_FLAGS_NOT_SUITABLE_FOR_STREAM_START) {
576                                 suitable_for_stream_start = NOT_SUITABLE_FOR_STREAM_START;
577                         } else {
578                                 suitable_for_stream_start = SUITABLE_FOR_STREAM_START;
579                         }
580                         for (size_t i = 0; i < stream_indices.size(); ++i) {
581                                 servers->add_data(stream_indices[i], inner_data, size, suitable_for_stream_start);
582                         }
583                 }
584
585                 // Consume the block. This isn't the most efficient way of dealing with things
586                 // should we have many blocks, but these routines don't need to be too efficient
587                 // anyway.
588                 pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube2_block_header) + size);
589                 has_metacube_header = false;
590         }
591 }
592
593 void HTTPInput::drop_pending_data(size_t num_bytes)
594 {
595         if (num_bytes == 0) {
596                 return;
597         }
598         log(WARNING, "[%s] Dropping %lld junk bytes from stream, maybe it is not a Metacube2 stream?",
599                 url.c_str(), (long long)num_bytes);
600         assert(pending_data.size() >= num_bytes);
601         pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
602 }
603
604 void HTTPInput::add_destination(int stream_index)
605 {
606         stream_indices.push_back(stream_index);
607         servers->set_header(stream_index, http_header, stream_header);
608 }
609
610 InputStats HTTPInput::get_stats() const
611 {
612         MutexLock lock(&stats_mutex);
613         return stats;
614 }