Add URL parsing.
[cubemap] / input.cpp
1 #include <stdio.h>
2 #include <string.h>
3 #include <stdint.h>
4 #include <assert.h>
5 #include <arpa/inet.h>
6 #include <sys/socket.h>
7 #include <pthread.h>
8 #include <sys/types.h>
9 #include <sys/ioctl.h>
10 #include <sys/types.h>
11 #include <sys/socket.h>
12 #include <netdb.h>
13 #include <poll.h>
14 #include <errno.h>
15 #include <vector>
16 #include <string>
17 #include <map>
18
19 #include "metacube.h"
20 #include "mutexlock.h"
21 #include "input.h"
22 #include "server.h"
23 #include "serverpool.h"
24 #include "parse.h"
25
26 using namespace std;
27
28 extern ServerPool *servers;
29           
30 // Extremely rudimentary URL parsing.
31 bool parse_url(const string &url, string *host, string *port, string *path)
32 {
33         if (url.find("http://") != 0) {
34                 return false;
35         }
36         
37         string rest = url.substr(strlen("http://"));
38         size_t split = rest.find_first_of(":/");
39         if (split == string::npos) {
40                 // http://foo
41                 *host = rest;
42                 *port = "http";
43                 *path = "/";
44                 return true;
45         }
46
47         *host = string(rest.begin(), rest.begin() + split);
48         char ch = rest[split];  // Colon or slash.
49         rest = string(rest.begin() + split + 1, rest.end());
50
51         if (ch == ':') {
52                 // Parse the port.
53                 split = rest.find_first_of('/');
54                 if (split == string::npos) {
55                         // http://foo:1234
56                         *port = rest;
57                         *path = "/";
58                         return true;
59                 } else {
60                         // http://foo:1234/bar
61                         *port = string(rest.begin(), rest.begin() + split);
62                         *path = string(rest.begin() + split, rest.end());
63                         return true;
64                 }
65         }
66
67         // http://foo/bar
68         *port = "http";
69         *path = rest;
70         return true;
71 }
72
73 Input::Input(const string &stream_id, const string &url)
74         : state(NOT_CONNECTED),
75           stream_id(stream_id),
76           url(url),
77           has_metacube_header(false),
78           sock(-1)
79 {
80 }
81
82 void Input::run()
83 {
84         should_stop = false;
85         
86         // Joinable is already the default, but it's good to be certain.
87         pthread_attr_t attr;
88         pthread_attr_init(&attr);
89         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
90         pthread_create(&worker_thread, &attr, Input::do_work_thunk, this);
91 }
92
93 void Input::stop()
94 {
95         should_stop = true;
96
97         if (pthread_join(worker_thread, NULL) == -1) {
98                 perror("pthread_join");
99                 exit(1);
100         }
101 }
102
103 void *Input::do_work_thunk(void *arg)
104 {
105         Input *input = static_cast<Input *>(arg);
106         input->do_work();
107         return NULL;
108 }
109
110 int Input::lookup_and_connect(const string &host, const string &port)
111 {
112         addrinfo *ai;
113         int err = getaddrinfo(host.c_str(), port.c_str(), NULL, &ai);
114         if (err == -1) {
115                 fprintf(stderr, "WARNING: Lookup of '%s' failed (%s).\n",
116                         host.c_str(), gai_strerror(err));
117                 freeaddrinfo(ai);
118                 return -1;
119         }
120
121         // Connect to everything in turn until we have a socket.
122         while (ai && !should_stop) {
123                 int sock = socket(ai->ai_family, SOCK_STREAM, IPPROTO_TCP);
124                 if (sock == -1) {
125                         // Could be e.g. EPROTONOSUPPORT. The show must go on.
126                         continue;
127                 }
128
129                 do {
130                         err = connect(sock, ai->ai_addr, ai->ai_addrlen);
131                 } while (err == -1 && errno == EINTR);
132
133                 if (err != -1) {
134                         freeaddrinfo(ai);
135                         return sock;
136                 }
137
138                 ai = ai->ai_next;
139         }
140
141         // Give the last one as error.
142         fprintf(stderr, "WARNING: Connect to '%s' failed (%s)\n",
143                 host.c_str(), strerror(errno));
144         freeaddrinfo(ai);
145         return -1;
146 }
147
148 void Input::do_work()
149 {
150         while (!should_stop) {
151                 if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
152                         // Since we are non-blocking, we need to wait for the right state first.
153                         // Wait up to 50 ms, then check should_stop.
154                         pollfd pfd;
155                         pfd.fd = sock;
156                         pfd.events = (state == SENDING_REQUEST) ? POLLOUT : POLLIN;
157                         pfd.events |= POLLRDHUP;
158
159                         int nfds = poll(&pfd, 1, 50);
160                         if (nfds == 0 || (nfds == -1 && errno == EAGAIN)) {
161                                 continue;
162                         }
163                         if (nfds == -1) {
164                                 perror("poll");
165                                 state = CLOSING_SOCKET;
166                         }
167                 }
168
169                 switch (state) {
170                 case NOT_CONNECTED:
171                         request.clear();
172                         request_bytes_sent = 0;
173                         response.clear();
174         
175                         if (!parse_url(url, &host, &port, &path)) {
176                                 fprintf(stderr, "Failed to parse URL '%s'\n", url.c_str());
177                                 break;
178                         }
179
180                         sock = lookup_and_connect(host, port);
181                         if (sock != -1) {
182                                 // Yay, successful connect. Try to set it as nonblocking.
183                                 int one = 1;
184                                 if (ioctl(sock, FIONBIO, &one) == -1) {
185                                         perror("ioctl(FIONBIO)");
186                                         state = CLOSING_SOCKET;
187                                 } else {
188                                         state = SENDING_REQUEST;
189                                         request = "GET " + path + " HTTP/1.0\r\nUser-Agent: cubemap\r\n\r\n";
190                                         request_bytes_sent = 0;
191                                 }
192                         }
193                         break;
194                 case SENDING_REQUEST: {
195                         size_t to_send = request.size() - request_bytes_sent;
196                         int ret;
197
198                         do {
199                                 ret = write(sock, request.data() + request_bytes_sent, to_send);
200                         } while (ret == -1 && errno == EINTR);
201
202                         if (ret == -1) {
203                                 perror("write");
204                                 state = CLOSING_SOCKET;
205                                 continue;
206                         }
207
208                         assert(ret >= 0);
209                         request_bytes_sent += ret;
210
211                         if (request_bytes_sent == request.size()) {
212                                 state = RECEIVING_HEADER;
213                         }
214                         break;
215                 }
216                 case RECEIVING_HEADER: {
217                         char buf[4096];
218                         int ret;
219
220                         do {
221                                 ret = read(sock, buf, sizeof(buf));
222                         } while (ret == -1 && errno == EINTR);
223
224                         if (ret == -1) {
225                                 perror("read");
226                                 state = CLOSING_SOCKET;
227                                 continue;
228                         }
229
230                         if (ret == 0) {
231                                 // This really shouldn't happen...
232                                 fprintf(stderr, "Socket unexpectedly closed while reading header\n");
233                                 state = CLOSING_SOCKET;
234                                 continue;
235                         }
236                         
237                         RequestParseStatus status = wait_for_double_newline(&response, buf, ret);
238                         
239                         if (status == RP_OUT_OF_SPACE) {
240                                 fprintf(stderr, "WARNING: fd %d sent overlong response!\n", sock);
241                                 state = CLOSING_SOCKET;
242                                 continue;
243                         } else if (status == RP_NOT_FINISHED_YET) {
244                                 continue;
245                         }
246         
247                         // OK, so we're fine, but there might be some of the actual data after the response.
248                         // We'll need to deal with that separately.
249                         string extra_data;
250                         if (status == RP_EXTRA_DATA) {
251                                 char *ptr = static_cast<char *>(
252                                         memmem(response.data(), response.size(), "\r\n\r\n", 4));
253                                 assert(ptr != NULL);
254                                 extra_data = string(ptr, &response[0] + response.size());
255                                 response.resize(ptr - response.data());
256                         }
257
258                         // TODO: Check that the response is 200, save the headers, etc.
259
260                         if (!extra_data.empty()) {
261                                 process_data(&extra_data[0], extra_data.size());
262                         }
263
264                         state = RECEIVING_DATA;
265                         break;
266                 }
267                 case RECEIVING_DATA: {
268                         char buf[4096];
269                         int ret;
270
271                         do {
272                                 ret = read(sock, buf, sizeof(buf));
273                         } while (ret == -1 && errno == EINTR);
274
275                         if (ret == -1) {
276                                 perror("read");
277                                 state = CLOSING_SOCKET;
278                                 continue;
279                         }
280
281                         if (ret == 0) {
282                                 // This really shouldn't happen...
283                                 fprintf(stderr, "Socket unexpectedly closed while reading header\n");
284                                 state = CLOSING_SOCKET;
285                                 continue;
286                         }
287
288                         process_data(buf, ret);
289                         break;
290                 }
291                 case CLOSING_SOCKET: {
292                         int err;
293                         do {
294                                 err = close(sock);
295                         } while (err == -1 && errno == EINTR);
296
297                         if (err == -1) {
298                                 perror("close");
299                         }
300
301                         state = NOT_CONNECTED;
302                         break;
303                 }
304                 default:
305                         assert(false);
306                 }
307
308                 // If we are still in NOT_CONNECTED, either something went wrong,
309                 // or the connection just got closed.
310                 // The earlier steps have already given the error message, if any.
311                 if (state == NOT_CONNECTED && !should_stop) {
312                         fprintf(stderr, "Waiting 0.2 second and restarting...\n");
313                         usleep(200000);
314                 }
315         }
316 }
317
318 void Input::process_data(char *ptr, size_t bytes)
319 {
320         pending_data.insert(pending_data.end(), ptr, ptr + bytes);
321
322         for ( ;; ) {
323                 // If we don't have enough data (yet) for even the Metacube header, just return.
324                 if (pending_data.size() < sizeof(metacube_block_header)) {
325                         return;
326                 }
327
328                 // Make sure we have the Metacube sync header at the start.
329                 // We may need to skip over junk data (it _should_ not happen, though).
330                 if (!has_metacube_header) {
331                         char *ptr = static_cast<char *>(
332                                 memmem(pending_data.data(), pending_data.size(),
333                                        METACUBE_SYNC, strlen(METACUBE_SYNC)));
334                         if (ptr == NULL) {
335                                 // OK, so we didn't find the sync marker. We know then that
336                                 // we do not have the _full_ marker in the buffer, but we
337                                 // could have N-1 bytes. Drop everything before that,
338                                 // and then give up.
339                                 drop_pending_data(pending_data.size() - (strlen(METACUBE_SYNC) - 1));
340                                 return;
341                         } else {
342                                 // Yay, we found the header. Drop everything (if anything) before it.
343                                 drop_pending_data(ptr - pending_data.data());
344                                 has_metacube_header = true;
345
346                                 // Re-check that we have the entire header; we could have dropped data.
347                                 if (pending_data.size() < sizeof(metacube_block_header)) {
348                                         return;
349                                 }
350                         }
351                 }
352
353                 // Now it's safe to read the header.
354                 metacube_block_header *hdr = reinterpret_cast<metacube_block_header *>(pending_data.data());    
355                 assert(memcmp(hdr->sync, METACUBE_SYNC, sizeof(hdr->sync)) == 0);
356                 uint32_t size = ntohl(hdr->size);
357                 uint32_t flags = ntohl(hdr->flags);
358
359                 // See if we have the entire block. If not, wait for more data.
360                 if (pending_data.size() < sizeof(metacube_block_header) + size) {
361                         return;
362                 }
363
364                 // Send this block on to the data.
365                 char *inner_data = pending_data.data() + sizeof(metacube_block_header);
366                 if (flags & METACUBE_FLAGS_HEADER) {
367                         string header(inner_data, inner_data + size);
368                         servers->set_header(stream_id, header);
369                 } else { 
370                         servers->add_data(stream_id, inner_data, size);
371                 }
372
373                 // Consume the block. This isn't the most efficient way of dealing with things
374                 // should we have many blocks, but these routines don't need to be too efficient
375                 // anyway.
376                 pending_data.erase(pending_data.begin(), pending_data.begin() + sizeof(metacube_block_header) + size);
377                 has_metacube_header = false;
378         }
379 }
380
381 void Input::drop_pending_data(size_t num_bytes)
382 {
383         if (num_bytes == 0) {
384                 return;
385         }
386         fprintf(stderr, "Warning: Dropping %lld junk bytes from stream, maybe it is not a Metacube stream?\n",
387                 (long long)num_bytes);
388         pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
389 }
390