1 /*
2  * Copyright (c) 2021 Omar Polo <op@omarpolo.com>
3  *
4  * Permission to use, copy, modify, and distribute this software for any
5  * purpose with or without fee is hereby granted, provided that the above
6  * copyright notice and this permission notice appear in all copies.
7  *
8  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
9  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
10  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
11  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
12  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
13  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
14  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
15  */
16 
17 #include <sys/mman.h>
18 #include <sys/stat.h>
19 
20 #include <netdb.h>
21 
22 #include <assert.h>
23 #include <errno.h>
24 #include <fcntl.h>
25 #include <string.h>
26 
27 #include "gmid.h"
28 
29 int connected_clients;
30 
31 int
32 check_path(struct client *c, const char *path, int *fd)
33 {
34 	struct stat sb;
35 
36 	assert(path != NULL);
37 	if ((*fd = openat(c->host->dirfd, *path ? path : ".",
38 	    O_RDONLY | O_NOFOLLOW)) == -1) {
39 		return FILE_MISSING;
40 	}
41 
42 	if (fstat(*fd, &sb) == -1) {
43 		LOGN(c, "failed stat for %s: %s", path, strerror(errno));
44 		return FILE_MISSING;
45 	}
46 
47 	if (S_ISDIR(sb.st_mode))
48 		return FILE_DIRECTORY;
49 
50 	if (sb.st_mode & S_IXUSR)
51 		return FILE_EXECUTABLE;
52 
53 	return FILE_EXISTS;
54 }
55 
56 int
57 open_file(char *fpath, char *query, struct pollfd *fds, struct client *c)
58 {
59 	switch (check_path(c, fpath, &c->fd)) {
60 	case FILE_EXECUTABLE:
61 		if (c->host->cgi != NULL && starts_with(fpath, c->host->cgi))
62 			return start_cgi(fpath, "", query, fds, c);
63 
64 		/* fallthrough */
65 
66 	case FILE_EXISTS:
67 		if ((c->len = filesize(c->fd)) == -1) {
68 			LOGE(c, "failed to get file size for %s", fpath);
69 			goodbye(fds, c);
70 			return 0;
71 		}
72 
73 		if ((c->buf = mmap(NULL, c->len, PROT_READ, MAP_PRIVATE,
74 			    c->fd, 0)) == MAP_FAILED) {
75 			LOGW(c, "mmap: %s: %s", fpath, strerror(errno));
76 			goodbye(fds, c);
77 			return 0;
78 		}
79 		c->i = c->buf;
80 		return start_reply(fds, c, SUCCESS, mime(fpath));
81 
82 	case FILE_DIRECTORY:
83 		LOGD(c, "%s is a directory, trying %s/index.gmi", fpath, fpath);
84 		close(c->fd);
85 		c->fd = -1;
86 		send_dir(fpath, fds, c);
87 		return 0;
88 
89 	case FILE_MISSING:
90 		if (c->host->cgi != NULL && starts_with(fpath, c->host->cgi))
91 			return check_for_cgi(fpath, query, fds, c);
92 
93 		if (!start_reply(fds, c, NOT_FOUND, "not found"))
94 			return 0;
95 		goodbye(fds, c);
96 		return 0;
97 
98 	default:
99 		/* unreachable */
100 		abort();
101 	}
102 }
103 
104 
105 /*
106  * the inverse of this algorithm, i.e. starting from the start of the
107  * path + strlen(cgi), and checking if each component, should be
108  * faster.  But it's tedious to write.  This does the opposite: starts
109  * from the end and strip one component at a time, until either an
110  * executable is found or we emptied the path.
111  */
112 int
113 check_for_cgi(char *path, char *query, struct pollfd *fds, struct client *c)
114 {
115 	char *end;
116 	end = strchr(path, '\0');
117 
118 	/* NB: assume CGI is enabled and path matches cgi */
119 
120 	while (end > path) {
121 		/* go up one level.  UNIX paths are simple and POSIX
122 		 * dirname, with its ambiguities on if the given path
123 		 * is changed or not, gives me headaches. */
124 		while (*end != '/')
125 			end--;
126 		*end = '\0';
127 
128 		switch (check_path(c, path, &c->fd)) {
129 		case FILE_EXECUTABLE:
130 			return start_cgi(path, end+1, query, fds,c);
131 		case FILE_MISSING:
132 			break;
133 		default:
134 			goto err;
135 		}
136 
137 		*end = '/';
138 		end--;
139 	}
140 
141 err:
142 	if (!start_reply(fds, c, NOT_FOUND, "not found"))
143 		return 0;
144 	goodbye(fds, c);
145 	return 0;
146 }
147 
148 void
149 mark_nonblock(int fd)
150 {
151 	int flags;
152 
153 	if ((flags = fcntl(fd, F_GETFL)) == -1)
154 		fatal("fcntl(F_GETFL): %s", strerror(errno));
155 	if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1)
156 		fatal("fcntl(F_SETFL): %s", strerror(errno));
157 }
158 
159 void
160 handle_handshake(struct pollfd *fds, struct client *c)
161 {
162 	struct vhost *h;
163 	const char *servname;
164 
165 	switch (tls_handshake(c->ctx)) {
166 	case 0:  /* success */
167 	case -1: /* already handshaked */
168 		break;
169 	case TLS_WANT_POLLIN:
170 		fds->events = POLLIN;
171 		return;
172 	case TLS_WANT_POLLOUT:
173 		fds->events = POLLOUT;
174 		return;
175 	default:
176 		/* unreachable */
177 		abort();
178 	}
179 
180 	servname = tls_conn_servername(c->ctx);
181 	if (servname == NULL)
182 		goto hostnotfound;
183 
184 	for (h = hosts; h->domain != NULL; ++h) {
185 		if (!strcmp(h->domain, servname) || !strcmp(h->domain, "*"))
186 			break;
187 	}
188 
189 	if (h->domain != NULL) {
190 		c->state = S_OPEN;
191 		c->host = h;
192 		handle_open_conn(fds, c);
193 		return;
194 	}
195 
196 hostnotfound:
197 	/* XXX: check the correct response */
198 	if (!start_reply(fds, c, BAD_REQUEST, "Wrong host or missing SNI"))
199 		return;
200 	goodbye(fds, c);
201 }
202 
203 void
204 handle_open_conn(struct pollfd *fds, struct client *c)
205 {
206 	char buf[GEMINI_URL_LEN];
207 	const char *parse_err = "invalid request";
208 	struct iri iri;
209 
210 	bzero(buf, sizeof(buf));
211 
212 	switch (tls_read(c->ctx, buf, sizeof(buf)-1)) {
213 	case -1:
214 		LOGE(c, "tls_read: %s", tls_error(c->ctx));
215 		goodbye(fds, c);
216 		return;
217 
218 	case TLS_WANT_POLLIN:
219 		fds->events = POLLIN;
220 		return;
221 
222 	case TLS_WANT_POLLOUT:
223 		fds->events = POLLOUT;
224 		return;
225 	}
226 
227 	if (!trim_req_iri(buf) || !parse_iri(buf, &iri, &parse_err)) {
228 		if (!start_reply(fds, c, BAD_REQUEST, parse_err))
229 			return;
230 		goodbye(fds, c);
231 		return;
232 	}
233 
234 	if (strcmp(iri.schema, "gemini") || iri.port_no != conf.port) {
235 		if (!start_reply(fds, c, PROXY_REFUSED, "won't proxy request"))
236 			return;
237 		goodbye(fds, c);
238 		return;
239 	}
240 
241 	LOGI(c, "GET %s%s%s",
242 	    *iri.path ? iri.path : "/",
243 	    *iri.query ? "?" : "",
244 	    *iri.query ? iri.query : "");
245 
246 	send_file(iri.path, iri.query, fds, c);
247 }
248 
249 int
250 start_reply(struct pollfd *pfd, struct client *client, int code, const char *reason)
251 {
252 	char buf[1030]; 	/* status + ' ' + max reply len + \r\n\0 */
253 	int len;
254 
255 	client->code = code;
256 	client->meta = reason;
257 	client->state = S_INITIALIZING;
258 
259 	len = snprintf(buf, sizeof(buf), "%d %s\r\n", code, reason);
260 	assert(len < (int)sizeof(buf));
261 
262 	switch (tls_write(client->ctx, buf, len)) {
263 	case TLS_WANT_POLLIN:
264 		pfd->events = POLLIN;
265 		return 0;
266 	case TLS_WANT_POLLOUT:
267 		pfd->events = POLLOUT;
268 		return 0;
269 	default:
270 		return 1;
271 	}
272 }
273 
274 int
275 start_cgi(const char *spath, const char *relpath, const char *query,
276     struct pollfd *fds, struct client *c)
277 {
278 	char addr[NI_MAXHOST];
279 	const char *ruser, *cissuer, *chash;
280 	int e;
281 
282 	e = getnameinfo((struct sockaddr*)&c->addr, sizeof(c->addr),
283 	    addr, sizeof(addr),
284 	    NULL, 0,
285 	    NI_NUMERICHOST);
286 	if (e != 0)
287 		goto err;
288 
289 	if (tls_peer_cert_provided(c->ctx)) {
290 		ruser = tls_peer_cert_subject(c->ctx);
291 		cissuer = tls_peer_cert_issuer(c->ctx);
292 		chash = tls_peer_cert_hash(c->ctx);
293 	} else {
294 		ruser = NULL;
295 		cissuer = NULL;
296 		chash = NULL;
297 	}
298 
299 	if (!send_string(exfd, spath)
300 	    || !send_string(exfd, relpath)
301 	    || !send_string(exfd, query)
302 	    || !send_string(exfd, addr)
303 	    || !send_string(exfd, ruser)
304 	    || !send_string(exfd, cissuer)
305 	    || !send_string(exfd, chash)
306 	    || !send_vhost(exfd, c->host))
307 		goto err;
308 
309 	close(c->fd);
310 	if ((c->fd = recv_fd(exfd)) == -1) {
311 		if (!start_reply(fds, c, TEMP_FAILURE, "internal server error"))
312 			return 0;
313 		goodbye(fds, c);
314 		return 0;
315 	}
316 	c->child = 1;
317 	c->state = S_SENDING;
318 	cgi_poll_on_child(fds, c);
319 	/* handle_cgi(fds, c); */
320 	return 0;
321 
322 err:
323 	/* fatal("cannot talk to the executor process: %s", strerror(errno)); */
324 	fatal("cannot talk to the executor process");
325 }
326 
327 void
328 send_file(char *path, char *query, struct pollfd *fds, struct client *c)
329 {
330 	ssize_t ret, len;
331 
332 	if (c->fd == -1) {
333 		if (!open_file(path, query, fds, c))
334 			return;
335 		c->state = S_SENDING;
336 	}
337 
338 	len = (c->buf + c->len) - c->i;
339 
340 	while (len > 0) {
341 		switch (ret = tls_write(c->ctx, c->i, len)) {
342 		case -1:
343 			LOGE(c, "tls_write: %s", tls_error(c->ctx));
344 			goodbye(fds, c);
345 			return;
346 
347 		case TLS_WANT_POLLIN:
348 			fds->events = POLLIN;
349 			return;
350 
351 		case TLS_WANT_POLLOUT:
352 			fds->events = POLLOUT;
353 			return;
354 
355 		default:
356 			c->i += ret;
357 			len -= ret;
358 			break;
359 		}
360 	}
361 
362 	goodbye(fds, c);
363 }
364 
365 void
366 send_dir(char *path, struct pollfd *fds, struct client *client)
367 {
368 	char fpath[PATHBUF];
369 	size_t len;
370 
371 	bzero(fpath, PATHBUF);
372 
373 	if (path[0] != '.')
374 		fpath[0] = '.';
375 
376 	/* this cannot fail since sizeof(fpath) > maxlen of path */
377 	strlcat(fpath, path, PATHBUF);
378 	len = strlen(fpath);
379 
380 	/* add a trailing / in case. */
381 	if (fpath[len-1] != '/') {
382 		fpath[len] = '/';
383 	}
384 
385 	strlcat(fpath, "index.gmi", sizeof(fpath));
386 
387 	send_file(fpath, NULL, fds, client);
388 }
389 
390 void
391 cgi_poll_on_child(struct pollfd *fds, struct client *c)
392 {
393 	int fd;
394 
395 	if (c->waiting_on_child)
396 		return;
397 	c->waiting_on_child = 1;
398 
399 	fds->events = POLLIN;
400 
401 	fd = fds->fd;
402 	fds->fd = c->fd;
403 	c->fd = fd;
404 }
405 
406 void
407 cgi_poll_on_client(struct pollfd *fds, struct client *c)
408 {
409 	int fd;
410 
411 	if (!c->waiting_on_child)
412 		return;
413 	c->waiting_on_child = 0;
414 
415 	fd = fds->fd;
416 	fds->fd = c->fd;
417 	c->fd = fd;
418 }
419 
420 void
421 handle_cgi(struct pollfd *fds, struct client *c)
422 {
423 	ssize_t r;
424 
425 	/* ensure c->fd is the child and fds->fd the client */
426 	cgi_poll_on_client(fds, c);
427 
428 	while (1) {
429 		if (c->len == 0) {
430 			if ((r = read(c->fd, c->sbuf, sizeof(c->sbuf))) == 0)
431 				goto end;
432 			if (r == -1) {
433 				if (errno == EAGAIN || errno == EWOULDBLOCK) {
434 					cgi_poll_on_child(fds, c);
435 					return;
436 				}
437                                 goto end;
438 			}
439 			c->len = r;
440 			c->off = 0;
441 		}
442 
443 		while (c->len > 0) {
444 			switch (r = tls_write(c->ctx, c->sbuf + c->off, c->len)) {
445 			case -1:
446 				goto end;
447 
448 			case TLS_WANT_POLLOUT:
449 				fds->events = POLLOUT;
450 				return;
451 
452 			case TLS_WANT_POLLIN:
453 				fds->events = POLLIN;
454 				return;
455 
456 			default:
457                                 c->off += r;
458 				c->len -= r;
459 				break;
460 			}
461 		}
462 	}
463 
464 end:
465 	goodbye(fds, c);
466 }
467 
468 void
469 goodbye(struct pollfd *pfd, struct client *c)
470 {
471 	c->state = S_CLOSING;
472 
473 	switch (tls_close(c->ctx)) {
474 	case TLS_WANT_POLLIN:
475 		pfd->events = POLLIN;
476 		return;
477 	case TLS_WANT_POLLOUT:
478 		pfd->events = POLLOUT;
479 		return;
480 	}
481 
482 	connected_clients--;
483 
484 	tls_free(c->ctx);
485 	c->ctx = NULL;
486 
487 	if (c->buf != MAP_FAILED)
488 		munmap(c->buf, c->len);
489 
490 	if (c->fd != -1)
491 		close(c->fd);
492 
493 	close(pfd->fd);
494 	pfd->fd = -1;
495 }
496 
497 void
498 do_accept(int sock, struct tls *ctx, struct pollfd *fds, struct client *clients)
499 {
500 	int i, fd;
501 	struct sockaddr_storage addr;
502 	socklen_t len;
503 
504 	len = sizeof(addr);
505 	if ((fd = accept(sock, (struct sockaddr*)&addr, &len)) == -1) {
506 		if (errno == EWOULDBLOCK)
507 			return;
508 		fatal("accept: %s", strerror(errno));
509 	}
510 
511 	mark_nonblock(fd);
512 
513 	for (i = 0; i < MAX_USERS; ++i) {
514 		if (fds[i].fd == -1) {
515 			bzero(&clients[i], sizeof(struct client));
516 			if (tls_accept_socket(ctx, &clients[i].ctx, fd) == -1)
517 				break; /* goodbye fd! */
518 
519 			fds[i].fd = fd;
520 			fds[i].events = POLLIN;
521 
522 			clients[i].state = S_HANDSHAKE;
523 			clients[i].fd = -1;
524 			clients[i].child = 0;
525 			clients[i].waiting_on_child = 0;
526 			clients[i].buf = MAP_FAILED;
527 			clients[i].af = AF_INET;
528 			clients[i].addr = addr;
529 
530 			connected_clients++;
531 			return;
532 		}
533 	}
534 
535 	close(fd);
536 }
537 
538 void
539 handle(struct pollfd *fds, struct client *client)
540 {
541 	switch (client->state) {
542 	case S_HANDSHAKE:
543 		handle_handshake(fds, client);
544 		break;
545 
546 	case S_OPEN:
547                 handle_open_conn(fds, client);
548 		break;
549 
550 	case S_INITIALIZING:
551 		if (!start_reply(fds, client, client->code, client->meta))
552 			return;
553 
554 		if (client->code != SUCCESS) {
555 			/* we don't need a body */
556 			goodbye(fds, client);
557 			return;
558 		}
559 
560 		client->state = S_SENDING;
561 
562 		/* fallthrough */
563 
564 	case S_SENDING:
565 		if (client->child)
566 			handle_cgi(fds, client);
567 		else
568 			send_file(NULL, NULL, fds, client);
569 		break;
570 
571 	case S_CLOSING:
572 		goodbye(fds, client);
573 		break;
574 
575 	default:
576 		/* unreachable */
577 		abort();
578 	}
579 }
580 
581 void
582 loop(struct tls *ctx, int sock4, int sock6)
583 {
584 	int i;
585 	struct client clients[MAX_USERS];
586 	struct pollfd fds[MAX_USERS];
587 
588 	connected_clients = 0;
589 
590 	for (i = 0; i < MAX_USERS; ++i) {
591 		fds[i].fd = -1;
592 		fds[i].events = POLLIN;
593 		bzero(&clients[i], sizeof(struct client));
594 	}
595 
596 	fds[0].fd = sock4;
597 	fds[1].fd = sock6;
598 
599 	for (;;) {
600 		if (poll(fds, MAX_USERS, INFTIM) == -1) {
601 			if (errno == EINTR) {
602                                 fprintf(stderr, "connected clients: %d\n",
603 				    connected_clients);
604 				continue;
605 			}
606 			fatal("poll: %s", strerror(errno));
607 		}
608 
609 		for (i = 0; i < MAX_USERS; i++) {
610 			if (fds[i].revents == 0)
611 				continue;
612 
613 			if (fds[i].revents & (POLLERR|POLLNVAL))
614 				fatal("bad fd %d: %s", fds[i].fd,
615 				    strerror(errno));
616 
617 			if (fds[i].revents & POLLHUP) {
618 				/* fds[i] may be the fd of the stdin
619 				 * of a cgi script that has exited. */
620 				if (!clients[i].waiting_on_child) {
621 					goodbye(&fds[i], &clients[i]);
622 					continue;
623 				}
624 			}
625 
626 			if (fds[i].fd == sock4)
627 				do_accept(sock4, ctx, fds, clients);
628 			else if (fds[i].fd == sock6)
629 				do_accept(sock6, ctx, fds, clients);
630 			else
631 				handle(&fds[i], &clients[i]);
632 		}
633 	}
634 }