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 <fnmatch.h>
26 #include <string.h>
27 
28 #include "gmid.h"
29 
30 int connected_clients;
31 
32 int
33 check_path(struct client *c, const char *path, int *fd)
34 {
35 	struct stat sb;
36 	const char *p;
37 
38 	assert(path != NULL);
39 
40 	if (*path == '\0')
41 		p = ".";
42 	else if (*path == '/')
43 		/* in send_dir we add an initial / (to be
44 		 * redirect-friendly), but here we want to skip it */
45 		p = path+1;
46 	else
47 		p = path;
48 
49 	if ((*fd = openat(c->host->dirfd, p, O_RDONLY | O_NOFOLLOW)) == -1) {
50 		return FILE_MISSING;
51 	}
52 
53 	if (fstat(*fd, &sb) == -1) {
54 		LOGN(c, "failed stat for %s: %s", path, strerror(errno));
55 		return FILE_MISSING;
56 	}
57 
58 	if (S_ISDIR(sb.st_mode))
59 		return FILE_DIRECTORY;
60 
61 	if (sb.st_mode & S_IXUSR)
62 		return FILE_EXECUTABLE;
63 
64 	return FILE_EXISTS;
65 }
66 
67 void
68 open_file(struct pollfd *fds, struct client *c)
69 {
70 	switch (check_path(c, c->iri.path, &c->fd)) {
71 	case FILE_EXECUTABLE:
72 		if (c->host->cgi != NULL && starts_with(c->iri.path, c->host->cgi)) {
73 			start_cgi(c->iri.path, "", c->iri.query, fds, c);
74 			return;
75 		}
76 
77 		/* fallthrough */
78 
79 	case FILE_EXISTS:
80 		if ((c->len = filesize(c->fd)) == -1) {
81 			LOGE(c, "failed to get file size for %s", c->iri.path);
82 			close_conn(fds, c);
83 			return;
84 		}
85 
86 		if ((c->buf = mmap(NULL, c->len, PROT_READ, MAP_PRIVATE,
87 			    c->fd, 0)) == MAP_FAILED) {
88 			LOGW(c, "mmap: %s: %s", c->iri.path, strerror(errno));
89 			close_conn(fds, c);
90 			return;
91 		}
92 		c->i = c->buf;
93 		c->next = S_SENDING_FILE;
94 		start_reply(fds, c, SUCCESS, mime(c->host, c->iri.path));
95 		return;
96 
97 	case FILE_DIRECTORY:
98 		close(c->fd);
99 		c->fd = -1;
100 		send_dir(fds, c);
101 		return;
102 
103 	case FILE_MISSING:
104 		if (c->host->cgi != NULL && starts_with(c->iri.path, c->host->cgi)) {
105 			check_for_cgi(c->iri.path, c->iri.query, fds, c);
106 			return;
107 		}
108 		start_reply(fds, c, NOT_FOUND, "not found");
109 		return;
110 
111 	default:
112 		/* unreachable */
113 		abort();
114 	}
115 }
116 
117 
118 /*
119  * the inverse of this algorithm, i.e. starting from the start of the
120  * path + strlen(cgi), and checking if each component, should be
121  * faster.  But it's tedious to write.  This does the opposite: starts
122  * from the end and strip one component at a time, until either an
123  * executable is found or we emptied the path.
124  */
125 void
126 check_for_cgi(char *path, char *query, struct pollfd *fds, struct client *c)
127 {
128 	char *end;
129 	end = strchr(path, '\0');
130 
131 	/* NB: assume CGI is enabled and path matches cgi */
132 
133 	while (end > path) {
134 		/* go up one level.  UNIX paths are simple and POSIX
135 		 * dirname, with its ambiguities on if the given path
136 		 * is changed or not, gives me headaches. */
137 		while (*end != '/')
138 			end--;
139 		*end = '\0';
140 
141 		switch (check_path(c, path, &c->fd)) {
142 		case FILE_EXECUTABLE:
143 			start_cgi(path, end+1, query, fds,c);
144 			return;
145 		case FILE_MISSING:
146 			break;
147 		default:
148 			goto err;
149 		}
150 
151 		*end = '/';
152 		end--;
153 	}
154 
155 err:
156 	start_reply(fds, c, NOT_FOUND, "not found");
157 	return;
158 }
159 
160 void
161 mark_nonblock(int fd)
162 {
163 	int flags;
164 
165 	if ((flags = fcntl(fd, F_GETFL)) == -1)
166 		fatal("fcntl(F_GETFL): %s", strerror(errno));
167 	if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1)
168 		fatal("fcntl(F_SETFL): %s", strerror(errno));
169 }
170 
171 void
172 handle_handshake(struct pollfd *fds, struct client *c)
173 {
174 	struct vhost *h;
175 	const char *servname;
176 
177 	switch (tls_handshake(c->ctx)) {
178 	case 0:  /* success */
179 	case -1: /* already handshaked */
180 		break;
181 	case TLS_WANT_POLLIN:
182 		fds->events = POLLIN;
183 		return;
184 	case TLS_WANT_POLLOUT:
185 		fds->events = POLLOUT;
186 		return;
187 	default:
188 		/* unreachable */
189 		abort();
190 	}
191 
192 	servname = tls_conn_servername(c->ctx);
193 
194 	for (h = hosts; h->domain != NULL; ++h) {
195 		if (!strcmp(h->domain, "*"))
196 			break;
197 
198 		if (servname != NULL && !fnmatch(h->domain, servname, 0))
199 			break;
200 	}
201 
202 	if (h->domain != NULL) {
203 		c->state = S_OPEN;
204 		c->host = h;
205 		handle_open_conn(fds, c);
206 		return;
207 	}
208 
209 	if (servname != NULL)
210 		strncpy(c->req, servname, sizeof(c->req));
211 	else
212 		strncpy(c->req, "null", sizeof(c->req));
213 
214 	start_reply(fds, c, BAD_REQUEST, "Wrong host or missing SNI");
215 }
216 
217 void
218 handle_open_conn(struct pollfd *fds, struct client *c)
219 {
220 	const char *parse_err = "invalid request";
221 
222 	bzero(c->req, sizeof(c->req));
223 	bzero(&c->iri, sizeof(c->iri));
224 
225 	switch (tls_read(c->ctx, c->req, sizeof(c->req)-1)) {
226 	case -1:
227 		LOGE(c, "tls_read: %s", tls_error(c->ctx));
228 		close_conn(fds, c);
229 		return;
230 
231 	case TLS_WANT_POLLIN:
232 		fds->events = POLLIN;
233 		return;
234 
235 	case TLS_WANT_POLLOUT:
236 		fds->events = POLLOUT;
237 		return;
238 	}
239 
240 	if (!trim_req_iri(c->req) || !parse_iri(c->req, &c->iri, &parse_err)) {
241 		start_reply(fds, c, BAD_REQUEST, parse_err);
242 		return;
243 	}
244 
245 	/* XXX: we should check that the SNI matches the requested host */
246 	if (strcmp(c->iri.schema, "gemini") || c->iri.port_no != conf.port) {
247 		start_reply(fds, c, PROXY_REFUSED, "won't proxy request");
248 		return;
249 	}
250 
251 	open_file(fds, c);
252 }
253 
254 void
255 start_reply(struct pollfd *pfd, struct client *c, int code, const char *meta)
256 {
257 	char buf[1030];		/* status + ' ' + max reply len + \r\n\0 */
258 	size_t len;
259 
260 	c->code = code;
261 	c->meta = meta;
262 	c->state = S_INITIALIZING;
263 
264 	snprintf(buf, sizeof(buf), "%d ", code);
265 	strlcat(buf, meta, sizeof(buf));
266 	if (!strcmp(meta, "text/gemini") && c->host->lang != NULL) {
267 		strlcat(buf, "; lang=", sizeof(buf));
268 		strlcat(buf, c->host->lang, sizeof(buf));
269 	}
270 
271 	len = strlcat(buf, "\r\n", sizeof(buf));
272 	assert(len < sizeof(buf));
273 
274 	switch (tls_write(c->ctx, buf, len)) {
275 	case -1:
276 		close_conn(pfd, c);
277 		return;
278 	case TLS_WANT_POLLIN:
279 		pfd->events = POLLIN;
280 		return;
281 	case TLS_WANT_POLLOUT:
282 		pfd->events = POLLOUT;
283 		return;
284 	}
285 
286 	log_request(c, buf, sizeof(buf));
287 
288 	/* we don't need a body */
289 	if (c->code != SUCCESS) {
290 		close_conn(pfd, c);
291 		return;
292 	}
293 
294 	/* advance the state machine */
295 	c->state = c->next;
296 	handle(pfd, c);
297 }
298 
299 void
300 start_cgi(const char *spath, const char *relpath, const char *query,
301     struct pollfd *fds, struct client *c)
302 {
303 	char addr[NI_MAXHOST];
304 	const char *ruser, *cissuer, *chash;
305 	int e;
306 
307 	e = getnameinfo((struct sockaddr*)&c->addr, sizeof(c->addr),
308 	    addr, sizeof(addr),
309 	    NULL, 0,
310 	    NI_NUMERICHOST);
311 	if (e != 0)
312 		goto err;
313 
314 	if (tls_peer_cert_provided(c->ctx)) {
315 		ruser = tls_peer_cert_subject(c->ctx);
316 		cissuer = tls_peer_cert_issuer(c->ctx);
317 		chash = tls_peer_cert_hash(c->ctx);
318 	} else {
319 		ruser = NULL;
320 		cissuer = NULL;
321 		chash = NULL;
322 	}
323 
324 	if (!send_string(exfd, spath)
325 	    || !send_string(exfd, relpath)
326 	    || !send_string(exfd, query)
327 	    || !send_string(exfd, addr)
328 	    || !send_string(exfd, ruser)
329 	    || !send_string(exfd, cissuer)
330 	    || !send_string(exfd, chash)
331 	    || !send_vhost(exfd, c->host))
332 		goto err;
333 
334 	close(c->fd);
335 	if ((c->fd = recv_fd(exfd)) == -1) {
336 		start_reply(fds, c, TEMP_FAILURE, "internal server error");
337 		return;
338 	}
339 	c->state = S_SENDING_CGI;
340 	cgi_poll_on_child(fds, c);
341 	c->code = -1;
342 	/* handle_cgi(fds, c); */
343 	return;
344 
345 err:
346 	/* fatal("cannot talk to the executor process: %s", strerror(errno)); */
347 	fatal("cannot talk to the executor process");
348 }
349 
350 void
351 send_file(struct pollfd *fds, struct client *c)
352 {
353 	ssize_t ret, len;
354 
355 	/* ensure the correct state */
356 	c->state = S_SENDING_FILE;
357 
358 	len = (c->buf + c->len) - c->i;
359 
360 	while (len > 0) {
361 		switch (ret = tls_write(c->ctx, c->i, len)) {
362 		case -1:
363 			LOGE(c, "tls_write: %s", tls_error(c->ctx));
364 			close_conn(fds, c);
365 			return;
366 
367 		case TLS_WANT_POLLIN:
368 			fds->events = POLLIN;
369 			return;
370 
371 		case TLS_WANT_POLLOUT:
372 			fds->events = POLLOUT;
373 			return;
374 
375 		default:
376 			c->i += ret;
377 			len -= ret;
378 			break;
379 		}
380 	}
381 
382 	close_conn(fds, c);
383 }
384 
385 void
386 send_dir(struct pollfd *fds, struct client *c)
387 {
388 	size_t len;
389 	const char *index = "index.gmi";
390 
391 	/* guard against a re-entrant call: open_file -> send_dir ->
392 	 * open_file -> send_dir.  This can happen only if:
393 	 *
394 	 *  - user requested a dir, say foo/
395 	 *  - we try to serve foo/$INDEX
396 	 *  - foo/$INDEX is a directory.
397 	 */
398 	if (c->iri.path == c->sbuf) {
399 		start_reply(fds, c, TEMP_REDIRECT, c->sbuf);
400 		return;
401 	}
402 
403 	strlcpy(c->sbuf, "/", sizeof(c->sbuf));
404 
405 	len = strlen(c->iri.path);
406 	if (len > 0 && c->iri.path[len-1] != '/') {
407 		/* redirect to url with the trailing / */
408 		strlcat(c->sbuf, c->iri.path, sizeof(c->sbuf));
409 		strlcat(c->sbuf, "/", sizeof(c->sbuf));
410 		start_reply(fds, c, TEMP_REDIRECT, c->sbuf);
411 		return;
412 	}
413 
414 	strlcat(c->sbuf, c->iri.path, sizeof(c->sbuf));
415 
416 	if (!ends_with(c->sbuf, "/"))
417 		strlcat(c->sbuf, "/", sizeof(c->sbuf));
418 
419 	if (c->host->index != NULL)
420 		index = c->host->index;
421 	len = strlcat(c->sbuf, index, sizeof(c->sbuf));
422 
423 	if (len >= sizeof(c->sbuf)) {
424 		start_reply(fds, c, TEMP_FAILURE, "internal server error");
425 		return;
426 	}
427 
428 	close(c->fd);
429 	c->iri.path = c->sbuf;
430 	open_file(fds, c);
431 }
432 
433 void
434 cgi_poll_on_child(struct pollfd *fds, struct client *c)
435 {
436 	int fd;
437 
438 	if (c->waiting_on_child)
439 		return;
440 	c->waiting_on_child = 1;
441 
442 	fds->events = POLLIN;
443 
444 	fd = fds->fd;
445 	fds->fd = c->fd;
446 	c->fd = fd;
447 }
448 
449 void
450 cgi_poll_on_client(struct pollfd *fds, struct client *c)
451 {
452 	int fd;
453 
454 	if (!c->waiting_on_child)
455 		return;
456 	c->waiting_on_child = 0;
457 
458 	fd = fds->fd;
459 	fds->fd = c->fd;
460 	c->fd = fd;
461 }
462 
463 /* handle the read from the child process.  Return like read(2) */
464 static ssize_t
465 read_from_cgi(struct client *c)
466 {
467 	void	*buf;
468 	size_t	 len;
469 	ssize_t	 r;
470 
471 	/* if we haven't read a whole response line, we want to
472 	 * continue reading. */
473 
474 	if (c->code == -1) {
475 		buf = c->sbuf + c->len;
476 		len = sizeof(c->sbuf) - c->len;
477 	} else {
478 		buf = c->sbuf;
479 		len = sizeof(c->sbuf);
480 	}
481 
482 	r = read(c->fd, buf, len);
483 	if (r == 0 || r == -1)
484 		return r;
485 
486 	c->len += r;
487 	c->off = 0;
488 
489 	if (c->code != -1)
490 		return r;
491 
492 	if (strchr(c->sbuf, '\n') || c->len == sizeof(c->sbuf)) {
493 		c->code = 0;
494 		log_request(c, c->sbuf, c->len);
495 	}
496 
497 	return r;
498 }
499 
500 void
501 handle_cgi(struct pollfd *fds, struct client *c)
502 {
503 	ssize_t r;
504 
505 	/* ensure c->fd is the child and fds->fd the client */
506 	cgi_poll_on_client(fds, c);
507 
508 	while (1) {
509 		if (c->code == -1 || c->len == 0) {
510 			switch (r = read_from_cgi(c)) {
511 			case 0:
512 				goto end;
513 
514 			case -1:
515 				if (errno == EAGAIN || errno == EWOULDBLOCK) {
516 					cgi_poll_on_child(fds, c);
517 					return;
518 				}
519 				goto end;
520 			}
521 		}
522 
523 		if (c->code == -1) {
524 			cgi_poll_on_child(fds, c);
525 			return;
526 		}
527 
528 		while (c->len > 0) {
529 			switch (r = tls_write(c->ctx, c->sbuf + c->off, c->len)) {
530 			case -1:
531 				goto end;
532 
533 			case TLS_WANT_POLLOUT:
534 				fds->events = POLLOUT;
535 				return;
536 
537 			case TLS_WANT_POLLIN:
538 				fds->events = POLLIN;
539 				return;
540 
541 			default:
542                                 c->off += r;
543 				c->len -= r;
544 				break;
545 			}
546 		}
547 	}
548 
549 end:
550 	close_conn(fds, c);
551 }
552 
553 void
554 close_conn(struct pollfd *pfd, struct client *c)
555 {
556 	c->state = S_CLOSING;
557 
558 	switch (tls_close(c->ctx)) {
559 	case TLS_WANT_POLLIN:
560 		pfd->events = POLLIN;
561 		return;
562 	case TLS_WANT_POLLOUT:
563 		pfd->events = POLLOUT;
564 		return;
565 	}
566 
567 	connected_clients--;
568 
569 	tls_free(c->ctx);
570 	c->ctx = NULL;
571 
572 	if (c->buf != MAP_FAILED)
573 		munmap(c->buf, c->len);
574 
575 	if (c->fd != -1)
576 		close(c->fd);
577 
578 	close(pfd->fd);
579 	pfd->fd = -1;
580 }
581 
582 void
583 do_accept(int sock, struct tls *ctx, struct pollfd *fds, struct client *clients)
584 {
585 	int i, fd;
586 	struct sockaddr_storage addr;
587 	socklen_t len;
588 
589 	len = sizeof(addr);
590 	if ((fd = accept(sock, (struct sockaddr*)&addr, &len)) == -1) {
591 		if (errno == EWOULDBLOCK)
592 			return;
593 		fatal("accept: %s", strerror(errno));
594 	}
595 
596 	mark_nonblock(fd);
597 
598 	for (i = 0; i < MAX_USERS; ++i) {
599 		if (fds[i].fd == -1) {
600 			bzero(&clients[i], sizeof(struct client));
601 			if (tls_accept_socket(ctx, &clients[i].ctx, fd) == -1)
602 				break; /* goodbye fd! */
603 
604 			fds[i].fd = fd;
605 			fds[i].events = POLLIN;
606 
607 			clients[i].state = S_HANDSHAKE;
608 			clients[i].next = S_SENDING_FILE;
609 			clients[i].fd = -1;
610 			clients[i].waiting_on_child = 0;
611 			clients[i].buf = MAP_FAILED;
612 			clients[i].addr = addr;
613 
614 			connected_clients++;
615 			return;
616 		}
617 	}
618 
619 	close(fd);
620 }
621 
622 void
623 handle(struct pollfd *fds, struct client *client)
624 {
625 	switch (client->state) {
626 	case S_HANDSHAKE:
627 		handle_handshake(fds, client);
628 		break;
629 
630 	case S_OPEN:
631                 handle_open_conn(fds, client);
632 		break;
633 
634 	case S_INITIALIZING:
635 		start_reply(fds, client, client->code, client->meta);
636                 break;
637 
638 	case S_SENDING_FILE:
639 		send_file(fds, client);
640 		break;
641 
642 	case S_SENDING_CGI:
643 		handle_cgi(fds, client);
644 		break;
645 
646 	case S_CLOSING:
647 		close_conn(fds, client);
648 		break;
649 
650 	default:
651 		/* unreachable */
652 		abort();
653 	}
654 }
655 
656 void
657 loop(struct tls *ctx, int sock4, int sock6)
658 {
659 	int i;
660 	struct client clients[MAX_USERS];
661 	struct pollfd fds[MAX_USERS];
662 
663 	connected_clients = 0;
664 
665 	for (i = 0; i < MAX_USERS; ++i) {
666 		fds[i].fd = -1;
667 		fds[i].events = POLLIN;
668 		bzero(&clients[i], sizeof(struct client));
669 	}
670 
671 	fds[0].fd = sock4;
672 	fds[1].fd = sock6;
673 
674 	for (;;) {
675 		if (poll(fds, MAX_USERS, INFTIM) == -1) {
676 			if (errno == EINTR) {
677                                 fprintf(stderr, "connected clients: %d\n",
678 				    connected_clients);
679 				continue;
680 			}
681 			fatal("poll: %s", strerror(errno));
682 		}
683 
684 		for (i = 0; i < MAX_USERS; i++) {
685 			if (fds[i].revents == 0)
686 				continue;
687 
688 			if (fds[i].revents & (POLLERR|POLLNVAL))
689 				fatal("bad fd %d: %s", fds[i].fd,
690 				    strerror(errno));
691 
692 			if (fds[i].revents & POLLHUP) {
693 				/* fds[i] may be the fd of the stdin
694 				 * of a cgi script that has exited. */
695 				if (!clients[i].waiting_on_child) {
696 					close_conn(&fds[i], &clients[i]);
697 					continue;
698 				}
699 			}
700 
701 			if (fds[i].fd == sock4)
702 				do_accept(sock4, ctx, fds, clients);
703 			else if (fds[i].fd == sock6)
704 				do_accept(sock6, ctx, fds, clients);
705 			else
706 				handle(&fds[i], &clients[i]);
707 		}
708 	}
709 }