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