1 /*
2  * Copyright (c) 2020-2022 Tracey Emery <tracey@traceyemery.net>
3  * Copyright (c) 2013 David Gwynne <dlg@openbsd.org>
4  * Copyright (c) 2013 Florian Obser <florian@openbsd.org>
5  *
6  * Permission to use, copy, modify, and distribute this software for any
7  * purpose with or without fee is hereby granted, provided that the above
8  * copyright notice and this permission notice appear in all copies.
9  *
10  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17  */
18 
19 #include <arpa/inet.h>
20 #include <sys/queue.h>
21 #include <sys/socket.h>
22 #include <sys/types.h>
23 #include <sys/uio.h>
24 
25 #include <errno.h>
26 #include <event.h>
27 #include <imsg.h>
28 #include <stdarg.h>
29 #include <stdlib.h>
30 #include <stdio.h>
31 #include <string.h>
32 #include <time.h>
33 #include <unistd.h>
34 
35 #include "got_error.h"
36 
37 #include "proc.h"
38 #include "gotwebd.h"
39 
40 size_t	 fcgi_parse_record(uint8_t *, size_t, struct request *);
41 void	 fcgi_parse_begin_request(uint8_t *, uint16_t, struct request *,
42 	    uint16_t);
43 void	 fcgi_parse_params(uint8_t *, uint16_t, struct request *, uint16_t);
44 int	 fcgi_send_response(struct request *, int, const void *, size_t);
45 
46 void	 dump_fcgi_record_header(const char *, struct fcgi_record_header *);
47 void	 dump_fcgi_begin_request_body(const char *,
48 	    struct fcgi_begin_request_body *);
49 void	 dump_fcgi_end_request_body(const char *,
50 	    struct fcgi_end_request_body *);
51 
52 extern int	 cgi_inflight;
53 extern volatile int client_cnt;
54 
55 void
56 fcgi_request(int fd, short events, void *arg)
57 {
58 	struct request *c = arg;
59 	ssize_t n;
60 	size_t parsed = 0;
61 
62 	n = read(fd, c->buf + c->buf_pos + c->buf_len,
63 	    FCGI_RECORD_SIZE - c->buf_pos-c->buf_len);
64 
65 	switch (n) {
66 	case -1:
67 		switch (errno) {
68 		case EINTR:
69 		case EAGAIN:
70 			return;
71 		default:
72 			goto fail;
73 		}
74 		break;
75 
76 	case 0:
77 		log_debug("closed connection");
78 		goto fail;
79 	default:
80 		break;
81 	}
82 
83 	c->buf_len += n;
84 
85 	/*
86 	 * Parse the records as they are received. Per the FastCGI
87 	 * specification, the server need only receive the FastCGI
88 	 * parameter records in full; it is free to begin execution
89 	 * at that point, which is what happens here.
90 	 */
91 	do {
92 		parsed = fcgi_parse_record(c->buf + c->buf_pos, c->buf_len, c);
93 		if (parsed != 0) {
94 			c->buf_pos += parsed;
95 			c->buf_len -= parsed;
96 		}
97 	} while (parsed > 0 && c->buf_len > 0);
98 
99 	/* Make space for further reads */
100 	if (parsed != 0)
101 		if (c->buf_len > 0) {
102 			bcopy(c->buf + c->buf_pos, c->buf, c->buf_len);
103 			c->buf_pos = 0;
104 		}
105 	return;
106 fail:
107 	fcgi_cleanup_request(c);
108 }
109 
110 size_t
111 fcgi_parse_record(uint8_t *buf, size_t n, struct request *c)
112 {
113 	struct fcgi_record_header *h;
114 
115 	if (n < sizeof(struct fcgi_record_header))
116 		 return 0;
117 
118 	h = (struct fcgi_record_header*) buf;
119 
120 	dump_fcgi_record("", h);
121 
122 	if (n < sizeof(struct fcgi_record_header) + ntohs(h->content_len)
123 	    + h->padding_len)
124 		 return 0;
125 
126 	if (h->version != 1)
127 		log_warn("wrong version");
128 
129 	switch (h->type) {
130 	case FCGI_BEGIN_REQUEST:
131 		fcgi_parse_begin_request(buf +
132 		    sizeof(struct fcgi_record_header),
133 		    ntohs(h->content_len), c, ntohs(h->id));
134 		break;
135 	case FCGI_PARAMS:
136 		fcgi_parse_params(buf + sizeof(struct fcgi_record_header),
137 		    ntohs(h->content_len), c, ntohs(h->id));
138 		break;
139 	case FCGI_STDIN:
140 	case FCGI_ABORT_REQUEST:
141 		if (c->sock->client_status != CLIENT_DISCONNECT &&
142 		    c->outbuf_len != 0) {
143 			fcgi_send_response(c, FCGI_STDOUT, c->outbuf,
144 			    c->outbuf_len);
145 		}
146 
147 		fcgi_create_end_record(c);
148 		fcgi_cleanup_request(c);
149 		return 0;
150 	default:
151 		log_warn("unimplemented type %d", h->type);
152 		break;
153 	}
154 
155 	return (sizeof(struct fcgi_record_header) + ntohs(h->content_len)
156 	    + h->padding_len);
157 }
158 
159 void
160 fcgi_parse_begin_request(uint8_t *buf, uint16_t n,
161     struct request *c, uint16_t id)
162 {
163 	/* XXX -- FCGI_CANT_MPX_CONN */
164 	if (c->request_started) {
165 		log_warn("unexpected FCGI_BEGIN_REQUEST, ignoring");
166 		return;
167 	}
168 
169 	if (n != sizeof(struct fcgi_begin_request_body)) {
170 		log_warn("wrong size %d != %lu", n,
171 		    sizeof(struct fcgi_begin_request_body));
172 		return;
173 	}
174 
175 	c->request_started = 1;
176 
177 	c->id = id;
178 	SLIST_INIT(&c->env);
179 	c->env_count = 0;
180 }
181 
182 void
183 fcgi_parse_params(uint8_t *buf, uint16_t n, struct request *c, uint16_t id)
184 {
185 	struct env_val *env_entry;
186 	uint32_t name_len, val_len;
187 	uint8_t *sd, *dr_buf;
188 
189 	if (!c->request_started) {
190 		log_warn("FCGI_PARAMS without FCGI_BEGIN_REQUEST, ignoring");
191 		return;
192 	}
193 
194 	if (c->id != id) {
195 		log_warn("unexpected id, ignoring");
196 		return;
197 	}
198 
199 	if (n == 0) {
200 		gotweb_process_request(c);
201 		return;
202 	}
203 
204 	while (n > 0) {
205 		if (buf[0] >> 7 == 0) {
206 			name_len = buf[0];
207 			n--;
208 			buf++;
209 		} else {
210 			if (n > 3) {
211 				name_len = ((buf[0] & 0x7f) << 24) +
212 				    (buf[1] << 16) + (buf[2] << 8) + buf[3];
213 				n -= 4;
214 				buf += 4;
215 			} else
216 				return;
217 		}
218 
219 		if (n > 0) {
220 			if (buf[0] >> 7 == 0) {
221 				val_len = buf[0];
222 				n--;
223 				buf++;
224 			} else {
225 				if (n > 3) {
226 					val_len = ((buf[0] & 0x7f) << 24) +
227 					    (buf[1] << 16) + (buf[2] << 8) +
228 					     buf[3];
229 					n -= 4;
230 					buf += 4;
231 				} else
232 					return;
233 			}
234 		} else
235 			return;
236 
237 		if (n < name_len + val_len)
238 			return;
239 
240 		if ((env_entry = malloc(sizeof(struct env_val))) == NULL) {
241 			log_warn("cannot malloc env_entry");
242 			return;
243 		}
244 
245 		if ((env_entry->val = calloc(sizeof(char), name_len + val_len +
246 		    2)) == NULL) {
247 			log_warn("cannot allocate env_entry->val");
248 			free(env_entry);
249 			return;
250 		}
251 
252 		bcopy(buf, env_entry->val, name_len);
253 		buf += name_len;
254 		n -= name_len;
255 
256 		env_entry->val[name_len] = '\0';
257 		if (val_len < MAX_QUERYSTRING && strcmp(env_entry->val,
258 		    "QUERY_STRING") == 0 && c->querystring[0] == '\0') {
259 			bcopy(buf, c->querystring, val_len);
260 			c->querystring[val_len] = '\0';
261 		}
262 		if (val_len < GOTWEBD_MAXTEXT && strcmp(env_entry->val,
263 		    "HTTP_HOST") == 0 && c->http_host[0] == '\0') {
264 
265 			/*
266 			 * lazily get subdomain
267 			 * will only get domain if no subdomain exists
268 			 * this can still work if gotweb server name is the same
269 			 */
270 			sd = strchr(buf, '.');
271 			if (sd)
272 				*sd = '\0';
273 
274 			bcopy(buf, c->http_host, val_len);
275 			c->http_host[val_len] = '\0';
276 		}
277 		if (val_len < MAX_SCRIPT_NAME && strcmp(env_entry->val,
278 		    "SCRIPT_NAME") == 0 && c->script_name[0] == '\0') {
279 			bcopy(dr_buf, c->script_name, val_len);
280 			c->script_name[val_len] = '\0';
281 		}
282 		if (val_len < MAX_SERVER_NAME && strcmp(env_entry->val,
283 		    "SERVER_NAME") == 0 && c->server_name[0] == '\0') {
284 			bcopy(buf, c->server_name, val_len);
285 			c->server_name[val_len] = '\0';
286 		}
287 		env_entry->val[name_len] = '=';
288 
289 		bcopy(buf, (env_entry->val) + name_len + 1, val_len);
290 		buf += val_len;
291 		n -= val_len;
292 
293 		SLIST_INSERT_HEAD(&c->env, env_entry, entry);
294 		log_debug("env[%d], %s", c->env_count, env_entry->val);
295 		c->env_count++;
296 	}
297 }
298 
299 void
300 fcgi_timeout(int fd, short events, void *arg)
301 {
302 	fcgi_cleanup_request((struct request*) arg);
303 }
304 
305 int
306 fcgi_printf(struct request *c, const char *fmt, ...)
307 {
308 	va_list ap;
309 	char *str;
310 	int r;
311 
312 	va_start(ap, fmt);
313 	r = vasprintf(&str, fmt, ap);
314 	va_end(ap);
315 
316 	if (r == -1) {
317 		log_warn("%s: asprintf", __func__);
318 		return -1;
319 	}
320 
321 	r = fcgi_gen_binary_response(c, str, r);
322 	free(str);
323 	return r;
324 }
325 
326 int
327 fcgi_gen_binary_response(struct request *c, const uint8_t *data, int len)
328 {
329 	int r;
330 
331 	if (c->sock->client_status == CLIENT_DISCONNECT)
332 		return -1;
333 
334 	if (data == NULL || len == 0)
335 		return 0;
336 
337 	/*
338 	 * special case: send big replies -like blobs- directly
339 	 * without copying.
340 	 */
341 	if (len > sizeof(c->outbuf)) {
342 		if (c->outbuf_len > 0) {
343 			fcgi_send_response(c, FCGI_STDOUT,
344 			    c->outbuf, c->outbuf_len);
345 			c->outbuf_len = 0;
346 		}
347 		return fcgi_send_response(c, FCGI_STDOUT, data, len);
348 	}
349 
350 	if (len < sizeof(c->outbuf) - c->outbuf_len) {
351 		memcpy(c->outbuf + c->outbuf_len, data, len);
352 		c->outbuf_len += len;
353 		return 0;
354 	}
355 
356 	r = fcgi_send_response(c, FCGI_STDOUT, c->outbuf, c->outbuf_len);
357 	if (r == -1)
358 		return -1;
359 
360 	memcpy(c->outbuf, data, len);
361 	c->outbuf_len = len;
362 	return 0;
363 }
364 
365 static int
366 send_response(struct request *c, int type, const uint8_t *data,
367     size_t len)
368 {
369 	static const uint8_t padding[FCGI_PADDING_SIZE];
370 	struct fcgi_record_header header;
371 	struct iovec iov[3];
372 	struct timespec ts;
373 	ssize_t nw;
374 	size_t padded_len, tot;
375 	int i, err = 0, th = 2000;
376 
377 	ts.tv_sec = 0;
378 	ts.tv_nsec = 50;
379 
380 	memset(&header, 0, sizeof(header));
381 	header.version = 1;
382 	header.type = type;
383 	header.id = htons(c->id);
384 	header.content_len = htons(len);
385 
386 	/* The FastCGI spec suggests to align the output buffer */
387 	tot = sizeof(header) + len;
388 	padded_len = FCGI_ALIGN(tot);
389 	if (padded_len > tot) {
390 		header.padding_len = padded_len - tot;
391 		tot += header.padding_len;
392 	}
393 
394 	iov[0].iov_base = &header;
395 	iov[0].iov_len = sizeof(header);
396 
397 	iov[1].iov_base = (void *)data;
398 	iov[1].iov_len = len;
399 
400 	iov[2].iov_base = (void *)padding;
401 	iov[2].iov_len = header.padding_len;
402 
403 	dump_fcgi_record("resp ", &header);
404 
405 	/*
406 	 * XXX: add some simple write heuristics here
407 	 * On slower VMs, spotty connections, etc., we don't want to go right to
408 	 * disconnect. Let's at least try to write the data a few times before
409 	 * giving up.
410 	 */
411 	while (tot > 0) {
412 		nw = writev(c->fd, iov, nitems(iov));
413 		if (nw == 0) {
414 			c->sock->client_status = CLIENT_DISCONNECT;
415 			break;
416 		}
417 		if (nw == -1) {
418 			err++;
419 			if (errno == EAGAIN && err < th) {
420 				nanosleep(&ts, NULL);
421 				continue;
422 			}
423 			log_warn("%s: write failure", __func__);
424 			c->sock->client_status = CLIENT_DISCONNECT;
425 			return -1;
426 		}
427 
428 		if (nw != tot)
429 			log_debug("%s: partial write: %zu vs %zu", __func__,
430 			    nw, tot);
431 
432 		tot -= nw;
433 		for (i = 0; i < nitems(iov); ++i) {
434 			if (nw < iov[i].iov_len) {
435 				iov[i].iov_base += nw;
436 				iov[i].iov_len -= nw;
437 				break;
438 			}
439 			nw -= iov[i].iov_len;
440 			iov[i].iov_len = 0;
441 		}
442 	}
443 
444 	return 0;
445 }
446 
447 int
448 fcgi_send_response(struct request *c, int type, const void *data,
449     size_t len)
450 {
451 	if (c->sock->client_status == CLIENT_DISCONNECT)
452 		return -1;
453 
454 	while (len > FCGI_CONTENT_SIZE) {
455 		if (send_response(c, type, data, len) == -1)
456 			return -1;
457 
458 		data += FCGI_CONTENT_SIZE;
459 		len -= FCGI_CONTENT_SIZE;
460 	}
461 
462 	if (len == 0)
463 		return 0;
464 
465 	return send_response(c, type, data, len);
466 }
467 
468 void
469 fcgi_create_end_record(struct request *c)
470 {
471 	struct fcgi_end_request_body end_request;
472 
473 	memset(&end_request, 0, sizeof(end_request));
474 	end_request.app_status = htonl(0); /* script status */
475 	end_request.protocol_status = FCGI_REQUEST_COMPLETE;
476 
477 	fcgi_send_response(c, FCGI_END_REQUEST, &end_request,
478 	    sizeof(end_request));
479 }
480 
481 void
482 fcgi_cleanup_request(struct request *c)
483 {
484 	cgi_inflight--;
485 	client_cnt--;
486 
487 	evtimer_del(&c->tmo);
488 	if (event_initialized(&c->ev))
489 		event_del(&c->ev);
490 
491 	close(c->fd);
492 	gotweb_free_transport(c->t);
493 	free(c);
494 }
495 
496 void
497 dump_fcgi_record(const char *p, struct fcgi_record_header *h)
498 {
499 	dump_fcgi_record_header(p, h);
500 
501 	if (h->type == FCGI_BEGIN_REQUEST)
502 		dump_fcgi_begin_request_body(p,
503 		    (struct fcgi_begin_request_body *)(h + 1));
504 	else if (h->type == FCGI_END_REQUEST)
505 		dump_fcgi_end_request_body(p,
506 		    (struct fcgi_end_request_body *)(h + 1));
507 }
508 
509 void
510 dump_fcgi_record_header(const char* p, struct fcgi_record_header *h)
511 {
512 	log_debug("%sversion:         %d", p, h->version);
513 	log_debug("%stype:            %d", p, h->type);
514 	log_debug("%srequestId:       %d", p, ntohs(h->id));
515 	log_debug("%scontentLength:   %d", p, ntohs(h->content_len));
516 	log_debug("%spaddingLength:   %d", p, h->padding_len);
517 	log_debug("%sreserved:        %d", p, h->reserved);
518 }
519 
520 void
521 dump_fcgi_begin_request_body(const char *p, struct fcgi_begin_request_body *b)
522 {
523 	log_debug("%srole             %d", p, ntohs(b->role));
524 	log_debug("%sflags            %d", p, b->flags);
525 }
526 
527 void
528 dump_fcgi_end_request_body(const char *p, struct fcgi_end_request_body *b)
529 {
530 	log_debug("%sappStatus:       %d", p, ntohl(b->app_status));
531 	log_debug("%sprotocolStatus:  %d", p, b->protocol_status);
532 }