htparser: Use bufio instead of stdio for greater responsiveness.
[ashd.git] / src / htparser.c
... / ...
CommitLineData
1/*
2 ashd - A Sane HTTP Daemon
3 Copyright (C) 2008 Fredrik Tolf <fredrik@dolda2000.com>
4
5 This program is free software: you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation, either version 3 of the License, or
8 (at your option) any later version.
9
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
14
15 You should have received a copy of the GNU General Public License
16 along with this program. If not, see <http://www.gnu.org/licenses/>.
17*/
18
19#include <stdlib.h>
20#include <unistd.h>
21#include <stdio.h>
22#include <fcntl.h>
23#include <string.h>
24#include <sys/socket.h>
25#include <pwd.h>
26#include <sys/signal.h>
27#include <errno.h>
28
29#ifdef HAVE_CONFIG_H
30#include <config.h>
31#endif
32#include <utils.h>
33#include <mt.h>
34#include <mtio.h>
35#include <log.h>
36#include <req.h>
37#include <proc.h>
38#include <bufio.h>
39
40#include "htparser.h"
41
42static int plex;
43static char *pidfile = NULL;
44static int daemonize, usesyslog;
45struct mtbuf listeners;
46
47static void trimx(struct hthead *req)
48{
49 int i;
50
51 i = 0;
52 while(i < req->noheaders) {
53 if(!strncasecmp(req->headers[i][0], "x-ash-", 6)) {
54 free(req->headers[i][0]);
55 free(req->headers[i][1]);
56 free(req->headers[i]);
57 memmove(req->headers + i, req->headers + i + 1, sizeof(*req->headers) * (--req->noheaders - i));
58 } else {
59 i++;
60 }
61 }
62}
63
64static struct hthead *parsereq(struct bufio *in)
65{
66 struct hthead *req;
67 struct charbuf method, url, ver;
68 int c;
69
70 req = NULL;
71 bufinit(method);
72 bufinit(url);
73 bufinit(ver);
74 while(1) {
75 c = biogetc(in);
76 if(c == ' ') {
77 break;
78 } else if((c == EOF) || (c < 32) || (c >= 128)) {
79 goto fail;
80 } else {
81 bufadd(method, c);
82 if(method.d >= 128)
83 goto fail;
84 }
85 }
86 while(1) {
87 c = biogetc(in);
88 if(c == ' ') {
89 break;
90 } else if((c == EOF) || (c < 32)) {
91 goto fail;
92 } else {
93 bufadd(url, c);
94 if(url.d >= 65536)
95 goto fail;
96 }
97 }
98 while(1) {
99 c = biogetc(in);
100 if(c == 10) {
101 break;
102 } else if(c == 13) {
103 } else if((c == EOF) || (c < 32) || (c >= 128)) {
104 goto fail;
105 } else {
106 bufadd(ver, c);
107 if(ver.d >= 128)
108 goto fail;
109 }
110 }
111 bufadd(method, 0);
112 bufadd(url, 0);
113 bufadd(ver, 0);
114 req = mkreq(method.b, url.b, ver.b);
115 if(parseheadersb(req, in))
116 goto fail;
117 trimx(req);
118 goto out;
119
120fail:
121 if(req != NULL) {
122 freehthead(req);
123 req = NULL;
124 }
125out:
126 buffree(method);
127 buffree(url);
128 buffree(ver);
129 return(req);
130}
131
132static off_t passdata(struct bufio *in, struct bufio *out, off_t max)
133{
134 ssize_t read;
135 off_t total;
136
137 total = 0;
138 while(!bioeof(in) && ((max < 0) || (total < max))) {
139 if((read = biordata(in)) > 0) {
140 if(max >= 0)
141 read = min(max - total, read);
142 if((read = biowritesome(out, in->rbuf.b + in->rh, read)) < 0)
143 return(-1);
144 in->rh += read;
145 total += read;
146 }
147 if(biorspace(in) && ((max < 0) || (biordata(in) < max - total)) && (biofillsome(in) < 0))
148 return(-1);
149 }
150 return(total);
151}
152
153static int recvchunks(struct bufio *in, struct bufio *out)
154{
155 size_t read, chlen;
156 int c, r;
157
158 while(1) {
159 chlen = 0;
160 r = 0;
161 while(1) {
162 c = biogetc(in);
163 if(c == 10) {
164 if(!r)
165 return(-1);
166 break;
167 } else if(c == 13) {
168 } else if((c >= '0') && (c <= '9')) {
169 chlen = (chlen << 4) + (c - '0');
170 r = 1;
171 } else if((c >= 'A') && (c <= 'F')) {
172 chlen = (chlen << 4) + (c + 10 - 'A');
173 r = 1;
174 } else if((c >= 'a') && (c <= 'f')) {
175 chlen = (chlen << 4) + (c + 10 - 'a');
176 r = 1;
177 } else {
178 /* XXX: Technically, there may be chunk extensions to
179 * be read, but since that will likely never actually
180 * happen in practice, I can just as well add support
181 * for that if it actually does become relevant. */
182 return(-1);
183 }
184 }
185 if(chlen == 0)
186 break;
187 while(chlen > 0) {
188 if((read = biordata(in)) > 0) {
189 if((read = biowritesome(out, in->rbuf.b + in->rh, min(read, chlen))) < 0)
190 return(-1);
191 in->rh += read;
192 chlen -= read;
193 }
194 if(biorspace(in) && (biordata(in) < chlen) && (biofillsome(in) <= 0))
195 return(-1);
196 }
197 if((biogetc(in) != 13) || (biogetc(in) != 10))
198 return(-1);
199 }
200 /* XXX: Technically, there may be trailers to be read, but that's
201 * just about as likely as chunk extensions. */
202 if((biogetc(in) != 13) || (biogetc(in) != 10))
203 return(-1);
204 return(0);
205}
206
207static int passchunks(struct bufio *in, struct bufio *out)
208{
209 size_t read;
210
211 while(!bioeof(in)) {
212 if((read = biordata(in)) > 0) {
213 bioprintf(out, "%zx\r\n", read);
214 if(biowrite(out, in->rbuf.b + in->rh, read) != read)
215 return(-1);
216 in->rh += read;
217 bioprintf(out, "\r\n");
218 if(bioflush(out) < 0)
219 return(-1);
220 }
221 if(biorspace(in) && (biofillsome(in) < 0))
222 return(-1);
223 }
224 bioprintf(out, "0\r\n\r\n");
225 return(0);
226}
227
228static int hasheader(struct hthead *head, char *name, char *val)
229{
230 char *hd;
231
232 if((hd = getheader(head, name)) == NULL)
233 return(0);
234 return(!strcasecmp(hd, val));
235}
236
237static int canonreq(struct hthead *req)
238{
239 char *p, *p2, *r;
240 int n;
241
242 if(req->url[0] == '/') {
243 replrest(req, req->url + 1);
244 if((p = strchr(req->rest, '?')) != NULL)
245 *p = 0;
246 return(1);
247 }
248 if((p = strstr(req->url, "://")) != NULL) {
249 n = p - req->url;
250 if(((n == 4) && !strncasecmp(req->url, "http", 4)) ||
251 ((n == 5) && !strncasecmp(req->url, "https", 5))) {
252 if(getheader(req, "host"))
253 return(0);
254 p += 3;
255 if((p2 = strchr(p, '/')) == NULL) {
256 headappheader(req, "Host", p);
257 free(req->url);
258 req->url = sstrdup("/");
259 } else {
260 r = sstrdup(p2);
261 *(p2++) = 0;
262 headappheader(req, "Host", p);
263 free(req->url);
264 req->url = r;
265 }
266 replrest(req, req->url + 1);
267 if((p = strchr(req->rest, '?')) != NULL)
268 *p = 0;
269 return(1);
270 }
271 }
272 return(0);
273}
274
275static int http10keep(struct hthead *req, struct hthead *resp)
276{
277 int fc;
278
279 fc = hasheader(resp, "connection", "close");
280 headrmheader(resp, "connection");
281 if(!fc && hasheader(req, "connection", "keep-alive")) {
282 headappheader(resp, "Connection", "Keep-Alive");
283 return(1);
284 } else {
285 return(0);
286 }
287}
288
289static char *connid(void)
290{
291 static struct charbuf cur;
292 int i;
293 char *ret;
294
295 for(i = 0; i < cur.d; i++) {
296 if((++cur.b[i]) > 'Z')
297 cur.b[i] = 'A';
298 else
299 goto done;
300 }
301 bufadd(cur, 'A');
302done:
303 ret = memcpy(smalloc(cur.d + 1), cur.b, cur.d);
304 ret[cur.d] = 0;
305 return(ret);
306}
307
308void serve(struct bufio *in, struct conn *conn)
309{
310 int pfds[2];
311 struct bufio *out;
312 struct hthead *req, *resp;
313 char *hd, *id;
314 off_t dlen;
315 int keep;
316
317 id = connid();
318 out = NULL;
319 req = resp = NULL;
320 while(plex >= 0) {
321 bioflush(in);
322 if((req = parsereq(in)) == NULL)
323 break;
324 if(!canonreq(req))
325 break;
326
327 headappheader(req, "X-Ash-Connection-ID", id);
328 if((conn->initreq != NULL) && conn->initreq(conn, req))
329 break;
330
331 if((plex < 0) || block(plex, EV_WRITE, 60) <= 0)
332 break;
333 if(socketpair(PF_UNIX, SOCK_STREAM, 0, pfds))
334 break;
335 if(sendreq(plex, req, pfds[0]))
336 break;
337 close(pfds[0]);
338 out = mtbioopen(pfds[1], 1, 600, "r+", NULL);
339
340 if(getheader(req, "content-type") != NULL) {
341 if((hd = getheader(req, "content-length")) != NULL) {
342 dlen = atoo(hd);
343 if(dlen > 0) {
344 if(passdata(in, out, dlen) != dlen)
345 break;
346 }
347 } else if(((hd = getheader(req, "transfer-encoding")) != NULL) && !strcasecmp(hd, "chunked")) {
348 if(recvchunks(in, out))
349 break;
350 } else {
351 /* Ignore rather than abort, to be kinder to broken clients. */
352 headrmheader(req, "content-type");
353 }
354 }
355 if(bioflush(out))
356 break;
357 /* Make sure to send EOF */
358 shutdown(pfds[1], SHUT_WR);
359
360 if((resp = parseresponseb(out)) == NULL)
361 break;
362 replstr(&resp->ver, req->ver);
363
364 if(!getheader(resp, "server"))
365 headappheader(resp, "Server", sprintf3("ashd/%s", VERSION));
366
367 if(!strcasecmp(req->ver, "HTTP/1.0")) {
368 if(!strcasecmp(req->method, "head")) {
369 keep = http10keep(req, resp);
370 writerespb(in, resp);
371 bioprintf(in, "\r\n");
372 } else if((hd = getheader(resp, "content-length")) != NULL) {
373 keep = http10keep(req, resp);
374 dlen = atoo(hd);
375 writerespb(in, resp);
376 bioprintf(in, "\r\n");
377 if(passdata(out, in, dlen) != dlen)
378 break;
379 } else {
380 headrmheader(resp, "connection");
381 writerespb(in, resp);
382 bioprintf(in, "\r\n");
383 passdata(out, in, -1);
384 break;
385 }
386 if(!keep)
387 break;
388 } else if(!strcasecmp(req->ver, "HTTP/1.1")) {
389 if(!strcasecmp(req->method, "head")) {
390 writerespb(in, resp);
391 bioprintf(in, "\r\n");
392 } else if((hd = getheader(resp, "content-length")) != NULL) {
393 writerespb(in, resp);
394 bioprintf(in, "\r\n");
395 dlen = atoo(hd);
396 if(passdata(out, in, dlen) != dlen)
397 break;
398 } else if(!getheader(resp, "transfer-encoding")) {
399 headappheader(resp, "Transfer-Encoding", "chunked");
400 writerespb(in, resp);
401 bioprintf(in, "\r\n");
402 if(passchunks(out, in))
403 break;
404 } else {
405 writerespb(in, resp);
406 bioprintf(in, "\r\n");
407 passdata(out, in, -1);
408 break;
409 }
410 if(hasheader(req, "connection", "close") || hasheader(resp, "connection", "close"))
411 break;
412 } else {
413 break;
414 }
415
416 bioclose(out);
417 out = NULL;
418 freehthead(req);
419 freehthead(resp);
420 req = resp = NULL;
421 }
422
423 if(out != NULL)
424 bioclose(out);
425 if(req != NULL)
426 freehthead(req);
427 if(resp != NULL)
428 freehthead(resp);
429 bioclose(in);
430 free(id);
431}
432
433static void plexwatch(struct muth *muth, va_list args)
434{
435 vavar(int, fd);
436 char *buf;
437 int i, s, ret;
438
439 s = 0;
440 while(1) {
441 if(block(fd, EV_READ, 0) == 0)
442 break;
443 buf = smalloc(65536);
444 ret = recv(fd, buf, 65536, 0);
445 if(ret < 0) {
446 flog(LOG_WARNING, "received error on rootplex read channel: %s", strerror(errno));
447 exit(1);
448 } else if(ret == 0) {
449 s = 1;
450 free(buf);
451 break;
452 }
453 /* Maybe I'd like to implement some protocol in this direction
454 * some day... */
455 free(buf);
456 }
457 shutdown(plex, SHUT_RDWR);
458 for(i = 0; i < listeners.d; i++) {
459 if(listeners.b[i] == muth)
460 bufdel(listeners, i);
461 }
462 if(s) {
463 flog(LOG_INFO, "root handler exited, so shutting down listening...");
464 while(listeners.d > 0)
465 resume(listeners.b[0], 0);
466 }
467}
468
469static void initroot(void *uu)
470{
471 int fd;
472
473 setsid();
474 if(daemonize) {
475 chdir("/");
476 if((fd = open("/dev/null", O_RDWR)) >= 0) {
477 dup2(fd, 0);
478 dup2(fd, 1);
479 dup2(fd, 2);
480 close(fd);
481 }
482 }
483 if(usesyslog)
484 putenv("ASHD_USESYSLOG=1");
485 else
486 unsetenv("ASHD_USESYSLOG");
487}
488
489static void usage(FILE *out)
490{
491 fprintf(out, "usage: htparser [-hSf] [-u USER] [-r ROOT] [-p PIDFILE] PORTSPEC... -- ROOT [ARGS...]\n");
492 fprintf(out, "\twhere PORTSPEC is HANDLER[:PAR[=VAL][(,PAR[=VAL])...]] (try HANDLER:help)\n");
493 fprintf(out, "\tavailable handlers are `plain' and `ssl'.\n");
494}
495
496static void addport(char *spec)
497{
498 char *nm, *p, *p2, *n;
499 struct charvbuf pars, vals;
500
501 bufinit(pars);
502 bufinit(vals);
503 if((p = strchr(spec, ':')) == NULL) {
504 nm = spec;
505 } else {
506 nm = spec;
507 *(p++) = 0;
508 do {
509 if((n = strchr(p, ',')) != NULL)
510 *(n++) = 0;
511 if((p2 = strchr(p, '=')) != NULL)
512 *(p2++) = 0;
513 if(!*p) {
514 usage(stderr);
515 exit(1);
516 }
517 bufadd(pars, p);
518 if(p2)
519 bufadd(vals, p2);
520 else
521 bufadd(vals, "");
522 } while((p = n) != NULL);
523 }
524
525 /* XXX: It would be nice to decentralize this, but, meh... */
526 if(!strcmp(nm, "plain")) {
527 handleplain(pars.d, pars.b, vals.b);
528#ifdef HAVE_GNUTLS
529 } else if(!strcmp(nm, "ssl")) {
530 handlegnussl(pars.d, pars.b, vals.b);
531#endif
532 } else {
533 flog(LOG_ERR, "htparser: unknown port handler `%s'", nm);
534 exit(1);
535 }
536
537 buffree(pars);
538 buffree(vals);
539}
540
541static void sighandler(int sig)
542{
543 exitioloop(1);
544}
545
546int main(int argc, char **argv)
547{
548 int c, d;
549 int i, s1;
550 char *root;
551 FILE *pidout;
552 struct passwd *pwent;
553
554 daemonize = usesyslog = 0;
555 root = NULL;
556 pwent = NULL;
557 while((c = getopt(argc, argv, "+hSfu:r:p:")) >= 0) {
558 switch(c) {
559 case 'h':
560 usage(stdout);
561 exit(0);
562 case 'f':
563 daemonize = 1;
564 break;
565 case 'S':
566 usesyslog = 1;
567 break;
568 case 'u':
569 if((pwent = getpwnam(optarg)) == NULL) {
570 flog(LOG_ERR, "could not find user %s", optarg);
571 exit(1);
572 }
573 break;
574 case 'r':
575 root = optarg;
576 break;
577 case 'p':
578 pidfile = optarg;
579 break;
580 default:
581 usage(stderr);
582 exit(1);
583 }
584 }
585 s1 = 0;
586 for(i = optind; i < argc; i++) {
587 if(!strcmp(argv[i], "--"))
588 break;
589 s1 = 1;
590 addport(argv[i]);
591 }
592 if(!s1 || (i == argc)) {
593 usage(stderr);
594 exit(1);
595 }
596 if((plex = stdmkchild(argv + ++i, initroot, NULL)) < 0) {
597 flog(LOG_ERR, "could not spawn root multiplexer: %s", strerror(errno));
598 return(1);
599 }
600 bufadd(listeners, mustart(plexwatch, plex));
601 pidout = NULL;
602 if(pidfile != NULL) {
603 if((pidout = fopen(pidfile, "w")) == NULL) {
604 flog(LOG_ERR, "could not open %s for writing: %s", pidfile, strerror(errno));
605 return(1);
606 }
607 }
608 if(usesyslog)
609 opensyslog();
610 if(root) {
611 if(chdir(root) || chroot(root)) {
612 flog(LOG_ERR, "could not chroot to %s: %s", root, strerror(errno));
613 exit(1);
614 }
615 }
616 if(pwent) {
617 if(setgid(pwent->pw_gid)) {
618 flog(LOG_ERR, "could not switch group to %i: %s", (int)pwent->pw_gid, strerror(errno));
619 exit(1);
620 }
621 if(setuid(pwent->pw_uid)) {
622 flog(LOG_ERR, "could not switch user to %i: %s", (int)pwent->pw_uid, strerror(errno));
623 exit(1);
624 }
625 }
626 signal(SIGPIPE, SIG_IGN);
627 signal(SIGCHLD, SIG_IGN);
628 signal(SIGINT, sighandler);
629 signal(SIGTERM, sighandler);
630 if(daemonize) {
631 daemon(0, 0);
632 }
633 if(pidout != NULL) {
634 fprintf(pidout, "%i\n", getpid());
635 fclose(pidout);
636 }
637 d = 0;
638 while(!d) {
639 switch(ioloop()) {
640 case 0:
641 d = 1;
642 break;
643 case 1:
644 if(listeners.d > 0) {
645 while(listeners.d > 0)
646 resume(listeners.b[0], 0);
647 flog(LOG_INFO, "no longer listening");
648 } else {
649 d = 1;
650 }
651 break;
652 }
653 }
654 return(0);
655}