2 * Copyright (c) 2020 Omar Polo <op@omarpolo.com>
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.
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.
18 #include <sys/socket.h>
21 #include <arpa/inet.h>
22 #include <netinet/in.h>
37 # define pledge(a, b) 0
38 # define unveil(a, b) 0
39 #endif /* __OpenBSD__ */
45 #define GEMINI_URL_LEN (1024+3) /* URL max len + \r\n + \0 */
47 /* large enough to hold a copy of a gemini URL and still have extra room */
51 #define TEMP_FAILURE 40
53 #define BAD_REQUEST 59
71 int fd, waiting_on_child;
73 char sbuf[1024]; /* static buffer */
74 void *buf, *i; /* mmap buffer */
75 ssize_t len, off; /* mmap/static buffer */
87 struct etm { /* file extension to mime */
91 {"application/pdf", "pdf"},
94 {"image/jpeg", "jpg"},
95 {"image/jpeg", "jpeg"},
97 {"image/svg+xml", "svg"},
99 {"text/gemini", "gemini"},
100 {"text/gemini", "gmi"},
101 {"text/markdown", "markdown"},
102 {"text/markdown", "md"},
103 {"text/plain", "txt"},
109 #define LOG(c, fmt, ...) \
111 char buf[INET_ADDRSTRLEN]; \
112 if (inet_ntop((c)->af, &(c)->addr, buf, sizeof(buf)) == NULL) \
113 err(1, "inet_ntop"); \
114 dprintf(logfd, "[%s] " fmt "\n", buf, __VA_ARGS__); \
117 const char *dir, *cgi;
120 int connected_clients;
122 void siginfo_handler(int);
123 int starts_with(const char*, const char*);
125 char *url_after_proto(char*);
126 char *url_start_of_request(char*);
127 int url_trim(struct client*, char*);
128 char *adjust_path(char*);
129 ssize_t filesize(int);
131 int start_reply(struct pollfd*, struct client*, int, const char*);
132 const char *path_ext(const char*);
133 const char *mime(const char*);
134 int check_path(const char*, int*);
135 int check_for_cgi(char *, char*, struct pollfd*, struct client*);
136 int open_file(char*, char*, struct pollfd*, struct client*);
137 int start_cgi(const char*, const char*, const char*, struct pollfd*, struct client*);
138 void cgi_setpoll_on_child(struct pollfd*, struct client*);
139 void cgi_setpoll_on_client(struct pollfd*, struct client*);
140 void handle_cgi(struct pollfd*, struct client*);
141 void send_file(char*, char*, struct pollfd*, struct client*);
142 void send_dir(char*, struct pollfd*, struct client*);
143 void handle(struct pollfd*, struct client*);
145 void mark_nonblock(int);
147 void do_accept(int, struct tls*, struct pollfd*, struct client*);
148 void goodbye(struct pollfd*, struct client*);
149 void loop(struct tls*, int);
151 void usage(const char*);
154 siginfo_handler(int sig)
160 starts_with(const char *str, const char *prefix)
164 for (i = 0; prefix[i] != '\0'; ++i)
165 if (str[i] != prefix[i])
171 url_after_proto(char *url)
174 const char *proto = "gemini";
175 const char *marker = "//";
178 if ((s = strstr(url, marker)) == NULL)
182 * if a protocol is not specified, gemini should be implied:
183 * this handles the case of //example.com
186 return s + strlen(marker);
188 if (s - strlen(proto) != url)
191 if (!starts_with(url, proto))
194 return s + strlen(marker);
198 url_start_of_request(char *url)
202 if ((s = url_after_proto(url)) == NULL)
205 /* non-absolute URL */
209 if ((t = strstr(s, "/")) == NULL)
210 return s + strlen(s);
215 url_trim(struct client *c, char *url)
217 const char *e = "\r\n";
220 if ((s = strstr(url, e)) == NULL)
226 LOG(c, "%s", "request longer than 1024 bytes\n");
234 adjust_path(char *path)
239 if ((query = strchr(path, '?')) != NULL) {
247 if (!strcmp(&path[len-3], "/..")) {
252 /* if the path is only `..` trim out and exit */
253 if (!strcmp(path, "..")) {
258 /* remove every ../ in the path */
260 if ((s = strstr(path, "../")) == NULL)
262 memmove(s, s+3, strlen(s)+1); /* copy also the \0 */
267 start_reply(struct pollfd *pfd, struct client *client, int code, const char *reason)
269 char buf[1030] = {0}; /* status + ' ' + max reply len + \r\n\0 */
274 client->meta = reason;
275 client->state = S_INITIALIZING;
277 len = snprintf(buf, sizeof(buf), "%d %s\r\n", code, reason);
278 assert(len < (int)sizeof(buf));
279 ret = tls_write(client->ctx, buf, len);
280 if (ret == TLS_WANT_POLLIN) {
281 pfd->events = POLLIN;
285 if (ret == TLS_WANT_POLLOUT) {
286 pfd->events = POLLOUT;
298 if ((len = lseek(fd, 0, SEEK_END)) == -1)
300 if (lseek(fd, 0, SEEK_SET) == -1)
306 path_ext(const char *path)
310 end = path + strlen(path)-1; /* the last byte before the NUL */
311 for (; end != path; --end) {
322 mime(const char *path)
324 const char *ext, *def = "application/octet-stream";
327 if ((ext = path_ext(path)) == NULL)
330 for (t = filetypes; t->mime != NULL; ++t)
331 if (!strcmp(ext, t->ext))
338 check_path(const char *path, int *fd)
342 assert(path != NULL);
343 if ((*fd = openat(dirfd, path,
344 O_RDONLY | O_NOFOLLOW | O_CLOEXEC)) == -1) {
348 if (fstat(*fd, &sb) == -1) {
349 dprintf(logfd, "failed stat for %s\n", path);
353 if (S_ISDIR(sb.st_mode))
354 return FILE_DIRECTORY;
356 if (sb.st_mode & S_IXUSR)
357 return FILE_EXECUTABLE;
363 * the inverse of this algorithm, i.e. starting from the start of the
364 * path + strlen(cgi), and checking if each component, should be
365 * faster. But it's tedious to write. This does the opposite: starts
366 * from the end and strip one component at a time, until either an
367 * executable is found or we emptied the path.
370 check_for_cgi(char *path, char *query, struct pollfd *fds, struct client *c)
373 end = strchr(path, '\0');
375 /* NB: assume CGI is enabled and path matches cgi */
378 /* go up one level. UNIX paths are simple and POSIX
379 * dirname, with its ambiguities on if the given path
380 * is changed or not, gives me headaches. */
385 switch (check_path(path, &c->fd)) {
386 case FILE_EXECUTABLE:
387 return start_cgi(path, end+1, query, fds,c);
399 if (!start_reply(fds, c, NOT_FOUND, "not found"))
407 open_file(char *path, char *query, struct pollfd *fds, struct client *c)
411 bzero(fpath, sizeof(fpath));
415 strlcat(fpath, path, PATHBUF);
417 switch (check_path(fpath, &c->fd)) {
418 case FILE_EXECUTABLE:
419 /* +2 to skip the ./ */
420 if (cgi != NULL && starts_with(fpath+2, cgi))
421 return start_cgi(fpath, "", query, fds, c);
426 if ((c->len = filesize(c->fd)) == -1) {
427 LOG(c, "failed to get file size for %s", fpath);
432 if ((c->buf = mmap(NULL, c->len, PROT_READ, MAP_PRIVATE,
433 c->fd, 0)) == MAP_FAILED) {
434 warn("mmap: %s", fpath);
439 return start_reply(fds, c, SUCCESS, mime(fpath));
442 LOG(c, "%s is a directory, trying %s/index.gmi", fpath, fpath);
445 send_dir(fpath, fds, c);
449 if (cgi != NULL && starts_with(fpath+2, cgi))
450 return check_for_cgi(fpath, query, fds, c);
452 if (!start_reply(fds, c, NOT_FOUND, "not found"))
464 start_cgi(const char *spath, const char *relpath, const char *query,
465 struct pollfd *fds, struct client *c)
468 int p[2]; /* read end, write end */
473 switch (pid = fork()) {
477 case 0: { /* child */
478 char *ex, *requri, *portno;
479 char addr[INET_ADDRSTRLEN];
480 char *argv[] = { NULL, NULL, NULL };
485 if (dup2(p[1], 1) == -1)
488 if (inet_ntop(c->af, &c->addr, addr, sizeof(addr)) == NULL)
491 if (asprintf(&portno, "%d", port) == -1)
494 if (asprintf(&ex, "%s%s", dir, spath+1) == -1)
497 if (asprintf(&requri, "%s%s%s", spath,
498 *relpath == '\0' ? "" : "/",
502 argv[0] = argv[1] = ex;
505 setenv("SERVER_SOFTWARE", "gmid", 1);
506 setenv("SERVER_PORT", portno, 1);
507 /* setenv("SERVER_NAME", "", 1); */
508 setenv("SCRIPT_NAME", spath, 1);
509 setenv("SCRIPT_EXECUTABLE", ex, 1);
510 setenv("REQUEST_URI", requri, 1);
511 setenv("REQUEST_RELATIVE", relpath, 1);
513 setenv("QUERY_STRING", query, 1);
514 setenv("REMOTE_HOST", addr, 1);
515 setenv("DOCUMENT_ROOT", dir, 1);
521 default: /* parent */
526 mark_nonblock(c->fd);
527 c->state = S_SENDING;
533 if (!start_reply(fds, c, TEMP_FAILURE, "internal server error"))
539 dprintf(p[1], "%d internal server error\r\n", TEMP_FAILURE);
545 cgi_setpoll_on_child(struct pollfd *fds, struct client *c)
549 if (c->waiting_on_child)
551 c->waiting_on_child = 1;
553 fds->events = POLLIN;
561 cgi_setpoll_on_client(struct pollfd *fds, struct client *c)
565 if (!c->waiting_on_child)
567 c->waiting_on_child = 0;
575 handle_cgi(struct pollfd *fds, struct client *c)
579 /* ensure c->fd is the child and fds->fd the client */
580 cgi_setpoll_on_client(fds, c);
584 if ((r = read(c->fd, c->sbuf, sizeof(c->sbuf))) == 0)
587 if (errno == EAGAIN || errno == EWOULDBLOCK) {
588 cgi_setpoll_on_child(fds, c);
598 switch (r = tls_write(c->ctx, c->sbuf + c->off, c->len)) {
602 case TLS_WANT_POLLOUT:
603 fds->events = POLLOUT;
606 case TLS_WANT_POLLIN:
607 fds->events = POLLIN;
623 send_file(char *path, char *query, struct pollfd *fds, struct client *c)
628 if (!open_file(path, query, fds, c))
630 c->state = S_SENDING;
633 len = (c->buf + c->len) - c->i;
636 switch (ret = tls_write(c->ctx, c->i, len)) {
638 LOG(c, "tls_write: %s", tls_error(c->ctx));
642 case TLS_WANT_POLLIN:
643 fds->events = POLLIN;
646 case TLS_WANT_POLLOUT:
647 fds->events = POLLOUT;
661 send_dir(char *path, struct pollfd *fds, struct client *client)
666 bzero(fpath, PATHBUF);
671 /* this cannot fail since sizeof(fpath) > maxlen of path */
672 strlcat(fpath, path, PATHBUF);
675 /* add a trailing / in case. */
676 if (fpath[len-1] != '/') {
680 strlcat(fpath, "index.gmi", sizeof(fpath));
682 send_file(fpath, NULL, fds, client);
686 handle(struct pollfd *fds, struct client *client)
688 char buf[GEMINI_URL_LEN];
692 switch (client->state) {
694 bzero(buf, GEMINI_URL_LEN);
695 switch (tls_read(client->ctx, buf, sizeof(buf)-1)) {
697 LOG(client, "tls_read: %s", tls_error(client->ctx));
698 goodbye(fds, client);
701 case TLS_WANT_POLLIN:
702 fds->events = POLLIN;
705 case TLS_WANT_POLLOUT:
706 fds->events = POLLOUT;
710 if (!url_trim(client, buf)) {
711 if (!start_reply(fds, client, BAD_REQUEST, "bad request"))
713 goodbye(fds, client);
717 if ((path = url_start_of_request(buf)) == NULL) {
718 if (!start_reply(fds, client, BAD_REQUEST, "bad request"))
720 goodbye(fds, client);
724 query = adjust_path(path);
725 LOG(client, "get %s%s%s", path,
729 send_file(path, query, fds, client);
733 if (!start_reply(fds, client, client->code, client->meta))
736 if (client->code != SUCCESS) {
737 /* we don't need a body */
738 goodbye(fds, client);
742 client->state = S_SENDING;
747 if (client->child != -1)
748 handle_cgi(fds, client);
750 send_file(NULL, NULL, fds, client);
754 goodbye(fds, client);
764 mark_nonblock(int fd)
768 if ((flags = fcntl(fd, F_GETFL)) == -1)
769 err(1, "fcntl(F_GETFL)");
770 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1)
771 err(1, "fcntl(F_SETFL)");
775 make_socket(int port, int family)
778 struct sockaddr_in addr4;
779 struct sockaddr_in6 addr6;
780 struct sockaddr *addr;
785 bzero(&addr4, sizeof(addr4));
786 addr4.sin_family = family;
787 addr4.sin_port = htons(port);
788 addr4.sin_addr.s_addr = INADDR_ANY;
789 addr = (struct sockaddr*)&addr4;
794 bzero(&addr6, sizeof(addr6));
795 addr6.sin6_family = AF_INET6;
796 addr6.sin6_port = htons(port);
797 addr6.sin6_addr = in6addr_any;
798 addr = (struct sockaddr*)&addr6;
807 if ((sock = socket(family, SOCK_STREAM, 0)) == -1)
811 if (setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &v, sizeof(v)) == -1)
812 err(1, "setsockopt(SO_REUSEADDR)");
815 if (setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, &v, sizeof(v)) == -1)
816 err(1, "setsockopt(SO_REUSEPORT)");
820 if (bind(sock, addr, len) == -1)
823 if (listen(sock, 16) == -1)
830 do_accept(int sock, struct tls *ctx, struct pollfd *fds, struct client *clients)
833 struct sockaddr_in addr;
837 if ((fd = accept(sock, (struct sockaddr*)&addr, &len)) == -1) {
838 if (errno == EWOULDBLOCK)
845 for (i = 0; i < MAX_USERS; ++i) {
846 if (fds[i].fd == -1) {
847 bzero(&clients[i], sizeof(struct client));
848 if (tls_accept_socket(ctx, &clients[i].ctx, fd) == -1)
849 break; /* goodbye fd! */
852 fds[i].events = POLLIN;
854 clients[i].state = S_OPEN;
856 clients[i].child = -1;
857 clients[i].buf = MAP_FAILED;
858 clients[i].af = AF_INET;
859 clients[i].addr = addr.sin_addr;
870 goodbye(struct pollfd *pfd, struct client *c)
874 c->state = S_CLOSING;
876 ret = tls_close(c->ctx);
877 if (ret == TLS_WANT_POLLIN) {
878 pfd->events = POLLIN;
881 if (ret == TLS_WANT_POLLOUT) {
882 pfd->events = POLLOUT;
891 if (c->buf != MAP_FAILED)
892 munmap(c->buf, c->len);
902 loop(struct tls *ctx, int sock)
905 struct client clients[MAX_USERS];
906 struct pollfd fds[MAX_USERS];
908 for (i = 0; i < MAX_USERS; ++i) {
910 fds[i].events = POLLIN;
911 bzero(&clients[i], sizeof(struct client));
917 if ((todo = poll(fds, MAX_USERS, INFTIM)) == -1) {
918 if (errno == EINTR) {
919 warnx("connected clients: %d", connected_clients);
925 for (i = 0; i < MAX_USERS; i++) {
926 assert(i < MAX_USERS);
928 if (fds[i].revents == 0)
931 if (fds[i].revents & (POLLERR|POLLNVAL))
932 err(1, "bad fd %d", fds[i].fd);
934 if (fds[i].revents & POLLHUP) {
935 /* fds[i] may be the fd of the stdin
936 * of a cgi script that has exited. */
937 if (!clients[i].waiting_on_child) {
938 goodbye(&fds[i], &clients[i]);
945 if (i == 0) { /* new client */
946 do_accept(sock, ctx, fds, clients);
950 handle(&fds[i], &clients[i]);
956 usage(const char *me)
959 "USAGE: %s [-h] [-c cert.pem] [-d docs] [-k key.pem] "
960 "[-l logfile] [-p port] [-x cgi-bin]\n",
965 main(int argc, char **argv)
967 const char *cert = "cert.pem", *key = "key.pem";
968 struct tls *ctx = NULL;
969 struct tls_config *conf;
972 signal(SIGPIPE, SIG_IGN);
973 signal(SIGCHLD, SIG_IGN);
976 signal(SIGINFO, siginfo_handler);
978 signal(SIGUSR2, siginfo_handler);
980 connected_clients = 0;
983 logfd = 2; /* stderr */
987 while ((ch = getopt(argc, argv, "c:d:hk:l:p:x:")) != -1) {
1006 /* open log file or create it with 644 */
1007 if ((logfd = open(optarg, O_WRONLY | O_CREAT | O_CLOEXEC,
1008 S_IRUSR | S_IWUSR | S_IRGRP | S_IWOTH)) == -1)
1009 err(1, "%s", optarg);
1017 lval = strtol(optarg, &ep, 10);
1018 if (optarg[0] == '\0' || *ep != '\0')
1019 err(1, "not a number: %s", optarg);
1020 if (lval < 0 || lval > UINT16_MAX)
1021 err(1, "port number out of range: %s", optarg);
1036 if ((conf = tls_config_new()) == NULL)
1037 err(1, "tls_config_new");
1039 if (tls_config_set_protocols(conf,
1040 TLS_PROTOCOL_TLSv1_2 | TLS_PROTOCOL_TLSv1_3) == -1)
1041 err(1, "tls_config_set_protocols");
1043 if (tls_config_set_cert_file(conf, cert) == -1)
1044 err(1, "tls_config_set_cert_file: %s", cert);
1046 if (tls_config_set_key_file(conf, key) == -1)
1047 err(1, "tls_config_set_key_file: %s", key);
1049 if ((ctx = tls_server()) == NULL)
1050 err(1, "tls_server");
1052 if (tls_configure(ctx, conf) == -1)
1053 errx(1, "tls_configure: %s", tls_error(ctx));
1055 sock = make_socket(port, AF_INET);
1057 if ((dirfd = open(dir, O_RDONLY | O_DIRECTORY)) == -1)
1058 err(1, "open: %s", dir);
1061 if (unveil(dir, "rx") == -1)
1063 if (pledge("stdio rpath inet proc exec", NULL) == -1)
1066 if (unveil(dir, "r") == -1)
1068 if (pledge("stdio rpath inet", NULL) == -1)
1076 tls_config_free(conf);