2 * Unix networking abstraction.
11 #include <sys/types.h>
12 #include <sys/socket.h>
13 #include <sys/ioctl.h>
14 #include <arpa/inet.h>
15 #include <netinet/in.h>
16 #include <netinet/tcp.h>
19 #define DEFINE_PLUG_METHOD_MACROS
25 struct socket_function_table *fn;
26 /* the above variable absolutely *must* be the first in this structure */
34 int frozen; /* this causes readability notifications to be ignored */
35 int frozen_readable; /* this means we missed at least one readability
36 * notification while we were frozen */
37 int localhost_only; /* for listening sockets */
40 int oobpending; /* is there OOB data available to read? */
42 int pending_error; /* in case send() returns error */
47 * We used to typedef struct Socket_tag *Socket.
49 * Since we have made the networking abstraction slightly more
50 * abstract, Socket no longer means a tcp socket (it could mean
51 * an ssl socket). So now we must use Actual_Socket when we know
52 * we are talking about a tcp socket.
54 typedef struct Socket_tag *Actual_Socket;
58 /* address family this belongs to, AF_INET for IPv4, AF_INET6 for IPv6. */
60 unsigned long address; /* Address IPv4 style. */
62 struct addrinfo *ai; /* Address IPv6 style. */
66 static tree234 *sktree;
68 static int cmpfortree(void *av, void *bv)
70 Actual_Socket a = (Actual_Socket) av, b = (Actual_Socket) bv;
71 int as = a->s, bs = b->s;
79 static int cmpforsearch(void *av, void *bv)
81 Actual_Socket b = (Actual_Socket) bv;
82 int as = (int) av, bs = b->s;
92 sktree = newtree234(cmpfortree);
101 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
107 char *error_string(int error)
111 return "Network error: Permission denied";
113 return "Network error: Address already in use";
115 return "Network error: Cannot assign requested address";
118 "Network error: Address family not supported by protocol family";
120 return "Network error: Operation already in progress";
122 return "Network error: Software caused connection abort";
124 return "Network error: Connection refused";
126 return "Network error: Connection reset by peer";
128 return "Network error: Destination address required";
130 return "Network error: Bad address";
132 return "Network error: Host is down";
134 return "Network error: No route to host";
136 return "Network error: Operation now in progress";
138 return "Network error: Interrupted function call";
140 return "Network error: Invalid argument";
142 return "Network error: Socket is already connected";
144 return "Network error: Too many open files";
146 return "Network error: Message too long";
148 return "Network error: Network is down";
150 return "Network error: Network dropped connection on reset";
152 return "Network error: Network is unreachable";
154 return "Network error: No buffer space available";
156 return "Network error: Bad protocol option";
158 return "Network error: Socket is not connected";
160 return "Network error: Socket operation on non-socket";
162 return "Network error: Operation not supported";
164 return "Network error: Protocol family not supported";
165 case EPROTONOSUPPORT:
166 return "Network error: Protocol not supported";
168 return "Network error: Protocol wrong type for socket";
170 return "Network error: Cannot send after socket shutdown";
171 case ESOCKTNOSUPPORT:
172 return "Network error: Socket type not supported";
174 return "Network error: Connection timed out";
176 return "Network error: Resource temporarily unavailable";
178 return "Unknown network error";
182 SockAddr sk_namelookup(char *host, char **canonicalname)
184 SockAddr ret = smalloc(sizeof(struct SockAddr_tag));
186 struct hostent *h = NULL;
189 /* Clear the structure and default to IPv4. */
190 memset(ret, 0, sizeof(struct SockAddr_tag));
191 ret->family = 0; /* We set this one when we have resolved the host. */
195 if ((a = inet_addr(host)) == (unsigned long) INADDR_NONE) {
197 if (getaddrinfo(host, NULL, NULL, &ret->ai) == 0) {
198 ret->family = ret->ai->ai_family;
203 * Otherwise use the IPv4-only gethostbyname... (NOTE:
204 * we don't use gethostbyname as a fallback!)
206 if (ret->family == 0) {
207 /*debug(("Resolving \"%s\" with gethostbyname() (IPv4 only)...\n", host)); */
208 if ( (h = gethostbyname(host)) )
209 ret->family = AF_INET;
211 if (ret->family == 0) {
212 ret->error = (h_errno == HOST_NOT_FOUND ||
213 h_errno == NO_DATA ||
214 h_errno == NO_ADDRESS ? "Host does not exist" :
215 h_errno == TRY_AGAIN ?
216 "Temporary name service failure" :
217 "gethostbyname: unknown error");
223 /* If we got an address info use that... */
226 /* Are we in IPv4 fallback mode? */
227 /* We put the IPv4 address into the a variable so we can further-on use the IPv4 code... */
228 if (ret->family == AF_INET)
230 (char *) &((struct sockaddr_in *) ret->ai->
231 ai_addr)->sin_addr, sizeof(a));
233 /* Now let's find that canonicalname... */
234 if (getnameinfo((struct sockaddr *) ret->ai->ai_addr,
236 AF_INET ? sizeof(struct sockaddr_in) :
237 sizeof(struct sockaddr_in6), realhost,
238 sizeof(realhost), NULL, 0, 0) != 0) {
239 strncpy(realhost, host, sizeof(realhost));
242 /* We used the IPv4-only gethostbyname()... */
246 memcpy(&a, h->h_addr, sizeof(a));
247 /* This way we are always sure the h->h_name is valid :) */
248 strncpy(realhost, h->h_name, sizeof(realhost));
252 * This must be a numeric IPv4 address because it caused a
253 * success return from inet_addr.
255 ret->family = AF_INET;
256 strncpy(realhost, host, sizeof(realhost));
258 ret->address = ntohl(a);
259 realhost[lenof(realhost)-1] = '\0';
260 *canonicalname = smalloc(1+strlen(realhost));
261 strcpy(*canonicalname, realhost);
265 void sk_getaddr(SockAddr addr, char *buf, int buflen)
268 if (addr->family == AF_INET) {
271 a.s_addr = htonl(addr->address);
272 strncpy(buf, inet_ntoa(a), buflen);
275 FIXME; /* I don't know how to get a text form of an IPv6 address. */
280 int sk_addrtype(SockAddr addr)
282 return (addr->family == AF_INET ? ADDRTYPE_IPV4 : ADDRTYPE_IPV6);
285 void sk_addrcopy(SockAddr addr, char *buf)
288 if (addr->family == AF_INET) {
291 a.s_addr = htonl(addr->address);
292 memcpy(buf, (char*) &a.s_addr, 4);
295 memcpy(buf, (char*) addr->ai, 16);
300 void sk_addr_free(SockAddr addr)
305 static Plug sk_tcp_plug(Socket sock, Plug p)
307 Actual_Socket s = (Actual_Socket) sock;
314 static void sk_tcp_flush(Socket s)
317 * We send data to the socket as soon as we can anyway,
318 * so we don't need to do anything here. :-)
322 static void sk_tcp_close(Socket s);
323 static int sk_tcp_write(Socket s, char *data, int len);
324 static int sk_tcp_write_oob(Socket s, char *data, int len);
325 static void sk_tcp_set_private_ptr(Socket s, void *ptr);
326 static void *sk_tcp_get_private_ptr(Socket s);
327 static void sk_tcp_set_frozen(Socket s, int is_frozen);
328 static char *sk_tcp_socket_error(Socket s);
330 Socket sk_register(void *sock, Plug plug)
332 static struct socket_function_table fn_table = {
338 sk_tcp_set_private_ptr,
339 sk_tcp_get_private_ptr,
347 * Create Socket structure.
349 ret = smalloc(sizeof(struct Socket_tag));
353 bufchain_init(&ret->output_data);
354 ret->writable = 1; /* to start with */
355 ret->sending_oob = 0;
357 ret->frozen_readable = 0;
358 ret->localhost_only = 0; /* unused, but best init anyway */
359 ret->pending_error = 0;
360 ret->oobpending = FALSE;
366 ret->error = error_string(errno);
377 Socket sk_new(SockAddr addr, int port, int privport, int oobinline,
378 int nodelay, Plug plug)
380 static struct socket_function_table fn_table = {
386 sk_tcp_set_private_ptr,
387 sk_tcp_get_private_ptr,
394 struct sockaddr_in6 a6;
396 struct sockaddr_in a;
402 * Create Socket structure.
404 ret = smalloc(sizeof(struct Socket_tag));
408 bufchain_init(&ret->output_data);
409 ret->connected = 0; /* to start with */
410 ret->writable = 0; /* to start with */
411 ret->sending_oob = 0;
413 ret->frozen_readable = 0;
414 ret->localhost_only = 0; /* unused, but best init anyway */
415 ret->pending_error = 0;
416 ret->oobpending = FALSE;
422 s = socket(addr->family, SOCK_STREAM, 0);
426 ret->error = error_string(errno);
430 ret->oobinline = oobinline;
433 setsockopt(s, SOL_SOCKET, SO_OOBINLINE, (void *) &b, sizeof(b));
438 setsockopt(s, IPPROTO_TCP, TCP_NODELAY, (void *) &b, sizeof(b));
442 * Bind to local address.
445 localport = 1023; /* count from 1023 downwards */
447 localport = 0; /* just use port 0 (ie kernel picks) */
449 /* Loop round trying to bind */
454 if (addr->family == AF_INET6) {
455 memset(&a6, 0, sizeof(a6));
456 a6.sin6_family = AF_INET6;
457 /*a6.sin6_addr = in6addr_any; *//* == 0 */
458 a6.sin6_port = htons(localport);
462 a.sin_family = AF_INET;
463 a.sin_addr.s_addr = htonl(INADDR_ANY);
464 a.sin_port = htons(localport);
467 retcode = bind(s, (addr->family == AF_INET6 ?
468 (struct sockaddr *) &a6 :
469 (struct sockaddr *) &a),
471 AF_INET6 ? sizeof(a6) : sizeof(a)));
473 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
480 if (err != EADDRINUSE) /* failed, for a bad reason */
485 break; /* we're only looping once */
488 break; /* we might have got to the end */
492 ret->error = error_string(err);
497 * Connect to remote address.
500 if (addr->family == AF_INET6) {
501 memset(&a, 0, sizeof(a));
502 a6.sin6_family = AF_INET6;
503 a6.sin6_port = htons((short) port);
505 ((struct sockaddr_in6 *) addr->ai->ai_addr)->sin6_addr;
509 a.sin_family = AF_INET;
510 a.sin_addr.s_addr = htonl(addr->address);
511 a.sin_port = htons((short) port);
516 connect(s, ((addr->family == AF_INET6) ?
517 (struct sockaddr *) &a6 : (struct sockaddr *) &a),
518 (addr->family == AF_INET6) ? sizeof(a6) : sizeof(a))
520 connect(s, (struct sockaddr *) &a, sizeof(a))
524 * FIXME: We are prepared to receive EWOULDBLOCK here,
525 * because we might want the connection to be made
526 * asynchronously; but how do we actually arrange this in
529 if ( errno != EWOULDBLOCK ) {
530 ret->error = error_string(errno);
535 * If we _don't_ get EWOULDBLOCK, the connect has completed
536 * and we should set the socket as connected and writable.
547 Socket sk_newlistener(int port, Plug plug, int local_host_only)
549 static struct socket_function_table fn_table = {
555 sk_tcp_set_private_ptr,
556 sk_tcp_get_private_ptr,
563 struct sockaddr_in6 a6;
565 struct sockaddr_in a;
572 * Create Socket structure.
574 ret = smalloc(sizeof(struct Socket_tag));
578 bufchain_init(&ret->output_data);
579 ret->writable = 0; /* to start with */
580 ret->sending_oob = 0;
582 ret->frozen_readable = 0;
583 ret->localhost_only = local_host_only;
584 ret->pending_error = 0;
585 ret->oobpending = FALSE;
591 s = socket(AF_INET, SOCK_STREAM, 0);
595 ret->error = error_string(errno);
601 setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (const char *)&on, sizeof(on));
604 if (addr->family == AF_INET6) {
605 memset(&a6, 0, sizeof(a6));
606 a6.sin6_family = AF_INET6;
608 a6.sin6_addr = in6addr_loopback;
610 a6.sin6_addr = in6addr_any;
611 a6.sin6_port = htons(port);
615 a.sin_family = AF_INET;
617 a.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
619 a.sin_addr.s_addr = htonl(INADDR_ANY);
620 a.sin_port = htons((short)port);
623 retcode = bind(s, (addr->family == AF_INET6 ?
624 (struct sockaddr *) &a6 :
625 (struct sockaddr *) &a),
627 AF_INET6 ? sizeof(a6) : sizeof(a)));
629 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
638 ret->error = error_string(err);
643 if (listen(s, SOMAXCONN) < 0) {
645 ret->error = error_string(errno);
654 static void sk_tcp_close(Socket sock)
656 Actual_Socket s = (Actual_Socket) sock;
664 * The function which tries to send on a socket once it's deemed
667 void try_send(Actual_Socket s)
669 while (s->sending_oob || bufchain_size(&s->output_data) > 0) {
675 if (s->sending_oob) {
676 urgentflag = MSG_OOB;
677 len = s->sending_oob;
681 bufchain_prefix(&s->output_data, &data, &len);
683 nsent = send(s->s, data, len, urgentflag);
684 noise_ultralight(nsent);
686 err = (nsent < 0 ? errno : 0);
687 if (err == EWOULDBLOCK) {
689 * Perfectly normal: we've sent all we can for the moment.
693 } else if (nsent == 0 ||
694 err == ECONNABORTED || err == ECONNRESET) {
696 * If send() returns CONNABORTED or CONNRESET, we
697 * unfortunately can't just call plug_closing(),
698 * because it's quite likely that we're currently
699 * _in_ a call from the code we'd be calling back
700 * to, so we'd have to make half the SSH code
701 * reentrant. Instead we flag a pending error on
702 * the socket, to be dealt with (by calling
703 * plug_closing()) at some suitable future moment.
705 s->pending_error = err;
708 /* We're inside the Unix frontend here, so we know
709 * that the frontend handle is unnecessary. */
710 logevent(NULL, error_string(err));
711 fatalbox("%s", error_string(err));
714 if (s->sending_oob) {
716 memmove(s->oobdata, s->oobdata+nsent, len-nsent);
717 s->sending_oob = len - nsent;
722 bufchain_consume(&s->output_data, nsent);
728 static int sk_tcp_write(Socket sock, char *buf, int len)
730 Actual_Socket s = (Actual_Socket) sock;
733 * Add the data to the buffer list on the socket.
735 bufchain_add(&s->output_data, buf, len);
738 * Now try sending from the start of the buffer list.
743 return bufchain_size(&s->output_data);
746 static int sk_tcp_write_oob(Socket sock, char *buf, int len)
748 Actual_Socket s = (Actual_Socket) sock;
751 * Replace the buffer list on the socket with the data.
753 bufchain_clear(&s->output_data);
754 assert(len <= sizeof(s->oobdata));
755 memcpy(s->oobdata, buf, len);
756 s->sending_oob = len;
759 * Now try sending from the start of the buffer list.
764 return s->sending_oob;
767 int select_result(int fd, int event)
771 char buf[20480]; /* nice big buffer for plenty of speed */
775 /* Find the Socket structure */
776 s = find234(sktree, (void *) fd, cmpforsearch);
778 return 1; /* boggle */
780 noise_ultralight(event);
783 #ifdef FIXME_NONBLOCKING_CONNECTIONS
784 case FIXME: /* connected */
785 s->connected = s->writable = 1;
788 case 4: /* exceptional */
791 * On a non-oobinline socket, this indicates that we
792 * can immediately perform an OOB read and get back OOB
793 * data, which we will send to the back end with
794 * type==2 (urgent data).
796 ret = recv(s->s, buf, sizeof(buf), MSG_OOB);
797 noise_ultralight(ret);
799 char *str = (ret == 0 ? "Internal networking trouble" :
800 error_string(errno));
801 /* We're inside the Unix frontend here, so we know
802 * that the frontend handle is unnecessary. */
806 return plug_receive(s->plug, 2, buf, ret);
812 * If we reach here, this is an oobinline socket, which
813 * means we should set s->oobpending and then deal with it
814 * when we get called for the readability event (which
815 * should also occur).
817 s->oobpending = TRUE;
819 case 1: /* readable; also acceptance */
822 * On a listening socket, the readability event means a
823 * connection is ready to be accepted.
825 struct sockaddr_in isa;
826 int addrlen = sizeof(struct sockaddr_in);
827 int t; /* socket of connection */
829 memset(&isa, 0, sizeof(struct sockaddr_in));
831 t = accept(s->s,(struct sockaddr *)&isa,&addrlen);
836 if (s->localhost_only &&
837 ntohl(isa.sin_addr.s_addr) != INADDR_LOOPBACK) {
838 close(t); /* someone let nonlocal through?! */
839 } else if (plug_accepting(s->plug, (void*)t)) {
840 close(t); /* denied or error */
846 * If we reach here, this is not a listening socket, so
847 * readability really means readability.
850 /* In the case the socket is still frozen, we don't even bother */
852 s->frozen_readable = 1;
857 * We have received data on the socket. For an oobinline
858 * socket, this might be data _before_ an urgent pointer,
859 * in which case we send it to the back end with type==1
860 * (data prior to urgent).
862 if (s->oobinline && s->oobpending) {
864 if (ioctl(s->s, SIOCATMARK, &atmark) == 0 && atmark)
865 s->oobpending = FALSE; /* clear this indicator */
869 ret = recv(s->s, buf, s->oobpending ? 1 : sizeof(buf), 0);
870 noise_ultralight(ret);
872 if (errno == EWOULDBLOCK) {
877 return plug_closing(s->plug, error_string(errno), errno, 0);
878 } else if (0 == ret) {
879 return plug_closing(s->plug, NULL, 0, 0);
881 return plug_receive(s->plug, atmark ? 0 : 1, buf, ret);
884 case 2: /* writable */
886 int bufsize_before, bufsize_after;
888 bufsize_before = s->sending_oob + bufchain_size(&s->output_data);
890 bufsize_after = s->sending_oob + bufchain_size(&s->output_data);
891 if (bufsize_after < bufsize_before)
892 plug_sent(s->plug, bufsize_after);
901 * Deal with socket errors detected in try_send().
903 void net_pending_errors(void)
909 * This might be a fiddly business, because it's just possible
910 * that handling a pending error on one socket might cause
911 * others to be closed. (I can't think of any reason this might
912 * happen in current SSH implementation, but to maintain
913 * generality of this network layer I'll assume the worst.)
915 * So what we'll do is search the socket list for _one_ socket
916 * with a pending error, and then handle it, and then search
917 * the list again _from the beginning_. Repeat until we make a
918 * pass with no socket errors present. That way we are
919 * protected against the socket list changing under our feet.
923 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
924 if (s->pending_error) {
926 * An error has occurred on this socket. Pass it to the
929 plug_closing(s->plug, error_string(s->pending_error),
930 s->pending_error, 0);
938 * Each socket abstraction contains a `void *' private field in
939 * which the client can keep state.
941 static void sk_tcp_set_private_ptr(Socket sock, void *ptr)
943 Actual_Socket s = (Actual_Socket) sock;
944 s->private_ptr = ptr;
947 static void *sk_tcp_get_private_ptr(Socket sock)
949 Actual_Socket s = (Actual_Socket) sock;
950 return s->private_ptr;
954 * Special error values are returned from sk_namelookup and sk_new
955 * if there's a problem. These functions extract an error message,
956 * or return NULL if there's no problem.
958 char *sk_addr_error(SockAddr addr)
962 static char *sk_tcp_socket_error(Socket sock)
964 Actual_Socket s = (Actual_Socket) sock;
968 static void sk_tcp_set_frozen(Socket sock, int is_frozen)
970 Actual_Socket s = (Actual_Socket) sock;
971 if (s->frozen == is_frozen)
973 s->frozen = is_frozen;
974 if (!is_frozen && s->frozen_readable) {
976 recv(s->s, &c, 1, MSG_PEEK);
978 s->frozen_readable = 0;
982 * For Unix select()-based frontends: enumerate all sockets
983 * currently active, and state whether we currently wish to receive
984 * select events on them for reading, writing and exceptional
987 static void set_rwx(Actual_Socket s, int *rwx)
990 if (s->connected && !s->frozen)
991 val |= 1 | 4; /* read, except */
992 if (bufchain_size(&s->output_data))
993 val |= 2; /* write */
995 val |= 1; /* read == accept */
999 int first_socket(int *state, int *rwx)
1003 s = index234(sktree, (*state)++);
1006 return s ? s->s : -1;
1009 int next_socket(int *state, int *rwx)
1011 Actual_Socket s = index234(sktree, (*state)++);
1014 return s ? s->s : -1;
1017 int net_service_lookup(char *service)
1020 se = getservbyname(service, NULL);
1022 return ntohs(se->s_port);