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
24 #define ipv4_is_loopback(addr) (inet_netof(addr) == IN_LOOPBACKNET)
27 struct socket_function_table *fn;
28 /* the above variable absolutely *must* be the first in this structure */
36 int frozen; /* this causes readability notifications to be ignored */
37 int frozen_readable; /* this means we missed at least one readability
38 * notification while we were frozen */
39 int localhost_only; /* for listening sockets */
42 int oobpending; /* is there OOB data available to read? */
44 int pending_error; /* in case send() returns error */
49 * We used to typedef struct Socket_tag *Socket.
51 * Since we have made the networking abstraction slightly more
52 * abstract, Socket no longer means a tcp socket (it could mean
53 * an ssl socket). So now we must use Actual_Socket when we know
54 * we are talking about a tcp socket.
56 typedef struct Socket_tag *Actual_Socket;
60 /* address family this belongs to, AF_INET for IPv4, AF_INET6 for IPv6. */
62 unsigned long address; /* Address IPv4 style. */
64 struct addrinfo *ai; /* Address IPv6 style. */
68 static tree234 *sktree;
70 static int cmpfortree(void *av, void *bv)
72 Actual_Socket a = (Actual_Socket) av, b = (Actual_Socket) bv;
73 int as = a->s, bs = b->s;
81 static int cmpforsearch(void *av, void *bv)
83 Actual_Socket b = (Actual_Socket) bv;
84 int as = (int) av, bs = b->s;
94 sktree = newtree234(cmpfortree);
103 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
109 char *error_string(int error)
111 return strerror(error);
114 SockAddr sk_namelookup(char *host, char **canonicalname)
116 SockAddr ret = smalloc(sizeof(struct SockAddr_tag));
118 struct hostent *h = NULL;
121 /* Clear the structure and default to IPv4. */
122 memset(ret, 0, sizeof(struct SockAddr_tag));
123 ret->family = 0; /* We set this one when we have resolved the host. */
127 if ((a = inet_addr(host)) == (unsigned long) INADDR_NONE) {
129 if (getaddrinfo(host, NULL, NULL, &ret->ai) == 0) {
130 ret->family = ret->ai->ai_family;
135 * Otherwise use the IPv4-only gethostbyname... (NOTE:
136 * we don't use gethostbyname as a fallback!)
138 if (ret->family == 0) {
139 /*debug(("Resolving \"%s\" with gethostbyname() (IPv4 only)...\n", host)); */
140 if ( (h = gethostbyname(host)) )
141 ret->family = AF_INET;
143 if (ret->family == 0) {
144 ret->error = (h_errno == HOST_NOT_FOUND ||
145 h_errno == NO_DATA ||
146 h_errno == NO_ADDRESS ? "Host does not exist" :
147 h_errno == TRY_AGAIN ?
148 "Temporary name service failure" :
149 "gethostbyname: unknown error");
155 /* If we got an address info use that... */
158 /* Are we in IPv4 fallback mode? */
159 /* We put the IPv4 address into the a variable so we can further-on use the IPv4 code... */
160 if (ret->family == AF_INET)
162 (char *) &((struct sockaddr_in *) ret->ai->
163 ai_addr)->sin_addr, sizeof(a));
165 /* Now let's find that canonicalname... */
166 if (getnameinfo((struct sockaddr *) ret->ai->ai_addr,
168 AF_INET ? sizeof(struct sockaddr_in) :
169 sizeof(struct sockaddr_in6), realhost,
170 sizeof(realhost), NULL, 0, 0) != 0) {
171 strncpy(realhost, host, sizeof(realhost));
174 /* We used the IPv4-only gethostbyname()... */
178 memcpy(&a, h->h_addr, sizeof(a));
179 /* This way we are always sure the h->h_name is valid :) */
180 strncpy(realhost, h->h_name, sizeof(realhost));
184 * This must be a numeric IPv4 address because it caused a
185 * success return from inet_addr.
187 ret->family = AF_INET;
188 strncpy(realhost, host, sizeof(realhost));
190 ret->address = ntohl(a);
191 realhost[lenof(realhost)-1] = '\0';
192 *canonicalname = smalloc(1+strlen(realhost));
193 strcpy(*canonicalname, realhost);
197 void sk_getaddr(SockAddr addr, char *buf, int buflen)
200 if (addr->family == AF_INET) {
203 a.s_addr = htonl(addr->address);
204 strncpy(buf, inet_ntoa(a), buflen);
207 FIXME; /* I don't know how to get a text form of an IPv6 address. */
212 int sk_hostname_is_local(char *name)
214 return !strcmp(name, "localhost");
217 int sk_address_is_local(SockAddr addr)
220 if (addr->family == AF_INET) {
223 a.s_addr = htonl(addr->address);
224 return ipv4_is_loopback(a);
227 FIXME; /* someone who can compile for IPV6 had better do this bit */
232 int sk_addrtype(SockAddr addr)
234 return (addr->family == AF_INET ? ADDRTYPE_IPV4 : ADDRTYPE_IPV6);
237 void sk_addrcopy(SockAddr addr, char *buf)
240 if (addr->family == AF_INET) {
243 a.s_addr = htonl(addr->address);
244 memcpy(buf, (char*) &a.s_addr, 4);
247 memcpy(buf, (char*) addr->ai, 16);
252 void sk_addr_free(SockAddr addr)
257 static Plug sk_tcp_plug(Socket sock, Plug p)
259 Actual_Socket s = (Actual_Socket) sock;
266 static void sk_tcp_flush(Socket s)
269 * We send data to the socket as soon as we can anyway,
270 * so we don't need to do anything here. :-)
274 static void sk_tcp_close(Socket s);
275 static int sk_tcp_write(Socket s, char *data, int len);
276 static int sk_tcp_write_oob(Socket s, char *data, int len);
277 static void sk_tcp_set_private_ptr(Socket s, void *ptr);
278 static void *sk_tcp_get_private_ptr(Socket s);
279 static void sk_tcp_set_frozen(Socket s, int is_frozen);
280 static char *sk_tcp_socket_error(Socket s);
282 Socket sk_register(void *sock, Plug plug)
284 static struct socket_function_table fn_table = {
290 sk_tcp_set_private_ptr,
291 sk_tcp_get_private_ptr,
299 * Create Socket structure.
301 ret = smalloc(sizeof(struct Socket_tag));
305 bufchain_init(&ret->output_data);
306 ret->writable = 1; /* to start with */
307 ret->sending_oob = 0;
309 ret->frozen_readable = 0;
310 ret->localhost_only = 0; /* unused, but best init anyway */
311 ret->pending_error = 0;
312 ret->oobpending = FALSE;
318 ret->error = error_string(errno);
329 Socket sk_new(SockAddr addr, int port, int privport, int oobinline,
330 int nodelay, Plug plug)
332 static struct socket_function_table fn_table = {
338 sk_tcp_set_private_ptr,
339 sk_tcp_get_private_ptr,
346 struct sockaddr_in6 a6;
348 struct sockaddr_in a;
354 * Create Socket structure.
356 ret = smalloc(sizeof(struct Socket_tag));
360 bufchain_init(&ret->output_data);
361 ret->connected = 0; /* to start with */
362 ret->writable = 0; /* to start with */
363 ret->sending_oob = 0;
365 ret->frozen_readable = 0;
366 ret->localhost_only = 0; /* unused, but best init anyway */
367 ret->pending_error = 0;
368 ret->oobpending = FALSE;
374 s = socket(addr->family, SOCK_STREAM, 0);
378 ret->error = error_string(errno);
382 ret->oobinline = oobinline;
385 setsockopt(s, SOL_SOCKET, SO_OOBINLINE, (void *) &b, sizeof(b));
390 setsockopt(s, IPPROTO_TCP, TCP_NODELAY, (void *) &b, sizeof(b));
394 * Bind to local address.
397 localport = 1023; /* count from 1023 downwards */
399 localport = 0; /* just use port 0 (ie kernel picks) */
401 /* Loop round trying to bind */
406 if (addr->family == AF_INET6) {
407 memset(&a6, 0, sizeof(a6));
408 a6.sin6_family = AF_INET6;
409 /*a6.sin6_addr = in6addr_any; *//* == 0 */
410 a6.sin6_port = htons(localport);
414 a.sin_family = AF_INET;
415 a.sin_addr.s_addr = htonl(INADDR_ANY);
416 a.sin_port = htons(localport);
419 retcode = bind(s, (addr->family == AF_INET6 ?
420 (struct sockaddr *) &a6 :
421 (struct sockaddr *) &a),
423 AF_INET6 ? sizeof(a6) : sizeof(a)));
425 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
432 if (err != EADDRINUSE) /* failed, for a bad reason */
437 break; /* we're only looping once */
440 break; /* we might have got to the end */
444 ret->error = error_string(err);
449 * Connect to remote address.
452 if (addr->family == AF_INET6) {
453 memset(&a, 0, sizeof(a));
454 a6.sin6_family = AF_INET6;
455 a6.sin6_port = htons((short) port);
457 ((struct sockaddr_in6 *) addr->ai->ai_addr)->sin6_addr;
461 a.sin_family = AF_INET;
462 a.sin_addr.s_addr = htonl(addr->address);
463 a.sin_port = htons((short) port);
468 connect(s, ((addr->family == AF_INET6) ?
469 (struct sockaddr *) &a6 : (struct sockaddr *) &a),
470 (addr->family == AF_INET6) ? sizeof(a6) : sizeof(a))
472 connect(s, (struct sockaddr *) &a, sizeof(a))
476 * FIXME: We are prepared to receive EWOULDBLOCK here,
477 * because we might want the connection to be made
478 * asynchronously; but how do we actually arrange this in
481 if ( errno != EWOULDBLOCK ) {
482 ret->error = error_string(errno);
487 * If we _don't_ get EWOULDBLOCK, the connect has completed
488 * and we should set the socket as connected and writable.
499 Socket sk_newlistener(char *srcaddr, int port, Plug plug, int local_host_only)
501 static struct socket_function_table fn_table = {
507 sk_tcp_set_private_ptr,
508 sk_tcp_get_private_ptr,
515 struct sockaddr_in6 a6;
517 struct sockaddr_in a;
524 * Create Socket structure.
526 ret = smalloc(sizeof(struct Socket_tag));
530 bufchain_init(&ret->output_data);
531 ret->writable = 0; /* to start with */
532 ret->sending_oob = 0;
534 ret->frozen_readable = 0;
535 ret->localhost_only = local_host_only;
536 ret->pending_error = 0;
537 ret->oobpending = FALSE;
543 s = socket(AF_INET, SOCK_STREAM, 0);
547 ret->error = error_string(errno);
553 setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (const char *)&on, sizeof(on));
556 if (addr->family == AF_INET6) {
557 memset(&a6, 0, sizeof(a6));
558 a6.sin6_family = AF_INET6;
559 /* FIXME: srcaddr is ignored for IPv6, because I (SGT) don't
560 * know how to do it. :-) */
562 a6.sin6_addr = in6addr_loopback;
564 a6.sin6_addr = in6addr_any;
565 a6.sin6_port = htons(port);
570 a.sin_family = AF_INET;
573 * Bind to source address. First try an explicitly
577 a.sin_addr.s_addr = inet_addr(srcaddr);
578 if (a.sin_addr.s_addr != INADDR_NONE) {
579 /* Override localhost_only with specified listen addr. */
580 ret->localhost_only = ipv4_is_loopback(a.sin_addr);
586 * ... and failing that, go with one of the standard ones.
590 a.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
592 a.sin_addr.s_addr = htonl(INADDR_ANY);
595 a.sin_port = htons((short)port);
598 retcode = bind(s, (addr->family == AF_INET6 ?
599 (struct sockaddr *) &a6 :
600 (struct sockaddr *) &a),
602 AF_INET6 ? sizeof(a6) : sizeof(a)));
604 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
613 ret->error = error_string(err);
618 if (listen(s, SOMAXCONN) < 0) {
620 ret->error = error_string(errno);
629 static void sk_tcp_close(Socket sock)
631 Actual_Socket s = (Actual_Socket) sock;
639 * The function which tries to send on a socket once it's deemed
642 void try_send(Actual_Socket s)
644 while (s->sending_oob || bufchain_size(&s->output_data) > 0) {
650 if (s->sending_oob) {
651 urgentflag = MSG_OOB;
652 len = s->sending_oob;
656 bufchain_prefix(&s->output_data, &data, &len);
658 nsent = send(s->s, data, len, urgentflag);
659 noise_ultralight(nsent);
661 err = (nsent < 0 ? errno : 0);
662 if (err == EWOULDBLOCK) {
664 * Perfectly normal: we've sent all we can for the moment.
668 } else if (nsent == 0 ||
669 err == ECONNABORTED || err == ECONNRESET) {
671 * If send() returns CONNABORTED or CONNRESET, we
672 * unfortunately can't just call plug_closing(),
673 * because it's quite likely that we're currently
674 * _in_ a call from the code we'd be calling back
675 * to, so we'd have to make half the SSH code
676 * reentrant. Instead we flag a pending error on
677 * the socket, to be dealt with (by calling
678 * plug_closing()) at some suitable future moment.
680 s->pending_error = err;
683 /* We're inside the Unix frontend here, so we know
684 * that the frontend handle is unnecessary. */
685 logevent(NULL, error_string(err));
686 fatalbox("%s", error_string(err));
689 if (s->sending_oob) {
691 memmove(s->oobdata, s->oobdata+nsent, len-nsent);
692 s->sending_oob = len - nsent;
697 bufchain_consume(&s->output_data, nsent);
703 static int sk_tcp_write(Socket sock, char *buf, int len)
705 Actual_Socket s = (Actual_Socket) sock;
708 * Add the data to the buffer list on the socket.
710 bufchain_add(&s->output_data, buf, len);
713 * Now try sending from the start of the buffer list.
718 return bufchain_size(&s->output_data);
721 static int sk_tcp_write_oob(Socket sock, char *buf, int len)
723 Actual_Socket s = (Actual_Socket) sock;
726 * Replace the buffer list on the socket with the data.
728 bufchain_clear(&s->output_data);
729 assert(len <= sizeof(s->oobdata));
730 memcpy(s->oobdata, buf, len);
731 s->sending_oob = len;
734 * Now try sending from the start of the buffer list.
739 return s->sending_oob;
742 int select_result(int fd, int event)
746 char buf[20480]; /* nice big buffer for plenty of speed */
750 /* Find the Socket structure */
751 s = find234(sktree, (void *) fd, cmpforsearch);
753 return 1; /* boggle */
755 noise_ultralight(event);
758 #ifdef FIXME_NONBLOCKING_CONNECTIONS
759 case FIXME: /* connected */
760 s->connected = s->writable = 1;
763 case 4: /* exceptional */
766 * On a non-oobinline socket, this indicates that we
767 * can immediately perform an OOB read and get back OOB
768 * data, which we will send to the back end with
769 * type==2 (urgent data).
771 ret = recv(s->s, buf, sizeof(buf), MSG_OOB);
772 noise_ultralight(ret);
774 char *str = (ret == 0 ? "Internal networking trouble" :
775 error_string(errno));
776 /* We're inside the Unix frontend here, so we know
777 * that the frontend handle is unnecessary. */
781 return plug_receive(s->plug, 2, buf, ret);
787 * If we reach here, this is an oobinline socket, which
788 * means we should set s->oobpending and then deal with it
789 * when we get called for the readability event (which
790 * should also occur).
792 s->oobpending = TRUE;
794 case 1: /* readable; also acceptance */
797 * On a listening socket, the readability event means a
798 * connection is ready to be accepted.
800 struct sockaddr_in isa;
801 int addrlen = sizeof(struct sockaddr_in);
802 int t; /* socket of connection */
804 memset(&isa, 0, sizeof(struct sockaddr_in));
806 t = accept(s->s,(struct sockaddr *)&isa,&addrlen);
811 if (s->localhost_only && !ipv4_is_loopback(isa.sin_addr)) {
812 close(t); /* someone let nonlocal through?! */
813 } else if (plug_accepting(s->plug, (void*)t)) {
814 close(t); /* denied or error */
820 * If we reach here, this is not a listening socket, so
821 * readability really means readability.
824 /* In the case the socket is still frozen, we don't even bother */
826 s->frozen_readable = 1;
831 * We have received data on the socket. For an oobinline
832 * socket, this might be data _before_ an urgent pointer,
833 * in which case we send it to the back end with type==1
834 * (data prior to urgent).
836 if (s->oobinline && s->oobpending) {
838 if (ioctl(s->s, SIOCATMARK, &atmark) == 0 && atmark)
839 s->oobpending = FALSE; /* clear this indicator */
843 ret = recv(s->s, buf, s->oobpending ? 1 : sizeof(buf), 0);
844 noise_ultralight(ret);
846 if (errno == EWOULDBLOCK) {
851 return plug_closing(s->plug, error_string(errno), errno, 0);
852 } else if (0 == ret) {
853 return plug_closing(s->plug, NULL, 0, 0);
855 return plug_receive(s->plug, atmark ? 0 : 1, buf, ret);
858 case 2: /* writable */
860 int bufsize_before, bufsize_after;
862 bufsize_before = s->sending_oob + bufchain_size(&s->output_data);
864 bufsize_after = s->sending_oob + bufchain_size(&s->output_data);
865 if (bufsize_after < bufsize_before)
866 plug_sent(s->plug, bufsize_after);
875 * Deal with socket errors detected in try_send().
877 void net_pending_errors(void)
883 * This might be a fiddly business, because it's just possible
884 * that handling a pending error on one socket might cause
885 * others to be closed. (I can't think of any reason this might
886 * happen in current SSH implementation, but to maintain
887 * generality of this network layer I'll assume the worst.)
889 * So what we'll do is search the socket list for _one_ socket
890 * with a pending error, and then handle it, and then search
891 * the list again _from the beginning_. Repeat until we make a
892 * pass with no socket errors present. That way we are
893 * protected against the socket list changing under our feet.
897 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
898 if (s->pending_error) {
900 * An error has occurred on this socket. Pass it to the
903 plug_closing(s->plug, error_string(s->pending_error),
904 s->pending_error, 0);
912 * Each socket abstraction contains a `void *' private field in
913 * which the client can keep state.
915 static void sk_tcp_set_private_ptr(Socket sock, void *ptr)
917 Actual_Socket s = (Actual_Socket) sock;
918 s->private_ptr = ptr;
921 static void *sk_tcp_get_private_ptr(Socket sock)
923 Actual_Socket s = (Actual_Socket) sock;
924 return s->private_ptr;
928 * Special error values are returned from sk_namelookup and sk_new
929 * if there's a problem. These functions extract an error message,
930 * or return NULL if there's no problem.
932 char *sk_addr_error(SockAddr addr)
936 static char *sk_tcp_socket_error(Socket sock)
938 Actual_Socket s = (Actual_Socket) sock;
942 static void sk_tcp_set_frozen(Socket sock, int is_frozen)
944 Actual_Socket s = (Actual_Socket) sock;
945 if (s->frozen == is_frozen)
947 s->frozen = is_frozen;
948 if (!is_frozen && s->frozen_readable) {
950 recv(s->s, &c, 1, MSG_PEEK);
952 s->frozen_readable = 0;
956 * For Unix select()-based frontends: enumerate all sockets
957 * currently active, and state whether we currently wish to receive
958 * select events on them for reading, writing and exceptional
961 static void set_rwx(Actual_Socket s, int *rwx)
964 if (s->connected && !s->frozen)
965 val |= 1 | 4; /* read, except */
966 if (bufchain_size(&s->output_data))
967 val |= 2; /* write */
969 val |= 1; /* read == accept */
973 int first_socket(int *state, int *rwx)
977 s = index234(sktree, (*state)++);
980 return s ? s->s : -1;
983 int next_socket(int *state, int *rwx)
985 Actual_Socket s = index234(sktree, (*state)++);
988 return s ? s->s : -1;
991 int net_service_lookup(char *service)
994 se = getservbyname(service, NULL);
996 return ntohs(se->s_port);