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;
61 * Which address family this address belongs to. AF_INET for
62 * IPv4; AF_INET6 for IPv6; AF_UNSPEC indicates that name
63 * resolution has not been done and a simple host name is held
64 * in this SockAddr structure.
68 struct addrinfo *ai; /* Address IPv6 style. */
70 unsigned long address; /* Address IPv4 style. */
72 char hostname[512]; /* Store an unresolved host name. */
75 static tree234 *sktree;
77 static void uxsel_tell(Actual_Socket s);
79 static int cmpfortree(void *av, void *bv)
81 Actual_Socket a = (Actual_Socket) av, b = (Actual_Socket) bv;
82 int as = a->s, bs = b->s;
90 static int cmpforsearch(void *av, void *bv)
92 Actual_Socket b = (Actual_Socket) bv;
93 int as = (int) av, bs = b->s;
103 sktree = newtree234(cmpfortree);
106 void sk_cleanup(void)
112 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
118 char *error_string(int error)
120 return strerror(error);
123 SockAddr sk_namelookup(const char *host, char **canonicalname)
125 SockAddr ret = snew(struct SockAddr_tag);
127 struct addrinfo hints;
131 struct hostent *h = NULL;
135 /* Clear the structure and default to IPv4. */
136 memset(ret, 0, sizeof(struct SockAddr_tag));
137 ret->family = 0; /* We set this one when we have resolved the host. */
142 hints.ai_flags = AI_CANONNAME;
143 hints.ai_family = AF_UNSPEC;
144 hints.ai_socktype = 0;
145 hints.ai_protocol = 0;
146 hints.ai_addrlen = 0;
147 hints.ai_addr = NULL;
148 hints.ai_canonname = NULL;
149 hints.ai_next = NULL;
150 err = getaddrinfo(host, NULL, NULL, &ret->ai);
152 ret->error = gai_strerror(err);
155 ret->family = ret->ai->ai_family;
157 if (ret->ai->ai_canonname != NULL)
158 strncat(realhost, ret->ai->ai_canonname, sizeof(realhost) - 1);
160 strncat(realhost, host, sizeof(realhost) - 1);
162 if ((a = inet_addr(host)) == (unsigned long) INADDR_NONE) {
164 * Otherwise use the IPv4-only gethostbyname... (NOTE:
165 * we don't use gethostbyname as a fallback!)
167 if (ret->family == 0) {
168 /*debug(("Resolving \"%s\" with gethostbyname() (IPv4 only)...\n", host)); */
169 if ( (h = gethostbyname(host)) )
170 ret->family = AF_INET;
172 if (ret->family == 0) {
173 ret->error = (h_errno == HOST_NOT_FOUND ||
174 h_errno == NO_DATA ||
175 h_errno == NO_ADDRESS ? "Host does not exist" :
176 h_errno == TRY_AGAIN ?
177 "Temporary name service failure" :
178 "gethostbyname: unknown error");
181 memcpy(&a, h->h_addr, sizeof(a));
182 /* This way we are always sure the h->h_name is valid :) */
183 strncpy(realhost, h->h_name, sizeof(realhost));
186 * This must be a numeric IPv4 address because it caused a
187 * success return from inet_addr.
189 ret->family = AF_INET;
190 strncpy(realhost, host, sizeof(realhost));
192 ret->address = ntohl(a);
194 realhost[lenof(realhost)-1] = '\0';
195 *canonicalname = snewn(1+strlen(realhost), char);
196 strcpy(*canonicalname, realhost);
200 SockAddr sk_nonamelookup(const char *host)
202 SockAddr ret = snew(struct SockAddr_tag);
204 ret->family = AF_UNSPEC;
205 strncpy(ret->hostname, host, lenof(ret->hostname));
206 ret->hostname[lenof(ret->hostname)-1] = '\0';
210 void sk_getaddr(SockAddr addr, char *buf, int buflen)
213 if (addr->family == AF_UNSPEC) {
214 strncpy(buf, addr->hostname, buflen);
215 buf[buflen-1] = '\0';
218 if (getnameinfo(addr->ai->ai_addr, addr->ai->ai_addrlen, buf, buflen,
219 NULL, 0, NI_NUMERICHOST) != 0) {
221 strncat(buf, "<unknown>", buflen - 1);
225 assert(addr->family == AF_INET);
226 a.s_addr = htonl(addr->address);
227 strncpy(buf, inet_ntoa(a), buflen);
228 buf[buflen-1] = '\0';
233 int sk_hostname_is_local(char *name)
235 return !strcmp(name, "localhost");
238 int sk_address_is_local(SockAddr addr)
241 if (addr->family == AF_UNSPEC)
242 return 0; /* we don't know; assume not */
245 if (addr->family == AF_INET)
246 return ipv4_is_loopback(
247 ((struct sockaddr_in *)addr->ai->ai_addr)->sin_addr);
248 else if (addr->family == AF_INET6)
249 return IN6_IS_ADDR_LOOPBACK(
250 &((struct sockaddr_in6 *)addr->ai->ai_addr)->sin6_addr);
255 assert(addr->family == AF_INET);
256 a.s_addr = htonl(addr->address);
257 return ipv4_is_loopback(a);
262 int sk_addrtype(SockAddr addr)
264 return (addr->family == AF_INET ? ADDRTYPE_IPV4 :
266 addr->family == AF_INET6 ? ADDRTYPE_IPV6 :
271 void sk_addrcopy(SockAddr addr, char *buf)
275 if (addr->family == AF_INET)
276 memcpy(buf, &((struct sockaddr_in *)addr->ai->ai_addr)->sin_addr,
277 sizeof(struct in_addr));
278 else if (addr->family == AF_INET6)
279 memcpy(buf, &((struct sockaddr_in6 *)addr->ai->ai_addr)->sin6_addr,
280 sizeof(struct in6_addr));
286 assert(addr->family == AF_INET);
287 a.s_addr = htonl(addr->address);
288 memcpy(buf, (char*) &a.s_addr, 4);
292 void sk_addr_free(SockAddr addr)
296 if (addr->ai != NULL)
297 freeaddrinfo(addr->ai);
302 static Plug sk_tcp_plug(Socket sock, Plug p)
304 Actual_Socket s = (Actual_Socket) sock;
311 static void sk_tcp_flush(Socket s)
314 * We send data to the socket as soon as we can anyway,
315 * so we don't need to do anything here. :-)
319 static void sk_tcp_close(Socket s);
320 static int sk_tcp_write(Socket s, const char *data, int len);
321 static int sk_tcp_write_oob(Socket s, const char *data, int len);
322 static void sk_tcp_set_private_ptr(Socket s, void *ptr);
323 static void *sk_tcp_get_private_ptr(Socket s);
324 static void sk_tcp_set_frozen(Socket s, int is_frozen);
325 static char *sk_tcp_socket_error(Socket s);
327 static struct socket_function_table tcp_fn_table = {
333 sk_tcp_set_private_ptr,
334 sk_tcp_get_private_ptr,
339 Socket sk_register(void *sock, Plug plug)
344 * Create Socket structure.
346 ret = snew(struct Socket_tag);
347 ret->fn = &tcp_fn_table;
350 bufchain_init(&ret->output_data);
351 ret->writable = 1; /* to start with */
352 ret->sending_oob = 0;
354 ret->frozen_readable = 0;
355 ret->localhost_only = 0; /* unused, but best init anyway */
356 ret->pending_error = 0;
357 ret->oobpending = FALSE;
363 ret->error = error_string(errno);
375 Socket sk_new(SockAddr addr, int port, int privport, int oobinline,
376 int nodelay, Plug plug)
380 struct sockaddr_in6 a6;
382 struct sockaddr_in a;
388 * Create Socket structure.
390 ret = snew(struct Socket_tag);
391 ret->fn = &tcp_fn_table;
394 bufchain_init(&ret->output_data);
395 ret->connected = 0; /* to start with */
396 ret->writable = 0; /* to start with */
397 ret->sending_oob = 0;
399 ret->frozen_readable = 0;
400 ret->localhost_only = 0; /* unused, but best init anyway */
401 ret->pending_error = 0;
402 ret->oobpending = FALSE;
408 assert(addr->family != AF_UNSPEC);
409 s = socket(addr->family, SOCK_STREAM, 0);
413 ret->error = error_string(errno);
417 ret->oobinline = oobinline;
420 setsockopt(s, SOL_SOCKET, SO_OOBINLINE, (void *) &b, sizeof(b));
425 setsockopt(s, IPPROTO_TCP, TCP_NODELAY, (void *) &b, sizeof(b));
429 * Bind to local address.
432 localport = 1023; /* count from 1023 downwards */
434 localport = 0; /* just use port 0 (ie kernel picks) */
436 /* Loop round trying to bind */
441 if (addr->family == AF_INET6) {
442 /* XXX use getaddrinfo to get a local address? */
443 a6.sin6_family = AF_INET6;
444 a6.sin6_addr = in6addr_any;
445 a6.sin6_port = htons(localport);
446 retcode = bind(s, (struct sockaddr *) &a6, sizeof(a6));
450 assert(addr->family == AF_INET);
451 a.sin_family = AF_INET;
452 a.sin_addr.s_addr = htonl(INADDR_ANY);
453 a.sin_port = htons(localport);
454 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
461 if (err != EADDRINUSE) /* failed, for a bad reason */
466 break; /* we're only looping once */
469 break; /* we might have got to the end */
473 ret->error = error_string(err);
478 * Connect to remote address.
481 /* XXX would be better to have got getaddrinfo() to fill in the port. */
482 if (addr->family == AF_INET)
483 ((struct sockaddr_in *)addr->ai->ai_addr)->sin_port =
486 assert(addr->family == AF_INET6);
487 ((struct sockaddr_in *)addr->ai->ai_addr)->sin_port =
491 a.sin_family = AF_INET;
492 a.sin_addr.s_addr = htonl(addr->address);
493 a.sin_port = htons((short) port);
497 ioctl(s, FIONBIO, &i);
502 connect(s, addr->ai->ai_addr, addr->ai->ai_addrlen)
504 connect(s, (struct sockaddr *) &a, sizeof(a))
507 if ( errno != EINPROGRESS ) {
508 ret->error = error_string(errno);
513 * If we _don't_ get EWOULDBLOCK, the connect has completed
514 * and we should set the socket as connected and writable.
526 Socket sk_newlistener(char *srcaddr, int port, Plug plug, int local_host_only)
531 struct sockaddr_in6 a6;
533 struct addrinfo hints, *ai;
536 struct sockaddr_in a;
543 * Create Socket structure.
545 ret = snew(struct Socket_tag);
546 ret->fn = &tcp_fn_table;
549 bufchain_init(&ret->output_data);
550 ret->writable = 0; /* to start with */
551 ret->sending_oob = 0;
553 ret->frozen_readable = 0;
554 ret->localhost_only = local_host_only;
555 ret->pending_error = 0;
556 ret->oobpending = FALSE;
562 s = socket(AF_INET, SOCK_STREAM, 0);
566 ret->error = error_string(errno);
572 setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (const char *)&on, sizeof(on));
575 hints.ai_flags = AI_NUMERICHOST;
576 hints.ai_family = AF_UNSPEC;
577 hints.ai_socktype = 0;
578 hints.ai_protocol = 0;
579 hints.ai_addrlen = 0;
580 hints.ai_addr = NULL;
581 hints.ai_canonname = NULL;
582 hints.ai_next = NULL;
583 sprintf(portstr, "%d", port);
584 if (srcaddr != NULL && getaddrinfo(srcaddr, portstr, &hints, &ai) == 0)
585 retcode = bind(s, ai->ai_addr, ai->ai_addrlen);
590 * FIXME: Need two listening sockets, in principle, one for v4
594 a6.sin6_addr = in6addr_loopback;
596 a6.sin6_addr = in6addr_any;
597 a6.sin6_port = htons(port);
603 a.sin_family = AF_INET;
606 * Bind to source address. First try an explicitly
610 a.sin_addr.s_addr = inet_addr(srcaddr);
611 if (a.sin_addr.s_addr != INADDR_NONE) {
612 /* Override localhost_only with specified listen addr. */
613 ret->localhost_only = ipv4_is_loopback(a.sin_addr);
619 * ... and failing that, go with one of the standard ones.
623 a.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
625 a.sin_addr.s_addr = htonl(INADDR_ANY);
628 a.sin_port = htons((short)port);
629 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
639 ret->error = error_string(err);
644 if (listen(s, SOMAXCONN) < 0) {
646 ret->error = error_string(errno);
656 static void sk_tcp_close(Socket sock)
658 Actual_Socket s = (Actual_Socket) sock;
666 int sk_getxdmdata(void *sock, unsigned long *ip, int *port)
668 Actual_Socket s = (Actual_Socket) sock;
669 struct sockaddr_in addr;
673 * We must check that this socket really _is_ an Actual_Socket.
675 if (s->fn != &tcp_fn_table)
676 return 0; /* failure */
679 * If we ever implement connecting to a local X server through
680 * a Unix socket, we return 0xFFFFFFFF for the IP address and
681 * our current pid for the port. Bizarre, but such is life.
684 addrlen = sizeof(addr);
685 if (getsockname(s->s, (struct sockaddr *)&addr, &addrlen) < 0 ||
686 addr.sin_family != AF_INET)
689 *ip = ntohl(addr.sin_addr.s_addr);
690 *port = ntohs(addr.sin_port);
696 * The function which tries to send on a socket once it's deemed
699 void try_send(Actual_Socket s)
701 while (s->sending_oob || bufchain_size(&s->output_data) > 0) {
707 if (s->sending_oob) {
708 urgentflag = MSG_OOB;
709 len = s->sending_oob;
713 bufchain_prefix(&s->output_data, &data, &len);
715 nsent = send(s->s, data, len, urgentflag);
716 noise_ultralight(nsent);
718 err = (nsent < 0 ? errno : 0);
719 if (err == EWOULDBLOCK) {
721 * Perfectly normal: we've sent all we can for the moment.
725 } else if (nsent == 0 ||
726 err == ECONNABORTED || err == ECONNRESET) {
728 * If send() returns CONNABORTED or CONNRESET, we
729 * unfortunately can't just call plug_closing(),
730 * because it's quite likely that we're currently
731 * _in_ a call from the code we'd be calling back
732 * to, so we'd have to make half the SSH code
733 * reentrant. Instead we flag a pending error on
734 * the socket, to be dealt with (by calling
735 * plug_closing()) at some suitable future moment.
737 s->pending_error = err;
740 /* We're inside the Unix frontend here, so we know
741 * that the frontend handle is unnecessary. */
742 logevent(NULL, error_string(err));
743 fatalbox("%s", error_string(err));
746 if (s->sending_oob) {
748 memmove(s->oobdata, s->oobdata+nsent, len-nsent);
749 s->sending_oob = len - nsent;
754 bufchain_consume(&s->output_data, nsent);
761 static int sk_tcp_write(Socket sock, const char *buf, int len)
763 Actual_Socket s = (Actual_Socket) sock;
766 * Add the data to the buffer list on the socket.
768 bufchain_add(&s->output_data, buf, len);
771 * Now try sending from the start of the buffer list.
776 return bufchain_size(&s->output_data);
779 static int sk_tcp_write_oob(Socket sock, const char *buf, int len)
781 Actual_Socket s = (Actual_Socket) sock;
784 * Replace the buffer list on the socket with the data.
786 bufchain_clear(&s->output_data);
787 assert(len <= sizeof(s->oobdata));
788 memcpy(s->oobdata, buf, len);
789 s->sending_oob = len;
792 * Now try sending from the start of the buffer list.
797 return s->sending_oob;
800 static int net_select_result(int fd, int event)
804 char buf[20480]; /* nice big buffer for plenty of speed */
808 /* Find the Socket structure */
809 s = find234(sktree, (void *) fd, cmpforsearch);
811 return 1; /* boggle */
813 noise_ultralight(event);
816 case 4: /* exceptional */
819 * On a non-oobinline socket, this indicates that we
820 * can immediately perform an OOB read and get back OOB
821 * data, which we will send to the back end with
822 * type==2 (urgent data).
824 ret = recv(s->s, buf, sizeof(buf), MSG_OOB);
825 noise_ultralight(ret);
827 char *str = (ret == 0 ? "Internal networking trouble" :
828 error_string(errno));
829 /* We're inside the Unix frontend here, so we know
830 * that the frontend handle is unnecessary. */
834 return plug_receive(s->plug, 2, buf, ret);
840 * If we reach here, this is an oobinline socket, which
841 * means we should set s->oobpending and then deal with it
842 * when we get called for the readability event (which
843 * should also occur).
845 s->oobpending = TRUE;
847 case 1: /* readable; also acceptance */
850 * On a listening socket, the readability event means a
851 * connection is ready to be accepted.
853 struct sockaddr_in isa;
854 int addrlen = sizeof(struct sockaddr_in);
855 int t; /* socket of connection */
857 memset(&isa, 0, sizeof(struct sockaddr_in));
859 t = accept(s->s,(struct sockaddr *)&isa,&addrlen);
864 if (s->localhost_only && !ipv4_is_loopback(isa.sin_addr)) {
865 close(t); /* someone let nonlocal through?! */
866 } else if (plug_accepting(s->plug, (void*)t)) {
867 close(t); /* denied or error */
873 * If we reach here, this is not a listening socket, so
874 * readability really means readability.
877 /* In the case the socket is still frozen, we don't even bother */
879 s->frozen_readable = 1;
884 * We have received data on the socket. For an oobinline
885 * socket, this might be data _before_ an urgent pointer,
886 * in which case we send it to the back end with type==1
887 * (data prior to urgent).
889 if (s->oobinline && s->oobpending) {
891 if (ioctl(s->s, SIOCATMARK, &atmark) == 0 && atmark)
892 s->oobpending = FALSE; /* clear this indicator */
896 ret = recv(s->s, buf, s->oobpending ? 1 : sizeof(buf), 0);
897 noise_ultralight(ret);
899 if (errno == EWOULDBLOCK) {
904 return plug_closing(s->plug, error_string(errno), errno, 0);
905 } else if (0 == ret) {
906 return plug_closing(s->plug, NULL, 0, 0);
908 return plug_receive(s->plug, atmark ? 0 : 1, buf, ret);
911 case 2: /* writable */
914 * select() reports a socket as _writable_ when an
915 * asynchronous connection is completed.
917 s->connected = s->writable = 1;
921 int bufsize_before, bufsize_after;
923 bufsize_before = s->sending_oob + bufchain_size(&s->output_data);
925 bufsize_after = s->sending_oob + bufchain_size(&s->output_data);
926 if (bufsize_after < bufsize_before)
927 plug_sent(s->plug, bufsize_after);
936 * Deal with socket errors detected in try_send().
938 void net_pending_errors(void)
944 * This might be a fiddly business, because it's just possible
945 * that handling a pending error on one socket might cause
946 * others to be closed. (I can't think of any reason this might
947 * happen in current SSH implementation, but to maintain
948 * generality of this network layer I'll assume the worst.)
950 * So what we'll do is search the socket list for _one_ socket
951 * with a pending error, and then handle it, and then search
952 * the list again _from the beginning_. Repeat until we make a
953 * pass with no socket errors present. That way we are
954 * protected against the socket list changing under our feet.
958 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
959 if (s->pending_error) {
961 * An error has occurred on this socket. Pass it to the
964 plug_closing(s->plug, error_string(s->pending_error),
965 s->pending_error, 0);
973 * Each socket abstraction contains a `void *' private field in
974 * which the client can keep state.
976 static void sk_tcp_set_private_ptr(Socket sock, void *ptr)
978 Actual_Socket s = (Actual_Socket) sock;
979 s->private_ptr = ptr;
982 static void *sk_tcp_get_private_ptr(Socket sock)
984 Actual_Socket s = (Actual_Socket) sock;
985 return s->private_ptr;
989 * Special error values are returned from sk_namelookup and sk_new
990 * if there's a problem. These functions extract an error message,
991 * or return NULL if there's no problem.
993 char *sk_addr_error(SockAddr addr)
997 static char *sk_tcp_socket_error(Socket sock)
999 Actual_Socket s = (Actual_Socket) sock;
1003 static void sk_tcp_set_frozen(Socket sock, int is_frozen)
1005 Actual_Socket s = (Actual_Socket) sock;
1006 if (s->frozen == is_frozen)
1008 s->frozen = is_frozen;
1009 if (!is_frozen && s->frozen_readable) {
1011 recv(s->s, &c, 1, MSG_PEEK);
1013 s->frozen_readable = 0;
1017 static void uxsel_tell(Actual_Socket s)
1021 rwx |= 2; /* write == connect */
1022 if (s->connected && !s->frozen)
1023 rwx |= 1 | 4; /* read, except */
1024 if (bufchain_size(&s->output_data))
1025 rwx |= 2; /* write */
1027 rwx |= 1; /* read == accept */
1028 uxsel_set(s->s, rwx, net_select_result);
1031 int net_service_lookup(char *service)
1034 se = getservbyname(service, NULL);
1036 return ntohs(se->s_port);