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)
109 char *ret = smalloc(strlen(strerror(error)) +
110 strlen("Network error: "));
111 sprintf(ret, "Network error: %s",strerror(error));
115 SockAddr sk_namelookup(char *host, char **canonicalname)
117 SockAddr ret = smalloc(sizeof(struct SockAddr_tag));
119 struct hostent *h = NULL;
122 /* Clear the structure and default to IPv4. */
123 memset(ret, 0, sizeof(struct SockAddr_tag));
124 ret->family = 0; /* We set this one when we have resolved the host. */
128 if ((a = inet_addr(host)) == (unsigned long) INADDR_NONE) {
130 if (getaddrinfo(host, NULL, NULL, &ret->ai) == 0) {
131 ret->family = ret->ai->ai_family;
136 * Otherwise use the IPv4-only gethostbyname... (NOTE:
137 * we don't use gethostbyname as a fallback!)
139 if (ret->family == 0) {
140 /*debug(("Resolving \"%s\" with gethostbyname() (IPv4 only)...\n", host)); */
141 if ( (h = gethostbyname(host)) )
142 ret->family = AF_INET;
144 if (ret->family == 0) {
145 ret->error = (h_errno == HOST_NOT_FOUND ||
146 h_errno == NO_DATA ||
147 h_errno == NO_ADDRESS ? "Host does not exist" :
148 h_errno == TRY_AGAIN ?
149 "Temporary name service failure" :
150 "gethostbyname: unknown error");
156 /* If we got an address info use that... */
159 /* Are we in IPv4 fallback mode? */
160 /* We put the IPv4 address into the a variable so we can further-on use the IPv4 code... */
161 if (ret->family == AF_INET)
163 (char *) &((struct sockaddr_in *) ret->ai->
164 ai_addr)->sin_addr, sizeof(a));
166 /* Now let's find that canonicalname... */
167 if (getnameinfo((struct sockaddr *) ret->ai->ai_addr,
169 AF_INET ? sizeof(struct sockaddr_in) :
170 sizeof(struct sockaddr_in6), realhost,
171 sizeof(realhost), NULL, 0, 0) != 0) {
172 strncpy(realhost, host, sizeof(realhost));
175 /* We used the IPv4-only gethostbyname()... */
179 memcpy(&a, h->h_addr, sizeof(a));
180 /* This way we are always sure the h->h_name is valid :) */
181 strncpy(realhost, h->h_name, sizeof(realhost));
185 * This must be a numeric IPv4 address because it caused a
186 * success return from inet_addr.
188 ret->family = AF_INET;
189 strncpy(realhost, host, sizeof(realhost));
191 ret->address = ntohl(a);
192 realhost[lenof(realhost)-1] = '\0';
193 *canonicalname = smalloc(1+strlen(realhost));
194 strcpy(*canonicalname, realhost);
198 void sk_getaddr(SockAddr addr, char *buf, int buflen)
201 if (addr->family == AF_INET) {
204 a.s_addr = htonl(addr->address);
205 strncpy(buf, inet_ntoa(a), buflen);
208 FIXME; /* I don't know how to get a text form of an IPv6 address. */
213 int sk_addrtype(SockAddr addr)
215 return (addr->family == AF_INET ? ADDRTYPE_IPV4 : ADDRTYPE_IPV6);
218 void sk_addrcopy(SockAddr addr, char *buf)
221 if (addr->family == AF_INET) {
224 a.s_addr = htonl(addr->address);
225 memcpy(buf, (char*) &a.s_addr, 4);
228 memcpy(buf, (char*) addr->ai, 16);
233 void sk_addr_free(SockAddr addr)
238 static Plug sk_tcp_plug(Socket sock, Plug p)
240 Actual_Socket s = (Actual_Socket) sock;
247 static void sk_tcp_flush(Socket s)
250 * We send data to the socket as soon as we can anyway,
251 * so we don't need to do anything here. :-)
255 static void sk_tcp_close(Socket s);
256 static int sk_tcp_write(Socket s, char *data, int len);
257 static int sk_tcp_write_oob(Socket s, char *data, int len);
258 static void sk_tcp_set_private_ptr(Socket s, void *ptr);
259 static void *sk_tcp_get_private_ptr(Socket s);
260 static void sk_tcp_set_frozen(Socket s, int is_frozen);
261 static char *sk_tcp_socket_error(Socket s);
263 Socket sk_register(void *sock, Plug plug)
265 static struct socket_function_table fn_table = {
271 sk_tcp_set_private_ptr,
272 sk_tcp_get_private_ptr,
280 * Create Socket structure.
282 ret = smalloc(sizeof(struct Socket_tag));
286 bufchain_init(&ret->output_data);
287 ret->writable = 1; /* to start with */
288 ret->sending_oob = 0;
290 ret->frozen_readable = 0;
291 ret->localhost_only = 0; /* unused, but best init anyway */
292 ret->pending_error = 0;
293 ret->oobpending = FALSE;
299 ret->error = error_string(errno);
310 Socket sk_new(SockAddr addr, int port, int privport, int oobinline,
311 int nodelay, Plug plug)
313 static struct socket_function_table fn_table = {
319 sk_tcp_set_private_ptr,
320 sk_tcp_get_private_ptr,
327 struct sockaddr_in6 a6;
329 struct sockaddr_in a;
335 * Create Socket structure.
337 ret = smalloc(sizeof(struct Socket_tag));
341 bufchain_init(&ret->output_data);
342 ret->connected = 0; /* to start with */
343 ret->writable = 0; /* to start with */
344 ret->sending_oob = 0;
346 ret->frozen_readable = 0;
347 ret->localhost_only = 0; /* unused, but best init anyway */
348 ret->pending_error = 0;
349 ret->oobpending = FALSE;
355 s = socket(addr->family, SOCK_STREAM, 0);
359 ret->error = error_string(errno);
363 ret->oobinline = oobinline;
366 setsockopt(s, SOL_SOCKET, SO_OOBINLINE, (void *) &b, sizeof(b));
371 setsockopt(s, IPPROTO_TCP, TCP_NODELAY, (void *) &b, sizeof(b));
375 * Bind to local address.
378 localport = 1023; /* count from 1023 downwards */
380 localport = 0; /* just use port 0 (ie kernel picks) */
382 /* Loop round trying to bind */
387 if (addr->family == AF_INET6) {
388 memset(&a6, 0, sizeof(a6));
389 a6.sin6_family = AF_INET6;
390 /*a6.sin6_addr = in6addr_any; *//* == 0 */
391 a6.sin6_port = htons(localport);
395 a.sin_family = AF_INET;
396 a.sin_addr.s_addr = htonl(INADDR_ANY);
397 a.sin_port = htons(localport);
400 retcode = bind(s, (addr->family == AF_INET6 ?
401 (struct sockaddr *) &a6 :
402 (struct sockaddr *) &a),
404 AF_INET6 ? sizeof(a6) : sizeof(a)));
406 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
413 if (err != EADDRINUSE) /* failed, for a bad reason */
418 break; /* we're only looping once */
421 break; /* we might have got to the end */
425 ret->error = error_string(err);
430 * Connect to remote address.
433 if (addr->family == AF_INET6) {
434 memset(&a, 0, sizeof(a));
435 a6.sin6_family = AF_INET6;
436 a6.sin6_port = htons((short) port);
438 ((struct sockaddr_in6 *) addr->ai->ai_addr)->sin6_addr;
442 a.sin_family = AF_INET;
443 a.sin_addr.s_addr = htonl(addr->address);
444 a.sin_port = htons((short) port);
449 connect(s, ((addr->family == AF_INET6) ?
450 (struct sockaddr *) &a6 : (struct sockaddr *) &a),
451 (addr->family == AF_INET6) ? sizeof(a6) : sizeof(a))
453 connect(s, (struct sockaddr *) &a, sizeof(a))
457 * FIXME: We are prepared to receive EWOULDBLOCK here,
458 * because we might want the connection to be made
459 * asynchronously; but how do we actually arrange this in
462 if ( errno != EWOULDBLOCK ) {
463 ret->error = error_string(errno);
468 * If we _don't_ get EWOULDBLOCK, the connect has completed
469 * and we should set the socket as connected and writable.
480 Socket sk_newlistener(int port, Plug plug, int local_host_only)
482 static struct socket_function_table fn_table = {
488 sk_tcp_set_private_ptr,
489 sk_tcp_get_private_ptr,
496 struct sockaddr_in6 a6;
498 struct sockaddr_in a;
505 * Create Socket structure.
507 ret = smalloc(sizeof(struct Socket_tag));
511 bufchain_init(&ret->output_data);
512 ret->writable = 0; /* to start with */
513 ret->sending_oob = 0;
515 ret->frozen_readable = 0;
516 ret->localhost_only = local_host_only;
517 ret->pending_error = 0;
518 ret->oobpending = FALSE;
524 s = socket(AF_INET, SOCK_STREAM, 0);
528 ret->error = error_string(errno);
534 setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (const char *)&on, sizeof(on));
537 if (addr->family == AF_INET6) {
538 memset(&a6, 0, sizeof(a6));
539 a6.sin6_family = AF_INET6;
541 a6.sin6_addr = in6addr_loopback;
543 a6.sin6_addr = in6addr_any;
544 a6.sin6_port = htons(port);
548 a.sin_family = AF_INET;
550 a.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
552 a.sin_addr.s_addr = htonl(INADDR_ANY);
553 a.sin_port = htons((short)port);
556 retcode = bind(s, (addr->family == AF_INET6 ?
557 (struct sockaddr *) &a6 :
558 (struct sockaddr *) &a),
560 AF_INET6 ? sizeof(a6) : sizeof(a)));
562 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
571 ret->error = error_string(err);
576 if (listen(s, SOMAXCONN) < 0) {
578 ret->error = error_string(errno);
587 static void sk_tcp_close(Socket sock)
589 Actual_Socket s = (Actual_Socket) sock;
597 * The function which tries to send on a socket once it's deemed
600 void try_send(Actual_Socket s)
602 while (s->sending_oob || bufchain_size(&s->output_data) > 0) {
608 if (s->sending_oob) {
609 urgentflag = MSG_OOB;
610 len = s->sending_oob;
614 bufchain_prefix(&s->output_data, &data, &len);
616 nsent = send(s->s, data, len, urgentflag);
617 noise_ultralight(nsent);
619 err = (nsent < 0 ? errno : 0);
620 if (err == EWOULDBLOCK) {
622 * Perfectly normal: we've sent all we can for the moment.
626 } else if (nsent == 0 ||
627 err == ECONNABORTED || err == ECONNRESET) {
629 * If send() returns CONNABORTED or CONNRESET, we
630 * unfortunately can't just call plug_closing(),
631 * because it's quite likely that we're currently
632 * _in_ a call from the code we'd be calling back
633 * to, so we'd have to make half the SSH code
634 * reentrant. Instead we flag a pending error on
635 * the socket, to be dealt with (by calling
636 * plug_closing()) at some suitable future moment.
638 s->pending_error = err;
641 /* We're inside the Unix frontend here, so we know
642 * that the frontend handle is unnecessary. */
643 logevent(NULL, error_string(err));
644 fatalbox("%s", error_string(err));
647 if (s->sending_oob) {
649 memmove(s->oobdata, s->oobdata+nsent, len-nsent);
650 s->sending_oob = len - nsent;
655 bufchain_consume(&s->output_data, nsent);
661 static int sk_tcp_write(Socket sock, char *buf, int len)
663 Actual_Socket s = (Actual_Socket) sock;
666 * Add the data to the buffer list on the socket.
668 bufchain_add(&s->output_data, buf, len);
671 * Now try sending from the start of the buffer list.
676 return bufchain_size(&s->output_data);
679 static int sk_tcp_write_oob(Socket sock, char *buf, int len)
681 Actual_Socket s = (Actual_Socket) sock;
684 * Replace the buffer list on the socket with the data.
686 bufchain_clear(&s->output_data);
687 assert(len <= sizeof(s->oobdata));
688 memcpy(s->oobdata, buf, len);
689 s->sending_oob = len;
692 * Now try sending from the start of the buffer list.
697 return s->sending_oob;
700 int select_result(int fd, int event)
704 char buf[20480]; /* nice big buffer for plenty of speed */
708 /* Find the Socket structure */
709 s = find234(sktree, (void *) fd, cmpforsearch);
711 return 1; /* boggle */
713 noise_ultralight(event);
716 #ifdef FIXME_NONBLOCKING_CONNECTIONS
717 case FIXME: /* connected */
718 s->connected = s->writable = 1;
721 case 4: /* exceptional */
724 * On a non-oobinline socket, this indicates that we
725 * can immediately perform an OOB read and get back OOB
726 * data, which we will send to the back end with
727 * type==2 (urgent data).
729 ret = recv(s->s, buf, sizeof(buf), MSG_OOB);
730 noise_ultralight(ret);
732 char *str = (ret == 0 ? "Internal networking trouble" :
733 error_string(errno));
734 /* We're inside the Unix frontend here, so we know
735 * that the frontend handle is unnecessary. */
739 return plug_receive(s->plug, 2, buf, ret);
745 * If we reach here, this is an oobinline socket, which
746 * means we should set s->oobpending and then deal with it
747 * when we get called for the readability event (which
748 * should also occur).
750 s->oobpending = TRUE;
752 case 1: /* readable; also acceptance */
755 * On a listening socket, the readability event means a
756 * connection is ready to be accepted.
758 struct sockaddr_in isa;
759 int addrlen = sizeof(struct sockaddr_in);
760 int t; /* socket of connection */
762 memset(&isa, 0, sizeof(struct sockaddr_in));
764 t = accept(s->s,(struct sockaddr *)&isa,&addrlen);
769 if (s->localhost_only &&
770 ntohl(isa.sin_addr.s_addr) != INADDR_LOOPBACK) {
771 close(t); /* someone let nonlocal through?! */
772 } else if (plug_accepting(s->plug, (void*)t)) {
773 close(t); /* denied or error */
779 * If we reach here, this is not a listening socket, so
780 * readability really means readability.
783 /* In the case the socket is still frozen, we don't even bother */
785 s->frozen_readable = 1;
790 * We have received data on the socket. For an oobinline
791 * socket, this might be data _before_ an urgent pointer,
792 * in which case we send it to the back end with type==1
793 * (data prior to urgent).
795 if (s->oobinline && s->oobpending) {
797 if (ioctl(s->s, SIOCATMARK, &atmark) == 0 && atmark)
798 s->oobpending = FALSE; /* clear this indicator */
802 ret = recv(s->s, buf, s->oobpending ? 1 : sizeof(buf), 0);
803 noise_ultralight(ret);
805 if (errno == EWOULDBLOCK) {
810 return plug_closing(s->plug, error_string(errno), errno, 0);
811 } else if (0 == ret) {
812 return plug_closing(s->plug, NULL, 0, 0);
814 return plug_receive(s->plug, atmark ? 0 : 1, buf, ret);
817 case 2: /* writable */
819 int bufsize_before, bufsize_after;
821 bufsize_before = s->sending_oob + bufchain_size(&s->output_data);
823 bufsize_after = s->sending_oob + bufchain_size(&s->output_data);
824 if (bufsize_after < bufsize_before)
825 plug_sent(s->plug, bufsize_after);
834 * Deal with socket errors detected in try_send().
836 void net_pending_errors(void)
842 * This might be a fiddly business, because it's just possible
843 * that handling a pending error on one socket might cause
844 * others to be closed. (I can't think of any reason this might
845 * happen in current SSH implementation, but to maintain
846 * generality of this network layer I'll assume the worst.)
848 * So what we'll do is search the socket list for _one_ socket
849 * with a pending error, and then handle it, and then search
850 * the list again _from the beginning_. Repeat until we make a
851 * pass with no socket errors present. That way we are
852 * protected against the socket list changing under our feet.
856 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
857 if (s->pending_error) {
859 * An error has occurred on this socket. Pass it to the
862 plug_closing(s->plug, error_string(s->pending_error),
863 s->pending_error, 0);
871 * Each socket abstraction contains a `void *' private field in
872 * which the client can keep state.
874 static void sk_tcp_set_private_ptr(Socket sock, void *ptr)
876 Actual_Socket s = (Actual_Socket) sock;
877 s->private_ptr = ptr;
880 static void *sk_tcp_get_private_ptr(Socket sock)
882 Actual_Socket s = (Actual_Socket) sock;
883 return s->private_ptr;
887 * Special error values are returned from sk_namelookup and sk_new
888 * if there's a problem. These functions extract an error message,
889 * or return NULL if there's no problem.
891 char *sk_addr_error(SockAddr addr)
895 static char *sk_tcp_socket_error(Socket sock)
897 Actual_Socket s = (Actual_Socket) sock;
901 static void sk_tcp_set_frozen(Socket sock, int is_frozen)
903 Actual_Socket s = (Actual_Socket) sock;
904 if (s->frozen == is_frozen)
906 s->frozen = is_frozen;
907 if (!is_frozen && s->frozen_readable) {
909 recv(s->s, &c, 1, MSG_PEEK);
911 s->frozen_readable = 0;
915 * For Unix select()-based frontends: enumerate all sockets
916 * currently active, and state whether we currently wish to receive
917 * select events on them for reading, writing and exceptional
920 static void set_rwx(Actual_Socket s, int *rwx)
923 if (s->connected && !s->frozen)
924 val |= 1 | 4; /* read, except */
925 if (bufchain_size(&s->output_data))
926 val |= 2; /* write */
928 val |= 1; /* read == accept */
932 int first_socket(int *state, int *rwx)
936 s = index234(sktree, (*state)++);
939 return s ? s->s : -1;
942 int next_socket(int *state, int *rwx)
944 Actual_Socket s = index234(sktree, (*state)++);
947 return s ? s->s : -1;
950 int net_service_lookup(char *service)
953 se = getservbyname(service, NULL);
955 return ntohs(se->s_port);