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.
67 unsigned long address; /* Address IPv4 style. */
69 struct addrinfo *ai; /* Address IPv6 style. */
71 char hostname[512]; /* Store an unresolved host name. */
74 static tree234 *sktree;
76 static int cmpfortree(void *av, void *bv)
78 Actual_Socket a = (Actual_Socket) av, b = (Actual_Socket) bv;
79 int as = a->s, bs = b->s;
87 static int cmpforsearch(void *av, void *bv)
89 Actual_Socket b = (Actual_Socket) bv;
90 int as = (int) av, bs = b->s;
100 sktree = newtree234(cmpfortree);
103 void sk_cleanup(void)
109 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
115 char *error_string(int error)
117 return strerror(error);
120 SockAddr sk_namelookup(char *host, char **canonicalname)
122 SockAddr ret = smalloc(sizeof(struct SockAddr_tag));
124 struct hostent *h = NULL;
127 /* Clear the structure and default to IPv4. */
128 memset(ret, 0, sizeof(struct SockAddr_tag));
129 ret->family = 0; /* We set this one when we have resolved the host. */
133 if ((a = inet_addr(host)) == (unsigned long) INADDR_NONE) {
135 if (getaddrinfo(host, NULL, NULL, &ret->ai) == 0) {
136 ret->family = ret->ai->ai_family;
141 * Otherwise use the IPv4-only gethostbyname... (NOTE:
142 * we don't use gethostbyname as a fallback!)
144 if (ret->family == 0) {
145 /*debug(("Resolving \"%s\" with gethostbyname() (IPv4 only)...\n", host)); */
146 if ( (h = gethostbyname(host)) )
147 ret->family = AF_INET;
149 if (ret->family == 0) {
150 ret->error = (h_errno == HOST_NOT_FOUND ||
151 h_errno == NO_DATA ||
152 h_errno == NO_ADDRESS ? "Host does not exist" :
153 h_errno == TRY_AGAIN ?
154 "Temporary name service failure" :
155 "gethostbyname: unknown error");
161 /* If we got an address info use that... */
164 /* Are we in IPv4 fallback mode? */
165 /* We put the IPv4 address into the a variable so we can further-on use the IPv4 code... */
166 if (ret->family == AF_INET)
168 (char *) &((struct sockaddr_in *) ret->ai->
169 ai_addr)->sin_addr, sizeof(a));
171 /* Now let's find that canonicalname... */
172 if (getnameinfo((struct sockaddr *) ret->ai->ai_addr,
174 AF_INET ? sizeof(struct sockaddr_in) :
175 sizeof(struct sockaddr_in6), realhost,
176 sizeof(realhost), NULL, 0, 0) != 0) {
177 strncpy(realhost, host, sizeof(realhost));
180 /* We used the IPv4-only gethostbyname()... */
184 memcpy(&a, h->h_addr, sizeof(a));
185 /* This way we are always sure the h->h_name is valid :) */
186 strncpy(realhost, h->h_name, sizeof(realhost));
190 * This must be a numeric IPv4 address because it caused a
191 * success return from inet_addr.
193 ret->family = AF_INET;
194 strncpy(realhost, host, sizeof(realhost));
196 ret->address = ntohl(a);
197 realhost[lenof(realhost)-1] = '\0';
198 *canonicalname = smalloc(1+strlen(realhost));
199 strcpy(*canonicalname, realhost);
203 SockAddr sk_nonamelookup(char *host)
205 SockAddr ret = smalloc(sizeof(struct SockAddr_tag));
207 ret->family = AF_UNSPEC;
208 strncpy(ret->hostname, host, lenof(ret->hostname));
209 ret->hostname[lenof(ret->hostname)-1] = '\0';
213 void sk_getaddr(SockAddr addr, char *buf, int buflen)
216 if (addr->family == AF_INET6) {
217 FIXME; /* I don't know how to get a text form of an IPv6 address. */
220 if (addr->family == AF_INET) {
222 a.s_addr = htonl(addr->address);
223 strncpy(buf, inet_ntoa(a), buflen);
224 buf[buflen-1] = '\0';
226 assert(addr->family == AF_UNSPEC);
227 strncpy(buf, addr->hostname, buflen);
228 buf[buflen-1] = '\0';
232 int sk_hostname_is_local(char *name)
234 return !strcmp(name, "localhost");
237 int sk_address_is_local(SockAddr addr)
240 if (addr->family == AF_INET6) {
241 FIXME; /* someone who can compile for IPV6 had better do this bit */
244 if (addr->family == AF_INET) {
246 a.s_addr = htonl(addr->address);
247 return ipv4_is_loopback(a);
249 assert(addr->family == AF_UNSPEC);
250 return 0; /* we don't know; assume not */
254 int sk_addrtype(SockAddr addr)
256 return (addr->family == AF_INET ? ADDRTYPE_IPV4 :
258 addr->family == AF_INET6 ? ADDRTYPE_IPV6 :
263 void sk_addrcopy(SockAddr addr, char *buf)
265 assert(addr->family != AF_UNSPEC);
267 if (addr->family == AF_INET6) {
268 memcpy(buf, (char*) addr->ai, 16);
271 if (addr->family == AF_INET) {
273 a.s_addr = htonl(addr->address);
274 memcpy(buf, (char*) &a.s_addr, 4);
278 void sk_addr_free(SockAddr addr)
283 static Plug sk_tcp_plug(Socket sock, Plug p)
285 Actual_Socket s = (Actual_Socket) sock;
292 static void sk_tcp_flush(Socket s)
295 * We send data to the socket as soon as we can anyway,
296 * so we don't need to do anything here. :-)
300 static void sk_tcp_close(Socket s);
301 static int sk_tcp_write(Socket s, const char *data, int len);
302 static int sk_tcp_write_oob(Socket s, const char *data, int len);
303 static void sk_tcp_set_private_ptr(Socket s, void *ptr);
304 static void *sk_tcp_get_private_ptr(Socket s);
305 static void sk_tcp_set_frozen(Socket s, int is_frozen);
306 static char *sk_tcp_socket_error(Socket s);
308 Socket sk_register(void *sock, Plug plug)
310 static struct socket_function_table fn_table = {
316 sk_tcp_set_private_ptr,
317 sk_tcp_get_private_ptr,
325 * Create Socket structure.
327 ret = smalloc(sizeof(struct Socket_tag));
331 bufchain_init(&ret->output_data);
332 ret->writable = 1; /* to start with */
333 ret->sending_oob = 0;
335 ret->frozen_readable = 0;
336 ret->localhost_only = 0; /* unused, but best init anyway */
337 ret->pending_error = 0;
338 ret->oobpending = FALSE;
344 ret->error = error_string(errno);
355 Socket sk_new(SockAddr addr, int port, int privport, int oobinline,
356 int nodelay, Plug plug)
358 static struct socket_function_table fn_table = {
364 sk_tcp_set_private_ptr,
365 sk_tcp_get_private_ptr,
372 struct sockaddr_in6 a6;
374 struct sockaddr_in a;
380 * Create Socket structure.
382 ret = smalloc(sizeof(struct Socket_tag));
386 bufchain_init(&ret->output_data);
387 ret->connected = 0; /* to start with */
388 ret->writable = 0; /* to start with */
389 ret->sending_oob = 0;
391 ret->frozen_readable = 0;
392 ret->localhost_only = 0; /* unused, but best init anyway */
393 ret->pending_error = 0;
394 ret->oobpending = FALSE;
400 assert(addr->family != AF_UNSPEC);
401 s = socket(addr->family, SOCK_STREAM, 0);
405 ret->error = error_string(errno);
409 ret->oobinline = oobinline;
412 setsockopt(s, SOL_SOCKET, SO_OOBINLINE, (void *) &b, sizeof(b));
417 setsockopt(s, IPPROTO_TCP, TCP_NODELAY, (void *) &b, sizeof(b));
421 * Bind to local address.
424 localport = 1023; /* count from 1023 downwards */
426 localport = 0; /* just use port 0 (ie kernel picks) */
428 /* Loop round trying to bind */
433 if (addr->family == AF_INET6) {
434 memset(&a6, 0, sizeof(a6));
435 a6.sin6_family = AF_INET6;
436 /*a6.sin6_addr = in6addr_any; *//* == 0 */
437 a6.sin6_port = htons(localport);
441 a.sin_family = AF_INET;
442 a.sin_addr.s_addr = htonl(INADDR_ANY);
443 a.sin_port = htons(localport);
446 retcode = bind(s, (addr->family == AF_INET6 ?
447 (struct sockaddr *) &a6 :
448 (struct sockaddr *) &a),
450 AF_INET6 ? sizeof(a6) : sizeof(a)));
452 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
459 if (err != EADDRINUSE) /* failed, for a bad reason */
464 break; /* we're only looping once */
467 break; /* we might have got to the end */
471 ret->error = error_string(err);
476 * Connect to remote address.
479 if (addr->family == AF_INET6) {
480 memset(&a, 0, sizeof(a));
481 a6.sin6_family = AF_INET6;
482 a6.sin6_port = htons((short) port);
484 ((struct sockaddr_in6 *) addr->ai->ai_addr)->sin6_addr;
488 a.sin_family = AF_INET;
489 a.sin_addr.s_addr = htonl(addr->address);
490 a.sin_port = htons((short) port);
495 ioctl(s, FIONBIO, &i);
500 connect(s, ((addr->family == AF_INET6) ?
501 (struct sockaddr *) &a6 : (struct sockaddr *) &a),
502 (addr->family == AF_INET6) ? sizeof(a6) : sizeof(a))
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.
525 Socket sk_newlistener(char *srcaddr, int port, Plug plug, int local_host_only)
527 static struct socket_function_table fn_table = {
533 sk_tcp_set_private_ptr,
534 sk_tcp_get_private_ptr,
541 struct sockaddr_in6 a6;
543 struct sockaddr_in a;
550 * Create Socket structure.
552 ret = smalloc(sizeof(struct Socket_tag));
556 bufchain_init(&ret->output_data);
557 ret->writable = 0; /* to start with */
558 ret->sending_oob = 0;
560 ret->frozen_readable = 0;
561 ret->localhost_only = local_host_only;
562 ret->pending_error = 0;
563 ret->oobpending = FALSE;
569 s = socket(AF_INET, SOCK_STREAM, 0);
573 ret->error = error_string(errno);
579 setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (const char *)&on, sizeof(on));
582 if (addr->family == AF_INET6) {
583 memset(&a6, 0, sizeof(a6));
584 a6.sin6_family = AF_INET6;
585 /* FIXME: srcaddr is ignored for IPv6, because I (SGT) don't
586 * know how to do it. :-) */
588 a6.sin6_addr = in6addr_loopback;
590 a6.sin6_addr = in6addr_any;
591 a6.sin6_port = htons(port);
596 a.sin_family = AF_INET;
599 * Bind to source address. First try an explicitly
603 a.sin_addr.s_addr = inet_addr(srcaddr);
604 if (a.sin_addr.s_addr != INADDR_NONE) {
605 /* Override localhost_only with specified listen addr. */
606 ret->localhost_only = ipv4_is_loopback(a.sin_addr);
612 * ... and failing that, go with one of the standard ones.
616 a.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
618 a.sin_addr.s_addr = htonl(INADDR_ANY);
621 a.sin_port = htons((short)port);
624 retcode = bind(s, (addr->family == AF_INET6 ?
625 (struct sockaddr *) &a6 :
626 (struct sockaddr *) &a),
628 AF_INET6 ? sizeof(a6) : sizeof(a)));
630 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);
655 static void sk_tcp_close(Socket sock)
657 Actual_Socket s = (Actual_Socket) sock;
665 * The function which tries to send on a socket once it's deemed
668 void try_send(Actual_Socket s)
670 while (s->sending_oob || bufchain_size(&s->output_data) > 0) {
676 if (s->sending_oob) {
677 urgentflag = MSG_OOB;
678 len = s->sending_oob;
682 bufchain_prefix(&s->output_data, &data, &len);
684 nsent = send(s->s, data, len, urgentflag);
685 noise_ultralight(nsent);
687 err = (nsent < 0 ? errno : 0);
688 if (err == EWOULDBLOCK) {
690 * Perfectly normal: we've sent all we can for the moment.
694 } else if (nsent == 0 ||
695 err == ECONNABORTED || err == ECONNRESET) {
697 * If send() returns CONNABORTED or CONNRESET, we
698 * unfortunately can't just call plug_closing(),
699 * because it's quite likely that we're currently
700 * _in_ a call from the code we'd be calling back
701 * to, so we'd have to make half the SSH code
702 * reentrant. Instead we flag a pending error on
703 * the socket, to be dealt with (by calling
704 * plug_closing()) at some suitable future moment.
706 s->pending_error = err;
709 /* We're inside the Unix frontend here, so we know
710 * that the frontend handle is unnecessary. */
711 logevent(NULL, error_string(err));
712 fatalbox("%s", error_string(err));
715 if (s->sending_oob) {
717 memmove(s->oobdata, s->oobdata+nsent, len-nsent);
718 s->sending_oob = len - nsent;
723 bufchain_consume(&s->output_data, nsent);
729 static int sk_tcp_write(Socket sock, const char *buf, int len)
731 Actual_Socket s = (Actual_Socket) sock;
734 * Add the data to the buffer list on the socket.
736 bufchain_add(&s->output_data, buf, len);
739 * Now try sending from the start of the buffer list.
744 return bufchain_size(&s->output_data);
747 static int sk_tcp_write_oob(Socket sock, const char *buf, int len)
749 Actual_Socket s = (Actual_Socket) sock;
752 * Replace the buffer list on the socket with the data.
754 bufchain_clear(&s->output_data);
755 assert(len <= sizeof(s->oobdata));
756 memcpy(s->oobdata, buf, len);
757 s->sending_oob = len;
760 * Now try sending from the start of the buffer list.
765 return s->sending_oob;
768 int select_result(int fd, int event)
772 char buf[20480]; /* nice big buffer for plenty of speed */
776 /* Find the Socket structure */
777 s = find234(sktree, (void *) fd, cmpforsearch);
779 return 1; /* boggle */
781 noise_ultralight(event);
784 case 4: /* exceptional */
787 * On a non-oobinline socket, this indicates that we
788 * can immediately perform an OOB read and get back OOB
789 * data, which we will send to the back end with
790 * type==2 (urgent data).
792 ret = recv(s->s, buf, sizeof(buf), MSG_OOB);
793 noise_ultralight(ret);
795 char *str = (ret == 0 ? "Internal networking trouble" :
796 error_string(errno));
797 /* We're inside the Unix frontend here, so we know
798 * that the frontend handle is unnecessary. */
802 return plug_receive(s->plug, 2, buf, ret);
808 * If we reach here, this is an oobinline socket, which
809 * means we should set s->oobpending and then deal with it
810 * when we get called for the readability event (which
811 * should also occur).
813 s->oobpending = TRUE;
815 case 1: /* readable; also acceptance */
818 * On a listening socket, the readability event means a
819 * connection is ready to be accepted.
821 struct sockaddr_in isa;
822 int addrlen = sizeof(struct sockaddr_in);
823 int t; /* socket of connection */
825 memset(&isa, 0, sizeof(struct sockaddr_in));
827 t = accept(s->s,(struct sockaddr *)&isa,&addrlen);
832 if (s->localhost_only && !ipv4_is_loopback(isa.sin_addr)) {
833 close(t); /* someone let nonlocal through?! */
834 } else if (plug_accepting(s->plug, (void*)t)) {
835 close(t); /* denied or error */
841 * If we reach here, this is not a listening socket, so
842 * readability really means readability.
845 /* In the case the socket is still frozen, we don't even bother */
847 s->frozen_readable = 1;
852 * We have received data on the socket. For an oobinline
853 * socket, this might be data _before_ an urgent pointer,
854 * in which case we send it to the back end with type==1
855 * (data prior to urgent).
857 if (s->oobinline && s->oobpending) {
859 if (ioctl(s->s, SIOCATMARK, &atmark) == 0 && atmark)
860 s->oobpending = FALSE; /* clear this indicator */
864 ret = recv(s->s, buf, s->oobpending ? 1 : sizeof(buf), 0);
865 noise_ultralight(ret);
867 if (errno == EWOULDBLOCK) {
872 return plug_closing(s->plug, error_string(errno), errno, 0);
873 } else if (0 == ret) {
874 return plug_closing(s->plug, NULL, 0, 0);
876 return plug_receive(s->plug, atmark ? 0 : 1, buf, ret);
879 case 2: /* writable */
882 * select() reports a socket as _writable_ when an
883 * asynchronous connection is completed.
885 s->connected = s->writable = 1;
888 int bufsize_before, bufsize_after;
890 bufsize_before = s->sending_oob + bufchain_size(&s->output_data);
892 bufsize_after = s->sending_oob + bufchain_size(&s->output_data);
893 if (bufsize_after < bufsize_before)
894 plug_sent(s->plug, bufsize_after);
903 * Deal with socket errors detected in try_send().
905 void net_pending_errors(void)
911 * This might be a fiddly business, because it's just possible
912 * that handling a pending error on one socket might cause
913 * others to be closed. (I can't think of any reason this might
914 * happen in current SSH implementation, but to maintain
915 * generality of this network layer I'll assume the worst.)
917 * So what we'll do is search the socket list for _one_ socket
918 * with a pending error, and then handle it, and then search
919 * the list again _from the beginning_. Repeat until we make a
920 * pass with no socket errors present. That way we are
921 * protected against the socket list changing under our feet.
925 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
926 if (s->pending_error) {
928 * An error has occurred on this socket. Pass it to the
931 plug_closing(s->plug, error_string(s->pending_error),
932 s->pending_error, 0);
940 * Each socket abstraction contains a `void *' private field in
941 * which the client can keep state.
943 static void sk_tcp_set_private_ptr(Socket sock, void *ptr)
945 Actual_Socket s = (Actual_Socket) sock;
946 s->private_ptr = ptr;
949 static void *sk_tcp_get_private_ptr(Socket sock)
951 Actual_Socket s = (Actual_Socket) sock;
952 return s->private_ptr;
956 * Special error values are returned from sk_namelookup and sk_new
957 * if there's a problem. These functions extract an error message,
958 * or return NULL if there's no problem.
960 char *sk_addr_error(SockAddr addr)
964 static char *sk_tcp_socket_error(Socket sock)
966 Actual_Socket s = (Actual_Socket) sock;
970 static void sk_tcp_set_frozen(Socket sock, int is_frozen)
972 Actual_Socket s = (Actual_Socket) sock;
973 if (s->frozen == is_frozen)
975 s->frozen = is_frozen;
976 if (!is_frozen && s->frozen_readable) {
978 recv(s->s, &c, 1, MSG_PEEK);
980 s->frozen_readable = 0;
984 * For Unix select()-based frontends: enumerate all sockets
985 * currently active, and state whether we currently wish to receive
986 * select events on them for reading, writing and exceptional
989 static void set_rwx(Actual_Socket s, int *rwx)
993 val |= 2; /* write == connect */
994 if (s->connected && !s->frozen)
995 val |= 1 | 4; /* read, except */
996 if (bufchain_size(&s->output_data))
997 val |= 2; /* write */
999 val |= 1; /* read == accept */
1003 int first_socket(int *state, int *rwx)
1007 s = index234(sktree, (*state)++);
1010 return s ? s->s : -1;
1013 int next_socket(int *state, int *rwx)
1015 Actual_Socket s = index234(sktree, (*state)++);
1018 return s ? s->s : -1;
1021 int net_service_lookup(char *service)
1024 se = getservbyname(service, NULL);
1026 return ntohs(se->s_port);