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
));
206 ret
->family
= AF_UNSPEC
;
207 strncpy(ret
->hostname
, host
, lenof(ret
->hostname
));
208 ret
->hostname
[lenof(ret
->hostname
)-1] = '\0';
212 void sk_getaddr(SockAddr addr
, char *buf
, int buflen
)
215 if (addr
->family
== AF_INET6
) {
216 FIXME
; /* I don't know how to get a text form of an IPv6 address. */
219 if (addr
->family
== AF_INET
) {
221 a
.s_addr
= htonl(addr
->address
);
222 strncpy(buf
, inet_ntoa(a
), buflen
);
223 buf
[buflen
-1] = '\0';
225 assert(addr
->family
== AF_UNSPEC
);
226 strncpy(buf
, addr
->hostname
, buflen
);
227 buf
[buflen
-1] = '\0';
231 int sk_hostname_is_local(char *name
)
233 return !strcmp(name
, "localhost");
236 int sk_address_is_local(SockAddr addr
)
239 if (addr
->family
== AF_INET6
) {
240 FIXME
; /* someone who can compile for IPV6 had better do this bit */
243 if (addr
->family
== AF_INET
) {
245 a
.s_addr
= htonl(addr
->address
);
246 return ipv4_is_loopback(a
);
248 assert(addr
->family
== AF_UNSPEC
);
249 return 0; /* we don't know; assume not */
253 int sk_addrtype(SockAddr addr
)
255 return (addr
->family
== AF_INET ? ADDRTYPE_IPV4
:
257 addr
->family
== AF_INET6 ? ADDRTYPE_IPV6
:
262 void sk_addrcopy(SockAddr addr
, char *buf
)
264 assert(addr
->family
!= AF_UNSPEC
);
266 if (addr
->family
== AF_INET6
) {
267 memcpy(buf
, (char*) addr
->ai
, 16);
270 if (addr
->family
== AF_INET
) {
272 a
.s_addr
= htonl(addr
->address
);
273 memcpy(buf
, (char*) &a
.s_addr
, 4);
277 void sk_addr_free(SockAddr addr
)
282 static Plug
sk_tcp_plug(Socket sock
, Plug p
)
284 Actual_Socket s
= (Actual_Socket
) sock
;
291 static void sk_tcp_flush(Socket s
)
294 * We send data to the socket as soon as we can anyway,
295 * so we don't need to do anything here. :-)
299 static void sk_tcp_close(Socket s
);
300 static int sk_tcp_write(Socket s
, char *data
, int len
);
301 static int sk_tcp_write_oob(Socket s
, char *data
, int len
);
302 static void sk_tcp_set_private_ptr(Socket s
, void *ptr
);
303 static void *sk_tcp_get_private_ptr(Socket s
);
304 static void sk_tcp_set_frozen(Socket s
, int is_frozen
);
305 static char *sk_tcp_socket_error(Socket s
);
307 Socket
sk_register(void *sock
, Plug plug
)
309 static struct socket_function_table fn_table
= {
315 sk_tcp_set_private_ptr
,
316 sk_tcp_get_private_ptr
,
324 * Create Socket structure.
326 ret
= smalloc(sizeof(struct Socket_tag
));
330 bufchain_init(&ret
->output_data
);
331 ret
->writable
= 1; /* to start with */
332 ret
->sending_oob
= 0;
334 ret
->frozen_readable
= 0;
335 ret
->localhost_only
= 0; /* unused, but best init anyway */
336 ret
->pending_error
= 0;
337 ret
->oobpending
= FALSE
;
343 ret
->error
= error_string(errno
);
354 Socket
sk_new(SockAddr addr
, int port
, int privport
, int oobinline
,
355 int nodelay
, Plug plug
)
357 static struct socket_function_table fn_table
= {
363 sk_tcp_set_private_ptr
,
364 sk_tcp_get_private_ptr
,
371 struct sockaddr_in6 a6
;
373 struct sockaddr_in a
;
379 * Create Socket structure.
381 ret
= smalloc(sizeof(struct Socket_tag
));
385 bufchain_init(&ret
->output_data
);
386 ret
->connected
= 0; /* to start with */
387 ret
->writable
= 0; /* to start with */
388 ret
->sending_oob
= 0;
390 ret
->frozen_readable
= 0;
391 ret
->localhost_only
= 0; /* unused, but best init anyway */
392 ret
->pending_error
= 0;
393 ret
->oobpending
= FALSE
;
399 assert(addr
->family
!= AF_UNSPEC
);
400 s
= socket(addr
->family
, SOCK_STREAM
, 0);
404 ret
->error
= error_string(errno
);
408 ret
->oobinline
= oobinline
;
411 setsockopt(s
, SOL_SOCKET
, SO_OOBINLINE
, (void *) &b
, sizeof(b
));
416 setsockopt(s
, IPPROTO_TCP
, TCP_NODELAY
, (void *) &b
, sizeof(b
));
420 * Bind to local address.
423 localport
= 1023; /* count from 1023 downwards */
425 localport
= 0; /* just use port 0 (ie kernel picks) */
427 /* Loop round trying to bind */
432 if (addr
->family
== AF_INET6
) {
433 memset(&a6
, 0, sizeof(a6
));
434 a6
.sin6_family
= AF_INET6
;
435 /*a6.sin6_addr = in6addr_any; *//* == 0 */
436 a6
.sin6_port
= htons(localport
);
440 a
.sin_family
= AF_INET
;
441 a
.sin_addr
.s_addr
= htonl(INADDR_ANY
);
442 a
.sin_port
= htons(localport
);
445 retcode
= bind(s
, (addr
->family
== AF_INET6 ?
446 (struct sockaddr
*) &a6
:
447 (struct sockaddr
*) &a
),
449 AF_INET6 ?
sizeof(a6
) : sizeof(a
)));
451 retcode
= bind(s
, (struct sockaddr
*) &a
, sizeof(a
));
458 if (err
!= EADDRINUSE
) /* failed, for a bad reason */
463 break; /* we're only looping once */
466 break; /* we might have got to the end */
470 ret
->error
= error_string(err
);
475 * Connect to remote address.
478 if (addr
->family
== AF_INET6
) {
479 memset(&a
, 0, sizeof(a
));
480 a6
.sin6_family
= AF_INET6
;
481 a6
.sin6_port
= htons((short) port
);
483 ((struct sockaddr_in6
*) addr
->ai
->ai_addr
)->sin6_addr
;
487 a
.sin_family
= AF_INET
;
488 a
.sin_addr
.s_addr
= htonl(addr
->address
);
489 a
.sin_port
= htons((short) port
);
494 connect(s
, ((addr
->family
== AF_INET6
) ?
495 (struct sockaddr
*) &a6
: (struct sockaddr
*) &a
),
496 (addr
->family
== AF_INET6
) ?
sizeof(a6
) : sizeof(a
))
498 connect(s
, (struct sockaddr
*) &a
, sizeof(a
))
502 * FIXME: We are prepared to receive EWOULDBLOCK here,
503 * because we might want the connection to be made
504 * asynchronously; but how do we actually arrange this in
507 if ( errno
!= EWOULDBLOCK
) {
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
, 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
, 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 #ifdef FIXME_NONBLOCKING_CONNECTIONS
785 case FIXME
: /* connected */
786 s
->connected
= s
->writable
= 1;
789 case 4: /* exceptional */
792 * On a non-oobinline socket, this indicates that we
793 * can immediately perform an OOB read and get back OOB
794 * data, which we will send to the back end with
795 * type==2 (urgent data).
797 ret
= recv(s
->s
, buf
, sizeof(buf
), MSG_OOB
);
798 noise_ultralight(ret
);
800 char *str
= (ret
== 0 ?
"Internal networking trouble" :
801 error_string(errno
));
802 /* We're inside the Unix frontend here, so we know
803 * that the frontend handle is unnecessary. */
807 return plug_receive(s
->plug
, 2, buf
, ret
);
813 * If we reach here, this is an oobinline socket, which
814 * means we should set s->oobpending and then deal with it
815 * when we get called for the readability event (which
816 * should also occur).
818 s
->oobpending
= TRUE
;
820 case 1: /* readable; also acceptance */
823 * On a listening socket, the readability event means a
824 * connection is ready to be accepted.
826 struct sockaddr_in isa
;
827 int addrlen
= sizeof(struct sockaddr_in
);
828 int t
; /* socket of connection */
830 memset(&isa
, 0, sizeof(struct sockaddr_in
));
832 t
= accept(s
->s
,(struct sockaddr
*)&isa
,&addrlen
);
837 if (s
->localhost_only
&& !ipv4_is_loopback(isa
.sin_addr
)) {
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
);