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 return strerror(error
);
112 SockAddr
sk_namelookup(char *host
, char **canonicalname
)
114 SockAddr ret
= smalloc(sizeof(struct SockAddr_tag
));
116 struct hostent
*h
= NULL
;
119 /* Clear the structure and default to IPv4. */
120 memset(ret
, 0, sizeof(struct SockAddr_tag
));
121 ret
->family
= 0; /* We set this one when we have resolved the host. */
125 if ((a
= inet_addr(host
)) == (unsigned long) INADDR_NONE
) {
127 if (getaddrinfo(host
, NULL
, NULL
, &ret
->ai
) == 0) {
128 ret
->family
= ret
->ai
->ai_family
;
133 * Otherwise use the IPv4-only gethostbyname... (NOTE:
134 * we don't use gethostbyname as a fallback!)
136 if (ret
->family
== 0) {
137 /*debug(("Resolving \"%s\" with gethostbyname() (IPv4 only)...\n", host)); */
138 if ( (h
= gethostbyname(host
)) )
139 ret
->family
= AF_INET
;
141 if (ret
->family
== 0) {
142 ret
->error
= (h_errno
== HOST_NOT_FOUND
||
143 h_errno
== NO_DATA
||
144 h_errno
== NO_ADDRESS ?
"Host does not exist" :
145 h_errno
== TRY_AGAIN ?
146 "Temporary name service failure" :
147 "gethostbyname: unknown error");
153 /* If we got an address info use that... */
156 /* Are we in IPv4 fallback mode? */
157 /* We put the IPv4 address into the a variable so we can further-on use the IPv4 code... */
158 if (ret
->family
== AF_INET
)
160 (char *) &((struct sockaddr_in
*) ret
->ai
->
161 ai_addr
)->sin_addr
, sizeof(a
));
163 /* Now let's find that canonicalname... */
164 if (getnameinfo((struct sockaddr
*) ret
->ai
->ai_addr
,
166 AF_INET ?
sizeof(struct sockaddr_in
) :
167 sizeof(struct sockaddr_in6
), realhost
,
168 sizeof(realhost
), NULL
, 0, 0) != 0) {
169 strncpy(realhost
, host
, sizeof(realhost
));
172 /* We used the IPv4-only gethostbyname()... */
176 memcpy(&a
, h
->h_addr
, sizeof(a
));
177 /* This way we are always sure the h->h_name is valid :) */
178 strncpy(realhost
, h
->h_name
, sizeof(realhost
));
182 * This must be a numeric IPv4 address because it caused a
183 * success return from inet_addr.
185 ret
->family
= AF_INET
;
186 strncpy(realhost
, host
, sizeof(realhost
));
188 ret
->address
= ntohl(a
);
189 realhost
[lenof(realhost
)-1] = '\0';
190 *canonicalname
= smalloc(1+strlen(realhost
));
191 strcpy(*canonicalname
, realhost
);
195 void sk_getaddr(SockAddr addr
, char *buf
, int buflen
)
198 if (addr
->family
== AF_INET
) {
201 a
.s_addr
= htonl(addr
->address
);
202 strncpy(buf
, inet_ntoa(a
), buflen
);
205 FIXME
; /* I don't know how to get a text form of an IPv6 address. */
210 int sk_addrtype(SockAddr addr
)
212 return (addr
->family
== AF_INET ? ADDRTYPE_IPV4
: ADDRTYPE_IPV6
);
215 void sk_addrcopy(SockAddr addr
, char *buf
)
218 if (addr
->family
== AF_INET
) {
221 a
.s_addr
= htonl(addr
->address
);
222 memcpy(buf
, (char*) &a
.s_addr
, 4);
225 memcpy(buf
, (char*) addr
->ai
, 16);
230 void sk_addr_free(SockAddr addr
)
235 static Plug
sk_tcp_plug(Socket sock
, Plug p
)
237 Actual_Socket s
= (Actual_Socket
) sock
;
244 static void sk_tcp_flush(Socket s
)
247 * We send data to the socket as soon as we can anyway,
248 * so we don't need to do anything here. :-)
252 static void sk_tcp_close(Socket s
);
253 static int sk_tcp_write(Socket s
, char *data
, int len
);
254 static int sk_tcp_write_oob(Socket s
, char *data
, int len
);
255 static void sk_tcp_set_private_ptr(Socket s
, void *ptr
);
256 static void *sk_tcp_get_private_ptr(Socket s
);
257 static void sk_tcp_set_frozen(Socket s
, int is_frozen
);
258 static char *sk_tcp_socket_error(Socket s
);
260 Socket
sk_register(void *sock
, Plug plug
)
262 static struct socket_function_table fn_table
= {
268 sk_tcp_set_private_ptr
,
269 sk_tcp_get_private_ptr
,
277 * Create Socket structure.
279 ret
= smalloc(sizeof(struct Socket_tag
));
283 bufchain_init(&ret
->output_data
);
284 ret
->writable
= 1; /* to start with */
285 ret
->sending_oob
= 0;
287 ret
->frozen_readable
= 0;
288 ret
->localhost_only
= 0; /* unused, but best init anyway */
289 ret
->pending_error
= 0;
290 ret
->oobpending
= FALSE
;
296 ret
->error
= error_string(errno
);
307 Socket
sk_new(SockAddr addr
, int port
, int privport
, int oobinline
,
308 int nodelay
, Plug plug
)
310 static struct socket_function_table fn_table
= {
316 sk_tcp_set_private_ptr
,
317 sk_tcp_get_private_ptr
,
324 struct sockaddr_in6 a6
;
326 struct sockaddr_in a
;
332 * Create Socket structure.
334 ret
= smalloc(sizeof(struct Socket_tag
));
338 bufchain_init(&ret
->output_data
);
339 ret
->connected
= 0; /* to start with */
340 ret
->writable
= 0; /* to start with */
341 ret
->sending_oob
= 0;
343 ret
->frozen_readable
= 0;
344 ret
->localhost_only
= 0; /* unused, but best init anyway */
345 ret
->pending_error
= 0;
346 ret
->oobpending
= FALSE
;
352 s
= socket(addr
->family
, SOCK_STREAM
, 0);
356 ret
->error
= error_string(errno
);
360 ret
->oobinline
= oobinline
;
363 setsockopt(s
, SOL_SOCKET
, SO_OOBINLINE
, (void *) &b
, sizeof(b
));
368 setsockopt(s
, IPPROTO_TCP
, TCP_NODELAY
, (void *) &b
, sizeof(b
));
372 * Bind to local address.
375 localport
= 1023; /* count from 1023 downwards */
377 localport
= 0; /* just use port 0 (ie kernel picks) */
379 /* Loop round trying to bind */
384 if (addr
->family
== AF_INET6
) {
385 memset(&a6
, 0, sizeof(a6
));
386 a6
.sin6_family
= AF_INET6
;
387 /*a6.sin6_addr = in6addr_any; *//* == 0 */
388 a6
.sin6_port
= htons(localport
);
392 a
.sin_family
= AF_INET
;
393 a
.sin_addr
.s_addr
= htonl(INADDR_ANY
);
394 a
.sin_port
= htons(localport
);
397 retcode
= bind(s
, (addr
->family
== AF_INET6 ?
398 (struct sockaddr
*) &a6
:
399 (struct sockaddr
*) &a
),
401 AF_INET6 ?
sizeof(a6
) : sizeof(a
)));
403 retcode
= bind(s
, (struct sockaddr
*) &a
, sizeof(a
));
410 if (err
!= EADDRINUSE
) /* failed, for a bad reason */
415 break; /* we're only looping once */
418 break; /* we might have got to the end */
422 ret
->error
= error_string(err
);
427 * Connect to remote address.
430 if (addr
->family
== AF_INET6
) {
431 memset(&a
, 0, sizeof(a
));
432 a6
.sin6_family
= AF_INET6
;
433 a6
.sin6_port
= htons((short) port
);
435 ((struct sockaddr_in6
*) addr
->ai
->ai_addr
)->sin6_addr
;
439 a
.sin_family
= AF_INET
;
440 a
.sin_addr
.s_addr
= htonl(addr
->address
);
441 a
.sin_port
= htons((short) port
);
446 connect(s
, ((addr
->family
== AF_INET6
) ?
447 (struct sockaddr
*) &a6
: (struct sockaddr
*) &a
),
448 (addr
->family
== AF_INET6
) ?
sizeof(a6
) : sizeof(a
))
450 connect(s
, (struct sockaddr
*) &a
, sizeof(a
))
454 * FIXME: We are prepared to receive EWOULDBLOCK here,
455 * because we might want the connection to be made
456 * asynchronously; but how do we actually arrange this in
459 if ( errno
!= EWOULDBLOCK
) {
460 ret
->error
= error_string(errno
);
465 * If we _don't_ get EWOULDBLOCK, the connect has completed
466 * and we should set the socket as connected and writable.
477 Socket
sk_newlistener(int port
, Plug plug
, int local_host_only
)
479 static struct socket_function_table fn_table
= {
485 sk_tcp_set_private_ptr
,
486 sk_tcp_get_private_ptr
,
493 struct sockaddr_in6 a6
;
495 struct sockaddr_in a
;
502 * Create Socket structure.
504 ret
= smalloc(sizeof(struct Socket_tag
));
508 bufchain_init(&ret
->output_data
);
509 ret
->writable
= 0; /* to start with */
510 ret
->sending_oob
= 0;
512 ret
->frozen_readable
= 0;
513 ret
->localhost_only
= local_host_only
;
514 ret
->pending_error
= 0;
515 ret
->oobpending
= FALSE
;
521 s
= socket(AF_INET
, SOCK_STREAM
, 0);
525 ret
->error
= error_string(errno
);
531 setsockopt(s
, SOL_SOCKET
, SO_REUSEADDR
, (const char *)&on
, sizeof(on
));
534 if (addr
->family
== AF_INET6
) {
535 memset(&a6
, 0, sizeof(a6
));
536 a6
.sin6_family
= AF_INET6
;
538 a6
.sin6_addr
= in6addr_loopback
;
540 a6
.sin6_addr
= in6addr_any
;
541 a6
.sin6_port
= htons(port
);
545 a
.sin_family
= AF_INET
;
547 a
.sin_addr
.s_addr
= htonl(INADDR_LOOPBACK
);
549 a
.sin_addr
.s_addr
= htonl(INADDR_ANY
);
550 a
.sin_port
= htons((short)port
);
553 retcode
= bind(s
, (addr
->family
== AF_INET6 ?
554 (struct sockaddr
*) &a6
:
555 (struct sockaddr
*) &a
),
557 AF_INET6 ?
sizeof(a6
) : sizeof(a
)));
559 retcode
= bind(s
, (struct sockaddr
*) &a
, sizeof(a
));
568 ret
->error
= error_string(err
);
573 if (listen(s
, SOMAXCONN
) < 0) {
575 ret
->error
= error_string(errno
);
584 static void sk_tcp_close(Socket sock
)
586 Actual_Socket s
= (Actual_Socket
) sock
;
594 * The function which tries to send on a socket once it's deemed
597 void try_send(Actual_Socket s
)
599 while (s
->sending_oob
|| bufchain_size(&s
->output_data
) > 0) {
605 if (s
->sending_oob
) {
606 urgentflag
= MSG_OOB
;
607 len
= s
->sending_oob
;
611 bufchain_prefix(&s
->output_data
, &data
, &len
);
613 nsent
= send(s
->s
, data
, len
, urgentflag
);
614 noise_ultralight(nsent
);
616 err
= (nsent
< 0 ? errno
: 0);
617 if (err
== EWOULDBLOCK
) {
619 * Perfectly normal: we've sent all we can for the moment.
623 } else if (nsent
== 0 ||
624 err
== ECONNABORTED
|| err
== ECONNRESET
) {
626 * If send() returns CONNABORTED or CONNRESET, we
627 * unfortunately can't just call plug_closing(),
628 * because it's quite likely that we're currently
629 * _in_ a call from the code we'd be calling back
630 * to, so we'd have to make half the SSH code
631 * reentrant. Instead we flag a pending error on
632 * the socket, to be dealt with (by calling
633 * plug_closing()) at some suitable future moment.
635 s
->pending_error
= err
;
638 /* We're inside the Unix frontend here, so we know
639 * that the frontend handle is unnecessary. */
640 logevent(NULL
, error_string(err
));
641 fatalbox("%s", error_string(err
));
644 if (s
->sending_oob
) {
646 memmove(s
->oobdata
, s
->oobdata
+nsent
, len
-nsent
);
647 s
->sending_oob
= len
- nsent
;
652 bufchain_consume(&s
->output_data
, nsent
);
658 static int sk_tcp_write(Socket sock
, char *buf
, int len
)
660 Actual_Socket s
= (Actual_Socket
) sock
;
663 * Add the data to the buffer list on the socket.
665 bufchain_add(&s
->output_data
, buf
, len
);
668 * Now try sending from the start of the buffer list.
673 return bufchain_size(&s
->output_data
);
676 static int sk_tcp_write_oob(Socket sock
, char *buf
, int len
)
678 Actual_Socket s
= (Actual_Socket
) sock
;
681 * Replace the buffer list on the socket with the data.
683 bufchain_clear(&s
->output_data
);
684 assert(len
<= sizeof(s
->oobdata
));
685 memcpy(s
->oobdata
, buf
, len
);
686 s
->sending_oob
= len
;
689 * Now try sending from the start of the buffer list.
694 return s
->sending_oob
;
697 int select_result(int fd
, int event
)
701 char buf
[20480]; /* nice big buffer for plenty of speed */
705 /* Find the Socket structure */
706 s
= find234(sktree
, (void *) fd
, cmpforsearch
);
708 return 1; /* boggle */
710 noise_ultralight(event
);
713 #ifdef FIXME_NONBLOCKING_CONNECTIONS
714 case FIXME
: /* connected */
715 s
->connected
= s
->writable
= 1;
718 case 4: /* exceptional */
721 * On a non-oobinline socket, this indicates that we
722 * can immediately perform an OOB read and get back OOB
723 * data, which we will send to the back end with
724 * type==2 (urgent data).
726 ret
= recv(s
->s
, buf
, sizeof(buf
), MSG_OOB
);
727 noise_ultralight(ret
);
729 char *str
= (ret
== 0 ?
"Internal networking trouble" :
730 error_string(errno
));
731 /* We're inside the Unix frontend here, so we know
732 * that the frontend handle is unnecessary. */
736 return plug_receive(s
->plug
, 2, buf
, ret
);
742 * If we reach here, this is an oobinline socket, which
743 * means we should set s->oobpending and then deal with it
744 * when we get called for the readability event (which
745 * should also occur).
747 s
->oobpending
= TRUE
;
749 case 1: /* readable; also acceptance */
752 * On a listening socket, the readability event means a
753 * connection is ready to be accepted.
755 struct sockaddr_in isa
;
756 int addrlen
= sizeof(struct sockaddr_in
);
757 int t
; /* socket of connection */
759 memset(&isa
, 0, sizeof(struct sockaddr_in
));
761 t
= accept(s
->s
,(struct sockaddr
*)&isa
,&addrlen
);
766 if (s
->localhost_only
&&
767 ntohl(isa
.sin_addr
.s_addr
) != INADDR_LOOPBACK
) {
768 close(t
); /* someone let nonlocal through?! */
769 } else if (plug_accepting(s
->plug
, (void*)t
)) {
770 close(t
); /* denied or error */
776 * If we reach here, this is not a listening socket, so
777 * readability really means readability.
780 /* In the case the socket is still frozen, we don't even bother */
782 s
->frozen_readable
= 1;
787 * We have received data on the socket. For an oobinline
788 * socket, this might be data _before_ an urgent pointer,
789 * in which case we send it to the back end with type==1
790 * (data prior to urgent).
792 if (s
->oobinline
&& s
->oobpending
) {
794 if (ioctl(s
->s
, SIOCATMARK
, &atmark
) == 0 && atmark
)
795 s
->oobpending
= FALSE
; /* clear this indicator */
799 ret
= recv(s
->s
, buf
, s
->oobpending ?
1 : sizeof(buf
), 0);
800 noise_ultralight(ret
);
802 if (errno
== EWOULDBLOCK
) {
807 return plug_closing(s
->plug
, error_string(errno
), errno
, 0);
808 } else if (0 == ret
) {
809 return plug_closing(s
->plug
, NULL
, 0, 0);
811 return plug_receive(s
->plug
, atmark ?
0 : 1, buf
, ret
);
814 case 2: /* writable */
816 int bufsize_before
, bufsize_after
;
818 bufsize_before
= s
->sending_oob
+ bufchain_size(&s
->output_data
);
820 bufsize_after
= s
->sending_oob
+ bufchain_size(&s
->output_data
);
821 if (bufsize_after
< bufsize_before
)
822 plug_sent(s
->plug
, bufsize_after
);
831 * Deal with socket errors detected in try_send().
833 void net_pending_errors(void)
839 * This might be a fiddly business, because it's just possible
840 * that handling a pending error on one socket might cause
841 * others to be closed. (I can't think of any reason this might
842 * happen in current SSH implementation, but to maintain
843 * generality of this network layer I'll assume the worst.)
845 * So what we'll do is search the socket list for _one_ socket
846 * with a pending error, and then handle it, and then search
847 * the list again _from the beginning_. Repeat until we make a
848 * pass with no socket errors present. That way we are
849 * protected against the socket list changing under our feet.
853 for (i
= 0; (s
= index234(sktree
, i
)) != NULL
; i
++) {
854 if (s
->pending_error
) {
856 * An error has occurred on this socket. Pass it to the
859 plug_closing(s
->plug
, error_string(s
->pending_error
),
860 s
->pending_error
, 0);
868 * Each socket abstraction contains a `void *' private field in
869 * which the client can keep state.
871 static void sk_tcp_set_private_ptr(Socket sock
, void *ptr
)
873 Actual_Socket s
= (Actual_Socket
) sock
;
874 s
->private_ptr
= ptr
;
877 static void *sk_tcp_get_private_ptr(Socket sock
)
879 Actual_Socket s
= (Actual_Socket
) sock
;
880 return s
->private_ptr
;
884 * Special error values are returned from sk_namelookup and sk_new
885 * if there's a problem. These functions extract an error message,
886 * or return NULL if there's no problem.
888 char *sk_addr_error(SockAddr addr
)
892 static char *sk_tcp_socket_error(Socket sock
)
894 Actual_Socket s
= (Actual_Socket
) sock
;
898 static void sk_tcp_set_frozen(Socket sock
, int is_frozen
)
900 Actual_Socket s
= (Actual_Socket
) sock
;
901 if (s
->frozen
== is_frozen
)
903 s
->frozen
= is_frozen
;
904 if (!is_frozen
&& s
->frozen_readable
) {
906 recv(s
->s
, &c
, 1, MSG_PEEK
);
908 s
->frozen_readable
= 0;
912 * For Unix select()-based frontends: enumerate all sockets
913 * currently active, and state whether we currently wish to receive
914 * select events on them for reading, writing and exceptional
917 static void set_rwx(Actual_Socket s
, int *rwx
)
920 if (s
->connected
&& !s
->frozen
)
921 val
|= 1 | 4; /* read, except */
922 if (bufchain_size(&s
->output_data
))
923 val
|= 2; /* write */
925 val
|= 1; /* read == accept */
929 int first_socket(int *state
, int *rwx
)
933 s
= index234(sktree
, (*state
)++);
936 return s ? s
->s
: -1;
939 int next_socket(int *state
, int *rwx
)
941 Actual_Socket s
= index234(sktree
, (*state
)++);
944 return s ? s
->s
: -1;
947 int net_service_lookup(char *service
)
950 se
= getservbyname(service
, NULL
);
952 return ntohs(se
->s_port
);