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
;
60 /* address family this belongs to, AF_INET for IPv4, AF_INET6 for IPv6. */
62 unsigned long address
; /* Address IPv4 style. */
64 struct addrinfo
*ai
; /* Address IPv6 style. */
68 static tree234
*sktree
;
70 static int cmpfortree(void *av
, void *bv
)
72 Actual_Socket a
= (Actual_Socket
) av
, b
= (Actual_Socket
) bv
;
73 int as
= a
->s
, bs
= b
->s
;
81 static int cmpforsearch(void *av
, void *bv
)
83 Actual_Socket b
= (Actual_Socket
) bv
;
84 int as
= (int) av
, bs
= b
->s
;
94 sktree
= newtree234(cmpfortree
);
103 for (i
= 0; (s
= index234(sktree
, i
)) != NULL
; i
++) {
109 char *error_string(int error
)
111 return strerror(error
);
114 SockAddr
sk_namelookup(char *host
, char **canonicalname
)
116 SockAddr ret
= smalloc(sizeof(struct SockAddr_tag
));
118 struct hostent
*h
= NULL
;
121 /* Clear the structure and default to IPv4. */
122 memset(ret
, 0, sizeof(struct SockAddr_tag
));
123 ret
->family
= 0; /* We set this one when we have resolved the host. */
127 if ((a
= inet_addr(host
)) == (unsigned long) INADDR_NONE
) {
129 if (getaddrinfo(host
, NULL
, NULL
, &ret
->ai
) == 0) {
130 ret
->family
= ret
->ai
->ai_family
;
135 * Otherwise use the IPv4-only gethostbyname... (NOTE:
136 * we don't use gethostbyname as a fallback!)
138 if (ret
->family
== 0) {
139 /*debug(("Resolving \"%s\" with gethostbyname() (IPv4 only)...\n", host)); */
140 if ( (h
= gethostbyname(host
)) )
141 ret
->family
= AF_INET
;
143 if (ret
->family
== 0) {
144 ret
->error
= (h_errno
== HOST_NOT_FOUND
||
145 h_errno
== NO_DATA
||
146 h_errno
== NO_ADDRESS ?
"Host does not exist" :
147 h_errno
== TRY_AGAIN ?
148 "Temporary name service failure" :
149 "gethostbyname: unknown error");
155 /* If we got an address info use that... */
158 /* Are we in IPv4 fallback mode? */
159 /* We put the IPv4 address into the a variable so we can further-on use the IPv4 code... */
160 if (ret
->family
== AF_INET
)
162 (char *) &((struct sockaddr_in
*) ret
->ai
->
163 ai_addr
)->sin_addr
, sizeof(a
));
165 /* Now let's find that canonicalname... */
166 if (getnameinfo((struct sockaddr
*) ret
->ai
->ai_addr
,
168 AF_INET ?
sizeof(struct sockaddr_in
) :
169 sizeof(struct sockaddr_in6
), realhost
,
170 sizeof(realhost
), NULL
, 0, 0) != 0) {
171 strncpy(realhost
, host
, sizeof(realhost
));
174 /* We used the IPv4-only gethostbyname()... */
178 memcpy(&a
, h
->h_addr
, sizeof(a
));
179 /* This way we are always sure the h->h_name is valid :) */
180 strncpy(realhost
, h
->h_name
, sizeof(realhost
));
184 * This must be a numeric IPv4 address because it caused a
185 * success return from inet_addr.
187 ret
->family
= AF_INET
;
188 strncpy(realhost
, host
, sizeof(realhost
));
190 ret
->address
= ntohl(a
);
191 realhost
[lenof(realhost
)-1] = '\0';
192 *canonicalname
= smalloc(1+strlen(realhost
));
193 strcpy(*canonicalname
, realhost
);
197 void sk_getaddr(SockAddr addr
, char *buf
, int buflen
)
200 if (addr
->family
== AF_INET
) {
203 a
.s_addr
= htonl(addr
->address
);
204 strncpy(buf
, inet_ntoa(a
), buflen
);
207 FIXME
; /* I don't know how to get a text form of an IPv6 address. */
212 int sk_addrtype(SockAddr addr
)
214 return (addr
->family
== AF_INET ? ADDRTYPE_IPV4
: ADDRTYPE_IPV6
);
217 void sk_addrcopy(SockAddr addr
, char *buf
)
220 if (addr
->family
== AF_INET
) {
223 a
.s_addr
= htonl(addr
->address
);
224 memcpy(buf
, (char*) &a
.s_addr
, 4);
227 memcpy(buf
, (char*) addr
->ai
, 16);
232 void sk_addr_free(SockAddr addr
)
237 static Plug
sk_tcp_plug(Socket sock
, Plug p
)
239 Actual_Socket s
= (Actual_Socket
) sock
;
246 static void sk_tcp_flush(Socket s
)
249 * We send data to the socket as soon as we can anyway,
250 * so we don't need to do anything here. :-)
254 static void sk_tcp_close(Socket s
);
255 static int sk_tcp_write(Socket s
, char *data
, int len
);
256 static int sk_tcp_write_oob(Socket s
, char *data
, int len
);
257 static void sk_tcp_set_private_ptr(Socket s
, void *ptr
);
258 static void *sk_tcp_get_private_ptr(Socket s
);
259 static void sk_tcp_set_frozen(Socket s
, int is_frozen
);
260 static char *sk_tcp_socket_error(Socket s
);
262 Socket
sk_register(void *sock
, Plug plug
)
264 static struct socket_function_table fn_table
= {
270 sk_tcp_set_private_ptr
,
271 sk_tcp_get_private_ptr
,
279 * Create Socket structure.
281 ret
= smalloc(sizeof(struct Socket_tag
));
285 bufchain_init(&ret
->output_data
);
286 ret
->writable
= 1; /* to start with */
287 ret
->sending_oob
= 0;
289 ret
->frozen_readable
= 0;
290 ret
->localhost_only
= 0; /* unused, but best init anyway */
291 ret
->pending_error
= 0;
292 ret
->oobpending
= FALSE
;
298 ret
->error
= error_string(errno
);
309 Socket
sk_new(SockAddr addr
, int port
, int privport
, int oobinline
,
310 int nodelay
, Plug plug
)
312 static struct socket_function_table fn_table
= {
318 sk_tcp_set_private_ptr
,
319 sk_tcp_get_private_ptr
,
326 struct sockaddr_in6 a6
;
328 struct sockaddr_in a
;
334 * Create Socket structure.
336 ret
= smalloc(sizeof(struct Socket_tag
));
340 bufchain_init(&ret
->output_data
);
341 ret
->connected
= 0; /* to start with */
342 ret
->writable
= 0; /* to start with */
343 ret
->sending_oob
= 0;
345 ret
->frozen_readable
= 0;
346 ret
->localhost_only
= 0; /* unused, but best init anyway */
347 ret
->pending_error
= 0;
348 ret
->oobpending
= FALSE
;
354 s
= socket(addr
->family
, SOCK_STREAM
, 0);
358 ret
->error
= error_string(errno
);
362 ret
->oobinline
= oobinline
;
365 setsockopt(s
, SOL_SOCKET
, SO_OOBINLINE
, (void *) &b
, sizeof(b
));
370 setsockopt(s
, IPPROTO_TCP
, TCP_NODELAY
, (void *) &b
, sizeof(b
));
374 * Bind to local address.
377 localport
= 1023; /* count from 1023 downwards */
379 localport
= 0; /* just use port 0 (ie kernel picks) */
381 /* Loop round trying to bind */
386 if (addr
->family
== AF_INET6
) {
387 memset(&a6
, 0, sizeof(a6
));
388 a6
.sin6_family
= AF_INET6
;
389 /*a6.sin6_addr = in6addr_any; *//* == 0 */
390 a6
.sin6_port
= htons(localport
);
394 a
.sin_family
= AF_INET
;
395 a
.sin_addr
.s_addr
= htonl(INADDR_ANY
);
396 a
.sin_port
= htons(localport
);
399 retcode
= bind(s
, (addr
->family
== AF_INET6 ?
400 (struct sockaddr
*) &a6
:
401 (struct sockaddr
*) &a
),
403 AF_INET6 ?
sizeof(a6
) : sizeof(a
)));
405 retcode
= bind(s
, (struct sockaddr
*) &a
, sizeof(a
));
412 if (err
!= EADDRINUSE
) /* failed, for a bad reason */
417 break; /* we're only looping once */
420 break; /* we might have got to the end */
424 ret
->error
= error_string(err
);
429 * Connect to remote address.
432 if (addr
->family
== AF_INET6
) {
433 memset(&a
, 0, sizeof(a
));
434 a6
.sin6_family
= AF_INET6
;
435 a6
.sin6_port
= htons((short) port
);
437 ((struct sockaddr_in6
*) addr
->ai
->ai_addr
)->sin6_addr
;
441 a
.sin_family
= AF_INET
;
442 a
.sin_addr
.s_addr
= htonl(addr
->address
);
443 a
.sin_port
= htons((short) port
);
448 connect(s
, ((addr
->family
== AF_INET6
) ?
449 (struct sockaddr
*) &a6
: (struct sockaddr
*) &a
),
450 (addr
->family
== AF_INET6
) ?
sizeof(a6
) : sizeof(a
))
452 connect(s
, (struct sockaddr
*) &a
, sizeof(a
))
456 * FIXME: We are prepared to receive EWOULDBLOCK here,
457 * because we might want the connection to be made
458 * asynchronously; but how do we actually arrange this in
461 if ( errno
!= EWOULDBLOCK
) {
462 ret
->error
= error_string(errno
);
467 * If we _don't_ get EWOULDBLOCK, the connect has completed
468 * and we should set the socket as connected and writable.
479 Socket
sk_newlistener(char *srcaddr
, int port
, Plug plug
, int local_host_only
)
481 static struct socket_function_table fn_table
= {
487 sk_tcp_set_private_ptr
,
488 sk_tcp_get_private_ptr
,
495 struct sockaddr_in6 a6
;
497 struct sockaddr_in a
;
504 * Create Socket structure.
506 ret
= smalloc(sizeof(struct Socket_tag
));
510 bufchain_init(&ret
->output_data
);
511 ret
->writable
= 0; /* to start with */
512 ret
->sending_oob
= 0;
514 ret
->frozen_readable
= 0;
515 ret
->localhost_only
= local_host_only
;
516 ret
->pending_error
= 0;
517 ret
->oobpending
= FALSE
;
523 s
= socket(AF_INET
, SOCK_STREAM
, 0);
527 ret
->error
= error_string(errno
);
533 setsockopt(s
, SOL_SOCKET
, SO_REUSEADDR
, (const char *)&on
, sizeof(on
));
536 if (addr
->family
== AF_INET6
) {
537 memset(&a6
, 0, sizeof(a6
));
538 a6
.sin6_family
= AF_INET6
;
539 /* FIXME: srcaddr is ignored for IPv6, because I (SGT) don't
540 * know how to do it. :-) */
542 a6
.sin6_addr
= in6addr_loopback
;
544 a6
.sin6_addr
= in6addr_any
;
545 a6
.sin6_port
= htons(port
);
550 a
.sin_family
= AF_INET
;
553 * Bind to source address. First try an explicitly
557 a
.sin_addr
.s_addr
= inet_addr(srcaddr
);
558 if (a
.sin_addr
.s_addr
!= INADDR_NONE
) {
559 /* Override localhost_only with specified listen addr. */
560 ret
->localhost_only
= ipv4_is_loopback(a
.sin_addr
);
566 * ... and failing that, go with one of the standard ones.
570 a
.sin_addr
.s_addr
= htonl(INADDR_LOOPBACK
);
572 a
.sin_addr
.s_addr
= htonl(INADDR_ANY
);
575 a
.sin_port
= htons((short)port
);
578 retcode
= bind(s
, (addr
->family
== AF_INET6 ?
579 (struct sockaddr
*) &a6
:
580 (struct sockaddr
*) &a
),
582 AF_INET6 ?
sizeof(a6
) : sizeof(a
)));
584 retcode
= bind(s
, (struct sockaddr
*) &a
, sizeof(a
));
593 ret
->error
= error_string(err
);
598 if (listen(s
, SOMAXCONN
) < 0) {
600 ret
->error
= error_string(errno
);
609 static void sk_tcp_close(Socket sock
)
611 Actual_Socket s
= (Actual_Socket
) sock
;
619 * The function which tries to send on a socket once it's deemed
622 void try_send(Actual_Socket s
)
624 while (s
->sending_oob
|| bufchain_size(&s
->output_data
) > 0) {
630 if (s
->sending_oob
) {
631 urgentflag
= MSG_OOB
;
632 len
= s
->sending_oob
;
636 bufchain_prefix(&s
->output_data
, &data
, &len
);
638 nsent
= send(s
->s
, data
, len
, urgentflag
);
639 noise_ultralight(nsent
);
641 err
= (nsent
< 0 ? errno
: 0);
642 if (err
== EWOULDBLOCK
) {
644 * Perfectly normal: we've sent all we can for the moment.
648 } else if (nsent
== 0 ||
649 err
== ECONNABORTED
|| err
== ECONNRESET
) {
651 * If send() returns CONNABORTED or CONNRESET, we
652 * unfortunately can't just call plug_closing(),
653 * because it's quite likely that we're currently
654 * _in_ a call from the code we'd be calling back
655 * to, so we'd have to make half the SSH code
656 * reentrant. Instead we flag a pending error on
657 * the socket, to be dealt with (by calling
658 * plug_closing()) at some suitable future moment.
660 s
->pending_error
= err
;
663 /* We're inside the Unix frontend here, so we know
664 * that the frontend handle is unnecessary. */
665 logevent(NULL
, error_string(err
));
666 fatalbox("%s", error_string(err
));
669 if (s
->sending_oob
) {
671 memmove(s
->oobdata
, s
->oobdata
+nsent
, len
-nsent
);
672 s
->sending_oob
= len
- nsent
;
677 bufchain_consume(&s
->output_data
, nsent
);
683 static int sk_tcp_write(Socket sock
, char *buf
, int len
)
685 Actual_Socket s
= (Actual_Socket
) sock
;
688 * Add the data to the buffer list on the socket.
690 bufchain_add(&s
->output_data
, buf
, len
);
693 * Now try sending from the start of the buffer list.
698 return bufchain_size(&s
->output_data
);
701 static int sk_tcp_write_oob(Socket sock
, char *buf
, int len
)
703 Actual_Socket s
= (Actual_Socket
) sock
;
706 * Replace the buffer list on the socket with the data.
708 bufchain_clear(&s
->output_data
);
709 assert(len
<= sizeof(s
->oobdata
));
710 memcpy(s
->oobdata
, buf
, len
);
711 s
->sending_oob
= len
;
714 * Now try sending from the start of the buffer list.
719 return s
->sending_oob
;
722 int select_result(int fd
, int event
)
726 char buf
[20480]; /* nice big buffer for plenty of speed */
730 /* Find the Socket structure */
731 s
= find234(sktree
, (void *) fd
, cmpforsearch
);
733 return 1; /* boggle */
735 noise_ultralight(event
);
738 #ifdef FIXME_NONBLOCKING_CONNECTIONS
739 case FIXME
: /* connected */
740 s
->connected
= s
->writable
= 1;
743 case 4: /* exceptional */
746 * On a non-oobinline socket, this indicates that we
747 * can immediately perform an OOB read and get back OOB
748 * data, which we will send to the back end with
749 * type==2 (urgent data).
751 ret
= recv(s
->s
, buf
, sizeof(buf
), MSG_OOB
);
752 noise_ultralight(ret
);
754 char *str
= (ret
== 0 ?
"Internal networking trouble" :
755 error_string(errno
));
756 /* We're inside the Unix frontend here, so we know
757 * that the frontend handle is unnecessary. */
761 return plug_receive(s
->plug
, 2, buf
, ret
);
767 * If we reach here, this is an oobinline socket, which
768 * means we should set s->oobpending and then deal with it
769 * when we get called for the readability event (which
770 * should also occur).
772 s
->oobpending
= TRUE
;
774 case 1: /* readable; also acceptance */
777 * On a listening socket, the readability event means a
778 * connection is ready to be accepted.
780 struct sockaddr_in isa
;
781 int addrlen
= sizeof(struct sockaddr_in
);
782 int t
; /* socket of connection */
784 memset(&isa
, 0, sizeof(struct sockaddr_in
));
786 t
= accept(s
->s
,(struct sockaddr
*)&isa
,&addrlen
);
791 if (s
->localhost_only
&& !ipv4_is_loopback(isa
.sin_addr
)) {
792 close(t
); /* someone let nonlocal through?! */
793 } else if (plug_accepting(s
->plug
, (void*)t
)) {
794 close(t
); /* denied or error */
800 * If we reach here, this is not a listening socket, so
801 * readability really means readability.
804 /* In the case the socket is still frozen, we don't even bother */
806 s
->frozen_readable
= 1;
811 * We have received data on the socket. For an oobinline
812 * socket, this might be data _before_ an urgent pointer,
813 * in which case we send it to the back end with type==1
814 * (data prior to urgent).
816 if (s
->oobinline
&& s
->oobpending
) {
818 if (ioctl(s
->s
, SIOCATMARK
, &atmark
) == 0 && atmark
)
819 s
->oobpending
= FALSE
; /* clear this indicator */
823 ret
= recv(s
->s
, buf
, s
->oobpending ?
1 : sizeof(buf
), 0);
824 noise_ultralight(ret
);
826 if (errno
== EWOULDBLOCK
) {
831 return plug_closing(s
->plug
, error_string(errno
), errno
, 0);
832 } else if (0 == ret
) {
833 return plug_closing(s
->plug
, NULL
, 0, 0);
835 return plug_receive(s
->plug
, atmark ?
0 : 1, buf
, ret
);
838 case 2: /* writable */
840 int bufsize_before
, bufsize_after
;
842 bufsize_before
= s
->sending_oob
+ bufchain_size(&s
->output_data
);
844 bufsize_after
= s
->sending_oob
+ bufchain_size(&s
->output_data
);
845 if (bufsize_after
< bufsize_before
)
846 plug_sent(s
->plug
, bufsize_after
);
855 * Deal with socket errors detected in try_send().
857 void net_pending_errors(void)
863 * This might be a fiddly business, because it's just possible
864 * that handling a pending error on one socket might cause
865 * others to be closed. (I can't think of any reason this might
866 * happen in current SSH implementation, but to maintain
867 * generality of this network layer I'll assume the worst.)
869 * So what we'll do is search the socket list for _one_ socket
870 * with a pending error, and then handle it, and then search
871 * the list again _from the beginning_. Repeat until we make a
872 * pass with no socket errors present. That way we are
873 * protected against the socket list changing under our feet.
877 for (i
= 0; (s
= index234(sktree
, i
)) != NULL
; i
++) {
878 if (s
->pending_error
) {
880 * An error has occurred on this socket. Pass it to the
883 plug_closing(s
->plug
, error_string(s
->pending_error
),
884 s
->pending_error
, 0);
892 * Each socket abstraction contains a `void *' private field in
893 * which the client can keep state.
895 static void sk_tcp_set_private_ptr(Socket sock
, void *ptr
)
897 Actual_Socket s
= (Actual_Socket
) sock
;
898 s
->private_ptr
= ptr
;
901 static void *sk_tcp_get_private_ptr(Socket sock
)
903 Actual_Socket s
= (Actual_Socket
) sock
;
904 return s
->private_ptr
;
908 * Special error values are returned from sk_namelookup and sk_new
909 * if there's a problem. These functions extract an error message,
910 * or return NULL if there's no problem.
912 char *sk_addr_error(SockAddr addr
)
916 static char *sk_tcp_socket_error(Socket sock
)
918 Actual_Socket s
= (Actual_Socket
) sock
;
922 static void sk_tcp_set_frozen(Socket sock
, int is_frozen
)
924 Actual_Socket s
= (Actual_Socket
) sock
;
925 if (s
->frozen
== is_frozen
)
927 s
->frozen
= is_frozen
;
928 if (!is_frozen
&& s
->frozen_readable
) {
930 recv(s
->s
, &c
, 1, MSG_PEEK
);
932 s
->frozen_readable
= 0;
936 * For Unix select()-based frontends: enumerate all sockets
937 * currently active, and state whether we currently wish to receive
938 * select events on them for reading, writing and exceptional
941 static void set_rwx(Actual_Socket s
, int *rwx
)
944 if (s
->connected
&& !s
->frozen
)
945 val
|= 1 | 4; /* read, except */
946 if (bufchain_size(&s
->output_data
))
947 val
|= 2; /* write */
949 val
|= 1; /* read == accept */
953 int first_socket(int *state
, int *rwx
)
957 s
= index234(sktree
, (*state
)++);
960 return s ? s
->s
: -1;
963 int next_socket(int *state
, int *rwx
)
965 Actual_Socket s
= index234(sktree
, (*state
)++);
968 return s ? s
->s
: -1;
971 int net_service_lookup(char *service
)
974 se
= getservbyname(service
, NULL
);
976 return ntohs(se
->s_port
);