2 * Support for SSH connection sharing, i.e. permitting one PuTTY to
3 * open its own channels over the SSH session being run by another.
7 * Discussion and technical documentation
8 * ======================================
10 * The basic strategy for PuTTY's implementation of SSH connection
11 * sharing is to have a single 'upstream' PuTTY process, which manages
12 * the real SSH connection and all the cryptography, and then zero or
13 * more 'downstream' PuTTYs, which never talk to the real host but
14 * only talk to the upstream through local IPC (Unix-domain sockets or
15 * Windows named pipes).
17 * The downstreams communicate with the upstream using a protocol
18 * derived from SSH itself, which I'll document in detail below. In
19 * brief, though: the downstream->upstream protocol uses a trivial
20 * binary packet protocol (just length/type/data) to encapsulate
21 * unencrypted SSH messages, and downstreams talk to the upstream more
22 * or less as if it was an SSH server itself. (So downstreams can
23 * themselves open multiple SSH channels, for example, by sending
24 * multiple SSH2_MSG_CHANNEL_OPENs; they can send CHANNEL_REQUESTs of
25 * their choice within each channel, and they handle their own
26 * WINDOW_ADJUST messages.)
28 * The upstream would ideally handle these downstreams by just putting
29 * their messages into the queue for proper SSH-2 encapsulation and
30 * encryption and sending them straight on to the server. However,
31 * that's not quite feasible as written, because client-side channel
32 * IDs could easily conflict (between multiple downstreams, or between
33 * a downstream and the upstream). To protect against that, the
34 * upstream rewrites the client-side channel IDs in messages it passes
35 * on to the server, so that it's performing what you might describe
36 * as 'channel-number NAT'. Then the upstream remembers which of its
37 * own channel IDs are channels it's managing itself, and which are
38 * placeholders associated with a particular downstream, so that when
39 * replies come in from the server they can be sent on to the relevant
40 * downstream (after un-NATting the channel number, of course).
42 * Global requests from downstreams are only accepted if the upstream
43 * knows what to do about them; currently the only such requests are
44 * the ones having to do with remote-to-local port forwarding (in
45 * which, again, the upstream remembers that some of the forwardings
46 * it's asked the server to set up were on behalf of particular
47 * downstreams, and sends the incoming CHANNEL_OPENs to those
48 * downstreams when connections come in).
50 * Other fiddly pieces of this mechanism are X forwarding and
51 * (OpenSSH-style) agent forwarding. Both of these have a fundamental
52 * problem arising from the protocol design: that the CHANNEL_OPEN
53 * from the server introducing a forwarded connection does not carry
54 * any indication of which session channel gave rise to it; so if
55 * session channels from multiple downstreams enable those forwarding
56 * methods, it's hard for the upstream to know which downstream to
57 * send the resulting connections back to.
59 * For X forwarding, we can work around this in a really painful way
60 * by using the fake X11 authorisation data sent to the server as part
61 * of the forwarding setup: upstream ensures that every X forwarding
62 * request carries distinguishable fake auth data, and then when X
63 * connections come in it waits to see the auth data in the X11 setup
64 * message before it decides which downstream to pass the connection
67 * For agent forwarding, that workaround is unavailable. As a result,
68 * this system (and, as far as I can think of, any other system too)
69 * has the fundamental constraint that it can only forward one SSH
70 * agent - it can't forward two agents to different session channels.
71 * So downstreams can request agent forwarding if they like, but if
72 * they do, they'll get whatever SSH agent is known to the upstream
73 * (if any) forwarded to their sessions.
75 * Downstream-to-upstream protocol
76 * -------------------------------
78 * Here I document in detail the protocol spoken between PuTTY
79 * downstreams and upstreams over local IPC. The IPC mechanism can
80 * vary between host platforms, but the protocol is the same.
82 * The protocol commences with a version exchange which is exactly
83 * like the SSH-2 one, in that each side sends a single line of text
86 * <protocol>-<version>-<softwareversion> [comments] \r\n
88 * The only difference is that in real SSH-2, <protocol> is the string
89 * "SSH", whereas in this protocol the string is
90 * "SSHCONNECTION@putty.projects.tartarus.org".
92 * (The SSH RFCs allow many protocol-level identifier namespaces to be
93 * extended by implementors without central standardisation as long as
94 * they suffix "@" and a domain name they control to their new ids.
95 * RFC 4253 does not define this particular name to be changeable at
96 * all, but I like to think this is obviously how it would have done
97 * so if the working group had foreseen the need :-)
99 * Thereafter, all data exchanged consists of a sequence of binary
100 * packets concatenated end-to-end, each of which is of the form
102 * uint32 length of packet, N
103 * byte[N] N bytes of packet data
105 * and, since these are SSH-2 messages, the first data byte is taken
106 * to be the packet type code.
108 * These messages are interpreted as those of an SSH connection, after
109 * userauth completes, and without any repeat key exchange.
110 * Specifically, any message from the SSH Connection Protocol is
111 * permitted, and also SSH_MSG_IGNORE, SSH_MSG_DEBUG,
112 * SSH_MSG_DISCONNECT and SSH_MSG_UNIMPLEMENTED from the SSH Transport
115 * This protocol imposes a few additional requirements, over and above
116 * those of the standard SSH Connection Protocol:
118 * Message sizes are not permitted to exceed 0x4010 (16400) bytes,
119 * including their length header.
121 * When the server (i.e. really the PuTTY upstream) sends
122 * SSH_MSG_CHANNEL_OPEN with channel type "x11", and the client
123 * (downstream) responds with SSH_MSG_CHANNEL_OPEN_CONFIRMATION, that
124 * confirmation message MUST include an initial window size of at
125 * least 256. (Rationale: this is a bit of a fudge which makes it
126 * easier, by eliminating the possibility of nasty edge cases, for an
127 * upstream to arrange not to pass the CHANNEL_OPEN on to downstream
128 * until after it's seen the X11 auth data to decide which downstream
129 * it needs to go to.)
141 struct ssh_sharing_state {
142 const struct plug_function_table *fn;
143 /* the above variable absolutely *must* be the first in this structure */
145 char *sockname; /* the socket name, kept for cleanup */
146 Socket listensock; /* the master listening Socket */
147 tree234 *connections; /* holds ssh_sharing_connstates */
148 unsigned nextid; /* preferred id for next connstate */
149 Ssh ssh; /* instance of the ssh backend */
150 char *server_verstring; /* server version string after "SSH-" */
153 struct share_globreq;
155 struct ssh_sharing_connstate {
156 const struct plug_function_table *fn;
157 /* the above variable absolutely *must* be the first in this structure */
159 unsigned id; /* used to identify this downstream in log messages */
161 Socket sock; /* the Socket for this connection */
162 struct ssh_sharing_state *parent;
164 int crLine; /* coroutine state for share_receive */
166 int sent_verstring, got_verstring, curr_packetlen;
168 unsigned char recvbuf[0x4010];
172 * Assorted state we have to remember about this downstream, so
173 * that we can clean it up appropriately when the downstream goes
177 /* Channels which don't have a downstream id, i.e. we've passed a
178 * CHANNEL_OPEN down from the server but not had an
179 * OPEN_CONFIRMATION or OPEN_FAILURE back. If downstream goes
180 * away, we respond to all of these with OPEN_FAILURE. */
181 tree234 *halfchannels; /* stores 'struct share_halfchannel' */
183 /* Channels which do have a downstream id. We need to index these
184 * by both server id and upstream id, so we can find a channel
185 * when handling either an upward or a downward message referring
187 tree234 *channels_by_us; /* stores 'struct share_channel' */
188 tree234 *channels_by_server; /* stores 'struct share_channel' */
190 /* Another class of channel which doesn't have a downstream id.
191 * The difference between these and halfchannels is that xchannels
192 * do have an *upstream* id, because upstream has already accepted
193 * the channel request from the server. This arises in the case of
194 * X forwarding, where we have to accept the request and read the
195 * X authorisation data before we know whether the channel needs
196 * to be forwarded to a downstream. */
197 tree234 *xchannels_by_us; /* stores 'struct share_xchannel' */
198 tree234 *xchannels_by_server; /* stores 'struct share_xchannel' */
200 /* Remote port forwarding requests in force. */
201 tree234 *forwardings; /* stores 'struct share_forwarding' */
203 /* Global requests we've sent on to the server, pending replies. */
204 struct share_globreq *globreq_head, *globreq_tail;
207 struct share_halfchannel {
211 /* States of a share_channel. */
216 /* Downstream has sent CHANNEL_OPEN but server hasn't replied yet.
217 * If downstream goes away when a channel is in this state, we
218 * must wait for the server's response before starting to send
219 * CLOSE. Channels in this state are also not held in
220 * channels_by_server, because their server_id field is
225 struct share_channel {
226 unsigned downstream_id, upstream_id, server_id;
227 int downstream_maxpkt;
230 * Some channels (specifically, channels on which downstream has
231 * sent "x11-req") have the additional function of storing a set
232 * of downstream X authorisation data and a handle to an upstream
235 struct X11FakeAuth *x11_auth_upstream;
238 int x11_auth_datalen;
242 struct share_forwarding {
245 int active; /* has the server sent REQUEST_SUCCESS? */
248 struct share_xchannel_message {
249 struct share_xchannel_message *next;
255 struct share_xchannel {
256 unsigned upstream_id, server_id;
259 * xchannels come in two flavours: live and dead. Live ones are
260 * waiting for an OPEN_CONFIRMATION or OPEN_FAILURE from
261 * downstream; dead ones have had an OPEN_FAILURE, so they only
262 * exist as a means of letting us conveniently respond to further
263 * channel messages from the server until such time as the server
264 * sends us CHANNEL_CLOSE.
269 * When we receive OPEN_CONFIRMATION, we will need to send a
270 * WINDOW_ADJUST to the server to synchronise the windows. For
271 * this purpose we need to know what window we have so far offered
272 * the server. We record this as exactly the value in the
273 * OPEN_CONFIRMATION that upstream sent us, adjusted by the amount
274 * by which the two X greetings differed in length.
279 * Linked list of SSH messages from the server relating to this
280 * channel, which we queue up until downstream sends us an
281 * OPEN_CONFIRMATION and we can belatedly send them all on.
283 struct share_xchannel_message *msghead, *msgtail;
287 GLOBREQ_TCPIP_FORWARD,
288 GLOBREQ_CANCEL_TCPIP_FORWARD
291 struct share_globreq {
292 struct share_globreq *next;
295 struct share_forwarding *fwd;
298 static int share_connstate_cmp(void *av, void *bv)
300 const struct ssh_sharing_connstate *a =
301 (const struct ssh_sharing_connstate *)av;
302 const struct ssh_sharing_connstate *b =
303 (const struct ssh_sharing_connstate *)bv;
307 else if (a->id > b->id)
313 static unsigned share_find_unused_id
314 (struct ssh_sharing_state *sharestate, unsigned first)
316 int low_orig, low, mid, high, high_orig;
317 struct ssh_sharing_connstate *cs;
321 * Find the lowest unused downstream ID greater or equal to
324 * Begin by seeing if 'first' itself is available. If it is, we'll
325 * just return it; if it's already in the tree, we'll find the
326 * tree index where it appears and use that for the next stage.
329 struct ssh_sharing_connstate dummy;
331 cs = findrelpos234(sharestate->connections, &dummy, NULL,
332 REL234_GE, &low_orig);
338 * Now binary-search using the counted B-tree, to find the largest
339 * ID which is in a contiguous sequence from the beginning of that
343 high = high_orig = count234(sharestate->connections);
344 while (high - low > 1) {
345 mid = (high + low) / 2;
346 cs = index234(sharestate->connections, mid);
347 if (cs->id == first + (mid - low_orig))
348 low = mid; /* this one is still in the sequence */
350 high = mid; /* this one is past the end */
354 * Now low is the tree index of the largest ID in the initial
355 * sequence. So the return value is one more than low's id, and we
356 * know low's id is given by the formula in the binary search loop
359 * (If an SSH connection went on for _enormously_ long, we might
360 * reach a point where all ids from 'first' to UINT_MAX were in
361 * use. In that situation the formula below would wrap round by
362 * one and return zero, which is conveniently the right way to
363 * signal 'no id available' from this function.)
365 ret = first + (low - low_orig) + 1;
367 struct ssh_sharing_connstate dummy;
369 assert(NULL == find234(sharestate->connections, &dummy, NULL));
374 static int share_halfchannel_cmp(void *av, void *bv)
376 const struct share_halfchannel *a = (const struct share_halfchannel *)av;
377 const struct share_halfchannel *b = (const struct share_halfchannel *)bv;
379 if (a->server_id < b->server_id)
381 else if (a->server_id > b->server_id)
387 static int share_channel_us_cmp(void *av, void *bv)
389 const struct share_channel *a = (const struct share_channel *)av;
390 const struct share_channel *b = (const struct share_channel *)bv;
392 if (a->upstream_id < b->upstream_id)
394 else if (a->upstream_id > b->upstream_id)
400 static int share_channel_server_cmp(void *av, void *bv)
402 const struct share_channel *a = (const struct share_channel *)av;
403 const struct share_channel *b = (const struct share_channel *)bv;
405 if (a->server_id < b->server_id)
407 else if (a->server_id > b->server_id)
413 static int share_xchannel_us_cmp(void *av, void *bv)
415 const struct share_xchannel *a = (const struct share_xchannel *)av;
416 const struct share_xchannel *b = (const struct share_xchannel *)bv;
418 if (a->upstream_id < b->upstream_id)
420 else if (a->upstream_id > b->upstream_id)
426 static int share_xchannel_server_cmp(void *av, void *bv)
428 const struct share_xchannel *a = (const struct share_xchannel *)av;
429 const struct share_xchannel *b = (const struct share_xchannel *)bv;
431 if (a->server_id < b->server_id)
433 else if (a->server_id > b->server_id)
439 static int share_forwarding_cmp(void *av, void *bv)
441 const struct share_forwarding *a = (const struct share_forwarding *)av;
442 const struct share_forwarding *b = (const struct share_forwarding *)bv;
445 if ((i = strcmp(a->host, b->host)) != 0)
447 else if (a->port < b->port)
449 else if (a->port > b->port)
455 static void share_xchannel_free(struct share_xchannel *xc)
457 while (xc->msghead) {
458 struct share_xchannel_message *tmp = xc->msghead;
459 xc->msghead = tmp->next;
465 static void share_connstate_free(struct ssh_sharing_connstate *cs)
467 struct share_halfchannel *hc;
468 struct share_xchannel *xc;
469 struct share_channel *chan;
470 struct share_forwarding *fwd;
472 while ((hc = (struct share_halfchannel *)
473 delpos234(cs->halfchannels, 0)) != NULL)
475 freetree234(cs->halfchannels);
477 /* All channels live in 'channels_by_us' but only some in
478 * 'channels_by_server', so we use the former to find the list of
480 freetree234(cs->channels_by_server);
481 while ((chan = (struct share_channel *)
482 delpos234(cs->channels_by_us, 0)) != NULL)
484 freetree234(cs->channels_by_us);
486 /* But every xchannel is in both trees, so it doesn't matter which
487 * we use to free them. */
488 while ((xc = (struct share_xchannel *)
489 delpos234(cs->xchannels_by_us, 0)) != NULL)
490 share_xchannel_free(xc);
491 freetree234(cs->xchannels_by_us);
492 freetree234(cs->xchannels_by_server);
494 while ((fwd = (struct share_forwarding *)
495 delpos234(cs->forwardings, 0)) != NULL)
497 freetree234(cs->forwardings);
499 while (cs->globreq_head) {
500 struct share_globreq *globreq = cs->globreq_head;
501 cs->globreq_head = cs->globreq_head->next;
508 void sharestate_free(void *v)
510 struct ssh_sharing_state *sharestate = (struct ssh_sharing_state *)v;
511 struct ssh_sharing_connstate *cs;
513 platform_ssh_share_cleanup(sharestate->sockname);
515 while ((cs = (struct ssh_sharing_connstate *)
516 delpos234(sharestate->connections, 0)) != NULL) {
517 share_connstate_free(cs);
519 freetree234(sharestate->connections);
520 sfree(sharestate->server_verstring);
521 sfree(sharestate->sockname);
525 static struct share_halfchannel *share_add_halfchannel
526 (struct ssh_sharing_connstate *cs, unsigned server_id)
528 struct share_halfchannel *hc = snew(struct share_halfchannel);
529 hc->server_id = server_id;
530 if (add234(cs->halfchannels, hc) != hc) {
539 static struct share_halfchannel *share_find_halfchannel
540 (struct ssh_sharing_connstate *cs, unsigned server_id)
542 struct share_halfchannel dummyhc;
543 dummyhc.server_id = server_id;
544 return find234(cs->halfchannels, &dummyhc, NULL);
547 static void share_remove_halfchannel(struct ssh_sharing_connstate *cs,
548 struct share_halfchannel *hc)
550 del234(cs->halfchannels, hc);
554 static struct share_channel *share_add_channel
555 (struct ssh_sharing_connstate *cs, unsigned downstream_id,
556 unsigned upstream_id, unsigned server_id, int state, int maxpkt)
558 struct share_channel *chan = snew(struct share_channel);
559 chan->downstream_id = downstream_id;
560 chan->upstream_id = upstream_id;
561 chan->server_id = server_id;
563 chan->downstream_maxpkt = maxpkt;
564 chan->x11_auth_upstream = NULL;
565 chan->x11_auth_data = NULL;
566 chan->x11_auth_proto = -1;
567 chan->x11_auth_datalen = 0;
568 chan->x11_one_shot = 0;
569 if (add234(cs->channels_by_us, chan) != chan) {
573 if (chan->state != UNACKNOWLEDGED) {
574 if (add234(cs->channels_by_server, chan) != chan) {
575 del234(cs->channels_by_us, chan);
583 static void share_channel_set_server_id(struct ssh_sharing_connstate *cs,
584 struct share_channel *chan,
585 unsigned server_id, int newstate)
587 chan->server_id = server_id;
588 chan->state = newstate;
589 assert(newstate != UNACKNOWLEDGED);
590 add234(cs->channels_by_server, chan);
593 static struct share_channel *share_find_channel_by_upstream
594 (struct ssh_sharing_connstate *cs, unsigned upstream_id)
596 struct share_channel dummychan;
597 dummychan.upstream_id = upstream_id;
598 return find234(cs->channels_by_us, &dummychan, NULL);
601 static struct share_channel *share_find_channel_by_server
602 (struct ssh_sharing_connstate *cs, unsigned server_id)
604 struct share_channel dummychan;
605 dummychan.server_id = server_id;
606 return find234(cs->channels_by_server, &dummychan, NULL);
609 static void share_remove_channel(struct ssh_sharing_connstate *cs,
610 struct share_channel *chan)
612 del234(cs->channels_by_us, chan);
613 del234(cs->channels_by_server, chan);
614 if (chan->x11_auth_upstream)
615 ssh_sharing_remove_x11_display(cs->parent->ssh,
616 chan->x11_auth_upstream);
617 sfree(chan->x11_auth_data);
621 static struct share_xchannel *share_add_xchannel
622 (struct ssh_sharing_connstate *cs,
623 unsigned upstream_id, unsigned server_id)
625 struct share_xchannel *xc = snew(struct share_xchannel);
626 xc->upstream_id = upstream_id;
627 xc->server_id = server_id;
629 xc->msghead = xc->msgtail = NULL;
630 if (add234(cs->xchannels_by_us, xc) != xc) {
634 if (add234(cs->xchannels_by_server, xc) != xc) {
635 del234(cs->xchannels_by_us, xc);
642 static struct share_xchannel *share_find_xchannel_by_upstream
643 (struct ssh_sharing_connstate *cs, unsigned upstream_id)
645 struct share_xchannel dummyxc;
646 dummyxc.upstream_id = upstream_id;
647 return find234(cs->xchannels_by_us, &dummyxc, NULL);
650 static struct share_xchannel *share_find_xchannel_by_server
651 (struct ssh_sharing_connstate *cs, unsigned server_id)
653 struct share_xchannel dummyxc;
654 dummyxc.server_id = server_id;
655 return find234(cs->xchannels_by_server, &dummyxc, NULL);
658 static void share_remove_xchannel(struct ssh_sharing_connstate *cs,
659 struct share_xchannel *xc)
661 del234(cs->xchannels_by_us, xc);
662 del234(cs->xchannels_by_server, xc);
663 share_xchannel_free(xc);
666 static struct share_forwarding *share_add_forwarding
667 (struct ssh_sharing_connstate *cs,
668 const char *host, int port)
670 struct share_forwarding *fwd = snew(struct share_forwarding);
671 fwd->host = dupstr(host);
674 if (add234(cs->forwardings, fwd) != fwd) {
682 static struct share_forwarding *share_find_forwarding
683 (struct ssh_sharing_connstate *cs, const char *host, int port)
685 struct share_forwarding dummyfwd, *ret;
686 dummyfwd.host = dupstr(host);
687 dummyfwd.port = port;
688 ret = find234(cs->forwardings, &dummyfwd, NULL);
689 sfree(dummyfwd.host);
693 static void share_remove_forwarding(struct ssh_sharing_connstate *cs,
694 struct share_forwarding *fwd)
696 del234(cs->forwardings, fwd);
700 static void send_packet_to_downstream(struct ssh_sharing_connstate *cs,
701 int type, const void *pkt, int pktlen,
702 struct share_channel *chan)
704 if (!cs->sock) /* throw away all packets destined for a dead downstream */
707 if (type == SSH2_MSG_CHANNEL_DATA) {
709 * Special case which we take care of at a low level, so as to
710 * be sure to apply it in all cases. On rare occasions we
711 * might find that we have a channel for which the
712 * downstream's maximum packet size exceeds the max packet
713 * size we presented to the server on its behalf. (This can
714 * occur in X11 forwarding, where we have to send _our_
715 * CHANNEL_OPEN_CONFIRMATION before we discover which if any
716 * downstream the channel is destined for, so if that
717 * downstream turns out to present a smaller max packet size
718 * then we're in this situation.)
720 * If that happens, we just chop up the packet into pieces and
721 * send them as separate CHANNEL_DATA packets.
723 const char *upkt = (const char *)pkt;
724 char header[13]; /* 4 length + 1 type + 4 channel id + 4 string len */
726 int len = toint(GET_32BIT(upkt + 4));
727 upkt += 8; /* skip channel id + length field */
729 if (len < 0 || len > pktlen - 8)
733 int this_len = (len > chan->downstream_maxpkt ?
734 chan->downstream_maxpkt : len);
735 PUT_32BIT(header, this_len + 9);
737 PUT_32BIT(header + 5, chan->downstream_id);
738 PUT_32BIT(header + 9, this_len);
739 sk_write(cs->sock, header, 13);
740 sk_write(cs->sock, upkt, this_len);
746 * Just do the obvious thing.
750 PUT_32BIT(header, pktlen + 1);
752 sk_write(cs->sock, header, 5);
753 sk_write(cs->sock, pkt, pktlen);
757 static void share_try_cleanup(struct ssh_sharing_connstate *cs)
760 struct share_halfchannel *hc;
761 struct share_channel *chan;
762 struct share_forwarding *fwd;
765 * Any half-open channels, i.e. those for which we'd received
766 * CHANNEL_OPEN from the server but not passed back a response
767 * from downstream, should be responded to with OPEN_FAILURE.
769 while ((hc = (struct share_halfchannel *)
770 index234(cs->halfchannels, 0)) != NULL) {
771 static const char reason[] = "PuTTY downstream no longer available";
772 static const char lang[] = "en";
773 unsigned char packet[256];
776 PUT_32BIT(packet + pos, hc->server_id); pos += 4;
777 PUT_32BIT(packet + pos, SSH2_OPEN_CONNECT_FAILED); pos += 4;
778 PUT_32BIT(packet + pos, strlen(reason)); pos += 4;
779 memcpy(packet + pos, reason, strlen(reason)); pos += strlen(reason);
780 PUT_32BIT(packet + pos, strlen(lang)); pos += 4;
781 memcpy(packet + pos, lang, strlen(lang)); pos += strlen(lang);
782 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
783 SSH2_MSG_CHANNEL_OPEN_FAILURE,
784 packet, pos, "cleanup after"
785 " downstream went away");
787 share_remove_halfchannel(cs, hc);
791 * Any actually open channels should have a CHANNEL_CLOSE sent for
792 * them, unless we've already done so. We won't be able to
793 * actually clean them up until CHANNEL_CLOSE comes back from the
794 * server, though (unless the server happens to have sent a CLOSE
797 * Another annoying exception is UNACKNOWLEDGED channels, i.e.
798 * we've _sent_ a CHANNEL_OPEN to the server but not received an
799 * OPEN_CONFIRMATION or OPEN_FAILURE. We must wait for a reply
800 * before closing the channel, because until we see that reply we
801 * won't have the server's channel id to put in the close message.
803 for (i = 0; (chan = (struct share_channel *)
804 index234(cs->channels_by_us, i)) != NULL; i++) {
805 unsigned char packet[256];
808 if (chan->state != SENT_CLOSE && chan->state != UNACKNOWLEDGED) {
809 PUT_32BIT(packet + pos, chan->server_id); pos += 4;
810 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
811 SSH2_MSG_CHANNEL_CLOSE,
812 packet, pos, "cleanup after"
813 " downstream went away");
814 if (chan->state != RCVD_CLOSE) {
815 chan->state = SENT_CLOSE;
817 /* In this case, we _can_ clear up the channel now. */
818 ssh_delete_sharing_channel(cs->parent->ssh, chan->upstream_id);
819 share_remove_channel(cs, chan);
820 i--; /* don't accidentally skip one as a result */
826 * Any remote port forwardings we're managing on behalf of this
827 * downstream should be cancelled. Again, we must defer those for
828 * which we haven't yet seen REQUEST_SUCCESS/FAILURE.
830 * We take a fire-and-forget approach during cleanup, not
831 * bothering to set want_reply.
833 for (i = 0; (fwd = (struct share_forwarding *)
834 index234(cs->forwardings, i)) != NULL; i++) {
836 static const char request[] = "cancel-tcpip-forward";
837 char *packet = snewn(256 + strlen(fwd->host), char);
840 PUT_32BIT(packet + pos, strlen(request)); pos += 4;
841 memcpy(packet + pos, request, strlen(request));
842 pos += strlen(request);
844 packet[pos++] = 0; /* !want_reply */
846 PUT_32BIT(packet + pos, strlen(fwd->host)); pos += 4;
847 memcpy(packet + pos, fwd->host, strlen(fwd->host));
848 pos += strlen(fwd->host);
850 PUT_32BIT(packet + pos, fwd->port); pos += 4;
852 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
853 SSH2_MSG_GLOBAL_REQUEST,
854 packet, pos, "cleanup after"
855 " downstream went away");
857 share_remove_forwarding(cs, fwd);
858 i--; /* don't accidentally skip one as a result */
862 if (count234(cs->halfchannels) == 0 &&
863 count234(cs->channels_by_us) == 0 &&
864 count234(cs->forwardings) == 0) {
866 * Now we're _really_ done, so we can get rid of cs completely.
868 del234(cs->parent->connections, cs);
869 ssh_sharing_downstream_disconnected(cs->parent->ssh, cs->id);
870 share_connstate_free(cs);
874 static void share_begin_cleanup(struct ssh_sharing_connstate *cs)
880 share_try_cleanup(cs);
883 static void share_disconnect(struct ssh_sharing_connstate *cs,
886 static const char lang[] = "en";
887 int msglen = strlen(message);
888 char *packet = snewn(msglen + 256, char);
891 PUT_32BIT(packet + pos, SSH2_DISCONNECT_PROTOCOL_ERROR); pos += 4;
893 PUT_32BIT(packet + pos, msglen); pos += 4;
894 memcpy(packet + pos, message, msglen);
897 PUT_32BIT(packet + pos, strlen(lang)); pos += 4;
898 memcpy(packet + pos, lang, strlen(lang)); pos += strlen(lang);
900 send_packet_to_downstream(cs, SSH2_MSG_DISCONNECT, packet, pos, NULL);
902 share_begin_cleanup(cs);
905 static int share_closing(Plug plug, const char *error_msg, int error_code,
908 struct ssh_sharing_connstate *cs = (struct ssh_sharing_connstate *)plug;
910 ssh_sharing_logf(cs->parent->ssh, cs->id, "%s", error_msg);
911 share_begin_cleanup(cs);
915 static int getstring_inner(const void *vdata, int datalen,
916 char **out, int *outlen)
918 const unsigned char *data = (const unsigned char *)vdata;
924 len = toint(GET_32BIT(data));
925 if (len < 0 || len > datalen - 4)
929 *outlen = len + 4; /* total size including length field */
931 *out = dupprintf("%.*s", len, (char *)data + 4);
935 static char *getstring(const void *data, int datalen)
938 if (getstring_inner(data, datalen, &ret, NULL))
944 static int getstring_size(const void *data, int datalen)
947 if (getstring_inner(data, datalen, NULL, &ret))
954 * Append a message to the end of an xchannel's queue, with the length
955 * and type code filled in and the data block allocated but
958 struct share_xchannel_message *share_xchannel_add_message
959 (struct share_xchannel *xc, int type, int len)
961 unsigned char *block;
962 struct share_xchannel_message *msg;
965 * Be a little tricksy here by allocating a single memory block
966 * containing both the 'struct share_xchannel_message' and the
967 * actual data. Simplifies freeing it later.
969 block = smalloc(sizeof(struct share_xchannel_message) + len);
970 msg = (struct share_xchannel_message *)block;
971 msg->data = block + sizeof(struct share_xchannel_message);
976 * Queue it in the xchannel.
979 xc->msgtail->next = msg;
988 void share_dead_xchannel_respond(struct ssh_sharing_connstate *cs,
989 struct share_xchannel *xc)
992 * Handle queued incoming messages from the server destined for an
993 * xchannel which is dead (i.e. downstream sent OPEN_FAILURE).
996 while (xc->msghead) {
997 struct share_xchannel_message *msg = xc->msghead;
998 xc->msghead = msg->next;
1000 if (msg->type == SSH2_MSG_CHANNEL_REQUEST && msg->datalen > 4) {
1002 * A CHANNEL_REQUEST is responded to by sending
1003 * CHANNEL_FAILURE, if it has want_reply set.
1005 int wantreplypos = getstring_size(msg->data, msg->datalen);
1006 if (wantreplypos > 0 && wantreplypos < msg->datalen &&
1007 msg->data[wantreplypos] != 0) {
1008 unsigned char id[4];
1009 PUT_32BIT(id, xc->server_id);
1010 ssh_send_packet_from_downstream
1011 (cs->parent->ssh, cs->id, SSH2_MSG_CHANNEL_FAILURE, id, 4,
1012 "downstream refused X channel open");
1014 } else if (msg->type == SSH2_MSG_CHANNEL_CLOSE) {
1016 * On CHANNEL_CLOSE we can discard the channel completely.
1025 ssh_delete_sharing_channel(cs->parent->ssh, xc->upstream_id);
1026 share_remove_xchannel(cs, xc);
1030 void share_xchannel_confirmation(struct ssh_sharing_connstate *cs,
1031 struct share_xchannel *xc,
1032 struct share_channel *chan,
1033 unsigned downstream_window)
1035 unsigned char window_adjust[8];
1038 * Send all the queued messages downstream.
1040 while (xc->msghead) {
1041 struct share_xchannel_message *msg = xc->msghead;
1042 xc->msghead = msg->next;
1044 if (msg->datalen >= 4)
1045 PUT_32BIT(msg->data, chan->downstream_id);
1046 send_packet_to_downstream(cs, msg->type,
1047 msg->data, msg->datalen, chan);
1053 * Send a WINDOW_ADJUST back upstream, to synchronise the window
1054 * size downstream thinks it's presented with the one we've
1055 * actually presented.
1057 PUT_32BIT(window_adjust, xc->server_id);
1058 PUT_32BIT(window_adjust + 4, downstream_window - xc->window);
1059 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
1060 SSH2_MSG_CHANNEL_WINDOW_ADJUST,
1061 window_adjust, 8, "window adjustment after"
1062 " downstream accepted X channel");
1065 void share_xchannel_failure(struct ssh_sharing_connstate *cs,
1066 struct share_xchannel *xc)
1069 * If downstream refuses to open our X channel at all for some
1070 * reason, we must respond by sending an emergency CLOSE upstream.
1072 unsigned char id[4];
1073 PUT_32BIT(id, xc->server_id);
1074 ssh_send_packet_from_downstream
1075 (cs->parent->ssh, cs->id, SSH2_MSG_CHANNEL_CLOSE, id, 4,
1076 "downstream refused X channel open");
1079 * Now mark the xchannel as dead, and respond to anything sent on
1080 * it until we see CLOSE for it in turn.
1083 share_dead_xchannel_respond(cs, xc);
1086 void share_setup_x11_channel(void *csv, void *chanv,
1087 unsigned upstream_id, unsigned server_id,
1088 unsigned server_currwin, unsigned server_maxpkt,
1089 unsigned client_adjusted_window,
1090 const char *peer_addr, int peer_port, int endian,
1091 int protomajor, int protominor,
1092 const void *initial_data, int initial_len)
1094 struct ssh_sharing_connstate *cs = (struct ssh_sharing_connstate *)csv;
1095 struct share_channel *chan = (struct share_channel *)chanv;
1096 struct share_xchannel *xc;
1097 struct share_xchannel_message *msg;
1104 * Create an xchannel containing data we've already received from
1105 * the X client, and preload it with a CHANNEL_DATA message
1106 * containing our own made-up authorisation greeting and any
1107 * additional data sent from the server so far.
1109 xc = share_add_xchannel(cs, upstream_id, server_id);
1110 greeting = x11_make_greeting(endian, protomajor, protominor,
1111 chan->x11_auth_proto,
1112 chan->x11_auth_data, chan->x11_auth_datalen,
1113 peer_addr, peer_port, &greeting_len);
1114 msg = share_xchannel_add_message(xc, SSH2_MSG_CHANNEL_DATA,
1115 8 + greeting_len + initial_len);
1116 /* leave the channel id field unfilled - we don't know the
1117 * downstream id yet, of course */
1118 PUT_32BIT(msg->data + 4, greeting_len + initial_len);
1119 memcpy(msg->data + 8, greeting, greeting_len);
1120 memcpy(msg->data + 8 + greeting_len, initial_data, initial_len);
1123 xc->window = client_adjusted_window + greeting_len;
1126 * Send on a CHANNEL_OPEN to downstream.
1128 pktlen = 27 + strlen(peer_addr);
1129 pkt = snewn(pktlen, unsigned char);
1130 PUT_32BIT(pkt, 3); /* strlen("x11") */
1131 memcpy(pkt+4, "x11", 3);
1132 PUT_32BIT(pkt+7, server_id);
1133 PUT_32BIT(pkt+11, server_currwin);
1134 PUT_32BIT(pkt+15, server_maxpkt);
1135 PUT_32BIT(pkt+19, strlen(peer_addr));
1136 memcpy(pkt+23, peer_addr, strlen(peer_addr));
1137 PUT_32BIT(pkt+23+strlen(peer_addr), peer_port);
1138 send_packet_to_downstream(cs, SSH2_MSG_CHANNEL_OPEN, pkt, pktlen, NULL);
1142 * If this was a once-only X forwarding, clean it up now.
1144 if (chan->x11_one_shot) {
1145 ssh_sharing_remove_x11_display(cs->parent->ssh,
1146 chan->x11_auth_upstream);
1147 chan->x11_auth_upstream = NULL;
1148 sfree(chan->x11_auth_data);
1149 chan->x11_auth_proto = -1;
1150 chan->x11_auth_datalen = 0;
1151 chan->x11_one_shot = 0;
1155 void share_got_pkt_from_server(void *csv, int type,
1156 unsigned char *pkt, int pktlen)
1158 struct ssh_sharing_connstate *cs = (struct ssh_sharing_connstate *)csv;
1159 struct share_globreq *globreq;
1161 unsigned upstream_id, server_id;
1162 struct share_channel *chan;
1163 struct share_xchannel *xc;
1166 case SSH2_MSG_REQUEST_SUCCESS:
1167 case SSH2_MSG_REQUEST_FAILURE:
1168 globreq = cs->globreq_head;
1169 if (globreq->type == GLOBREQ_TCPIP_FORWARD) {
1170 if (type == SSH2_MSG_REQUEST_FAILURE) {
1171 share_remove_forwarding(cs, globreq->fwd);
1173 globreq->fwd->active = TRUE;
1175 } else if (globreq->type == GLOBREQ_CANCEL_TCPIP_FORWARD) {
1176 if (type == SSH2_MSG_REQUEST_SUCCESS) {
1177 share_remove_forwarding(cs, globreq->fwd);
1180 if (globreq->want_reply) {
1181 send_packet_to_downstream(cs, type, pkt, pktlen, NULL);
1183 cs->globreq_head = globreq->next;
1185 if (cs->globreq_head == NULL)
1186 cs->globreq_tail = NULL;
1189 /* Retry cleaning up this connection, in case that reply
1190 * was the last thing we were waiting for. */
1191 share_try_cleanup(cs);
1196 case SSH2_MSG_CHANNEL_OPEN:
1197 id_pos = getstring_size(pkt, pktlen);
1198 assert(id_pos >= 0);
1199 server_id = GET_32BIT(pkt + id_pos);
1200 share_add_halfchannel(cs, server_id);
1202 send_packet_to_downstream(cs, type, pkt, pktlen, NULL);
1205 case SSH2_MSG_CHANNEL_OPEN_CONFIRMATION:
1206 case SSH2_MSG_CHANNEL_OPEN_FAILURE:
1207 case SSH2_MSG_CHANNEL_CLOSE:
1208 case SSH2_MSG_CHANNEL_WINDOW_ADJUST:
1209 case SSH2_MSG_CHANNEL_DATA:
1210 case SSH2_MSG_CHANNEL_EXTENDED_DATA:
1211 case SSH2_MSG_CHANNEL_EOF:
1212 case SSH2_MSG_CHANNEL_REQUEST:
1213 case SSH2_MSG_CHANNEL_SUCCESS:
1214 case SSH2_MSG_CHANNEL_FAILURE:
1216 * All these messages have the recipient channel id as the
1217 * first uint32 field in the packet. Substitute the downstream
1218 * channel id for our one and pass the packet downstream.
1220 assert(pktlen >= 4);
1221 upstream_id = GET_32BIT(pkt);
1222 if ((chan = share_find_channel_by_upstream(cs, upstream_id)) != NULL) {
1224 * The normal case: this id refers to an open channel.
1226 PUT_32BIT(pkt, chan->downstream_id);
1227 send_packet_to_downstream(cs, type, pkt, pktlen, chan);
1230 * Update the channel state, for messages that need it.
1232 if (type == SSH2_MSG_CHANNEL_OPEN_CONFIRMATION) {
1233 if (chan->state == UNACKNOWLEDGED && pktlen >= 8) {
1234 share_channel_set_server_id(cs, chan, GET_32BIT(pkt+4),
1237 /* Retry cleaning up this connection, so that we
1238 * can send an immediate CLOSE on this channel for
1239 * which we now know the server id. */
1240 share_try_cleanup(cs);
1243 } else if (type == SSH2_MSG_CHANNEL_OPEN_FAILURE) {
1244 ssh_delete_sharing_channel(cs->parent->ssh, chan->upstream_id);
1245 share_remove_channel(cs, chan);
1246 } else if (type == SSH2_MSG_CHANNEL_CLOSE) {
1247 if (chan->state == SENT_CLOSE) {
1248 ssh_delete_sharing_channel(cs->parent->ssh,
1250 share_remove_channel(cs, chan);
1252 /* Retry cleaning up this connection, in case this
1253 * channel closure was the last thing we were
1255 share_try_cleanup(cs);
1258 chan->state = RCVD_CLOSE;
1261 } else if ((xc = share_find_xchannel_by_upstream(cs, upstream_id))
1264 * The unusual case: this id refers to an xchannel. Add it
1265 * to the xchannel's queue.
1267 struct share_xchannel_message *msg;
1269 msg = share_xchannel_add_message(xc, type, pktlen);
1270 memcpy(msg->data, pkt, pktlen);
1272 /* If the xchannel is dead, then also respond to it (which
1273 * may involve deleting the channel). */
1275 share_dead_xchannel_respond(cs, xc);
1280 assert(!"This packet type should never have come from ssh.c");
1285 static void share_got_pkt_from_downstream(struct ssh_sharing_connstate *cs,
1287 unsigned char *pkt, int pktlen)
1290 struct share_forwarding *fwd;
1292 unsigned old_id, new_id, server_id;
1293 struct share_globreq *globreq;
1294 struct share_channel *chan;
1295 struct share_halfchannel *hc;
1296 struct share_xchannel *xc;
1300 case SSH2_MSG_DISCONNECT:
1302 * This message stops here: if downstream is disconnecting
1303 * from us, that doesn't mean we want to disconnect from the
1304 * SSH server. Close the downstream connection and start
1307 share_begin_cleanup(cs);
1310 case SSH2_MSG_GLOBAL_REQUEST:
1312 * The only global requests we understand are "tcpip-forward"
1313 * and "cancel-tcpip-forward". Since those require us to
1314 * maintain state, we must assume that other global requests
1315 * will probably require that too, and so we don't forward on
1316 * any request we don't understand.
1318 request_name = getstring(pkt, pktlen);
1319 if (request_name == NULL) {
1320 err = dupprintf("Truncated GLOBAL_REQUEST packet");
1324 if (!strcmp(request_name, "tcpip-forward")) {
1325 int wantreplypos, orig_wantreply, port, ret;
1328 sfree(request_name);
1331 * Pick the packet apart to find the want_reply field and
1332 * the host/port we're going to ask to listen on.
1334 wantreplypos = getstring_size(pkt, pktlen);
1335 if (wantreplypos < 0 || wantreplypos >= pktlen) {
1336 err = dupprintf("Truncated GLOBAL_REQUEST packet");
1339 orig_wantreply = pkt[wantreplypos];
1340 port = getstring_size(pkt + (wantreplypos + 1),
1341 pktlen - (wantreplypos + 1));
1342 port += (wantreplypos + 1);
1343 if (port < 0 || port > pktlen - 4) {
1344 err = dupprintf("Truncated GLOBAL_REQUEST packet");
1347 host = getstring(pkt + (wantreplypos + 1),
1348 pktlen - (wantreplypos + 1));
1349 assert(host != NULL);
1350 port = GET_32BIT(pkt + port);
1353 * See if we can allocate space in ssh.c's tree of remote
1354 * port forwardings. If we can't, it's because another
1355 * client sharing this connection has already allocated
1356 * the identical port forwarding, so we take it on
1357 * ourselves to manufacture a failure packet and send it
1358 * back to downstream.
1360 ret = ssh_alloc_sharing_rportfwd(cs->parent->ssh, host, port, cs);
1362 if (orig_wantreply) {
1363 send_packet_to_downstream(cs, SSH2_MSG_REQUEST_FAILURE,
1368 * We've managed to make space for this forwarding
1369 * locally. Pass the request on to the SSH server, but
1370 * set want_reply even if it wasn't originally set, so
1371 * that we know whether this forwarding needs to be
1372 * cleaned up if downstream goes away.
1374 int old_wantreply = pkt[wantreplypos];
1375 pkt[wantreplypos] = 1;
1376 ssh_send_packet_from_downstream
1377 (cs->parent->ssh, cs->id, type, pkt, pktlen,
1378 old_wantreply ? NULL : "upstream added want_reply flag");
1379 fwd = share_add_forwarding(cs, host, port);
1380 ssh_sharing_queue_global_request(cs->parent->ssh, cs);
1383 globreq = snew(struct share_globreq);
1384 globreq->next = NULL;
1385 if (cs->globreq_tail)
1386 cs->globreq_tail->next = globreq;
1388 cs->globreq_head = globreq;
1390 globreq->want_reply = orig_wantreply;
1391 globreq->type = GLOBREQ_TCPIP_FORWARD;
1396 } else if (!strcmp(request_name, "cancel-tcpip-forward")) {
1397 int wantreplypos, orig_wantreply, port;
1399 struct share_forwarding *fwd;
1401 sfree(request_name);
1404 * Pick the packet apart to find the want_reply field and
1405 * the host/port we're going to ask to listen on.
1407 wantreplypos = getstring_size(pkt, pktlen);
1408 if (wantreplypos < 0 || wantreplypos >= pktlen) {
1409 err = dupprintf("Truncated GLOBAL_REQUEST packet");
1412 orig_wantreply = pkt[wantreplypos];
1413 port = getstring_size(pkt + (wantreplypos + 1),
1414 pktlen - (wantreplypos + 1));
1415 port += (wantreplypos + 1);
1416 if (port < 0 || port > pktlen - 4) {
1417 err = dupprintf("Truncated GLOBAL_REQUEST packet");
1420 host = getstring(pkt + (wantreplypos + 1),
1421 pktlen - (wantreplypos + 1));
1422 assert(host != NULL);
1423 port = GET_32BIT(pkt + port);
1426 * Look up the existing forwarding with these details.
1428 fwd = share_find_forwarding(cs, host, port);
1430 if (orig_wantreply) {
1431 send_packet_to_downstream(cs, SSH2_MSG_REQUEST_FAILURE,
1436 * Pass the cancel request on to the SSH server, but
1437 * set want_reply even if it wasn't originally set, so
1438 * that _we_ know whether the forwarding has been
1439 * deleted even if downstream doesn't want to know.
1441 int old_wantreply = pkt[wantreplypos];
1442 pkt[wantreplypos] = 1;
1443 ssh_send_packet_from_downstream
1444 (cs->parent->ssh, cs->id, type, pkt, pktlen,
1445 old_wantreply ? NULL : "upstream added want_reply flag");
1446 ssh_sharing_queue_global_request(cs->parent->ssh, cs);
1452 * Request we don't understand. Manufacture a failure
1453 * message if an answer was required.
1457 sfree(request_name);
1459 wantreplypos = getstring_size(pkt, pktlen);
1460 if (wantreplypos < 0 || wantreplypos >= pktlen) {
1461 err = dupprintf("Truncated GLOBAL_REQUEST packet");
1464 if (pkt[wantreplypos])
1465 send_packet_to_downstream(cs, SSH2_MSG_REQUEST_FAILURE,
1470 case SSH2_MSG_CHANNEL_OPEN:
1471 /* Sender channel id comes after the channel type string */
1472 id_pos = getstring_size(pkt, pktlen);
1473 if (id_pos < 0 || id_pos > pktlen - 12) {
1474 err = dupprintf("Truncated CHANNEL_OPEN packet");
1478 old_id = GET_32BIT(pkt + id_pos);
1479 new_id = ssh_alloc_sharing_channel(cs->parent->ssh, cs);
1480 share_add_channel(cs, old_id, new_id, 0, UNACKNOWLEDGED,
1481 GET_32BIT(pkt + id_pos + 8));
1482 PUT_32BIT(pkt + id_pos, new_id);
1483 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
1484 type, pkt, pktlen, NULL);
1487 case SSH2_MSG_CHANNEL_OPEN_CONFIRMATION:
1489 err = dupprintf("Truncated CHANNEL_OPEN_CONFIRMATION packet");
1493 id_pos = 4; /* sender channel id is 2nd uint32 field in packet */
1494 old_id = GET_32BIT(pkt + id_pos);
1496 server_id = GET_32BIT(pkt);
1497 /* This server id may refer to either a halfchannel or an xchannel. */
1498 hc = NULL, xc = NULL; /* placate optimiser */
1499 if ((hc = share_find_halfchannel(cs, server_id)) != NULL) {
1500 new_id = ssh_alloc_sharing_channel(cs->parent->ssh, cs);
1501 } else if ((xc = share_find_xchannel_by_server(cs, server_id))
1503 new_id = xc->upstream_id;
1505 err = dupprintf("CHANNEL_OPEN_CONFIRMATION packet cited unknown channel %u", (unsigned)server_id);
1509 PUT_32BIT(pkt + id_pos, new_id);
1511 chan = share_add_channel(cs, old_id, new_id, server_id, OPEN,
1512 GET_32BIT(pkt + 12));
1515 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
1516 type, pkt, pktlen, NULL);
1517 share_remove_halfchannel(cs, hc);
1519 unsigned downstream_window = GET_32BIT(pkt + 8);
1520 if (downstream_window < 256) {
1521 err = dupprintf("Initial window size for x11 channel must be at least 256 (got %u)", downstream_window);
1524 share_xchannel_confirmation(cs, xc, chan, downstream_window);
1525 share_remove_xchannel(cs, xc);
1530 case SSH2_MSG_CHANNEL_OPEN_FAILURE:
1532 err = dupprintf("Truncated CHANNEL_OPEN_FAILURE packet");
1536 server_id = GET_32BIT(pkt);
1537 /* This server id may refer to either a halfchannel or an xchannel. */
1538 if ((hc = share_find_halfchannel(cs, server_id)) != NULL) {
1539 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
1540 type, pkt, pktlen, NULL);
1541 share_remove_halfchannel(cs, hc);
1542 } else if ((xc = share_find_xchannel_by_server(cs, server_id))
1544 share_xchannel_failure(cs, xc);
1546 err = dupprintf("CHANNEL_OPEN_FAILURE packet cited unknown channel %u", (unsigned)server_id);
1552 case SSH2_MSG_CHANNEL_WINDOW_ADJUST:
1553 case SSH2_MSG_CHANNEL_DATA:
1554 case SSH2_MSG_CHANNEL_EXTENDED_DATA:
1555 case SSH2_MSG_CHANNEL_EOF:
1556 case SSH2_MSG_CHANNEL_CLOSE:
1557 case SSH2_MSG_CHANNEL_REQUEST:
1558 case SSH2_MSG_CHANNEL_SUCCESS:
1559 case SSH2_MSG_CHANNEL_FAILURE:
1560 case SSH2_MSG_IGNORE:
1561 case SSH2_MSG_DEBUG:
1562 if (type == SSH2_MSG_CHANNEL_REQUEST &&
1563 (request_name = getstring(pkt + 4, pktlen - 4)) != NULL) {
1565 * Agent forwarding requests from downstream are treated
1566 * specially. Because OpenSSHD doesn't let us enable agent
1567 * forwarding independently per session channel, and in
1568 * particular because the OpenSSH-defined agent forwarding
1569 * protocol does not mark agent-channel requests with the
1570 * id of the session channel they originate from, the only
1571 * way we can implement agent forwarding in a
1572 * connection-shared PuTTY is to forward the _upstream_
1573 * agent. Hence, we unilaterally deny agent forwarding
1574 * requests from downstreams if we aren't prepared to
1575 * forward an agent ourselves.
1577 * (If we are, then we dutifully pass agent forwarding
1578 * requests upstream. OpenSSHD has the curious behaviour
1579 * that all but the first such request will be rejected,
1580 * but all session channels opened after the first request
1581 * get agent forwarding enabled whether they ask for it or
1582 * not; but that's not our concern, since other SSH
1583 * servers supporting the same piece of protocol might in
1584 * principle at least manage to enable agent forwarding on
1585 * precisely the channels that requested it, even if the
1586 * subsequent CHANNEL_OPENs still can't be associated with
1587 * a parent session channel.)
1589 if (!strcmp(request_name, "auth-agent-req@openssh.com") &&
1590 !ssh_agent_forwarding_permitted(cs->parent->ssh)) {
1591 unsigned server_id = GET_32BIT(pkt);
1592 unsigned char recipient_id[4];
1593 chan = share_find_channel_by_server(cs, server_id);
1595 PUT_32BIT(recipient_id, chan->downstream_id);
1596 send_packet_to_downstream(cs, SSH2_MSG_CHANNEL_FAILURE,
1597 recipient_id, 4, NULL);
1599 char *buf = dupprintf("Agent forwarding request for "
1600 "unrecognised channel %u", server_id);
1601 share_disconnect(cs, buf);
1609 * Another thing we treat specially is X11 forwarding
1610 * requests. For these, we have to make up another set of
1611 * X11 auth data, and enter it into our SSH connection's
1612 * list of possible X11 authorisation credentials so that
1613 * when we see an X11 channel open request we can know
1614 * whether it's one to handle locally or one to pass on to
1615 * a downstream, and if the latter, which one.
1617 if (!strcmp(request_name, "x11-req")) {
1618 unsigned server_id = GET_32BIT(pkt);
1619 int want_reply, single_connection, screen;
1620 char *auth_proto_str, *auth_data;
1621 int auth_proto, protolen, datalen;
1624 chan = share_find_channel_by_server(cs, server_id);
1626 char *buf = dupprintf("X11 forwarding request for "
1627 "unrecognised channel %u", server_id);
1628 share_disconnect(cs, buf);
1634 * Pick apart the whole message to find the downstream
1637 /* we have already seen: 4 bytes channel id, 4+7 request name */
1639 err = dupprintf("Truncated CHANNEL_REQUEST(\"x11\") packet");
1642 want_reply = pkt[15] != 0;
1643 single_connection = pkt[16] != 0;
1644 auth_proto_str = getstring(pkt+17, pktlen-17);
1645 pos = 17 + getstring_size(pkt+17, pktlen-17);
1646 auth_data = getstring(pkt+pos, pktlen-pos);
1647 pos += getstring_size(pkt+pos, pktlen-pos);
1648 if (pktlen < pos+4) {
1649 err = dupprintf("Truncated CHANNEL_REQUEST(\"x11\") packet");
1652 screen = GET_32BIT(pkt+pos);
1654 auth_proto = x11_identify_auth_proto(auth_proto_str);
1655 if (auth_proto < 0) {
1656 /* Reject due to not understanding downstream's
1657 * requested authorisation method. */
1658 unsigned char recipient_id[4];
1659 PUT_32BIT(recipient_id, chan->downstream_id);
1660 send_packet_to_downstream(cs, SSH2_MSG_CHANNEL_FAILURE,
1661 recipient_id, 4, NULL);
1664 chan->x11_auth_proto = auth_proto;
1665 chan->x11_auth_data = x11_dehexify(auth_data,
1666 &chan->x11_auth_datalen);
1667 chan->x11_auth_upstream =
1668 ssh_sharing_add_x11_display(cs->parent->ssh, auth_proto,
1670 chan->x11_one_shot = single_connection;
1673 * Now construct a replacement X forwarding request,
1674 * containing our own auth data, and send that to the
1677 protolen = strlen(chan->x11_auth_upstream->protoname);
1678 datalen = strlen(chan->x11_auth_upstream->datastring);
1679 pktlen = 29+protolen+datalen;
1680 pkt = snewn(pktlen, unsigned char);
1681 PUT_32BIT(pkt, server_id);
1682 PUT_32BIT(pkt+4, 7); /* strlen("x11-req") */
1683 memcpy(pkt+8, "x11-req", 7);
1684 pkt[15] = want_reply;
1685 pkt[16] = single_connection;
1686 PUT_32BIT(pkt+17, protolen);
1687 memcpy(pkt+21, chan->x11_auth_upstream->protoname, protolen);
1688 PUT_32BIT(pkt+21+protolen, datalen);
1689 memcpy(pkt+25+protolen, chan->x11_auth_upstream->datastring,
1691 PUT_32BIT(pkt+25+protolen+datalen, screen);
1692 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
1693 SSH2_MSG_CHANNEL_REQUEST,
1701 ssh_send_packet_from_downstream(cs->parent->ssh, cs->id,
1702 type, pkt, pktlen, NULL);
1703 if (type == SSH2_MSG_CHANNEL_CLOSE && pktlen >= 4) {
1704 server_id = GET_32BIT(pkt);
1705 chan = share_find_channel_by_server(cs, server_id);
1707 if (chan->state == RCVD_CLOSE) {
1708 ssh_delete_sharing_channel(cs->parent->ssh,
1710 share_remove_channel(cs, chan);
1712 chan->state = SENT_CLOSE;
1719 err = dupprintf("Unexpected packet type %d\n", type);
1723 * Any other packet type is unexpected. In particular, we
1724 * never pass GLOBAL_REQUESTs downstream, so we never expect
1725 * to see SSH2_MSG_REQUEST_{SUCCESS,FAILURE}.
1728 assert(err != NULL);
1729 share_disconnect(cs, err);
1736 * Coroutine macros similar to, but simplified from, those in ssh.c.
1738 #define crBegin(v) { int *crLine = &v; switch(v) { case 0:;
1739 #define crFinish(z) } *crLine = 0; return (z); }
1740 #define crGetChar(c) do \
1742 while (len == 0) { \
1743 *crLine =__LINE__; return 1; case __LINE__:; \
1746 (c) = (unsigned char)*data++; \
1749 static int share_receive(Plug plug, int urgent, char *data, int len)
1751 struct ssh_sharing_connstate *cs = (struct ssh_sharing_connstate *)plug;
1752 static const char expected_verstring_prefix[] =
1753 "SSHCONNECTION@putty.projects.tartarus.org-2.0-";
1756 crBegin(cs->crLine);
1759 * First read the version string from downstream.
1766 if (cs->recvlen > sizeof(cs->recvbuf)) {
1767 char *buf = dupprintf("Version string far too long\n");
1768 share_disconnect(cs, buf);
1772 cs->recvbuf[cs->recvlen++] = c;
1776 * Now parse the version string to make sure it's at least vaguely
1777 * sensible, and log it.
1779 if (cs->recvlen < sizeof(expected_verstring_prefix)-1 ||
1780 memcmp(cs->recvbuf, expected_verstring_prefix,
1781 sizeof(expected_verstring_prefix) - 1)) {
1782 char *buf = dupprintf("Version string did not have expected prefix\n");
1783 share_disconnect(cs, buf);
1787 if (cs->recvlen > 0 && cs->recvbuf[cs->recvlen-1] == '\015')
1788 cs->recvlen--; /* trim off \r before \n */
1789 ssh_sharing_logf(cs->parent->ssh, cs->id,
1790 "Downstream version string: %.*s",
1791 cs->recvlen, cs->recvbuf);
1794 * Loop round reading packets.
1798 while (cs->recvlen < 4) {
1800 cs->recvbuf[cs->recvlen++] = c;
1802 cs->curr_packetlen = toint(GET_32BIT(cs->recvbuf) + 4);
1803 if (cs->curr_packetlen < 5 ||
1804 cs->curr_packetlen > sizeof(cs->recvbuf)) {
1805 char *buf = dupprintf("Bad packet length %u\n",
1806 (unsigned)cs->curr_packetlen);
1807 share_disconnect(cs, buf);
1811 while (cs->recvlen < cs->curr_packetlen) {
1813 cs->recvbuf[cs->recvlen++] = c;
1816 share_got_pkt_from_downstream(cs, cs->recvbuf[4],
1817 cs->recvbuf + 5, cs->recvlen - 5);
1824 static void share_sent(Plug plug, int bufsize)
1826 /* struct ssh_sharing_connstate *cs = (struct ssh_sharing_connstate *)plug; */
1829 * We do nothing here, because we expect that there won't be a
1830 * need to throttle and unthrottle the connection to a downstream.
1831 * It should automatically throttle itself: if the SSH server
1832 * sends huge amounts of data on all channels then it'll run out
1833 * of window until our downstream sends it back some
1838 static int share_listen_closing(Plug plug, const char *error_msg,
1839 int error_code, int calling_back)
1841 struct ssh_sharing_state *sharestate = (struct ssh_sharing_state *)plug;
1843 ssh_sharing_logf(sharestate->ssh, 0,
1844 "listening socket: %s", error_msg);
1845 sk_close(sharestate->listensock);
1849 static void share_send_verstring(struct ssh_sharing_connstate *cs)
1851 char *fullstring = dupcat("SSHCONNECTION@putty.projects.tartarus.org-2.0-",
1852 cs->parent->server_verstring, "\015\012", NULL);
1853 sk_write(cs->sock, fullstring, strlen(fullstring));
1856 cs->sent_verstring = TRUE;
1859 int share_ndownstreams(void *state)
1861 struct ssh_sharing_state *sharestate = (struct ssh_sharing_state *)state;
1862 return count234(sharestate->connections);
1865 void share_activate(void *state, const char *server_verstring)
1868 * Indication from ssh.c that we are now ready to begin serving
1869 * any downstreams that have already connected to us.
1871 struct ssh_sharing_state *sharestate = (struct ssh_sharing_state *)state;
1872 struct ssh_sharing_connstate *cs;
1876 * Trim the server's version string down to just the software
1877 * version component, removing "SSH-2.0-" or whatever at the
1880 for (i = 0; i < 2; i++) {
1881 server_verstring += strcspn(server_verstring, "-");
1882 if (*server_verstring)
1886 sharestate->server_verstring = dupstr(server_verstring);
1888 for (i = 0; (cs = (struct ssh_sharing_connstate *)
1889 index234(sharestate->connections, i)) != NULL; i++) {
1890 assert(!cs->sent_verstring);
1891 share_send_verstring(cs);
1895 static int share_listen_accepting(Plug plug,
1896 accept_fn_t constructor, accept_ctx_t ctx)
1898 static const struct plug_function_table connection_fn_table = {
1899 NULL, /* no log function, because that's for outgoing connections */
1903 NULL /* no accepting function, because we've already done it */
1905 struct ssh_sharing_state *sharestate = (struct ssh_sharing_state *)plug;
1906 struct ssh_sharing_connstate *cs;
1910 * A new downstream has connected to us.
1912 cs = snew(struct ssh_sharing_connstate);
1913 cs->fn = &connection_fn_table;
1914 cs->parent = sharestate;
1916 if ((cs->id = share_find_unused_id(sharestate, sharestate->nextid)) == 0 &&
1917 (cs->id = share_find_unused_id(sharestate, 1)) == 0) {
1921 sharestate->nextid = cs->id + 1;
1922 if (sharestate->nextid == 0)
1923 sharestate->nextid++; /* only happens in VERY long-running upstreams */
1925 cs->sock = constructor(ctx, (Plug) cs);
1926 if ((err = sk_socket_error(cs->sock)) != NULL) {
1931 sk_set_frozen(cs->sock, 0);
1933 add234(cs->parent->connections, cs);
1935 cs->sent_verstring = FALSE;
1936 if (sharestate->server_verstring)
1937 share_send_verstring(cs);
1939 cs->got_verstring = FALSE;
1942 cs->halfchannels = newtree234(share_halfchannel_cmp);
1943 cs->channels_by_us = newtree234(share_channel_us_cmp);
1944 cs->channels_by_server = newtree234(share_channel_server_cmp);
1945 cs->xchannels_by_us = newtree234(share_xchannel_us_cmp);
1946 cs->xchannels_by_server = newtree234(share_xchannel_server_cmp);
1947 cs->forwardings = newtree234(share_forwarding_cmp);
1948 cs->globreq_head = cs->globreq_tail = NULL;
1950 ssh_sharing_downstream_connected(sharestate->ssh, cs->id);
1955 /* Per-application overrides for what roles we can take (e.g. pscp
1956 * will never be an upstream) */
1957 extern const int share_can_be_downstream;
1958 extern const int share_can_be_upstream;
1961 * Init function for connection sharing. We either open a listening
1962 * socket and become an upstream, or connect to an existing one and
1963 * become a downstream, or do neither. We are responsible for deciding
1964 * which of these to do (including checking the Conf to see if
1965 * connection sharing is even enabled in the first place). If we
1966 * become a downstream, we return the Socket with which we connected
1967 * to the upstream; otherwise (whether or not we have established an
1968 * upstream) we return NULL.
1970 Socket ssh_connection_sharing_init(const char *host, int port,
1971 Conf *conf, Ssh ssh, void **state)
1973 static const struct plug_function_table listen_fn_table = {
1974 NULL, /* no log function, because that's for outgoing connections */
1975 share_listen_closing,
1976 NULL, /* no receive function on a listening socket */
1977 NULL, /* no sent function on a listening socket */
1978 share_listen_accepting
1981 int result, can_upstream, can_downstream;
1982 char *logtext, *ds_err, *us_err;
1985 struct ssh_sharing_state *sharestate;
1987 if (!conf_get_int(conf, CONF_ssh_connection_sharing))
1988 return NULL; /* do not share anything */
1989 can_upstream = share_can_be_upstream &&
1990 conf_get_int(conf, CONF_ssh_connection_sharing_upstream);
1991 can_downstream = share_can_be_downstream &&
1992 conf_get_int(conf, CONF_ssh_connection_sharing_downstream);
1993 if (!can_upstream && !can_downstream)
1997 * Decide on the string used to identify the connection point
1998 * between upstream and downstream (be it a Windows named pipe or
1999 * a Unix-domain socket or whatever else).
2001 * I wondered about making this a SHA hash of all sorts of pieces
2002 * of the PuTTY configuration - essentially everything PuTTY uses
2003 * to know where and how to make a connection, including all the
2004 * proxy details (or rather, all the _relevant_ ones - only
2005 * including settings that other settings didn't prevent from
2006 * having any effect), plus the username. However, I think it's
2007 * better to keep it really simple: the connection point
2008 * identifier is derived from the hostname and port used to index
2009 * the host-key cache (not necessarily where we _physically_
2010 * connected to, in cases involving proxies or CONF_loghost), plus
2011 * the username if one is specified.
2014 char *username = get_remote_username(conf);
2018 sockname = dupprintf("%s@%s", username, host);
2020 sockname = dupprintf("%s", host);
2023 sockname = dupprintf("%s@%s:%d", username, host, port);
2025 sockname = dupprintf("%s:%d", host, port);
2031 * The platform-specific code may transform this further in
2032 * order to conform to local namespace conventions (e.g. not
2033 * using slashes in filenames), but that's its job and not
2039 * Create a data structure for the listening plug if we turn out
2040 * to be an upstream.
2042 sharestate = snew(struct ssh_sharing_state);
2043 sharestate->fn = &listen_fn_table;
2044 sharestate->listensock = NULL;
2047 * Now hand off to a per-platform routine that either connects to
2048 * an existing upstream (using 'ssh' as the plug), establishes our
2049 * own upstream (using 'sharestate' as the plug), or forks off a
2050 * separate upstream and then connects to that. It will return a
2051 * code telling us which kind of socket it put in 'sock'.
2054 logtext = ds_err = us_err = NULL;
2055 result = platform_ssh_share(sockname, conf, (Plug)ssh,
2056 (Plug)sharestate, &sock, &logtext, &ds_err,
2057 &us_err, can_upstream, can_downstream);
2058 ssh_connshare_log(ssh, result, logtext, ds_err, us_err);
2065 * We aren't sharing our connection at all (e.g. something
2066 * went wrong setting the socket up). Free the upstream
2067 * structure and return NULL.
2069 assert(sock == NULL);
2075 case SHARE_DOWNSTREAM:
2077 * We are downstream, so free sharestate which it turns out we
2078 * don't need after all, and return the downstream socket as a
2079 * replacement for an ordinary SSH connection.
2086 case SHARE_UPSTREAM:
2088 * We are upstream. Set up sharestate properly and pass a copy
2089 * to the caller; return NULL, to tell ssh.c that it has to
2090 * make an ordinary connection after all.
2092 *state = sharestate;
2093 sharestate->listensock = sock;
2094 sharestate->connections = newtree234(share_connstate_cmp);
2095 sharestate->ssh = ssh;
2096 sharestate->server_verstring = NULL;
2097 sharestate->sockname = dupstr(sockname);
2098 sharestate->nextid = 1;