1 /* Copyright (c) 2001 Matej Pfajfar.
2 * Copyright (c) 2001-2004, Roger Dingledine.
3 * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
4 * Copyright (c) 2007-2017, The Tor Project, Inc. */
5 /* See LICENSE for licensing information */
9 * \brief General high-level functions to handle reading and writing
12 * Each connection (ideally) represents a TLS connection, a TCP socket, a unix
13 * socket, or a UDP socket on which reads and writes can occur. (But see
14 * connection_edge.c for cases where connections can also represent streams
15 * that do not have a corresponding socket.)
17 * The module implements the abstract type, connection_t. The subtypes are:
19 * <li>listener_connection_t, implemented here in connection.c
20 * <li>dir_connection_t, implemented in directory.c
21 * <li>or_connection_t, implemented in connection_or.c
22 * <li>edge_connection_t, implemented in connection_edge.c, along with
24 * <ul><li>entry_connection_t, also implemented in connection_edge.c
26 * <li>control_connection_t, implemented in control.c
29 * The base type implemented in this module is responsible for basic
30 * rate limiting, flow control, and marshalling bytes onto and off of the
31 * network (either directly or via TLS).
33 * Connections are registered with the main loop with connection_add(). As
34 * they become able to read or write register the fact with the event main
35 * loop by calling connection_watch_events(), connection_start_reading(), or
36 * connection_start_writing(). When they no longer want to read or write,
37 * they call connection_stop_reading() or connection_stop_writing().
39 * To queue data to be written on a connection, call
40 * connection_write_to_buf(). When data arrives, the
41 * connection_process_inbuf() callback is invoked, which dispatches to a
42 * type-specific function (such as connection_edge_process_inbuf() for
43 * example). Connection types that need notice of when data has been written
44 * receive notification via connection_flushed_some() and
45 * connection_finished_flushing(). These functions all delegate to
46 * type-specific implementations.
48 * Additionally, beyond the core of connection_t, this module also implements:
50 * <li>Listeners, which wait for incoming sockets and launch connections
51 * <li>Outgoing SOCKS proxy support
52 * <li>Outgoing HTTP proxy support
53 * <li>An out-of-sockets handler for dealing with socket exhaustion
57 #define CONNECTION_PRIVATE
62 * Define this so we get channel internal functions, since we're implementing
63 * part of a subclass (channel_tls_t).
65 #define TOR_CHANNEL_INTERNAL_
66 #define CONNECTION_PRIVATE
67 #include "backtrace.h"
69 #include "channeltls.h"
70 #include "circuitbuild.h"
71 #include "circuitlist.h"
72 #include "circuituse.h"
74 #include "connection.h"
75 #include "connection_edge.h"
76 #include "connection_or.h"
78 #include "directory.h"
82 #include "entrynodes.h"
83 #include "ext_orport.h"
86 #include "hs_common.h"
92 #include "rendclient.h"
93 #include "rendcommon.h"
96 #include "routerlist.h"
97 #include "transports.h"
98 #include "routerparse.h"
100 #include "transports.h"
107 #include <sys/socket.h>
111 static connection_t
*connection_listener_new(
112 const struct sockaddr
*listensockaddr
,
113 socklen_t listensocklen
, int type
,
115 const port_cfg_t
*portcfg
);
116 static void connection_init(time_t now
, connection_t
*conn
, int type
,
118 static int connection_init_accepted_conn(connection_t
*conn
,
119 const listener_connection_t
*listener
);
120 static int connection_handle_listener_read(connection_t
*conn
, int new_type
);
121 static int connection_bucket_should_increase(int bucket
,
122 or_connection_t
*conn
);
123 static int connection_finished_flushing(connection_t
*conn
);
124 static int connection_flushed_some(connection_t
*conn
);
125 static int connection_finished_connecting(connection_t
*conn
);
126 static int connection_reached_eof(connection_t
*conn
);
127 static int connection_read_to_buf(connection_t
*conn
, ssize_t
*max_to_read
,
129 static int connection_process_inbuf(connection_t
*conn
, int package_partial
);
130 static void client_check_address_changed(tor_socket_t sock
);
131 static void set_constrained_socket_buffers(tor_socket_t sock
, int size
);
133 static const char *connection_proxy_state_to_string(int state
);
134 static int connection_read_https_proxy_response(connection_t
*conn
);
135 static void connection_send_socks5_connect(connection_t
*conn
);
136 static const char *proxy_type_to_string(int proxy_type
);
137 static int get_proxy_type(void);
138 const tor_addr_t
*conn_get_outbound_address(sa_family_t family
,
139 const or_options_t
*options
, unsigned int conn_type
);
141 /** The last addresses that our network interface seemed to have been
142 * binding to. We use this as one way to detect when our IP changes.
144 * XXXX+ We should really use the entire list of interfaces here.
146 static tor_addr_t
*last_interface_ipv4
= NULL
;
147 /* DOCDOC last_interface_ipv6 */
148 static tor_addr_t
*last_interface_ipv6
= NULL
;
149 /** A list of tor_addr_t for addresses we've used in outgoing connections.
150 * Used to detect IP address changes. */
151 static smartlist_t
*outgoing_addrs
= NULL
;
153 #define CASE_ANY_LISTENER_TYPE \
154 case CONN_TYPE_OR_LISTENER: \
155 case CONN_TYPE_EXT_OR_LISTENER: \
156 case CONN_TYPE_AP_LISTENER: \
157 case CONN_TYPE_DIR_LISTENER: \
158 case CONN_TYPE_CONTROL_LISTENER: \
159 case CONN_TYPE_AP_TRANS_LISTENER: \
160 case CONN_TYPE_AP_NATD_LISTENER: \
161 case CONN_TYPE_AP_DNS_LISTENER
163 /**************************************************************/
166 * Return the human-readable name for the connection type <b>type</b>
169 conn_type_to_string(int type
)
173 case CONN_TYPE_OR_LISTENER
: return "OR listener";
174 case CONN_TYPE_OR
: return "OR";
175 case CONN_TYPE_EXIT
: return "Exit";
176 case CONN_TYPE_AP_LISTENER
: return "Socks listener";
177 case CONN_TYPE_AP_TRANS_LISTENER
:
178 return "Transparent pf/netfilter listener";
179 case CONN_TYPE_AP_NATD_LISTENER
: return "Transparent natd listener";
180 case CONN_TYPE_AP_DNS_LISTENER
: return "DNS listener";
181 case CONN_TYPE_AP
: return "Socks";
182 case CONN_TYPE_DIR_LISTENER
: return "Directory listener";
183 case CONN_TYPE_DIR
: return "Directory";
184 case CONN_TYPE_CONTROL_LISTENER
: return "Control listener";
185 case CONN_TYPE_CONTROL
: return "Control";
186 case CONN_TYPE_EXT_OR
: return "Extended OR";
187 case CONN_TYPE_EXT_OR_LISTENER
: return "Extended OR listener";
189 log_warn(LD_BUG
, "unknown connection type %d", type
);
190 tor_snprintf(buf
, sizeof(buf
), "unknown [%d]", type
);
196 * Return the human-readable name for the connection state <b>state</b>
197 * for the connection type <b>type</b>
200 conn_state_to_string(int type
, int state
)
204 CASE_ANY_LISTENER_TYPE
:
205 if (state
== LISTENER_STATE_READY
)
210 case OR_CONN_STATE_CONNECTING
: return "connect()ing";
211 case OR_CONN_STATE_PROXY_HANDSHAKING
: return "handshaking (proxy)";
212 case OR_CONN_STATE_TLS_HANDSHAKING
: return "handshaking (TLS)";
213 case OR_CONN_STATE_TLS_CLIENT_RENEGOTIATING
:
214 return "renegotiating (TLS, v2 handshake)";
215 case OR_CONN_STATE_TLS_SERVER_RENEGOTIATING
:
216 return "waiting for renegotiation or V3 handshake";
217 case OR_CONN_STATE_OR_HANDSHAKING_V2
:
218 return "handshaking (Tor, v2 handshake)";
219 case OR_CONN_STATE_OR_HANDSHAKING_V3
:
220 return "handshaking (Tor, v3 handshake)";
221 case OR_CONN_STATE_OPEN
: return "open";
224 case CONN_TYPE_EXT_OR
:
226 case EXT_OR_CONN_STATE_AUTH_WAIT_AUTH_TYPE
:
227 return "waiting for authentication type";
228 case EXT_OR_CONN_STATE_AUTH_WAIT_CLIENT_NONCE
:
229 return "waiting for client nonce";
230 case EXT_OR_CONN_STATE_AUTH_WAIT_CLIENT_HASH
:
231 return "waiting for client hash";
232 case EXT_OR_CONN_STATE_OPEN
: return "open";
233 case EXT_OR_CONN_STATE_FLUSHING
: return "flushing final OKAY";
238 case EXIT_CONN_STATE_RESOLVING
: return "waiting for dest info";
239 case EXIT_CONN_STATE_CONNECTING
: return "connecting";
240 case EXIT_CONN_STATE_OPEN
: return "open";
241 case EXIT_CONN_STATE_RESOLVEFAILED
: return "resolve failed";
246 case AP_CONN_STATE_SOCKS_WAIT
: return "waiting for socks info";
247 case AP_CONN_STATE_NATD_WAIT
: return "waiting for natd dest info";
248 case AP_CONN_STATE_RENDDESC_WAIT
: return "waiting for rendezvous desc";
249 case AP_CONN_STATE_CONTROLLER_WAIT
: return "waiting for controller";
250 case AP_CONN_STATE_CIRCUIT_WAIT
: return "waiting for circuit";
251 case AP_CONN_STATE_CONNECT_WAIT
: return "waiting for connect response";
252 case AP_CONN_STATE_RESOLVE_WAIT
: return "waiting for resolve response";
253 case AP_CONN_STATE_OPEN
: return "open";
258 case DIR_CONN_STATE_CONNECTING
: return "connecting";
259 case DIR_CONN_STATE_CLIENT_SENDING
: return "client sending";
260 case DIR_CONN_STATE_CLIENT_READING
: return "client reading";
261 case DIR_CONN_STATE_CLIENT_FINISHED
: return "client finished";
262 case DIR_CONN_STATE_SERVER_COMMAND_WAIT
: return "waiting for command";
263 case DIR_CONN_STATE_SERVER_WRITING
: return "writing";
266 case CONN_TYPE_CONTROL
:
268 case CONTROL_CONN_STATE_OPEN
: return "open (protocol v1)";
269 case CONTROL_CONN_STATE_NEEDAUTH
:
270 return "waiting for authentication (protocol v1)";
275 log_warn(LD_BUG
, "unknown connection state %d (type %d)", state
, type
);
276 tor_snprintf(buf
, sizeof(buf
),
277 "unknown state [%d] on unknown [%s] connection",
278 state
, conn_type_to_string(type
));
282 /** Allocate and return a new dir_connection_t, initialized as by
283 * connection_init(). */
285 dir_connection_new(int socket_family
)
287 dir_connection_t
*dir_conn
= tor_malloc_zero(sizeof(dir_connection_t
));
288 connection_init(time(NULL
), TO_CONN(dir_conn
), CONN_TYPE_DIR
, socket_family
);
292 /** Allocate and return a new or_connection_t, initialized as by
295 * Initialize active_circuit_pqueue.
297 * Set active_circuit_pqueue_last_recalibrated to current cell_ewma tick.
300 or_connection_new(int type
, int socket_family
)
302 or_connection_t
*or_conn
= tor_malloc_zero(sizeof(or_connection_t
));
303 time_t now
= time(NULL
);
304 tor_assert(type
== CONN_TYPE_OR
|| type
== CONN_TYPE_EXT_OR
);
305 connection_init(now
, TO_CONN(or_conn
), type
, socket_family
);
307 connection_or_set_canonical(or_conn
, 0);
309 if (type
== CONN_TYPE_EXT_OR
)
310 connection_or_set_ext_or_identifier(or_conn
);
315 /** Allocate and return a new entry_connection_t, initialized as by
318 * Allocate space to store the socks_request.
321 entry_connection_new(int type
, int socket_family
)
323 entry_connection_t
*entry_conn
= tor_malloc_zero(sizeof(entry_connection_t
));
324 tor_assert(type
== CONN_TYPE_AP
);
325 connection_init(time(NULL
), ENTRY_TO_CONN(entry_conn
), type
, socket_family
);
326 entry_conn
->socks_request
= socks_request_new();
327 /* If this is coming from a listener, we'll set it up based on the listener
328 * in a little while. Otherwise, we're doing this as a linked connection
329 * of some kind, and we should set it up here based on the socket family */
330 if (socket_family
== AF_INET
)
331 entry_conn
->entry_cfg
.ipv4_traffic
= 1;
332 else if (socket_family
== AF_INET6
)
333 entry_conn
->entry_cfg
.ipv6_traffic
= 1;
334 else if (socket_family
== AF_UNIX
)
335 entry_conn
->is_socks_socket
= 1;
339 /** Allocate and return a new edge_connection_t, initialized as by
340 * connection_init(). */
342 edge_connection_new(int type
, int socket_family
)
344 edge_connection_t
*edge_conn
= tor_malloc_zero(sizeof(edge_connection_t
));
345 tor_assert(type
== CONN_TYPE_EXIT
);
346 connection_init(time(NULL
), TO_CONN(edge_conn
), type
, socket_family
);
350 /** Allocate and return a new control_connection_t, initialized as by
351 * connection_init(). */
352 control_connection_t
*
353 control_connection_new(int socket_family
)
355 control_connection_t
*control_conn
=
356 tor_malloc_zero(sizeof(control_connection_t
));
357 connection_init(time(NULL
),
358 TO_CONN(control_conn
), CONN_TYPE_CONTROL
, socket_family
);
362 /** Allocate and return a new listener_connection_t, initialized as by
363 * connection_init(). */
364 listener_connection_t
*
365 listener_connection_new(int type
, int socket_family
)
367 listener_connection_t
*listener_conn
=
368 tor_malloc_zero(sizeof(listener_connection_t
));
369 connection_init(time(NULL
), TO_CONN(listener_conn
), type
, socket_family
);
370 return listener_conn
;
373 /** Allocate, initialize, and return a new connection_t subtype of <b>type</b>
374 * to make or receive connections of address family <b>socket_family</b>. The
375 * type should be one of the CONN_TYPE_* constants. */
377 connection_new(int type
, int socket_family
)
381 case CONN_TYPE_EXT_OR
:
382 return TO_CONN(or_connection_new(type
, socket_family
));
385 return TO_CONN(edge_connection_new(type
, socket_family
));
388 return ENTRY_TO_CONN(entry_connection_new(type
, socket_family
));
391 return TO_CONN(dir_connection_new(socket_family
));
393 case CONN_TYPE_CONTROL
:
394 return TO_CONN(control_connection_new(socket_family
));
396 CASE_ANY_LISTENER_TYPE
:
397 return TO_CONN(listener_connection_new(type
, socket_family
));
400 connection_t
*conn
= tor_malloc_zero(sizeof(connection_t
));
401 connection_init(time(NULL
), conn
, type
, socket_family
);
407 /** Initializes conn. (you must call connection_add() to link it into the main
410 * Set conn-\>magic to the correct value.
412 * Set conn-\>type to <b>type</b>. Set conn-\>s and conn-\>conn_array_index to
413 * -1 to signify they are not yet assigned.
415 * Initialize conn's timestamps to now.
418 connection_init(time_t now
, connection_t
*conn
, int type
, int socket_family
)
420 static uint64_t n_connections_allocated
= 1;
424 case CONN_TYPE_EXT_OR
:
425 conn
->magic
= OR_CONNECTION_MAGIC
;
428 conn
->magic
= EDGE_CONNECTION_MAGIC
;
431 conn
->magic
= ENTRY_CONNECTION_MAGIC
;
434 conn
->magic
= DIR_CONNECTION_MAGIC
;
436 case CONN_TYPE_CONTROL
:
437 conn
->magic
= CONTROL_CONNECTION_MAGIC
;
439 CASE_ANY_LISTENER_TYPE
:
440 conn
->magic
= LISTENER_CONNECTION_MAGIC
;
443 conn
->magic
= BASE_CONNECTION_MAGIC
;
447 conn
->s
= TOR_INVALID_SOCKET
; /* give it a default of 'not used' */
448 conn
->conn_array_index
= -1; /* also default to 'not used' */
449 conn
->global_identifier
= n_connections_allocated
++;
452 conn
->socket_family
= socket_family
;
453 if (!connection_is_listener(conn
)) {
454 /* listeners never use their buf */
455 conn
->inbuf
= buf_new();
456 conn
->outbuf
= buf_new();
459 conn
->timestamp_created
= now
;
460 conn
->timestamp_lastread
= now
;
461 conn
->timestamp_lastwritten
= now
;
464 /** Create a link between <b>conn_a</b> and <b>conn_b</b>. */
466 connection_link_connections(connection_t
*conn_a
, connection_t
*conn_b
)
468 tor_assert(! SOCKET_OK(conn_a
->s
));
469 tor_assert(! SOCKET_OK(conn_b
->s
));
473 conn_a
->linked_conn
= conn_b
;
474 conn_b
->linked_conn
= conn_a
;
477 /** Return true iff the provided connection listener type supports AF_UNIX
480 conn_listener_type_supports_af_unix(int type
)
482 /* For now only control ports or SOCKS ports can be Unix domain sockets
483 * and listeners at the same time */
485 case CONN_TYPE_CONTROL_LISTENER
:
486 case CONN_TYPE_AP_LISTENER
:
493 /** Deallocate memory used by <b>conn</b>. Deallocate its buffers if
494 * necessary, close its socket if necessary, and mark the directory as dirty
495 * if <b>conn</b> is an OR or OP connection.
498 connection_free_(connection_t
*conn
)
505 switch (conn
->type
) {
507 case CONN_TYPE_EXT_OR
:
508 tor_assert(conn
->magic
== OR_CONNECTION_MAGIC
);
509 mem
= TO_OR_CONN(conn
);
510 memlen
= sizeof(or_connection_t
);
513 tor_assert(conn
->magic
== ENTRY_CONNECTION_MAGIC
);
514 mem
= TO_ENTRY_CONN(conn
);
515 memlen
= sizeof(entry_connection_t
);
518 tor_assert(conn
->magic
== EDGE_CONNECTION_MAGIC
);
519 mem
= TO_EDGE_CONN(conn
);
520 memlen
= sizeof(edge_connection_t
);
523 tor_assert(conn
->magic
== DIR_CONNECTION_MAGIC
);
524 mem
= TO_DIR_CONN(conn
);
525 memlen
= sizeof(dir_connection_t
);
527 case CONN_TYPE_CONTROL
:
528 tor_assert(conn
->magic
== CONTROL_CONNECTION_MAGIC
);
529 mem
= TO_CONTROL_CONN(conn
);
530 memlen
= sizeof(control_connection_t
);
532 CASE_ANY_LISTENER_TYPE
:
533 tor_assert(conn
->magic
== LISTENER_CONNECTION_MAGIC
);
534 mem
= TO_LISTENER_CONN(conn
);
535 memlen
= sizeof(listener_connection_t
);
538 tor_assert(conn
->magic
== BASE_CONNECTION_MAGIC
);
540 memlen
= sizeof(connection_t
);
545 log_info(LD_GENERAL
, "Freeing linked %s connection [%s] with %d "
546 "bytes on inbuf, %d on outbuf.",
547 conn_type_to_string(conn
->type
),
548 conn_state_to_string(conn
->type
, conn
->state
),
549 (int)connection_get_inbuf_len(conn
),
550 (int)connection_get_outbuf_len(conn
));
553 if (!connection_is_listener(conn
)) {
554 buf_free(conn
->inbuf
);
555 buf_free(conn
->outbuf
);
557 if (conn
->socket_family
== AF_UNIX
) {
558 /* For now only control and SOCKS ports can be Unix domain sockets
559 * and listeners at the same time */
560 tor_assert(conn_listener_type_supports_af_unix(conn
->type
));
562 if (unlink(conn
->address
) < 0 && errno
!= ENOENT
) {
563 log_warn(LD_NET
, "Could not unlink %s: %s", conn
->address
,
569 tor_free(conn
->address
);
571 if (connection_speaks_cells(conn
)) {
572 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
573 tor_tls_free(or_conn
->tls
);
575 or_handshake_state_free(or_conn
->handshake_state
);
576 or_conn
->handshake_state
= NULL
;
577 tor_free(or_conn
->nickname
);
579 /* Owww, this shouldn't happen, but... */
581 "Freeing orconn at %p, saw channel %p with ID "
582 U64_FORMAT
" left un-NULLed",
583 or_conn
, TLS_CHAN_TO_BASE(or_conn
->chan
),
585 TLS_CHAN_TO_BASE(or_conn
->chan
)->global_identifier
));
586 if (!CHANNEL_FINISHED(TLS_CHAN_TO_BASE(or_conn
->chan
))) {
587 channel_close_for_error(TLS_CHAN_TO_BASE(or_conn
->chan
));
590 or_conn
->chan
->conn
= NULL
;
591 or_conn
->chan
= NULL
;
594 if (conn
->type
== CONN_TYPE_AP
) {
595 entry_connection_t
*entry_conn
= TO_ENTRY_CONN(conn
);
596 tor_free(entry_conn
->chosen_exit_name
);
597 tor_free(entry_conn
->original_dest_address
);
598 if (entry_conn
->socks_request
)
599 socks_request_free(entry_conn
->socks_request
);
600 if (entry_conn
->pending_optimistic_data
) {
601 buf_free(entry_conn
->pending_optimistic_data
);
603 if (entry_conn
->sending_optimistic_data
) {
604 buf_free(entry_conn
->sending_optimistic_data
);
607 if (CONN_IS_EDGE(conn
)) {
608 rend_data_free(TO_EDGE_CONN(conn
)->rend_data
);
609 hs_ident_edge_conn_free(TO_EDGE_CONN(conn
)->hs_ident
);
611 if (conn
->type
== CONN_TYPE_CONTROL
) {
612 control_connection_t
*control_conn
= TO_CONTROL_CONN(conn
);
613 tor_free(control_conn
->safecookie_client_hash
);
614 tor_free(control_conn
->incoming_cmd
);
615 if (control_conn
->ephemeral_onion_services
) {
616 SMARTLIST_FOREACH(control_conn
->ephemeral_onion_services
, char *, cp
, {
617 memwipe(cp
, 0, strlen(cp
));
620 smartlist_free(control_conn
->ephemeral_onion_services
);
624 /* Probably already freed by connection_free. */
625 tor_event_free(conn
->read_event
);
626 tor_event_free(conn
->write_event
);
627 conn
->read_event
= conn
->write_event
= NULL
;
629 if (conn
->type
== CONN_TYPE_DIR
) {
630 dir_connection_t
*dir_conn
= TO_DIR_CONN(conn
);
631 tor_free(dir_conn
->requested_resource
);
633 tor_compress_free(dir_conn
->compress_state
);
634 if (dir_conn
->spool
) {
635 SMARTLIST_FOREACH(dir_conn
->spool
, spooled_resource_t
*, spooled
,
636 spooled_resource_free(spooled
));
637 smartlist_free(dir_conn
->spool
);
640 rend_data_free(dir_conn
->rend_data
);
641 hs_ident_dir_conn_free(dir_conn
->hs_ident
);
642 if (dir_conn
->guard_state
) {
643 /* Cancel before freeing, if it's still there. */
644 entry_guard_cancel(&dir_conn
->guard_state
);
646 circuit_guard_state_free(dir_conn
->guard_state
);
649 if (SOCKET_OK(conn
->s
)) {
650 log_debug(LD_NET
,"closing fd %d.",(int)conn
->s
);
651 tor_close_socket(conn
->s
);
652 conn
->s
= TOR_INVALID_SOCKET
;
655 if (conn
->type
== CONN_TYPE_OR
&&
656 !tor_digest_is_zero(TO_OR_CONN(conn
)->identity_digest
)) {
657 log_warn(LD_BUG
, "called on OR conn with non-zeroed identity_digest");
658 connection_or_clear_identity(TO_OR_CONN(conn
));
660 if (conn
->type
== CONN_TYPE_OR
|| conn
->type
== CONN_TYPE_EXT_OR
) {
661 connection_or_remove_from_ext_or_id_map(TO_OR_CONN(conn
));
662 tor_free(TO_OR_CONN(conn
)->ext_or_conn_id
);
663 tor_free(TO_OR_CONN(conn
)->ext_or_auth_correct_client_hash
);
664 tor_free(TO_OR_CONN(conn
)->ext_or_transport
);
667 memwipe(mem
, 0xCC, memlen
); /* poison memory */
671 /** Make sure <b>conn</b> isn't in any of the global conn lists; then free it.
674 connection_free
,(connection_t
*conn
))
678 tor_assert(!connection_is_on_closeable_list(conn
));
679 tor_assert(!connection_in_array(conn
));
680 if (BUG(conn
->linked_conn
)) {
681 conn
->linked_conn
->linked_conn
= NULL
;
682 if (! conn
->linked_conn
->marked_for_close
&&
683 conn
->linked_conn
->reading_from_linked_conn
)
684 connection_start_reading(conn
->linked_conn
);
685 conn
->linked_conn
= NULL
;
687 if (connection_speaks_cells(conn
)) {
688 if (!tor_digest_is_zero(TO_OR_CONN(conn
)->identity_digest
)) {
689 connection_or_clear_identity(TO_OR_CONN(conn
));
692 if (conn
->type
== CONN_TYPE_CONTROL
) {
693 connection_control_closed(TO_CONTROL_CONN(conn
));
697 if (conn
->type
== CONN_TYPE_AP
) {
698 connection_ap_warn_and_unmark_if_pending_circ(TO_ENTRY_CONN(conn
),
702 connection_unregister_events(conn
);
703 connection_free_(conn
);
707 * Called when we're about to finally unlink and free a connection:
708 * perform necessary accounting and cleanup
709 * - Directory conns that failed to fetch a rendezvous descriptor
710 * need to inform pending rendezvous streams.
711 * - OR conns need to call rep_hist_note_*() to record status.
712 * - AP conns need to send a socks reject if necessary.
713 * - Exit conns need to call connection_dns_remove() if necessary.
714 * - AP and Exit conns need to send an end cell if they can.
715 * - DNS conns need to fail any resolves that are pending on them.
716 * - OR and edge connections need to be unlinked from circuits.
719 connection_about_to_close_connection(connection_t
*conn
)
721 tor_assert(conn
->marked_for_close
);
723 switch (conn
->type
) {
725 connection_dir_about_to_close(TO_DIR_CONN(conn
));
728 case CONN_TYPE_EXT_OR
:
729 connection_or_about_to_close(TO_OR_CONN(conn
));
732 connection_ap_about_to_close(TO_ENTRY_CONN(conn
));
735 connection_exit_about_to_close(TO_EDGE_CONN(conn
));
740 /** Return true iff connection_close_immediate() has been called on this
742 #define CONN_IS_CLOSED(c) \
743 ((c)->linked ? ((c)->linked_conn_is_closed) : (! SOCKET_OK(c->s)))
745 /** Close the underlying socket for <b>conn</b>, so we don't try to
746 * flush it. Must be used in conjunction with (right before)
747 * connection_mark_for_close().
750 connection_close_immediate(connection_t
*conn
)
752 assert_connection_ok(conn
,0);
753 if (CONN_IS_CLOSED(conn
)) {
754 log_err(LD_BUG
,"Attempt to close already-closed connection.");
755 tor_fragile_assert();
758 if (conn
->outbuf_flushlen
) {
759 log_info(LD_NET
,"fd %d, type %s, state %s, %d bytes on outbuf.",
760 (int)conn
->s
, conn_type_to_string(conn
->type
),
761 conn_state_to_string(conn
->type
, conn
->state
),
762 (int)conn
->outbuf_flushlen
);
765 connection_unregister_events(conn
);
767 if (SOCKET_OK(conn
->s
))
768 tor_close_socket(conn
->s
);
769 conn
->s
= TOR_INVALID_SOCKET
;
771 conn
->linked_conn_is_closed
= 1;
773 buf_clear(conn
->outbuf
);
774 conn
->outbuf_flushlen
= 0;
777 /** Mark <b>conn</b> to be closed next time we loop through
778 * conn_close_if_marked() in main.c. */
780 connection_mark_for_close_(connection_t
*conn
, int line
, const char *file
)
782 assert_connection_ok(conn
,0);
784 tor_assert(line
< 1<<16); /* marked_for_close can only fit a uint16_t. */
787 if (conn
->type
== CONN_TYPE_OR
) {
789 * An or_connection should have been closed through one of the channel-
790 * aware functions in connection_or.c. We'll assume this is an error
791 * close and do that, and log a bug warning.
793 log_warn(LD_CHANNEL
| LD_BUG
,
794 "Something tried to close an or_connection_t without going "
795 "through channels at %s:%d",
797 connection_or_close_for_error(TO_OR_CONN(conn
), 0);
799 /* Pass it down to the real function */
800 connection_mark_for_close_internal_(conn
, line
, file
);
804 /** Mark <b>conn</b> to be closed next time we loop through
805 * conn_close_if_marked() in main.c; the _internal version bypasses the
806 * CONN_TYPE_OR checks; this should be called when you either are sure that
807 * if this is an or_connection_t the controlling channel has been notified
808 * (e.g. with connection_or_notify_error()), or you actually are the
809 * connection_or_close_for_error() or connection_or_close_normally function.
810 * For all other cases, use connection_mark_and_flush() instead, which
811 * checks for or_connection_t properly, instead. See below.
814 connection_mark_for_close_internal_
, (connection_t
*conn
,
815 int line
, const char *file
))
817 assert_connection_ok(conn
,0);
819 tor_assert(line
< 1<<16); /* marked_for_close can only fit a uint16_t. */
822 if (conn
->marked_for_close
) {
823 log_warn(LD_BUG
,"Duplicate call to connection_mark_for_close at %s:%d"
824 " (first at %s:%d)", file
, line
, conn
->marked_for_close_file
,
825 conn
->marked_for_close
);
826 tor_fragile_assert();
830 if (conn
->type
== CONN_TYPE_OR
) {
832 * Bad news if this happens without telling the controlling channel; do
833 * this so we can find things that call this wrongly when the asserts hit.
835 log_debug(LD_CHANNEL
,
836 "Calling connection_mark_for_close_internal_() on an OR conn "
841 conn
->marked_for_close
= line
;
842 conn
->marked_for_close_file
= file
;
843 add_connection_to_closeable_list(conn
);
845 /* in case we're going to be held-open-til-flushed, reset
846 * the number of seconds since last successful write, so
847 * we get our whole 15 seconds */
848 conn
->timestamp_lastwritten
= time(NULL
);
851 /** Find each connection that has hold_open_until_flushed set to
852 * 1 but hasn't written in the past 15 seconds, and set
853 * hold_open_until_flushed to 0. This means it will get cleaned
854 * up in the next loop through close_if_marked() in main.c.
857 connection_expire_held_open(void)
860 smartlist_t
*conns
= get_connection_array();
864 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, conn
) {
865 /* If we've been holding the connection open, but we haven't written
868 if (conn
->hold_open_until_flushed
) {
869 tor_assert(conn
->marked_for_close
);
870 if (now
- conn
->timestamp_lastwritten
>= 15) {
872 if (conn
->type
== CONN_TYPE_EXIT
||
873 (conn
->type
== CONN_TYPE_DIR
&&
874 conn
->purpose
== DIR_PURPOSE_SERVER
))
877 severity
= LOG_NOTICE
;
878 log_fn(severity
, LD_NET
,
879 "Giving up on marked_for_close conn that's been flushing "
880 "for 15s (fd %d, type %s, state %s).",
881 (int)conn
->s
, conn_type_to_string(conn
->type
),
882 conn_state_to_string(conn
->type
, conn
->state
));
883 conn
->hold_open_until_flushed
= 0;
886 } SMARTLIST_FOREACH_END(conn
);
889 #if defined(HAVE_SYS_UN_H) || defined(RUNNING_DOXYGEN)
890 /** Create an AF_UNIX listenaddr struct.
891 * <b>listenaddress</b> provides the path to the Unix socket.
893 * Eventually <b>listenaddress</b> will also optionally contain user, group,
894 * and file permissions for the new socket. But not yet. XXX
895 * Also, since we do not create the socket here the information doesn't help
898 * If not NULL <b>readable_address</b> will contain a copy of the path part of
899 * <b>listenaddress</b>.
901 * The listenaddr struct has to be freed by the caller.
903 static struct sockaddr_un
*
904 create_unix_sockaddr(const char *listenaddress
, char **readable_address
,
907 struct sockaddr_un
*sockaddr
= NULL
;
909 sockaddr
= tor_malloc_zero(sizeof(struct sockaddr_un
));
910 sockaddr
->sun_family
= AF_UNIX
;
911 if (strlcpy(sockaddr
->sun_path
, listenaddress
, sizeof(sockaddr
->sun_path
))
912 >= sizeof(sockaddr
->sun_path
)) {
913 log_warn(LD_CONFIG
, "Unix socket path '%s' is too long to fit.",
914 escaped(listenaddress
));
919 if (readable_address
)
920 *readable_address
= tor_strdup(listenaddress
);
922 *len_out
= sizeof(struct sockaddr_un
);
926 static struct sockaddr
*
927 create_unix_sockaddr(const char *listenaddress
, char **readable_address
,
931 (void)readable_address
;
932 log_fn(LOG_ERR
, LD_BUG
,
933 "Unix domain sockets not supported, yet we tried to create one.");
935 tor_fragile_assert();
938 #endif /* HAVE_SYS_UN_H */
940 /** Warn that an accept or a connect has failed because we're running out of
941 * TCP sockets we can use on current system. Rate-limit these warnings so
942 * that we don't spam the log. */
944 warn_too_many_conns(void)
946 #define WARN_TOO_MANY_CONNS_INTERVAL (6*60*60)
947 static ratelim_t last_warned
= RATELIM_INIT(WARN_TOO_MANY_CONNS_INTERVAL
);
949 if ((m
= rate_limit_log(&last_warned
, approx_time()))) {
950 int n_conns
= get_n_open_sockets();
951 log_warn(LD_NET
,"Failing because we have %d connections already. Please "
952 "read doc/TUNING for guidance.%s", n_conns
, m
);
954 control_event_general_status(LOG_WARN
, "TOO_MANY_CONNECTIONS CURRENT=%d",
961 #define UNIX_SOCKET_PURPOSE_CONTROL_SOCKET 0
962 #define UNIX_SOCKET_PURPOSE_SOCKS_SOCKET 1
964 /** Check if the purpose isn't one of the ones we know what to do with */
967 is_valid_unix_socket_purpose(int purpose
)
972 case UNIX_SOCKET_PURPOSE_CONTROL_SOCKET
:
973 case UNIX_SOCKET_PURPOSE_SOCKS_SOCKET
:
981 /** Return a string description of a unix socket purpose */
983 unix_socket_purpose_to_string(int purpose
)
985 const char *s
= "unknown-purpose socket";
988 case UNIX_SOCKET_PURPOSE_CONTROL_SOCKET
:
989 s
= "control socket";
991 case UNIX_SOCKET_PURPOSE_SOCKS_SOCKET
:
999 /** Check whether we should be willing to open an AF_UNIX socket in
1000 * <b>path</b>. Return 0 if we should go ahead and -1 if we shouldn't. */
1002 check_location_for_unix_socket(const or_options_t
*options
, const char *path
,
1003 int purpose
, const port_cfg_t
*port
)
1008 tor_assert(is_valid_unix_socket_purpose(purpose
));
1010 p
= tor_strdup(path
);
1011 cpd_check_t flags
= CPD_CHECK_MODE_ONLY
;
1012 if (get_parent_directory(p
)<0 || p
[0] != '/') {
1013 log_warn(LD_GENERAL
, "Bad unix socket address '%s'. Tor does not support "
1014 "relative paths for unix sockets.", path
);
1018 if (port
->is_world_writable
) {
1019 /* World-writable sockets can go anywhere. */
1024 if (port
->is_group_writable
) {
1025 flags
|= CPD_GROUP_OK
;
1028 if (port
->relax_dirmode_check
) {
1029 flags
|= CPD_RELAX_DIRMODE_CHECK
;
1032 if (check_private_dir(p
, flags
, options
->User
) < 0) {
1033 char *escpath
, *escdir
;
1034 escpath
= esc_for_log(path
);
1035 escdir
= esc_for_log(p
);
1036 log_warn(LD_GENERAL
, "Before Tor can create a %s in %s, the directory "
1037 "%s needs to exist, and to be accessible only by the user%s "
1038 "account that is running Tor. (On some Unix systems, anybody "
1039 "who can list a socket can connect to it, so Tor is being "
1041 unix_socket_purpose_to_string(purpose
), escpath
, escdir
,
1042 port
->is_group_writable
? " and group" : "");
1055 /** Tell the TCP stack that it shouldn't wait for a long time after
1056 * <b>sock</b> has closed before reusing its port. Return 0 on success,
1059 make_socket_reuseable(tor_socket_t sock
)
1067 /* REUSEADDR on normal places means you can rebind to the port
1068 * right after somebody else has let it go. But REUSEADDR on win32
1069 * means you can bind to the port _even when somebody else
1070 * already has it bound_. So, don't do that on Win32. */
1071 if (setsockopt(sock
, SOL_SOCKET
, SO_REUSEADDR
, (void*) &one
,
1072 (socklen_t
)sizeof(one
)) == -1) {
1080 /** Tell the Windows TCP stack to prevent other applications from receiving
1081 * traffic from tor's open ports. Return 0 on success, -1 on failure. */
1083 make_win32_socket_exclusive(tor_socket_t sock
)
1085 #ifdef SO_EXCLUSIVEADDRUSE
1088 /* Any socket that sets REUSEADDR on win32 can bind to a port _even when
1089 * somebody else already has it bound_, and _even if the original socket
1090 * didn't set REUSEADDR_. Use EXCLUSIVEADDRUSE to prevent this port-stealing
1092 if (setsockopt(sock
, SOL_SOCKET
, SO_EXCLUSIVEADDRUSE
, (void*) &one
,
1093 (socklen_t
)sizeof(one
))) {
1104 /** Max backlog to pass to listen. We start at */
1105 static int listen_limit
= INT_MAX
;
1107 /* Listen on <b>fd</b> with appropriate backlog. Return as for listen. */
1109 tor_listen(tor_socket_t fd
)
1113 if ((r
= listen(fd
, listen_limit
)) < 0) {
1114 if (listen_limit
== SOMAXCONN
)
1116 if ((r
= listen(fd
, SOMAXCONN
)) == 0) {
1117 listen_limit
= SOMAXCONN
;
1118 log_warn(LD_NET
, "Setting listen backlog to INT_MAX connections "
1119 "didn't work, but SOMAXCONN did. Lowering backlog limit.");
1125 /** Bind a new non-blocking socket listening to the socket described
1126 * by <b>listensockaddr</b>.
1128 * <b>address</b> is only used for logging purposes and to add the information
1131 static connection_t
*
1132 connection_listener_new(const struct sockaddr
*listensockaddr
,
1134 int type
, const char *address
,
1135 const port_cfg_t
*port_cfg
)
1137 listener_connection_t
*lis_conn
;
1138 connection_t
*conn
= NULL
;
1139 tor_socket_t s
= TOR_INVALID_SOCKET
; /* the socket we're going to make */
1140 or_options_t
const *options
= get_options();
1141 (void) options
; /* Windows doesn't use this. */
1142 #if defined(HAVE_PWD_H) && defined(HAVE_SYS_UN_H)
1143 const struct passwd
*pw
= NULL
;
1145 uint16_t usePort
= 0, gotPort
= 0;
1146 int start_reading
= 0;
1147 static int global_next_session_group
= SESSION_GROUP_FIRST_AUTO
;
1151 if (listensockaddr
->sa_family
== AF_INET
||
1152 listensockaddr
->sa_family
== AF_INET6
) {
1153 int is_stream
= (type
!= CONN_TYPE_AP_DNS_LISTENER
);
1157 tor_addr_from_sockaddr(&addr
, listensockaddr
, &usePort
);
1158 log_notice(LD_NET
, "Opening %s on %s",
1159 conn_type_to_string(type
), fmt_addrport(&addr
, usePort
));
1161 s
= tor_open_socket_nonblocking(tor_addr_family(&addr
),
1162 is_stream
? SOCK_STREAM
: SOCK_DGRAM
,
1163 is_stream
? IPPROTO_TCP
: IPPROTO_UDP
);
1164 if (!SOCKET_OK(s
)) {
1165 int e
= tor_socket_errno(s
);
1166 if (ERRNO_IS_RESOURCE_LIMIT(e
)) {
1167 warn_too_many_conns();
1169 * We'll call the OOS handler at the error exit, so set the
1170 * exhaustion flag for it.
1174 log_warn(LD_NET
, "Socket creation failed: %s",
1175 tor_socket_strerror(e
));
1180 if (make_socket_reuseable(s
) < 0) {
1181 log_warn(LD_NET
, "Error setting SO_REUSEADDR flag on %s: %s",
1182 conn_type_to_string(type
),
1183 tor_socket_strerror(errno
));
1187 if (make_win32_socket_exclusive(s
) < 0) {
1188 log_warn(LD_NET
, "Error setting SO_EXCLUSIVEADDRUSE flag on %s: %s",
1189 conn_type_to_string(type
),
1190 tor_socket_strerror(errno
));
1194 #if defined(USE_TRANSPARENT) && defined(IP_TRANSPARENT)
1195 if (options
->TransProxyType_parsed
== TPT_TPROXY
&&
1196 type
== CONN_TYPE_AP_TRANS_LISTENER
) {
1198 if (setsockopt(s
, SOL_IP
, IP_TRANSPARENT
, (void*)&one
,
1199 (socklen_t
)sizeof(one
)) < 0) {
1200 const char *extra
= "";
1201 int e
= tor_socket_errno(s
);
1203 extra
= "TransTPROXY requires root privileges or similar"
1205 log_warn(LD_NET
, "Error setting IP_TRANSPARENT flag: %s.%s",
1206 tor_socket_strerror(e
), extra
);
1212 if (listensockaddr
->sa_family
== AF_INET6
) {
1214 /* We need to set IPV6_V6ONLY so that this socket can't get used for
1215 * IPv4 connections. */
1216 if (setsockopt(s
,IPPROTO_IPV6
, IPV6_V6ONLY
,
1217 (void*)&one
, (socklen_t
)sizeof(one
)) < 0) {
1218 int e
= tor_socket_errno(s
);
1219 log_warn(LD_NET
, "Error setting IPV6_V6ONLY flag: %s",
1220 tor_socket_strerror(e
));
1221 /* Keep going; probably not harmful. */
1226 if (bind(s
,listensockaddr
,socklen
) < 0) {
1227 const char *helpfulhint
= "";
1228 int e
= tor_socket_errno(s
);
1229 if (ERRNO_IS_EADDRINUSE(e
))
1230 helpfulhint
= ". Is Tor already running?";
1231 log_warn(LD_NET
, "Could not bind to %s:%u: %s%s", address
, usePort
,
1232 tor_socket_strerror(e
), helpfulhint
);
1237 if (tor_listen(s
) < 0) {
1238 log_warn(LD_NET
, "Could not listen on %s:%u: %s", address
, usePort
,
1239 tor_socket_strerror(tor_socket_errno(s
)));
1248 struct sockaddr_storage ss
;
1249 socklen_t ss_len
=sizeof(ss
);
1250 if (getsockname(s
, (struct sockaddr
*)&ss
, &ss_len
)<0) {
1251 log_warn(LD_NET
, "getsockname() couldn't learn address for %s: %s",
1252 conn_type_to_string(type
),
1253 tor_socket_strerror(tor_socket_errno(s
)));
1256 tor_addr_from_sockaddr(&addr2
, (struct sockaddr
*)&ss
, &gotPort
);
1258 #ifdef HAVE_SYS_UN_H
1260 * AF_UNIX generic setup stuff
1262 } else if (listensockaddr
->sa_family
== AF_UNIX
) {
1263 /* We want to start reading for both AF_UNIX cases */
1266 tor_assert(conn_listener_type_supports_af_unix(type
));
1268 if (check_location_for_unix_socket(options
, address
,
1269 (type
== CONN_TYPE_CONTROL_LISTENER
) ?
1270 UNIX_SOCKET_PURPOSE_CONTROL_SOCKET
:
1271 UNIX_SOCKET_PURPOSE_SOCKS_SOCKET
, port_cfg
) < 0) {
1275 log_notice(LD_NET
, "Opening %s on %s",
1276 conn_type_to_string(type
), address
);
1278 tor_addr_make_unspec(&addr
);
1280 if (unlink(address
) < 0 && errno
!= ENOENT
) {
1281 log_warn(LD_NET
, "Could not unlink %s: %s", address
,
1286 s
= tor_open_socket_nonblocking(AF_UNIX
, SOCK_STREAM
, 0);
1287 if (! SOCKET_OK(s
)) {
1288 int e
= tor_socket_errno(s
);
1289 if (ERRNO_IS_RESOURCE_LIMIT(e
)) {
1290 warn_too_many_conns();
1292 * We'll call the OOS handler at the error exit, so set the
1293 * exhaustion flag for it.
1297 log_warn(LD_NET
,"Socket creation failed: %s.", strerror(e
));
1302 if (bind(s
, listensockaddr
,
1303 (socklen_t
)sizeof(struct sockaddr_un
)) == -1) {
1304 log_warn(LD_NET
,"Bind to %s failed: %s.", address
,
1305 tor_socket_strerror(tor_socket_errno(s
)));
1310 if (options
->User
) {
1311 pw
= tor_getpwnam(options
->User
);
1314 log_warn(LD_NET
,"Unable to chown() %s socket: user %s not found.",
1315 address
, options
->User
);
1317 } else if (fstat(s
, &st
) == 0 &&
1318 st
.st_uid
== pw
->pw_uid
&& st
.st_gid
== pw
->pw_gid
) {
1319 /* No change needed */
1320 } else if (chown(sandbox_intern_string(address
),
1321 pw
->pw_uid
, pw
->pw_gid
) < 0) {
1322 log_warn(LD_NET
,"Unable to chown() %s socket: %s.",
1323 address
, strerror(errno
));
1333 if (port_cfg
->is_world_writable
) {
1335 status
= "world-writable";
1336 } else if (port_cfg
->is_group_writable
) {
1338 status
= "group-writable";
1343 /* We need to use chmod; fchmod doesn't work on sockets on all
1345 if (fstat(s
, &st
) == 0 && (st
.st_mode
& 0777) == mode
) {
1346 /* no change needed */
1347 } else if (chmod(sandbox_intern_string(address
), mode
) < 0) {
1348 log_warn(LD_FS
,"Unable to make %s %s.", address
, status
);
1353 if (listen(s
, SOMAXCONN
) < 0) {
1354 log_warn(LD_NET
, "Could not listen on %s: %s", address
,
1355 tor_socket_strerror(tor_socket_errno(s
)));
1358 #endif /* HAVE_SYS_UN_H */
1360 log_err(LD_BUG
, "Got unexpected address family %d.",
1361 listensockaddr
->sa_family
);
1365 lis_conn
= listener_connection_new(type
, listensockaddr
->sa_family
);
1366 conn
= TO_CONN(lis_conn
);
1367 conn
->socket_family
= listensockaddr
->sa_family
;
1369 s
= TOR_INVALID_SOCKET
; /* Prevent double-close */
1370 conn
->address
= tor_strdup(address
);
1371 conn
->port
= gotPort
;
1372 tor_addr_copy(&conn
->addr
, &addr
);
1374 memcpy(&lis_conn
->entry_cfg
, &port_cfg
->entry_cfg
, sizeof(entry_port_cfg_t
));
1376 if (port_cfg
->entry_cfg
.isolation_flags
) {
1377 lis_conn
->entry_cfg
.isolation_flags
= port_cfg
->entry_cfg
.isolation_flags
;
1378 if (port_cfg
->entry_cfg
.session_group
>= 0) {
1379 lis_conn
->entry_cfg
.session_group
= port_cfg
->entry_cfg
.session_group
;
1381 /* This can wrap after around INT_MAX listeners are opened. But I don't
1382 * believe that matters, since you would need to open a ridiculous
1383 * number of listeners while keeping the early ones open before you ever
1384 * hit this. An OR with a dozen ports open, for example, would have to
1385 * close and re-open its listeners every second for 4 years nonstop.
1387 lis_conn
->entry_cfg
.session_group
= global_next_session_group
--;
1391 if (type
!= CONN_TYPE_AP_LISTENER
) {
1392 lis_conn
->entry_cfg
.ipv4_traffic
= 1;
1393 lis_conn
->entry_cfg
.ipv6_traffic
= 1;
1394 lis_conn
->entry_cfg
.prefer_ipv6
= 0;
1397 if (connection_add(conn
) < 0) { /* no space, forget it */
1398 log_warn(LD_NET
,"connection_add for listener failed. Giving up.");
1402 log_fn(usePort
==gotPort
? LOG_DEBUG
: LOG_NOTICE
, LD_NET
,
1403 "%s listening on port %u.",
1404 conn_type_to_string(type
), gotPort
);
1406 conn
->state
= LISTENER_STATE_READY
;
1407 if (start_reading
) {
1408 connection_start_reading(conn
);
1410 tor_assert(type
== CONN_TYPE_AP_DNS_LISTENER
);
1411 dnsserv_configure_listener(conn
);
1415 * Normal exit; call the OOS handler since connection count just changed;
1416 * the exhaustion flag will always be zero here though.
1418 connection_check_oos(get_n_open_sockets(), 0);
1424 tor_close_socket(s
);
1426 connection_free(conn
);
1428 /* Call the OOS handler, indicate if we saw an exhaustion-related error */
1429 connection_check_oos(get_n_open_sockets(), exhaustion
);
1434 /** Do basic sanity checking on a newly received socket. Return 0
1435 * if it looks ok, else return -1.
1437 * Notably, some TCP stacks can erroneously have accept() return successfully
1438 * with socklen 0, when the client sends an RST before the accept call (as
1439 * nmap does). We want to detect that, and not go on with the connection.
1442 check_sockaddr(const struct sockaddr
*sa
, int len
, int level
)
1446 if (sa
->sa_family
== AF_INET
) {
1447 struct sockaddr_in
*sin
=(struct sockaddr_in
*)sa
;
1448 if (len
!= sizeof(struct sockaddr_in
)) {
1449 log_fn(level
, LD_NET
, "Length of address not as expected: %d vs %d",
1450 len
,(int)sizeof(struct sockaddr_in
));
1453 if (sin
->sin_addr
.s_addr
== 0 || sin
->sin_port
== 0) {
1454 log_fn(level
, LD_NET
,
1455 "Address for new connection has address/port equal to zero.");
1458 } else if (sa
->sa_family
== AF_INET6
) {
1459 struct sockaddr_in6
*sin6
=(struct sockaddr_in6
*)sa
;
1460 if (len
!= sizeof(struct sockaddr_in6
)) {
1461 log_fn(level
, LD_NET
, "Length of address not as expected: %d vs %d",
1462 len
,(int)sizeof(struct sockaddr_in6
));
1465 if (tor_mem_is_zero((void*)sin6
->sin6_addr
.s6_addr
, 16) ||
1466 sin6
->sin6_port
== 0) {
1467 log_fn(level
, LD_NET
,
1468 "Address for new connection has address/port equal to zero.");
1471 } else if (sa
->sa_family
== AF_UNIX
) {
1479 /** Check whether the socket family from an accepted socket <b>got</b> is the
1480 * same as the one that <b>listener</b> is waiting for. If it isn't, log
1481 * a useful message and return -1. Else return 0.
1483 * This is annoying, but can apparently happen on some Darwins. */
1485 check_sockaddr_family_match(sa_family_t got
, connection_t
*listener
)
1487 if (got
!= listener
->socket_family
) {
1488 log_info(LD_BUG
, "A listener connection returned a socket with a "
1489 "mismatched family. %s for addr_family %d gave us a socket "
1490 "with address family %d. Dropping.",
1491 conn_type_to_string(listener
->type
),
1492 (int)listener
->socket_family
,
1499 /** The listener connection <b>conn</b> told poll() it wanted to read.
1500 * Call accept() on conn-\>s, and add the new connection if necessary.
1503 connection_handle_listener_read(connection_t
*conn
, int new_type
)
1505 tor_socket_t news
; /* the new socket */
1506 connection_t
*newconn
= 0;
1507 /* information about the remote peer when connecting to other routers */
1508 struct sockaddr_storage addrbuf
;
1509 struct sockaddr
*remote
= (struct sockaddr
*)&addrbuf
;
1510 /* length of the remote address. Must be whatever accept() needs. */
1511 socklen_t remotelen
= (socklen_t
)sizeof(addrbuf
);
1512 const or_options_t
*options
= get_options();
1514 tor_assert((size_t)remotelen
>= sizeof(struct sockaddr_in
));
1515 memset(&addrbuf
, 0, sizeof(addrbuf
));
1517 news
= tor_accept_socket_nonblocking(conn
->s
,remote
,&remotelen
);
1518 if (!SOCKET_OK(news
)) { /* accept() error */
1519 int e
= tor_socket_errno(conn
->s
);
1520 if (ERRNO_IS_ACCEPT_EAGAIN(e
)) {
1522 * they hung up before we could accept(). that's fine.
1524 * give the OOS handler a chance to run though
1526 connection_check_oos(get_n_open_sockets(), 0);
1528 } else if (ERRNO_IS_RESOURCE_LIMIT(e
)) {
1529 warn_too_many_conns();
1530 /* Exhaustion; tell the OOS handler */
1531 connection_check_oos(get_n_open_sockets(), 1);
1534 /* else there was a real error. */
1535 log_warn(LD_NET
,"accept() failed: %s. Closing listener.",
1536 tor_socket_strerror(e
));
1537 connection_mark_for_close(conn
);
1538 /* Tell the OOS handler about this too */
1539 connection_check_oos(get_n_open_sockets(), 0);
1543 "Connection accepted on socket %d (child of fd %d).",
1544 (int)news
,(int)conn
->s
);
1546 /* We accepted a new conn; run OOS handler */
1547 connection_check_oos(get_n_open_sockets(), 0);
1549 if (make_socket_reuseable(news
) < 0) {
1550 if (tor_socket_errno(news
) == EINVAL
) {
1551 /* This can happen on OSX if we get a badly timed shutdown. */
1552 log_debug(LD_NET
, "make_socket_reuseable returned EINVAL");
1554 log_warn(LD_NET
, "Error setting SO_REUSEADDR flag on %s: %s",
1555 conn_type_to_string(new_type
),
1556 tor_socket_strerror(errno
));
1558 tor_close_socket(news
);
1562 if (options
->ConstrainedSockets
)
1563 set_constrained_socket_buffers(news
, (int)options
->ConstrainedSockSize
);
1565 if (check_sockaddr_family_match(remote
->sa_family
, conn
) < 0) {
1566 tor_close_socket(news
);
1570 if (conn
->socket_family
== AF_INET
|| conn
->socket_family
== AF_INET6
||
1571 (conn
->socket_family
== AF_UNIX
&& new_type
== CONN_TYPE_AP
)) {
1574 if (check_sockaddr(remote
, remotelen
, LOG_INFO
)<0) {
1576 "accept() returned a strange address; closing connection.");
1577 tor_close_socket(news
);
1581 tor_addr_from_sockaddr(&addr
, remote
, &port
);
1583 /* process entrance policies here, before we even create the connection */
1584 if (new_type
== CONN_TYPE_AP
) {
1585 /* check sockspolicy to see if we should accept it */
1586 if (socks_policy_permits_address(&addr
) == 0) {
1588 "Denying socks connection from untrusted address %s.",
1589 fmt_and_decorate_addr(&addr
));
1590 tor_close_socket(news
);
1594 if (new_type
== CONN_TYPE_DIR
) {
1595 /* check dirpolicy to see if we should accept it */
1596 if (dir_policy_permits_address(&addr
) == 0) {
1597 log_notice(LD_DIRSERV
,"Denying dir connection from address %s.",
1598 fmt_and_decorate_addr(&addr
));
1599 tor_close_socket(news
);
1604 newconn
= connection_new(new_type
, conn
->socket_family
);
1607 /* remember the remote address */
1608 tor_addr_copy(&newconn
->addr
, &addr
);
1609 if (new_type
== CONN_TYPE_AP
&& conn
->socket_family
== AF_UNIX
) {
1611 newconn
->address
= tor_strdup(conn
->address
);
1613 newconn
->port
= port
;
1614 newconn
->address
= tor_addr_to_str_dup(&addr
);
1617 if (new_type
== CONN_TYPE_AP
&& conn
->socket_family
!= AF_UNIX
) {
1618 log_info(LD_NET
, "New SOCKS connection opened from %s.",
1619 fmt_and_decorate_addr(&addr
));
1621 if (new_type
== CONN_TYPE_AP
&& conn
->socket_family
== AF_UNIX
) {
1622 log_info(LD_NET
, "New SOCKS AF_UNIX connection opened");
1624 if (new_type
== CONN_TYPE_CONTROL
) {
1625 log_notice(LD_CONTROL
, "New control connection opened from %s.",
1626 fmt_and_decorate_addr(&addr
));
1629 } else if (conn
->socket_family
== AF_UNIX
&& conn
->type
!= CONN_TYPE_AP
) {
1630 tor_assert(conn
->type
== CONN_TYPE_CONTROL_LISTENER
);
1631 tor_assert(new_type
== CONN_TYPE_CONTROL
);
1632 log_notice(LD_CONTROL
, "New control connection opened.");
1634 newconn
= connection_new(new_type
, conn
->socket_family
);
1637 /* remember the remote address -- do we have anything sane to put here? */
1638 tor_addr_make_unspec(&newconn
->addr
);
1640 newconn
->address
= tor_strdup(conn
->address
);
1645 if (connection_add(newconn
) < 0) { /* no space, forget it */
1646 connection_free(newconn
);
1647 return 0; /* no need to tear down the parent */
1650 if (connection_init_accepted_conn(newconn
, TO_LISTENER_CONN(conn
)) < 0) {
1651 if (! newconn
->marked_for_close
)
1652 connection_mark_for_close(newconn
);
1658 /** Initialize states for newly accepted connection <b>conn</b>.
1659 * If conn is an OR, start the TLS handshake.
1660 * If conn is a transparent AP, get its original destination
1661 * and place it in circuit_wait.
1664 connection_init_accepted_conn(connection_t
*conn
,
1665 const listener_connection_t
*listener
)
1669 connection_start_reading(conn
);
1671 switch (conn
->type
) {
1672 case CONN_TYPE_EXT_OR
:
1673 /* Initiate Extended ORPort authentication. */
1674 return connection_ext_or_start_auth(TO_OR_CONN(conn
));
1676 control_event_or_conn_status(TO_OR_CONN(conn
), OR_CONN_EVENT_NEW
, 0);
1677 rv
= connection_tls_start_handshake(TO_OR_CONN(conn
), 1);
1679 connection_or_close_for_error(TO_OR_CONN(conn
), 0);
1684 memcpy(&TO_ENTRY_CONN(conn
)->entry_cfg
, &listener
->entry_cfg
,
1685 sizeof(entry_port_cfg_t
));
1686 TO_ENTRY_CONN(conn
)->nym_epoch
= get_signewnym_epoch();
1687 TO_ENTRY_CONN(conn
)->socks_request
->listener_type
= listener
->base_
.type
;
1689 switch (TO_CONN(listener
)->type
) {
1690 case CONN_TYPE_AP_LISTENER
:
1691 conn
->state
= AP_CONN_STATE_SOCKS_WAIT
;
1692 TO_ENTRY_CONN(conn
)->socks_request
->socks_prefer_no_auth
=
1693 listener
->entry_cfg
.socks_prefer_no_auth
;
1695 case CONN_TYPE_AP_TRANS_LISTENER
:
1696 TO_ENTRY_CONN(conn
)->is_transparent_ap
= 1;
1697 /* XXXX028 -- is this correct still, with the addition of
1698 * pending_entry_connections ? */
1699 conn
->state
= AP_CONN_STATE_CIRCUIT_WAIT
;
1700 return connection_ap_process_transparent(TO_ENTRY_CONN(conn
));
1701 case CONN_TYPE_AP_NATD_LISTENER
:
1702 TO_ENTRY_CONN(conn
)->is_transparent_ap
= 1;
1703 conn
->state
= AP_CONN_STATE_NATD_WAIT
;
1708 conn
->purpose
= DIR_PURPOSE_SERVER
;
1709 conn
->state
= DIR_CONN_STATE_SERVER_COMMAND_WAIT
;
1711 case CONN_TYPE_CONTROL
:
1712 conn
->state
= CONTROL_CONN_STATE_NEEDAUTH
;
1718 /** Take conn, make a nonblocking socket; try to connect to
1719 * sa, binding to bindaddr if sa is not localhost. If fail, return -1 and if
1720 * applicable put your best guess about errno into *<b>socket_error</b>.
1721 * If connected return 1, if EAGAIN return 0.
1723 MOCK_IMPL(STATIC
int,
1724 connection_connect_sockaddr
,(connection_t
*conn
,
1725 const struct sockaddr
*sa
,
1727 const struct sockaddr
*bindaddr
,
1728 socklen_t bindaddr_len
,
1733 const or_options_t
*options
= get_options();
1737 tor_assert(socket_error
);
1739 if (get_options()->DisableNetwork
) {
1740 /* We should never even try to connect anyplace if DisableNetwork is set.
1741 * Warn if we do, and refuse to make the connection. */
1742 static ratelim_t disablenet_violated
= RATELIM_INIT(30*60);
1743 *socket_error
= SOCK_ERRNO(ENETUNREACH
);
1744 log_fn_ratelim(&disablenet_violated
, LOG_WARN
, LD_BUG
,
1745 "Tried to open a socket with DisableNetwork set.");
1746 tor_fragile_assert();
1750 const int protocol_family
= sa
->sa_family
;
1751 const int proto
= (sa
->sa_family
== AF_INET6
||
1752 sa
->sa_family
== AF_INET
) ? IPPROTO_TCP
: 0;
1754 s
= tor_open_socket_nonblocking(protocol_family
, SOCK_STREAM
, proto
);
1755 if (! SOCKET_OK(s
)) {
1757 * Early OOS handler calls; it matters if it's an exhaustion-related
1760 *socket_error
= tor_socket_errno(s
);
1761 if (ERRNO_IS_RESOURCE_LIMIT(*socket_error
)) {
1762 warn_too_many_conns();
1763 connection_check_oos(get_n_open_sockets(), 1);
1765 log_warn(LD_NET
,"Error creating network socket: %s",
1766 tor_socket_strerror(*socket_error
));
1767 connection_check_oos(get_n_open_sockets(), 0);
1772 if (make_socket_reuseable(s
) < 0) {
1773 log_warn(LD_NET
, "Error setting SO_REUSEADDR flag on new connection: %s",
1774 tor_socket_strerror(errno
));
1778 * We've got the socket open; give the OOS handler a chance to check
1779 * against configured maximum socket number, but tell it no exhaustion
1782 connection_check_oos(get_n_open_sockets(), 0);
1784 if (bindaddr
&& bind(s
, bindaddr
, bindaddr_len
) < 0) {
1785 *socket_error
= tor_socket_errno(s
);
1786 log_warn(LD_NET
,"Error binding network socket: %s",
1787 tor_socket_strerror(*socket_error
));
1788 tor_close_socket(s
);
1792 tor_assert(options
);
1793 if (options
->ConstrainedSockets
)
1794 set_constrained_socket_buffers(s
, (int)options
->ConstrainedSockSize
);
1796 if (connect(s
, sa
, sa_len
) < 0) {
1797 int e
= tor_socket_errno(s
);
1798 if (!ERRNO_IS_CONN_EINPROGRESS(e
)) {
1799 /* yuck. kill it. */
1802 "connect() to socket failed: %s",
1803 tor_socket_strerror(e
));
1804 tor_close_socket(s
);
1811 /* it succeeded. we're connected. */
1812 log_fn(inprogress
? LOG_DEBUG
: LOG_INFO
, LD_NET
,
1813 "Connection to socket %s (sock "TOR_SOCKET_T_FORMAT
").",
1814 inprogress
? "in progress" : "established", s
);
1816 if (connection_add_connecting(conn
) < 0) {
1817 /* no space, forget it */
1818 *socket_error
= SOCK_ERRNO(ENOBUFS
);
1822 return inprogress
? 0 : 1;
1825 /* Log a message if connection attempt is made when IPv4 or IPv6 is disabled.
1826 * Log a less severe message if we couldn't conform to ClientPreferIPv6ORPort
1827 * or ClientPreferIPv6ORPort. */
1829 connection_connect_log_client_use_ip_version(const connection_t
*conn
)
1831 const or_options_t
*options
= get_options();
1833 /* Only clients care about ClientUseIPv4/6, bail out early on servers, and
1834 * on connections we don't care about */
1835 if (server_mode(options
) || !conn
|| conn
->type
== CONN_TYPE_EXIT
) {
1839 /* We're only prepared to log OR and DIR connections here */
1840 if (conn
->type
!= CONN_TYPE_OR
&& conn
->type
!= CONN_TYPE_DIR
) {
1844 const int must_ipv4
= !fascist_firewall_use_ipv6(options
);
1845 const int must_ipv6
= (options
->ClientUseIPv4
== 0);
1846 const int pref_ipv6
= (conn
->type
== CONN_TYPE_OR
1847 ? fascist_firewall_prefer_ipv6_orport(options
)
1848 : fascist_firewall_prefer_ipv6_dirport(options
));
1849 tor_addr_t real_addr
;
1850 tor_addr_make_null(&real_addr
, AF_UNSPEC
);
1852 /* OR conns keep the original address in real_addr, as addr gets overwritten
1853 * with the descriptor address */
1854 if (conn
->type
== CONN_TYPE_OR
) {
1855 const or_connection_t
*or_conn
= TO_OR_CONN((connection_t
*)conn
);
1856 tor_addr_copy(&real_addr
, &or_conn
->real_addr
);
1857 } else if (conn
->type
== CONN_TYPE_DIR
) {
1858 tor_addr_copy(&real_addr
, &conn
->addr
);
1861 /* Check if we broke a mandatory address family restriction */
1862 if ((must_ipv4
&& tor_addr_family(&real_addr
) == AF_INET6
)
1863 || (must_ipv6
&& tor_addr_family(&real_addr
) == AF_INET
)) {
1864 static int logged_backtrace
= 0;
1865 log_info(LD_BUG
, "Outgoing %s connection to %s violated ClientUseIPv%s 0.",
1866 conn
->type
== CONN_TYPE_OR
? "OR" : "Dir",
1867 fmt_addr(&real_addr
),
1868 options
->ClientUseIPv4
== 0 ? "4" : "6");
1869 if (!logged_backtrace
) {
1870 log_backtrace(LOG_INFO
, LD_BUG
, "Address came from");
1871 logged_backtrace
= 1;
1875 /* Bridges are allowed to break IPv4/IPv6 ORPort preferences to connect to
1876 * the node's configured address when ClientPreferIPv6ORPort is auto */
1877 if (options
->UseBridges
&& conn
->type
== CONN_TYPE_OR
1878 && options
->ClientPreferIPv6ORPort
== -1) {
1882 /* Check if we couldn't satisfy an address family preference */
1883 if ((!pref_ipv6
&& tor_addr_family(&real_addr
) == AF_INET6
)
1884 || (pref_ipv6
&& tor_addr_family(&real_addr
) == AF_INET
)) {
1885 log_info(LD_NET
, "Outgoing connection to %s doesn't satisfy "
1886 "ClientPreferIPv6%sPort %d, with ClientUseIPv4 %d, and "
1887 "fascist_firewall_use_ipv6 %d (ClientUseIPv6 %d and UseBridges "
1889 fmt_addr(&real_addr
),
1890 conn
->type
== CONN_TYPE_OR
? "OR" : "Dir",
1891 conn
->type
== CONN_TYPE_OR
? options
->ClientPreferIPv6ORPort
1892 : options
->ClientPreferIPv6DirPort
,
1893 options
->ClientUseIPv4
, fascist_firewall_use_ipv6(options
),
1894 options
->ClientUseIPv6
, options
->UseBridges
);
1898 /** Retrieve the outbound address depending on the protocol (IPv4 or IPv6)
1899 * and the connection type (relay, exit, ...)
1900 * Return a socket address or NULL in case nothing is configured.
1903 conn_get_outbound_address(sa_family_t family
,
1904 const or_options_t
*options
, unsigned int conn_type
)
1906 const tor_addr_t
*ext_addr
= NULL
;
1920 // If an exit connection, use the exit address (if present)
1921 if (conn_type
== CONN_TYPE_EXIT
) {
1922 if (!tor_addr_is_null(
1923 &options
->OutboundBindAddresses
[OUTBOUND_ADDR_EXIT
][fam_index
])) {
1924 ext_addr
= &options
->OutboundBindAddresses
[OUTBOUND_ADDR_EXIT
]
1926 } else if (!tor_addr_is_null(
1927 &options
->OutboundBindAddresses
[OUTBOUND_ADDR_EXIT_AND_OR
]
1929 ext_addr
= &options
->OutboundBindAddresses
[OUTBOUND_ADDR_EXIT_AND_OR
]
1932 } else { // All non-exit connections
1933 if (!tor_addr_is_null(
1934 &options
->OutboundBindAddresses
[OUTBOUND_ADDR_OR
][fam_index
])) {
1935 ext_addr
= &options
->OutboundBindAddresses
[OUTBOUND_ADDR_OR
]
1937 } else if (!tor_addr_is_null(
1938 &options
->OutboundBindAddresses
[OUTBOUND_ADDR_EXIT_AND_OR
]
1940 ext_addr
= &options
->OutboundBindAddresses
[OUTBOUND_ADDR_EXIT_AND_OR
]
1947 /** Take conn, make a nonblocking socket; try to connect to
1948 * addr:port (port arrives in *host order*). If fail, return -1 and if
1949 * applicable put your best guess about errno into *<b>socket_error</b>.
1950 * Else assign s to conn-\>s: if connected return 1, if EAGAIN return 0.
1952 * addr:port can be different to conn->addr:conn->port if connecting through
1955 * address is used to make the logs useful.
1957 * On success, add conn to the list of polled connections.
1960 connection_connect(connection_t
*conn
, const char *address
,
1961 const tor_addr_t
*addr
, uint16_t port
, int *socket_error
)
1963 struct sockaddr_storage addrbuf
;
1964 struct sockaddr_storage bind_addr_ss
;
1965 struct sockaddr
*bind_addr
= NULL
;
1966 struct sockaddr
*dest_addr
;
1967 int dest_addr_len
, bind_addr_len
= 0;
1969 /* Log if we didn't stick to ClientUseIPv4/6 or ClientPreferIPv6OR/DirPort
1971 connection_connect_log_client_use_ip_version(conn
);
1973 if (!tor_addr_is_loopback(addr
)) {
1974 const tor_addr_t
*ext_addr
= NULL
;
1975 ext_addr
= conn_get_outbound_address(tor_addr_family(addr
), get_options(),
1978 memset(&bind_addr_ss
, 0, sizeof(bind_addr_ss
));
1979 bind_addr_len
= tor_addr_to_sockaddr(ext_addr
, 0,
1980 (struct sockaddr
*) &bind_addr_ss
,
1981 sizeof(bind_addr_ss
));
1982 if (bind_addr_len
== 0) {
1984 "Error converting OutboundBindAddress %s into sockaddr. "
1985 "Ignoring.", fmt_and_decorate_addr(ext_addr
));
1987 bind_addr
= (struct sockaddr
*)&bind_addr_ss
;
1992 memset(&addrbuf
,0,sizeof(addrbuf
));
1993 dest_addr
= (struct sockaddr
*) &addrbuf
;
1994 dest_addr_len
= tor_addr_to_sockaddr(addr
, port
, dest_addr
, sizeof(addrbuf
));
1995 tor_assert(dest_addr_len
> 0);
1997 log_debug(LD_NET
, "Connecting to %s:%u.",
1998 escaped_safe_str_client(address
), port
);
2000 return connection_connect_sockaddr(conn
, dest_addr
, dest_addr_len
,
2001 bind_addr
, bind_addr_len
, socket_error
);
2004 #ifdef HAVE_SYS_UN_H
2006 /** Take conn, make a nonblocking socket; try to connect to
2007 * an AF_UNIX socket at socket_path. If fail, return -1 and if applicable
2008 * put your best guess about errno into *<b>socket_error</b>. Else assign s
2009 * to conn-\>s: if connected return 1, if EAGAIN return 0.
2011 * On success, add conn to the list of polled connections.
2014 connection_connect_unix(connection_t
*conn
, const char *socket_path
,
2017 struct sockaddr_un dest_addr
;
2019 tor_assert(socket_path
);
2021 /* Check that we'll be able to fit it into dest_addr later */
2022 if (strlen(socket_path
) + 1 > sizeof(dest_addr
.sun_path
)) {
2024 "Path %s is too long for an AF_UNIX socket\n",
2025 escaped_safe_str_client(socket_path
));
2026 *socket_error
= SOCK_ERRNO(ENAMETOOLONG
);
2030 memset(&dest_addr
, 0, sizeof(dest_addr
));
2031 dest_addr
.sun_family
= AF_UNIX
;
2032 strlcpy(dest_addr
.sun_path
, socket_path
, sizeof(dest_addr
.sun_path
));
2035 "Connecting to AF_UNIX socket at %s.",
2036 escaped_safe_str_client(socket_path
));
2038 return connection_connect_sockaddr(conn
,
2039 (struct sockaddr
*)&dest_addr
, sizeof(dest_addr
),
2040 NULL
, 0, socket_error
);
2043 #endif /* defined(HAVE_SYS_UN_H) */
2045 /** Convert state number to string representation for logging purposes.
2048 connection_proxy_state_to_string(int state
)
2050 static const char *unknown
= "???";
2051 static const char *states
[] = {
2054 "PROXY_HTTPS_WANT_CONNECT_OK",
2055 "PROXY_SOCKS4_WANT_CONNECT_OK",
2056 "PROXY_SOCKS5_WANT_AUTH_METHOD_NONE",
2057 "PROXY_SOCKS5_WANT_AUTH_METHOD_RFC1929",
2058 "PROXY_SOCKS5_WANT_AUTH_RFC1929_OK",
2059 "PROXY_SOCKS5_WANT_CONNECT_OK",
2063 if (state
< PROXY_NONE
|| state
> PROXY_CONNECTED
)
2066 return states
[state
];
2069 /** Returns the global proxy type used by tor. Use this function for
2070 * logging or high-level purposes, don't use it to fill the
2071 * <b>proxy_type</b> field of or_connection_t; use the actual proxy
2072 * protocol instead.*/
2074 get_proxy_type(void)
2076 const or_options_t
*options
= get_options();
2078 if (options
->ClientTransportPlugin
)
2079 return PROXY_PLUGGABLE
;
2080 else if (options
->HTTPSProxy
)
2081 return PROXY_CONNECT
;
2082 else if (options
->Socks4Proxy
)
2083 return PROXY_SOCKS4
;
2084 else if (options
->Socks5Proxy
)
2085 return PROXY_SOCKS5
;
2090 /* One byte for the version, one for the command, two for the
2091 port, and four for the addr... and, one more for the
2093 #define SOCKS4_STANDARD_BUFFER_SIZE (1 + 1 + 2 + 4 + 1)
2095 /** Write a proxy request of <b>type</b> (socks4, socks5, https) to conn
2096 * for conn->addr:conn->port, authenticating with the auth details given
2097 * in the configuration (if available). SOCKS 5 and HTTP CONNECT proxies
2098 * support authentication.
2100 * Returns -1 if conn->addr is incompatible with the proxy protocol, and
2103 * Use connection_read_proxy_handshake() to complete the handshake.
2106 connection_proxy_connect(connection_t
*conn
, int type
)
2108 const or_options_t
*options
;
2112 options
= get_options();
2115 case PROXY_CONNECT
: {
2117 char *base64_authenticator
=NULL
;
2118 const char *authenticator
= options
->HTTPSProxyAuthenticator
;
2120 /* Send HTTP CONNECT and authentication (if available) in
2123 if (authenticator
) {
2124 base64_authenticator
= alloc_http_authenticator(authenticator
);
2125 if (!base64_authenticator
)
2126 log_warn(LD_OR
, "Encoding https authenticator failed");
2129 if (base64_authenticator
) {
2130 const char *addrport
= fmt_addrport(&conn
->addr
, conn
->port
);
2131 tor_snprintf(buf
, sizeof(buf
), "CONNECT %s HTTP/1.1\r\n"
2133 "Proxy-Authorization: Basic %s\r\n\r\n",
2136 base64_authenticator
);
2137 tor_free(base64_authenticator
);
2139 tor_snprintf(buf
, sizeof(buf
), "CONNECT %s HTTP/1.0\r\n\r\n",
2140 fmt_addrport(&conn
->addr
, conn
->port
));
2143 connection_write_to_buf(buf
, strlen(buf
), conn
);
2144 conn
->proxy_state
= PROXY_HTTPS_WANT_CONNECT_OK
;
2148 case PROXY_SOCKS4
: {
2152 size_t buf_size
= 0;
2153 char *socks_args_string
= NULL
;
2155 /* Send a SOCKS4 connect request */
2157 if (tor_addr_family(&conn
->addr
) != AF_INET
) {
2158 log_warn(LD_NET
, "SOCKS4 client is incompatible with IPv6");
2162 { /* If we are here because we are trying to connect to a
2163 pluggable transport proxy, check if we have any SOCKS
2164 arguments to transmit. If we do, compress all arguments to
2165 a single string in 'socks_args_string': */
2167 if (get_proxy_type() == PROXY_PLUGGABLE
) {
2169 pt_get_socks_args_for_proxy_addrport(&conn
->addr
, conn
->port
);
2170 if (socks_args_string
)
2171 log_debug(LD_NET
, "Sending out '%s' as our SOCKS argument string.",
2176 { /* Figure out the buffer size we need for the SOCKS message: */
2178 buf_size
= SOCKS4_STANDARD_BUFFER_SIZE
;
2180 /* If we have a SOCKS argument string, consider its size when
2181 calculating the buffer size: */
2182 if (socks_args_string
)
2183 buf_size
+= strlen(socks_args_string
);
2186 buf
= tor_malloc_zero(buf_size
);
2188 ip4addr
= tor_addr_to_ipv4n(&conn
->addr
);
2189 portn
= htons(conn
->port
);
2191 buf
[0] = 4; /* version */
2192 buf
[1] = SOCKS_COMMAND_CONNECT
; /* command */
2193 memcpy(buf
+ 2, &portn
, 2); /* port */
2194 memcpy(buf
+ 4, &ip4addr
, 4); /* addr */
2196 /* Next packet field is the userid. If we have pluggable
2197 transport SOCKS arguments, we have to embed them
2198 there. Otherwise, we use an empty userid. */
2199 if (socks_args_string
) { /* place the SOCKS args string: */
2200 tor_assert(strlen(socks_args_string
) > 0);
2201 tor_assert(buf_size
>=
2202 SOCKS4_STANDARD_BUFFER_SIZE
+ strlen(socks_args_string
));
2203 strlcpy((char *)buf
+ 8, socks_args_string
, buf_size
- 8);
2204 tor_free(socks_args_string
);
2206 buf
[8] = 0; /* no userid */
2209 connection_write_to_buf((char *)buf
, buf_size
, conn
);
2212 conn
->proxy_state
= PROXY_SOCKS4_WANT_CONNECT_OK
;
2216 case PROXY_SOCKS5
: {
2217 unsigned char buf
[4]; /* fields: vers, num methods, method list */
2219 /* Send a SOCKS5 greeting (connect request must wait) */
2221 buf
[0] = 5; /* version */
2223 /* We have to use SOCKS5 authentication, if we have a
2224 Socks5ProxyUsername or if we want to pass arguments to our
2225 pluggable transport proxy: */
2226 if ((options
->Socks5ProxyUsername
) ||
2227 (get_proxy_type() == PROXY_PLUGGABLE
&&
2228 (get_socks_args_by_bridge_addrport(&conn
->addr
, conn
->port
)))) {
2229 /* number of auth methods */
2231 buf
[2] = 0x00; /* no authentication */
2232 buf
[3] = 0x02; /* rfc1929 Username/Passwd auth */
2233 conn
->proxy_state
= PROXY_SOCKS5_WANT_AUTH_METHOD_RFC1929
;
2236 buf
[2] = 0x00; /* no authentication */
2237 conn
->proxy_state
= PROXY_SOCKS5_WANT_AUTH_METHOD_NONE
;
2240 connection_write_to_buf((char *)buf
, 2 + buf
[1], conn
);
2245 log_err(LD_BUG
, "Invalid proxy protocol, %d", type
);
2246 tor_fragile_assert();
2250 log_debug(LD_NET
, "set state %s",
2251 connection_proxy_state_to_string(conn
->proxy_state
));
2256 /** Read conn's inbuf. If the http response from the proxy is all
2257 * here, make sure it's good news, then return 1. If it's bad news,
2258 * return -1. Else return 0 and hope for better luck next time.
2261 connection_read_https_proxy_response(connection_t
*conn
)
2268 switch (fetch_from_buf_http(conn
->inbuf
,
2269 &headers
, MAX_HEADERS_SIZE
,
2270 NULL
, NULL
, 10000, 0)) {
2271 case -1: /* overflow */
2272 log_warn(LD_PROTOCOL
,
2273 "Your https proxy sent back an oversized response. Closing.");
2276 log_info(LD_NET
,"https proxy response not all here yet. Waiting.");
2278 /* case 1, fall through */
2281 if (parse_http_response(headers
, &status_code
, &date_header
,
2282 NULL
, &reason
) < 0) {
2284 "Unparseable headers from proxy (connecting to '%s'). Closing.",
2290 if (!reason
) reason
= tor_strdup("[no reason given]");
2292 if (status_code
== 200) {
2294 "HTTPS connect to '%s' successful! (200 %s) Starting TLS.",
2295 conn
->address
, escaped(reason
));
2299 /* else, bad news on the status code */
2300 switch (status_code
) {
2303 "The https proxy refused to allow connection to %s "
2304 "(status code %d, %s). Closing.",
2305 conn
->address
, status_code
, escaped(reason
));
2309 "The https proxy sent back an unexpected status code %d (%s). "
2311 status_code
, escaped(reason
));
2318 /** Send SOCKS5 CONNECT command to <b>conn</b>, copying <b>conn->addr</b>
2319 * and <b>conn->port</b> into the request.
2322 connection_send_socks5_connect(connection_t
*conn
)
2324 unsigned char buf
[1024];
2326 uint16_t port
= htons(conn
->port
);
2328 buf
[0] = 5; /* version */
2329 buf
[1] = SOCKS_COMMAND_CONNECT
; /* command */
2330 buf
[2] = 0; /* reserved */
2332 if (tor_addr_family(&conn
->addr
) == AF_INET
) {
2333 uint32_t addr
= tor_addr_to_ipv4n(&conn
->addr
);
2337 memcpy(buf
+ 4, &addr
, 4);
2338 memcpy(buf
+ 8, &port
, 2);
2339 } else { /* AF_INET6 */
2342 memcpy(buf
+ 4, tor_addr_to_in6_addr8(&conn
->addr
), 16);
2343 memcpy(buf
+ 20, &port
, 2);
2346 connection_write_to_buf((char *)buf
, reqsize
, conn
);
2348 conn
->proxy_state
= PROXY_SOCKS5_WANT_CONNECT_OK
;
2351 /** Wrapper around fetch_from_buf_socks_client: see that functions
2352 * for documentation of its behavior. */
2354 connection_fetch_from_buf_socks_client(connection_t
*conn
,
2355 int state
, char **reason
)
2357 return fetch_from_buf_socks_client(conn
->inbuf
, state
, reason
);
2360 /** Call this from connection_*_process_inbuf() to advance the proxy
2363 * No matter what proxy protocol is used, if this function returns 1, the
2364 * handshake is complete, and the data remaining on inbuf may contain the
2365 * start of the communication with the requested server.
2367 * Returns 0 if the current buffer contains an incomplete response, and -1
2371 connection_read_proxy_handshake(connection_t
*conn
)
2374 char *reason
= NULL
;
2376 log_debug(LD_NET
, "enter state %s",
2377 connection_proxy_state_to_string(conn
->proxy_state
));
2379 switch (conn
->proxy_state
) {
2380 case PROXY_HTTPS_WANT_CONNECT_OK
:
2381 ret
= connection_read_https_proxy_response(conn
);
2383 conn
->proxy_state
= PROXY_CONNECTED
;
2386 case PROXY_SOCKS4_WANT_CONNECT_OK
:
2387 ret
= connection_fetch_from_buf_socks_client(conn
,
2391 conn
->proxy_state
= PROXY_CONNECTED
;
2394 case PROXY_SOCKS5_WANT_AUTH_METHOD_NONE
:
2395 ret
= connection_fetch_from_buf_socks_client(conn
,
2398 /* no auth needed, do connect */
2400 connection_send_socks5_connect(conn
);
2405 case PROXY_SOCKS5_WANT_AUTH_METHOD_RFC1929
:
2406 ret
= connection_fetch_from_buf_socks_client(conn
,
2410 /* send auth if needed, otherwise do connect */
2412 connection_send_socks5_connect(conn
);
2414 } else if (ret
== 2) {
2415 unsigned char buf
[1024];
2416 size_t reqsize
, usize
, psize
;
2417 const char *user
, *pass
;
2418 char *socks_args_string
= NULL
;
2420 if (get_proxy_type() == PROXY_PLUGGABLE
) {
2422 pt_get_socks_args_for_proxy_addrport(&conn
->addr
, conn
->port
);
2423 if (!socks_args_string
) {
2424 log_warn(LD_NET
, "Could not create SOCKS args string.");
2429 log_debug(LD_NET
, "SOCKS5 arguments: %s", socks_args_string
);
2430 tor_assert(strlen(socks_args_string
) > 0);
2431 tor_assert(strlen(socks_args_string
) <= MAX_SOCKS5_AUTH_SIZE_TOTAL
);
2433 if (strlen(socks_args_string
) > MAX_SOCKS5_AUTH_FIELD_SIZE
) {
2434 user
= socks_args_string
;
2435 usize
= MAX_SOCKS5_AUTH_FIELD_SIZE
;
2436 pass
= socks_args_string
+ MAX_SOCKS5_AUTH_FIELD_SIZE
;
2437 psize
= strlen(socks_args_string
) - MAX_SOCKS5_AUTH_FIELD_SIZE
;
2439 user
= socks_args_string
;
2440 usize
= strlen(socks_args_string
);
2444 } else if (get_options()->Socks5ProxyUsername
) {
2445 user
= get_options()->Socks5ProxyUsername
;
2446 pass
= get_options()->Socks5ProxyPassword
;
2447 tor_assert(user
&& pass
);
2448 usize
= strlen(user
);
2449 psize
= strlen(pass
);
2451 log_err(LD_BUG
, "We entered %s for no reason!", __func__
);
2452 tor_fragile_assert();
2457 /* Username and password lengths should have been checked
2458 above and during torrc parsing. */
2459 tor_assert(usize
<= MAX_SOCKS5_AUTH_FIELD_SIZE
&&
2460 psize
<= MAX_SOCKS5_AUTH_FIELD_SIZE
);
2461 reqsize
= 3 + usize
+ psize
;
2463 buf
[0] = 1; /* negotiation version */
2465 memcpy(buf
+ 2, user
, usize
);
2466 buf
[2 + usize
] = psize
;
2467 memcpy(buf
+ 3 + usize
, pass
, psize
);
2469 if (socks_args_string
)
2470 tor_free(socks_args_string
);
2472 connection_write_to_buf((char *)buf
, reqsize
, conn
);
2474 conn
->proxy_state
= PROXY_SOCKS5_WANT_AUTH_RFC1929_OK
;
2479 case PROXY_SOCKS5_WANT_AUTH_RFC1929_OK
:
2480 ret
= connection_fetch_from_buf_socks_client(conn
,
2483 /* send the connect request */
2485 connection_send_socks5_connect(conn
);
2490 case PROXY_SOCKS5_WANT_CONNECT_OK
:
2491 ret
= connection_fetch_from_buf_socks_client(conn
,
2495 conn
->proxy_state
= PROXY_CONNECTED
;
2499 log_err(LD_BUG
, "Invalid proxy_state for reading, %d",
2501 tor_fragile_assert();
2506 log_debug(LD_NET
, "leaving state %s",
2507 connection_proxy_state_to_string(conn
->proxy_state
));
2511 log_warn(LD_NET
, "Proxy Client: unable to connect to %s:%d (%s)",
2512 conn
->address
, conn
->port
, escaped(reason
));
2515 log_warn(LD_NET
, "Proxy Client: unable to connect to %s:%d",
2516 conn
->address
, conn
->port
);
2518 } else if (ret
== 1) {
2519 log_info(LD_NET
, "Proxy Client: connection to %s:%d successful",
2520 conn
->address
, conn
->port
);
2526 /** Given a list of listener connections in <b>old_conns</b>, and list of
2527 * port_cfg_t entries in <b>ports</b>, open a new listener for every port in
2528 * <b>ports</b> that does not already have a listener in <b>old_conns</b>.
2530 * Remove from <b>old_conns</b> every connection that has a corresponding
2531 * entry in <b>ports</b>. Add to <b>new_conns</b> new every connection we
2534 * If <b>control_listeners_only</b> is true, then we only open control
2535 * listeners, and we do not remove any noncontrol listeners from old_conns.
2537 * Return 0 on success, -1 on failure.
2540 retry_listener_ports(smartlist_t
*old_conns
,
2541 const smartlist_t
*ports
,
2542 smartlist_t
*new_conns
,
2543 int control_listeners_only
)
2545 smartlist_t
*launch
= smartlist_new();
2548 if (control_listeners_only
) {
2549 SMARTLIST_FOREACH(ports
, port_cfg_t
*, p
, {
2550 if (p
->type
== CONN_TYPE_CONTROL_LISTENER
)
2551 smartlist_add(launch
, p
);
2554 smartlist_add_all(launch
, ports
);
2557 /* Iterate through old_conns, comparing it to launch: remove from both lists
2558 * each pair of elements that corresponds to the same port. */
2559 SMARTLIST_FOREACH_BEGIN(old_conns
, connection_t
*, conn
) {
2560 const port_cfg_t
*found_port
= NULL
;
2562 /* Okay, so this is a listener. Is it configured? */
2563 SMARTLIST_FOREACH_BEGIN(launch
, const port_cfg_t
*, wanted
) {
2564 if (conn
->type
!= wanted
->type
)
2566 if ((conn
->socket_family
!= AF_UNIX
&& wanted
->is_unix_addr
) ||
2567 (conn
->socket_family
== AF_UNIX
&& ! wanted
->is_unix_addr
))
2570 if (wanted
->server_cfg
.no_listen
)
2571 continue; /* We don't want to open a listener for this one */
2573 if (wanted
->is_unix_addr
) {
2574 if (conn
->socket_family
== AF_UNIX
&&
2575 !strcmp(wanted
->unix_addr
, conn
->address
)) {
2576 found_port
= wanted
;
2581 if (wanted
->port
== CFG_AUTO_PORT
) {
2584 port_matches
= (wanted
->port
== conn
->port
);
2586 if (port_matches
&& tor_addr_eq(&wanted
->addr
, &conn
->addr
)) {
2587 found_port
= wanted
;
2591 } SMARTLIST_FOREACH_END(wanted
);
2594 /* This listener is already running; we don't need to launch it. */
2595 //log_debug(LD_NET, "Already have %s on %s:%d",
2596 // conn_type_to_string(found_port->type), conn->address, conn->port);
2597 smartlist_remove(launch
, found_port
);
2598 /* And we can remove the connection from old_conns too. */
2599 SMARTLIST_DEL_CURRENT(old_conns
, conn
);
2601 } SMARTLIST_FOREACH_END(conn
);
2603 /* Now open all the listeners that are configured but not opened. */
2604 SMARTLIST_FOREACH_BEGIN(launch
, const port_cfg_t
*, port
) {
2605 struct sockaddr
*listensockaddr
;
2606 socklen_t listensocklen
= 0;
2609 int real_port
= port
->port
== CFG_AUTO_PORT
? 0 : port
->port
;
2610 tor_assert(real_port
<= UINT16_MAX
);
2611 if (port
->server_cfg
.no_listen
)
2615 /* We don't need to be root to create a UNIX socket, so defer until after
2617 const or_options_t
*options
= get_options();
2618 if (port
->is_unix_addr
&& !geteuid() && (options
->User
) &&
2619 strcmp(options
->User
, "root"))
2623 if (port
->is_unix_addr
) {
2624 listensockaddr
= (struct sockaddr
*)
2625 create_unix_sockaddr(port
->unix_addr
,
2626 &address
, &listensocklen
);
2628 listensockaddr
= tor_malloc(sizeof(struct sockaddr_storage
));
2629 listensocklen
= tor_addr_to_sockaddr(&port
->addr
,
2632 sizeof(struct sockaddr_storage
));
2633 address
= tor_addr_to_str_dup(&port
->addr
);
2636 if (listensockaddr
) {
2637 conn
= connection_listener_new(listensockaddr
, listensocklen
,
2638 port
->type
, address
, port
);
2639 tor_free(listensockaddr
);
2649 smartlist_add(new_conns
, conn
);
2651 } SMARTLIST_FOREACH_END(port
);
2653 smartlist_free(launch
);
2658 /** Launch listeners for each port you should have open. Only launch
2659 * listeners who are not already open, and only close listeners we no longer
2662 * Add all old conns that should be closed to <b>replaced_conns</b>.
2663 * Add all new connections to <b>new_conns</b>.
2665 * If <b>close_all_noncontrol</b> is true, then we only open control
2666 * listeners, and we close all other listeners.
2669 retry_all_listeners(smartlist_t
*replaced_conns
,
2670 smartlist_t
*new_conns
, int close_all_noncontrol
)
2672 smartlist_t
*listeners
= smartlist_new();
2673 const or_options_t
*options
= get_options();
2675 const uint16_t old_or_port
= router_get_advertised_or_port(options
);
2676 const uint16_t old_or_port_ipv6
=
2677 router_get_advertised_or_port_by_af(options
,AF_INET6
);
2678 const uint16_t old_dir_port
= router_get_advertised_dir_port(options
, 0);
2680 SMARTLIST_FOREACH_BEGIN(get_connection_array(), connection_t
*, conn
) {
2681 if (connection_is_listener(conn
) && !conn
->marked_for_close
)
2682 smartlist_add(listeners
, conn
);
2683 } SMARTLIST_FOREACH_END(conn
);
2685 if (retry_listener_ports(listeners
,
2686 get_configured_ports(),
2688 close_all_noncontrol
) < 0)
2691 /* Any members that were still in 'listeners' don't correspond to
2692 * any configured port. Kill 'em. */
2693 SMARTLIST_FOREACH_BEGIN(listeners
, connection_t
*, conn
) {
2694 log_notice(LD_NET
, "Closing no-longer-configured %s on %s:%d",
2695 conn_type_to_string(conn
->type
), conn
->address
, conn
->port
);
2696 if (replaced_conns
) {
2697 smartlist_add(replaced_conns
, conn
);
2699 connection_close_immediate(conn
);
2700 connection_mark_for_close(conn
);
2702 } SMARTLIST_FOREACH_END(conn
);
2704 smartlist_free(listeners
);
2706 if (old_or_port
!= router_get_advertised_or_port(options
) ||
2707 old_or_port_ipv6
!= router_get_advertised_or_port_by_af(options
,
2709 old_dir_port
!= router_get_advertised_dir_port(options
, 0)) {
2710 /* Our chosen ORPort or DirPort is not what it used to be: the
2711 * descriptor we had (if any) should be regenerated. (We won't
2712 * automatically notice this because of changes in the option,
2713 * since the value could be "auto".) */
2714 mark_my_descriptor_dirty("Chosen Or/DirPort changed");
2720 /** Mark every listener of type other than CONTROL_LISTENER to be closed. */
2722 connection_mark_all_noncontrol_listeners(void)
2724 SMARTLIST_FOREACH_BEGIN(get_connection_array(), connection_t
*, conn
) {
2725 if (conn
->marked_for_close
)
2727 if (conn
->type
== CONN_TYPE_CONTROL_LISTENER
)
2729 if (connection_is_listener(conn
))
2730 connection_mark_for_close(conn
);
2731 } SMARTLIST_FOREACH_END(conn
);
2734 /** Mark every external connection not used for controllers for close. */
2736 connection_mark_all_noncontrol_connections(void)
2738 SMARTLIST_FOREACH_BEGIN(get_connection_array(), connection_t
*, conn
) {
2739 if (conn
->marked_for_close
)
2741 switch (conn
->type
) {
2742 case CONN_TYPE_CONTROL_LISTENER
:
2743 case CONN_TYPE_CONTROL
:
2746 connection_mark_unattached_ap(TO_ENTRY_CONN(conn
),
2747 END_STREAM_REASON_HIBERNATING
);
2751 or_connection_t
*orconn
= TO_OR_CONN(conn
);
2753 connection_or_close_normally(orconn
, 0);
2756 * There should have been one, but mark for close and hope
2759 connection_mark_for_close(conn
);
2764 connection_mark_for_close(conn
);
2767 } SMARTLIST_FOREACH_END(conn
);
2770 /** Return 1 if we should apply rate limiting to <b>conn</b>, and 0
2772 * Right now this just checks if it's an internal IP address or an
2773 * internal connection. We also should, but don't, check if the connection
2774 * uses pluggable transports, since we should then limit it even if it
2775 * comes from an internal IP address. */
2777 connection_is_rate_limited(connection_t
*conn
)
2779 const or_options_t
*options
= get_options();
2781 return 0; /* Internal connection */
2782 else if (! options
->CountPrivateBandwidth
&&
2783 (tor_addr_family(&conn
->addr
) == AF_UNSPEC
|| /* no address */
2784 tor_addr_family(&conn
->addr
) == AF_UNIX
|| /* no address */
2785 tor_addr_is_internal(&conn
->addr
, 0)))
2786 return 0; /* Internal address */
2791 /** Did either global write bucket run dry last second? If so,
2792 * we are likely to run dry again this second, so be stingy with the
2793 * tokens we just put in. */
2794 static int write_buckets_empty_last_second
= 0;
2796 /** How many seconds of no active local circuits will make the
2797 * connection revert to the "relayed" bandwidth class? */
2798 #define CLIENT_IDLE_TIME_FOR_PRIORITY 30
2800 /** Return 1 if <b>conn</b> should use tokens from the "relayed"
2801 * bandwidth rates, else 0. Currently, only OR conns with bandwidth
2802 * class 1, and directory conns that are serving data out, count.
2805 connection_counts_as_relayed_traffic(connection_t
*conn
, time_t now
)
2807 if (conn
->type
== CONN_TYPE_OR
&&
2808 connection_or_client_used(TO_OR_CONN(conn
)) +
2809 CLIENT_IDLE_TIME_FOR_PRIORITY
< now
)
2811 if (conn
->type
== CONN_TYPE_DIR
&& DIR_CONN_IS_SERVER(conn
))
2816 /** Helper function to decide how many bytes out of <b>global_bucket</b>
2817 * we're willing to use for this transaction. <b>base</b> is the size
2818 * of a cell on the network; <b>priority</b> says whether we should
2819 * write many of them or just a few; and <b>conn_bucket</b> (if
2820 * non-negative) provides an upper limit for our answer. */
2822 connection_bucket_round_robin(int base
, int priority
,
2823 ssize_t global_bucket
, ssize_t conn_bucket
)
2826 ssize_t num_bytes_high
= (priority
? 32 : 16) * base
;
2827 ssize_t num_bytes_low
= (priority
? 4 : 2) * base
;
2829 /* Do a rudimentary round-robin so one circuit can't hog a connection.
2830 * Pick at most 32 cells, at least 4 cells if possible, and if we're in
2831 * the middle pick 1/8 of the available bandwidth. */
2832 at_most
= global_bucket
/ 8;
2833 at_most
-= (at_most
% base
); /* round down */
2834 if (at_most
> num_bytes_high
) /* 16 KB, or 8 KB for low-priority */
2835 at_most
= num_bytes_high
;
2836 else if (at_most
< num_bytes_low
) /* 2 KB, or 1 KB for low-priority */
2837 at_most
= num_bytes_low
;
2839 if (at_most
> global_bucket
)
2840 at_most
= global_bucket
;
2842 if (conn_bucket
>= 0 && at_most
> conn_bucket
)
2843 at_most
= conn_bucket
;
2850 /** How many bytes at most can we read onto this connection? */
2852 connection_bucket_read_limit(connection_t
*conn
, time_t now
)
2854 int base
= RELAY_PAYLOAD_SIZE
;
2855 int priority
= conn
->type
!= CONN_TYPE_DIR
;
2856 int conn_bucket
= -1;
2857 int global_bucket
= global_read_bucket
;
2859 if (connection_speaks_cells(conn
)) {
2860 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
2861 if (conn
->state
== OR_CONN_STATE_OPEN
)
2862 conn_bucket
= or_conn
->read_bucket
;
2863 base
= get_cell_network_size(or_conn
->wide_circ_ids
);
2866 if (!connection_is_rate_limited(conn
)) {
2867 /* be willing to read on local conns even if our buckets are empty */
2868 return conn_bucket
>=0 ? conn_bucket
: 1<<14;
2871 if (connection_counts_as_relayed_traffic(conn
, now
) &&
2872 global_relayed_read_bucket
<= global_read_bucket
)
2873 global_bucket
= global_relayed_read_bucket
;
2875 return connection_bucket_round_robin(base
, priority
,
2876 global_bucket
, conn_bucket
);
2879 /** How many bytes at most can we write onto this connection? */
2881 connection_bucket_write_limit(connection_t
*conn
, time_t now
)
2883 int base
= RELAY_PAYLOAD_SIZE
;
2884 int priority
= conn
->type
!= CONN_TYPE_DIR
;
2885 int conn_bucket
= (int)conn
->outbuf_flushlen
;
2886 int global_bucket
= global_write_bucket
;
2888 if (!connection_is_rate_limited(conn
)) {
2889 /* be willing to write to local conns even if our buckets are empty */
2890 return conn
->outbuf_flushlen
;
2893 if (connection_speaks_cells(conn
)) {
2894 /* use the per-conn write limit if it's lower, but if it's less
2895 * than zero just use zero */
2896 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
2897 if (conn
->state
== OR_CONN_STATE_OPEN
)
2898 if (or_conn
->write_bucket
< conn_bucket
)
2899 conn_bucket
= or_conn
->write_bucket
>= 0 ?
2900 or_conn
->write_bucket
: 0;
2901 base
= get_cell_network_size(or_conn
->wide_circ_ids
);
2904 if (connection_counts_as_relayed_traffic(conn
, now
) &&
2905 global_relayed_write_bucket
<= global_write_bucket
)
2906 global_bucket
= global_relayed_write_bucket
;
2908 return connection_bucket_round_robin(base
, priority
,
2909 global_bucket
, conn_bucket
);
2912 /** Return 1 if the global write buckets are low enough that we
2913 * shouldn't send <b>attempt</b> bytes of low-priority directory stuff
2914 * out to <b>conn</b>. Else return 0.
2916 * Priority was 1 for v1 requests (directories and running-routers),
2917 * and 2 for v2 requests and later (statuses and descriptors).
2919 * There are a lot of parameters we could use here:
2920 * - global_relayed_write_bucket. Low is bad.
2921 * - global_write_bucket. Low is bad.
2922 * - bandwidthrate. Low is bad.
2923 * - bandwidthburst. Not a big factor?
2924 * - attempt. High is bad.
2925 * - total bytes queued on outbufs. High is bad. But I'm wary of
2926 * using this, since a few slow-flushing queues will pump up the
2927 * number without meaning what we meant to mean. What we really
2928 * mean is "total directory bytes added to outbufs recently", but
2929 * that's harder to quantify and harder to keep track of.
2932 global_write_bucket_low(connection_t
*conn
, size_t attempt
, int priority
)
2934 int smaller_bucket
= global_write_bucket
< global_relayed_write_bucket
?
2935 global_write_bucket
: global_relayed_write_bucket
;
2936 if (authdir_mode(get_options()) && priority
>1)
2937 return 0; /* there's always room to answer v2 if we're an auth dir */
2939 if (!connection_is_rate_limited(conn
))
2940 return 0; /* local conns don't get limited */
2942 if (smaller_bucket
< (int)attempt
)
2943 return 1; /* not enough space no matter the priority */
2945 if (write_buckets_empty_last_second
)
2946 return 1; /* we're already hitting our limits, no more please */
2948 if (priority
== 1) { /* old-style v1 query */
2949 /* Could we handle *two* of these requests within the next two seconds? */
2950 const or_options_t
*options
= get_options();
2951 int64_t can_write
= (int64_t)smaller_bucket
2952 + 2*(options
->RelayBandwidthRate
? options
->RelayBandwidthRate
:
2953 options
->BandwidthRate
);
2954 if (can_write
< 2*(int64_t)attempt
)
2956 } else { /* v2 query */
2957 /* no further constraints yet */
2962 /** Helper: adjusts our bandwidth history and informs the controller as
2963 * appropriate, given that we have just read <b>num_read</b> bytes and written
2964 * <b>num_written</b> bytes on <b>conn</b>. */
2966 record_num_bytes_transferred_impl(connection_t
*conn
,
2967 time_t now
, size_t num_read
, size_t num_written
)
2969 /* Count bytes of answering direct and tunneled directory requests */
2970 if (conn
->type
== CONN_TYPE_DIR
&& conn
->purpose
== DIR_PURPOSE_SERVER
) {
2972 rep_hist_note_dir_bytes_read(num_read
, now
);
2973 if (num_written
> 0)
2974 rep_hist_note_dir_bytes_written(num_written
, now
);
2977 if (!connection_is_rate_limited(conn
))
2978 return; /* local IPs are free */
2980 if (conn
->type
== CONN_TYPE_OR
)
2981 rep_hist_note_or_conn_bytes(conn
->global_identifier
, num_read
,
2985 rep_hist_note_bytes_read(num_read
, now
);
2987 if (num_written
> 0) {
2988 rep_hist_note_bytes_written(num_written
, now
);
2990 if (conn
->type
== CONN_TYPE_EXIT
)
2991 rep_hist_note_exit_bytes(conn
->port
, num_written
, num_read
);
2994 /** Helper: convert given <b>tvnow</b> time value to milliseconds since
2997 msec_since_midnight(const struct timeval
*tvnow
)
2999 return (uint32_t)(((tvnow
->tv_sec
% 86400L) * 1000L) +
3000 ((uint32_t)tvnow
->tv_usec
/ (uint32_t)1000L));
3003 /** Helper: return the time in milliseconds since <b>last_empty_time</b>
3004 * when a bucket ran empty that previously had <b>tokens_before</b> tokens
3005 * now has <b>tokens_after</b> tokens after refilling at timestamp
3006 * <b>tvnow</b>, capped at <b>milliseconds_elapsed</b> milliseconds since
3007 * last refilling that bucket. Return 0 if the bucket has not been empty
3008 * since the last refill or has not been refilled. */
3010 bucket_millis_empty(int tokens_before
, uint32_t last_empty_time
,
3011 int tokens_after
, int milliseconds_elapsed
,
3012 const struct timeval
*tvnow
)
3014 uint32_t result
= 0, refilled
;
3015 if (tokens_before
<= 0 && tokens_after
> tokens_before
) {
3016 refilled
= msec_since_midnight(tvnow
);
3017 result
= (uint32_t)((refilled
+ 86400L * 1000L - last_empty_time
) %
3019 if (result
> (uint32_t)milliseconds_elapsed
)
3020 result
= (uint32_t)milliseconds_elapsed
;
3025 /** Check if a bucket which had <b>tokens_before</b> tokens and which got
3026 * <b>tokens_removed</b> tokens removed at timestamp <b>tvnow</b> has run
3027 * out of tokens, and if so, note the milliseconds since midnight in
3028 * <b>timestamp_var</b> for the next TB_EMPTY event. */
3030 connection_buckets_note_empty_ts(uint32_t *timestamp_var
,
3031 int tokens_before
, size_t tokens_removed
,
3032 const struct timeval
*tvnow
)
3034 if (tokens_before
> 0 && (uint32_t)tokens_before
<= tokens_removed
)
3035 *timestamp_var
= msec_since_midnight(tvnow
);
3038 /** Last time at which the global or relay buckets were emptied in msec
3039 * since midnight. */
3040 static uint32_t global_relayed_read_emptied
= 0,
3041 global_relayed_write_emptied
= 0,
3042 global_read_emptied
= 0,
3043 global_write_emptied
= 0;
3045 /** We just read <b>num_read</b> and wrote <b>num_written</b> bytes
3046 * onto <b>conn</b>. Decrement buckets appropriately. */
3048 connection_buckets_decrement(connection_t
*conn
, time_t now
,
3049 size_t num_read
, size_t num_written
)
3051 if (num_written
>= INT_MAX
|| num_read
>= INT_MAX
) {
3052 log_err(LD_BUG
, "Value out of range. num_read=%lu, num_written=%lu, "
3053 "connection type=%s, state=%s",
3054 (unsigned long)num_read
, (unsigned long)num_written
,
3055 conn_type_to_string(conn
->type
),
3056 conn_state_to_string(conn
->type
, conn
->state
));
3057 if (num_written
>= INT_MAX
) num_written
= 1;
3058 if (num_read
>= INT_MAX
) num_read
= 1;
3059 tor_fragile_assert();
3062 record_num_bytes_transferred_impl(conn
, now
, num_read
, num_written
);
3064 if (!connection_is_rate_limited(conn
))
3065 return; /* local IPs are free */
3067 /* If one or more of our token buckets ran dry just now, note the
3068 * timestamp for TB_EMPTY events. */
3069 if (get_options()->TestingEnableTbEmptyEvent
) {
3070 struct timeval tvnow
;
3071 tor_gettimeofday_cached(&tvnow
);
3072 if (connection_counts_as_relayed_traffic(conn
, now
)) {
3073 connection_buckets_note_empty_ts(&global_relayed_read_emptied
,
3074 global_relayed_read_bucket
, num_read
, &tvnow
);
3075 connection_buckets_note_empty_ts(&global_relayed_write_emptied
,
3076 global_relayed_write_bucket
, num_written
, &tvnow
);
3078 connection_buckets_note_empty_ts(&global_read_emptied
,
3079 global_read_bucket
, num_read
, &tvnow
);
3080 connection_buckets_note_empty_ts(&global_write_emptied
,
3081 global_write_bucket
, num_written
, &tvnow
);
3082 if (connection_speaks_cells(conn
) && conn
->state
== OR_CONN_STATE_OPEN
) {
3083 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
3084 connection_buckets_note_empty_ts(&or_conn
->read_emptied_time
,
3085 or_conn
->read_bucket
, num_read
, &tvnow
);
3086 connection_buckets_note_empty_ts(&or_conn
->write_emptied_time
,
3087 or_conn
->write_bucket
, num_written
, &tvnow
);
3091 if (connection_counts_as_relayed_traffic(conn
, now
)) {
3092 global_relayed_read_bucket
-= (int)num_read
;
3093 global_relayed_write_bucket
-= (int)num_written
;
3095 global_read_bucket
-= (int)num_read
;
3096 global_write_bucket
-= (int)num_written
;
3097 if (connection_speaks_cells(conn
) && conn
->state
== OR_CONN_STATE_OPEN
) {
3098 TO_OR_CONN(conn
)->read_bucket
-= (int)num_read
;
3099 TO_OR_CONN(conn
)->write_bucket
-= (int)num_written
;
3103 /** If we have exhausted our global buckets, or the buckets for conn,
3106 connection_consider_empty_read_buckets(connection_t
*conn
)
3110 if (!connection_is_rate_limited(conn
))
3111 return; /* Always okay. */
3113 if (global_read_bucket
<= 0) {
3114 reason
= "global read bucket exhausted. Pausing.";
3115 } else if (connection_counts_as_relayed_traffic(conn
, approx_time()) &&
3116 global_relayed_read_bucket
<= 0) {
3117 reason
= "global relayed read bucket exhausted. Pausing.";
3118 } else if (connection_speaks_cells(conn
) &&
3119 conn
->state
== OR_CONN_STATE_OPEN
&&
3120 TO_OR_CONN(conn
)->read_bucket
<= 0) {
3121 reason
= "connection read bucket exhausted. Pausing.";
3123 return; /* all good, no need to stop it */
3125 LOG_FN_CONN(conn
, (LOG_DEBUG
, LD_NET
, "%s", reason
));
3126 conn
->read_blocked_on_bw
= 1;
3127 connection_stop_reading(conn
);
3130 /** If we have exhausted our global buckets, or the buckets for conn,
3133 connection_consider_empty_write_buckets(connection_t
*conn
)
3137 if (!connection_is_rate_limited(conn
))
3138 return; /* Always okay. */
3140 if (global_write_bucket
<= 0) {
3141 reason
= "global write bucket exhausted. Pausing.";
3142 } else if (connection_counts_as_relayed_traffic(conn
, approx_time()) &&
3143 global_relayed_write_bucket
<= 0) {
3144 reason
= "global relayed write bucket exhausted. Pausing.";
3145 } else if (connection_speaks_cells(conn
) &&
3146 conn
->state
== OR_CONN_STATE_OPEN
&&
3147 TO_OR_CONN(conn
)->write_bucket
<= 0) {
3148 reason
= "connection write bucket exhausted. Pausing.";
3150 return; /* all good, no need to stop it */
3152 LOG_FN_CONN(conn
, (LOG_DEBUG
, LD_NET
, "%s", reason
));
3153 conn
->write_blocked_on_bw
= 1;
3154 connection_stop_writing(conn
);
3157 /** Initialize the global read bucket to options-\>BandwidthBurst. */
3159 connection_bucket_init(void)
3161 const or_options_t
*options
= get_options();
3162 /* start it at max traffic */
3163 global_read_bucket
= (int)options
->BandwidthBurst
;
3164 global_write_bucket
= (int)options
->BandwidthBurst
;
3165 if (options
->RelayBandwidthRate
) {
3166 global_relayed_read_bucket
= (int)options
->RelayBandwidthBurst
;
3167 global_relayed_write_bucket
= (int)options
->RelayBandwidthBurst
;
3169 global_relayed_read_bucket
= (int)options
->BandwidthBurst
;
3170 global_relayed_write_bucket
= (int)options
->BandwidthBurst
;
3174 /** Refill a single <b>bucket</b> called <b>name</b> with bandwidth rate per
3175 * second <b>rate</b> and bandwidth burst <b>burst</b>, assuming that
3176 * <b>milliseconds_elapsed</b> milliseconds have passed since the last
3179 connection_bucket_refill_helper(int *bucket
, int rate
, int burst
,
3180 int milliseconds_elapsed
,
3183 int starting_bucket
= *bucket
;
3184 if (starting_bucket
< burst
&& milliseconds_elapsed
> 0) {
3185 int64_t incr
= (((int64_t)rate
) * milliseconds_elapsed
) / 1000;
3186 if ((burst
- starting_bucket
) < incr
) {
3187 *bucket
= burst
; /* We would overflow the bucket; just set it to
3190 *bucket
+= (int)incr
;
3191 if (*bucket
> burst
|| *bucket
< starting_bucket
) {
3192 /* If we overflow the burst, or underflow our starting bucket,
3193 * cap the bucket value to burst. */
3194 /* XXXX this might be redundant now, but it doesn't show up
3195 * in profiles. Remove it after analysis. */
3199 log_debug(LD_NET
,"%s now %d.", name
, *bucket
);
3203 /** Time has passed; increment buckets appropriately. */
3205 connection_bucket_refill(int milliseconds_elapsed
, time_t now
)
3207 const or_options_t
*options
= get_options();
3208 smartlist_t
*conns
= get_connection_array();
3209 int bandwidthrate
, bandwidthburst
, relayrate
, relayburst
;
3211 int prev_global_read
= global_read_bucket
;
3212 int prev_global_write
= global_write_bucket
;
3213 int prev_relay_read
= global_relayed_read_bucket
;
3214 int prev_relay_write
= global_relayed_write_bucket
;
3215 struct timeval tvnow
; /*< Only used if TB_EMPTY events are enabled. */
3217 bandwidthrate
= (int)options
->BandwidthRate
;
3218 bandwidthburst
= (int)options
->BandwidthBurst
;
3220 if (options
->RelayBandwidthRate
) {
3221 relayrate
= (int)options
->RelayBandwidthRate
;
3222 relayburst
= (int)options
->RelayBandwidthBurst
;
3224 relayrate
= bandwidthrate
;
3225 relayburst
= bandwidthburst
;
3228 tor_assert(milliseconds_elapsed
>= 0);
3230 write_buckets_empty_last_second
=
3231 global_relayed_write_bucket
<= 0 || global_write_bucket
<= 0;
3233 /* refill the global buckets */
3234 connection_bucket_refill_helper(&global_read_bucket
,
3235 bandwidthrate
, bandwidthburst
,
3236 milliseconds_elapsed
,
3237 "global_read_bucket");
3238 connection_bucket_refill_helper(&global_write_bucket
,
3239 bandwidthrate
, bandwidthburst
,
3240 milliseconds_elapsed
,
3241 "global_write_bucket");
3242 connection_bucket_refill_helper(&global_relayed_read_bucket
,
3243 relayrate
, relayburst
,
3244 milliseconds_elapsed
,
3245 "global_relayed_read_bucket");
3246 connection_bucket_refill_helper(&global_relayed_write_bucket
,
3247 relayrate
, relayburst
,
3248 milliseconds_elapsed
,
3249 "global_relayed_write_bucket");
3251 /* If buckets were empty before and have now been refilled, tell any
3252 * interested controllers. */
3253 if (get_options()->TestingEnableTbEmptyEvent
) {
3254 uint32_t global_read_empty_time
, global_write_empty_time
,
3255 relay_read_empty_time
, relay_write_empty_time
;
3256 tor_gettimeofday_cached(&tvnow
);
3257 global_read_empty_time
= bucket_millis_empty(prev_global_read
,
3258 global_read_emptied
, global_read_bucket
,
3259 milliseconds_elapsed
, &tvnow
);
3260 global_write_empty_time
= bucket_millis_empty(prev_global_write
,
3261 global_write_emptied
, global_write_bucket
,
3262 milliseconds_elapsed
, &tvnow
);
3263 control_event_tb_empty("GLOBAL", global_read_empty_time
,
3264 global_write_empty_time
, milliseconds_elapsed
);
3265 relay_read_empty_time
= bucket_millis_empty(prev_relay_read
,
3266 global_relayed_read_emptied
,
3267 global_relayed_read_bucket
,
3268 milliseconds_elapsed
, &tvnow
);
3269 relay_write_empty_time
= bucket_millis_empty(prev_relay_write
,
3270 global_relayed_write_emptied
,
3271 global_relayed_write_bucket
,
3272 milliseconds_elapsed
, &tvnow
);
3273 control_event_tb_empty("RELAY", relay_read_empty_time
,
3274 relay_write_empty_time
, milliseconds_elapsed
);
3277 /* refill the per-connection buckets */
3278 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, conn
) {
3279 if (connection_speaks_cells(conn
)) {
3280 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
3281 int orbandwidthrate
= or_conn
->bandwidthrate
;
3282 int orbandwidthburst
= or_conn
->bandwidthburst
;
3284 int prev_conn_read
= or_conn
->read_bucket
;
3285 int prev_conn_write
= or_conn
->write_bucket
;
3287 if (connection_bucket_should_increase(or_conn
->read_bucket
, or_conn
)) {
3288 connection_bucket_refill_helper(&or_conn
->read_bucket
,
3291 milliseconds_elapsed
,
3292 "or_conn->read_bucket");
3294 if (connection_bucket_should_increase(or_conn
->write_bucket
, or_conn
)) {
3295 connection_bucket_refill_helper(&or_conn
->write_bucket
,
3298 milliseconds_elapsed
,
3299 "or_conn->write_bucket");
3302 /* If buckets were empty before and have now been refilled, tell any
3303 * interested controllers. */
3304 if (get_options()->TestingEnableTbEmptyEvent
) {
3306 uint32_t conn_read_empty_time
, conn_write_empty_time
;
3307 tor_asprintf(&bucket
, "ORCONN ID="U64_FORMAT
,
3308 U64_PRINTF_ARG(or_conn
->base_
.global_identifier
));
3309 conn_read_empty_time
= bucket_millis_empty(prev_conn_read
,
3310 or_conn
->read_emptied_time
,
3311 or_conn
->read_bucket
,
3312 milliseconds_elapsed
, &tvnow
);
3313 conn_write_empty_time
= bucket_millis_empty(prev_conn_write
,
3314 or_conn
->write_emptied_time
,
3315 or_conn
->write_bucket
,
3316 milliseconds_elapsed
, &tvnow
);
3317 control_event_tb_empty(bucket
, conn_read_empty_time
,
3318 conn_write_empty_time
,
3319 milliseconds_elapsed
);
3324 if (conn
->read_blocked_on_bw
== 1 /* marked to turn reading back on now */
3325 && global_read_bucket
> 0 /* and we're allowed to read */
3326 && (!connection_counts_as_relayed_traffic(conn
, now
) ||
3327 global_relayed_read_bucket
> 0) /* even if we're relayed traffic */
3328 && (!connection_speaks_cells(conn
) ||
3329 conn
->state
!= OR_CONN_STATE_OPEN
||
3330 TO_OR_CONN(conn
)->read_bucket
> 0)) {
3331 /* and either a non-cell conn or a cell conn with non-empty bucket */
3332 LOG_FN_CONN(conn
, (LOG_DEBUG
,LD_NET
,
3333 "waking up conn (fd %d) for read", (int)conn
->s
));
3334 conn
->read_blocked_on_bw
= 0;
3335 connection_start_reading(conn
);
3338 if (conn
->write_blocked_on_bw
== 1
3339 && global_write_bucket
> 0 /* and we're allowed to write */
3340 && (!connection_counts_as_relayed_traffic(conn
, now
) ||
3341 global_relayed_write_bucket
> 0) /* even if it's relayed traffic */
3342 && (!connection_speaks_cells(conn
) ||
3343 conn
->state
!= OR_CONN_STATE_OPEN
||
3344 TO_OR_CONN(conn
)->write_bucket
> 0)) {
3345 LOG_FN_CONN(conn
, (LOG_DEBUG
,LD_NET
,
3346 "waking up conn (fd %d) for write", (int)conn
->s
));
3347 conn
->write_blocked_on_bw
= 0;
3348 connection_start_writing(conn
);
3350 } SMARTLIST_FOREACH_END(conn
);
3353 /** Is the <b>bucket</b> for connection <b>conn</b> low enough that we
3354 * should add another pile of tokens to it?
3357 connection_bucket_should_increase(int bucket
, or_connection_t
*conn
)
3361 if (conn
->base_
.state
!= OR_CONN_STATE_OPEN
)
3362 return 0; /* only open connections play the rate limiting game */
3363 if (bucket
>= conn
->bandwidthburst
)
3369 /** Read bytes from conn-\>s and process them.
3371 * It calls connection_read_to_buf() to bring in any new bytes,
3372 * and then calls connection_process_inbuf() to process them.
3374 * Mark the connection and return -1 if you want to close it, else
3378 connection_handle_read_impl(connection_t
*conn
)
3380 ssize_t max_to_read
=-1, try_to_read
;
3381 size_t before
, n_read
= 0;
3382 int socket_error
= 0;
3384 if (conn
->marked_for_close
)
3385 return 0; /* do nothing */
3387 conn
->timestamp_lastread
= approx_time();
3389 switch (conn
->type
) {
3390 case CONN_TYPE_OR_LISTENER
:
3391 return connection_handle_listener_read(conn
, CONN_TYPE_OR
);
3392 case CONN_TYPE_EXT_OR_LISTENER
:
3393 return connection_handle_listener_read(conn
, CONN_TYPE_EXT_OR
);
3394 case CONN_TYPE_AP_LISTENER
:
3395 case CONN_TYPE_AP_TRANS_LISTENER
:
3396 case CONN_TYPE_AP_NATD_LISTENER
:
3397 return connection_handle_listener_read(conn
, CONN_TYPE_AP
);
3398 case CONN_TYPE_DIR_LISTENER
:
3399 return connection_handle_listener_read(conn
, CONN_TYPE_DIR
);
3400 case CONN_TYPE_CONTROL_LISTENER
:
3401 return connection_handle_listener_read(conn
, CONN_TYPE_CONTROL
);
3402 case CONN_TYPE_AP_DNS_LISTENER
:
3403 /* This should never happen; eventdns.c handles the reads here. */
3404 tor_fragile_assert();
3409 try_to_read
= max_to_read
;
3410 tor_assert(!conn
->marked_for_close
);
3412 before
= buf_datalen(conn
->inbuf
);
3413 if (connection_read_to_buf(conn
, &max_to_read
, &socket_error
) < 0) {
3414 /* There's a read error; kill the connection.*/
3415 if (conn
->type
== CONN_TYPE_OR
) {
3416 connection_or_notify_error(TO_OR_CONN(conn
),
3418 errno_to_orconn_end_reason(socket_error
) :
3419 END_OR_CONN_REASON_CONNRESET
,
3421 tor_socket_strerror(socket_error
) :
3422 "(unknown, errno was 0)");
3424 if (CONN_IS_EDGE(conn
)) {
3425 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
3426 connection_edge_end_errno(edge_conn
);
3427 if (conn
->type
== CONN_TYPE_AP
&& TO_ENTRY_CONN(conn
)->socks_request
) {
3428 /* broken, don't send a socks reply back */
3429 TO_ENTRY_CONN(conn
)->socks_request
->has_finished
= 1;
3432 connection_close_immediate(conn
); /* Don't flush; connection is dead. */
3434 * This can bypass normal channel checking since we did
3435 * connection_or_notify_error() above.
3437 connection_mark_for_close_internal(conn
);
3440 n_read
+= buf_datalen(conn
->inbuf
) - before
;
3441 if (CONN_IS_EDGE(conn
) && try_to_read
!= max_to_read
) {
3442 /* instruct it not to try to package partial cells. */
3443 if (connection_process_inbuf(conn
, 0) < 0) {
3446 if (!conn
->marked_for_close
&&
3447 connection_is_reading(conn
) &&
3448 !conn
->inbuf_reached_eof
&&
3450 goto loop_again
; /* try reading again, in case more is here now */
3452 /* one last try, packaging partial cells and all. */
3453 if (!conn
->marked_for_close
&&
3454 connection_process_inbuf(conn
, 1) < 0) {
3457 if (conn
->linked_conn
) {
3458 /* The other side's handle_write() will never actually get called, so
3459 * we need to invoke the appropriate callbacks ourself. */
3460 connection_t
*linked
= conn
->linked_conn
;
3463 /* Probably a no-op, since linked conns typically don't count for
3464 * bandwidth rate limiting. But do it anyway so we can keep stats
3465 * accurately. Note that since we read the bytes from conn, and
3466 * we're writing the bytes onto the linked connection, we count
3467 * these as <i>written</i> bytes. */
3468 connection_buckets_decrement(linked
, approx_time(), 0, n_read
);
3470 if (connection_flushed_some(linked
) < 0)
3471 connection_mark_for_close(linked
);
3472 if (!connection_wants_to_flush(linked
))
3473 connection_finished_flushing(linked
);
3476 if (!buf_datalen(linked
->outbuf
) && conn
->active_on_link
)
3477 connection_stop_reading_from_linked_conn(conn
);
3479 /* If we hit the EOF, call connection_reached_eof(). */
3480 if (!conn
->marked_for_close
&&
3481 conn
->inbuf_reached_eof
&&
3482 connection_reached_eof(conn
) < 0) {
3488 /* DOCDOC connection_handle_read */
3490 connection_handle_read(connection_t
*conn
)
3494 tor_gettimeofday_cache_clear();
3495 res
= connection_handle_read_impl(conn
);
3499 /** Pull in new bytes from conn-\>s or conn-\>linked_conn onto conn-\>inbuf,
3500 * either directly or via TLS. Reduce the token buckets by the number of bytes
3503 * If *max_to_read is -1, then decide it ourselves, else go with the
3504 * value passed to us. When returning, if it's changed, subtract the
3505 * number of bytes we read from *max_to_read.
3507 * Return -1 if we want to break conn, else return 0.
3510 connection_read_to_buf(connection_t
*conn
, ssize_t
*max_to_read
,
3514 ssize_t at_most
= *max_to_read
;
3515 size_t slack_in_buf
, more_to_read
;
3516 size_t n_read
= 0, n_written
= 0;
3518 if (at_most
== -1) { /* we need to initialize it */
3519 /* how many bytes are we allowed to read? */
3520 at_most
= connection_bucket_read_limit(conn
, approx_time());
3523 slack_in_buf
= buf_slack(conn
->inbuf
);
3525 if ((size_t)at_most
> slack_in_buf
&& slack_in_buf
>= 1024) {
3526 more_to_read
= at_most
- slack_in_buf
;
3527 at_most
= slack_in_buf
;
3532 if (connection_speaks_cells(conn
) &&
3533 conn
->state
> OR_CONN_STATE_PROXY_HANDSHAKING
) {
3535 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
3536 size_t initial_size
;
3537 if (conn
->state
== OR_CONN_STATE_TLS_HANDSHAKING
||
3538 conn
->state
== OR_CONN_STATE_TLS_CLIENT_RENEGOTIATING
) {
3539 /* continue handshaking even if global token bucket is empty */
3540 return connection_tls_continue_handshake(or_conn
);
3544 "%d: starting, inbuf_datalen %ld (%d pending in tls object)."
3546 (int)conn
->s
,(long)buf_datalen(conn
->inbuf
),
3547 tor_tls_get_pending_bytes(or_conn
->tls
), (long)at_most
);
3549 initial_size
= buf_datalen(conn
->inbuf
);
3550 /* else open, or closing */
3551 result
= read_to_buf_tls(or_conn
->tls
, at_most
, conn
->inbuf
);
3552 if (TOR_TLS_IS_ERROR(result
) || result
== TOR_TLS_CLOSE
)
3553 or_conn
->tls_error
= result
;
3555 or_conn
->tls_error
= 0;
3559 case TOR_TLS_ERROR_IO
:
3560 log_debug(LD_NET
,"TLS connection closed %son read. Closing. "
3561 "(Nickname %s, address %s)",
3562 result
== TOR_TLS_CLOSE
? "cleanly " : "",
3563 or_conn
->nickname
? or_conn
->nickname
: "not set",
3566 CASE_TOR_TLS_ERROR_ANY_NONIO
:
3567 log_debug(LD_NET
,"tls error [%s]. breaking (nickname %s, address %s).",
3568 tor_tls_err_to_string(result
),
3569 or_conn
->nickname
? or_conn
->nickname
: "not set",
3572 case TOR_TLS_WANTWRITE
:
3573 connection_start_writing(conn
);
3575 case TOR_TLS_WANTREAD
:
3576 if (conn
->in_connection_handle_write
) {
3577 /* We've been invoked from connection_handle_write, because we're
3578 * waiting for a TLS renegotiation, the renegotiation started, and
3579 * SSL_read returned WANTWRITE. But now SSL_read is saying WANTREAD
3580 * again. Stop waiting for write events now, or else we'll
3581 * busy-loop until data arrives for us to read. */
3582 connection_stop_writing(conn
);
3583 if (!connection_is_reading(conn
))
3584 connection_start_reading(conn
);
3586 /* we're already reading, one hopes */
3589 case TOR_TLS_DONE
: /* no data read, so nothing to process */
3591 break; /* so we call bucket_decrement below */
3595 pending
= tor_tls_get_pending_bytes(or_conn
->tls
);
3597 /* If we have any pending bytes, we read them now. This *can*
3598 * take us over our read allotment, but really we shouldn't be
3599 * believing that SSL bytes are the same as TCP bytes anyway. */
3600 int r2
= read_to_buf_tls(or_conn
->tls
, pending
, conn
->inbuf
);
3602 log_warn(LD_BUG
, "apparently, reading pending bytes can fail.");
3606 result
= (int)(buf_datalen(conn
->inbuf
)-initial_size
);
3607 tor_tls_get_n_raw_bytes(or_conn
->tls
, &n_read
, &n_written
);
3608 log_debug(LD_GENERAL
, "After TLS read of %d: %ld read, %ld written",
3609 result
, (long)n_read
, (long)n_written
);
3610 } else if (conn
->linked
) {
3611 if (conn
->linked_conn
) {
3612 result
= move_buf_to_buf(conn
->inbuf
, conn
->linked_conn
->outbuf
,
3613 &conn
->linked_conn
->outbuf_flushlen
);
3617 //log_notice(LD_GENERAL, "Moved %d bytes on an internal link!", result);
3618 /* If the other side has disappeared, or if it's been marked for close and
3619 * we flushed its outbuf, then we should set our inbuf_reached_eof. */
3620 if (!conn
->linked_conn
||
3621 (conn
->linked_conn
->marked_for_close
&&
3622 buf_datalen(conn
->linked_conn
->outbuf
) == 0))
3623 conn
->inbuf_reached_eof
= 1;
3625 n_read
= (size_t) result
;
3627 /* !connection_speaks_cells, !conn->linked_conn. */
3628 int reached_eof
= 0;
3629 CONN_LOG_PROTECT(conn
,
3630 result
= read_to_buf(conn
->s
, at_most
, conn
->inbuf
, &reached_eof
,
3633 conn
->inbuf_reached_eof
= 1;
3635 // log_fn(LOG_DEBUG,"read_to_buf returned %d.",read_result);
3639 n_read
= (size_t) result
;
3643 /* change *max_to_read */
3644 *max_to_read
= at_most
- n_read
;
3646 /* Update edge_conn->n_read and ocirc->n_read_circ_bw */
3647 if (conn
->type
== CONN_TYPE_AP
) {
3648 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
3649 circuit_t
*circ
= circuit_get_by_edge_conn(edge_conn
);
3650 origin_circuit_t
*ocirc
;
3652 /* Check for overflow: */
3653 if (PREDICT_LIKELY(UINT32_MAX
- edge_conn
->n_read
> n_read
))
3654 edge_conn
->n_read
+= (int)n_read
;
3656 edge_conn
->n_read
= UINT32_MAX
;
3658 if (circ
&& CIRCUIT_IS_ORIGIN(circ
)) {
3659 ocirc
= TO_ORIGIN_CIRCUIT(circ
);
3660 if (PREDICT_LIKELY(UINT32_MAX
- ocirc
->n_read_circ_bw
> n_read
))
3661 ocirc
->n_read_circ_bw
+= (int)n_read
;
3663 ocirc
->n_read_circ_bw
= UINT32_MAX
;
3667 /* If CONN_BW events are enabled, update conn->n_read_conn_bw for
3668 * OR/DIR/EXIT connections, checking for overflow. */
3669 if (get_options()->TestingEnableConnBwEvent
&&
3670 (conn
->type
== CONN_TYPE_OR
||
3671 conn
->type
== CONN_TYPE_DIR
||
3672 conn
->type
== CONN_TYPE_EXIT
)) {
3673 if (PREDICT_LIKELY(UINT32_MAX
- conn
->n_read_conn_bw
> n_read
))
3674 conn
->n_read_conn_bw
+= (int)n_read
;
3676 conn
->n_read_conn_bw
= UINT32_MAX
;
3680 connection_buckets_decrement(conn
, approx_time(), n_read
, n_written
);
3682 if (more_to_read
&& result
== at_most
) {
3683 slack_in_buf
= buf_slack(conn
->inbuf
);
3684 at_most
= more_to_read
;
3688 /* Call even if result is 0, since the global read bucket may
3689 * have reached 0 on a different conn, and this connection needs to
3690 * know to stop reading. */
3691 connection_consider_empty_read_buckets(conn
);
3692 if (n_written
> 0 && connection_is_writing(conn
))
3693 connection_consider_empty_write_buckets(conn
);
3698 /** A pass-through to fetch_from_buf. */
3700 connection_fetch_from_buf(char *string
, size_t len
, connection_t
*conn
)
3702 return fetch_from_buf(string
, len
, conn
->inbuf
);
3705 /** As fetch_from_buf_line(), but read from a connection's input buffer. */
3707 connection_fetch_from_buf_line(connection_t
*conn
, char *data
,
3710 return fetch_from_buf_line(conn
->inbuf
, data
, data_len
);
3713 /** As fetch_from_buf_http, but fetches from a connection's input buffer_t as
3716 connection_fetch_from_buf_http(connection_t
*conn
,
3717 char **headers_out
, size_t max_headerlen
,
3718 char **body_out
, size_t *body_used
,
3719 size_t max_bodylen
, int force_complete
)
3721 return fetch_from_buf_http(conn
->inbuf
, headers_out
, max_headerlen
,
3722 body_out
, body_used
, max_bodylen
, force_complete
);
3725 /** Return conn-\>outbuf_flushlen: how many bytes conn wants to flush
3726 * from its outbuf. */
3728 connection_wants_to_flush(connection_t
*conn
)
3730 return conn
->outbuf_flushlen
> 0;
3733 /** Are there too many bytes on edge connection <b>conn</b>'s outbuf to
3734 * send back a relay-level sendme yet? Return 1 if so, 0 if not. Used by
3735 * connection_edge_consider_sending_sendme().
3738 connection_outbuf_too_full(connection_t
*conn
)
3740 return (conn
->outbuf_flushlen
> 10*CELL_PAYLOAD_SIZE
);
3743 /** Try to flush more bytes onto <b>conn</b>-\>s.
3745 * This function gets called either from conn_write_callback() in main.c
3746 * when libevent tells us that conn wants to write, or below
3747 * from connection_write_to_buf() when an entire TLS record is ready.
3749 * Update <b>conn</b>-\>timestamp_lastwritten to now, and call flush_buf
3750 * or flush_buf_tls appropriately. If it succeeds and there are no more
3751 * more bytes on <b>conn</b>-\>outbuf, then call connection_finished_flushing
3754 * If <b>force</b>, then write as many bytes as possible, ignoring bandwidth
3755 * limits. (Used for flushing messages to controller connections on fatal
3758 * Mark the connection and return -1 if you want to close it, else
3762 connection_handle_write_impl(connection_t
*conn
, int force
)
3765 socklen_t len
=(socklen_t
)sizeof(e
);
3767 ssize_t max_to_write
;
3768 time_t now
= approx_time();
3769 size_t n_read
= 0, n_written
= 0;
3770 int dont_stop_writing
= 0;
3772 tor_assert(!connection_is_listener(conn
));
3774 if (conn
->marked_for_close
|| !SOCKET_OK(conn
->s
))
3775 return 0; /* do nothing */
3777 if (conn
->in_flushed_some
) {
3778 log_warn(LD_BUG
, "called recursively from inside conn->in_flushed_some");
3782 conn
->timestamp_lastwritten
= now
;
3784 /* Sometimes, "writable" means "connected". */
3785 if (connection_state_is_connecting(conn
)) {
3786 if (getsockopt(conn
->s
, SOL_SOCKET
, SO_ERROR
, (void*)&e
, &len
) < 0) {
3787 log_warn(LD_BUG
, "getsockopt() syscall failed");
3788 if (conn
->type
== CONN_TYPE_OR
) {
3789 or_connection_t
*orconn
= TO_OR_CONN(conn
);
3790 connection_or_close_for_error(orconn
, 0);
3792 if (CONN_IS_EDGE(conn
)) {
3793 connection_edge_end_errno(TO_EDGE_CONN(conn
));
3795 connection_mark_for_close(conn
);
3800 /* some sort of error, but maybe just inprogress still */
3801 if (!ERRNO_IS_CONN_EINPROGRESS(e
)) {
3802 log_info(LD_NET
,"in-progress connect failed. Removing. (%s)",
3803 tor_socket_strerror(e
));
3804 if (CONN_IS_EDGE(conn
))
3805 connection_edge_end_errno(TO_EDGE_CONN(conn
));
3806 if (conn
->type
== CONN_TYPE_OR
)
3807 connection_or_notify_error(TO_OR_CONN(conn
),
3808 errno_to_orconn_end_reason(e
),
3809 tor_socket_strerror(e
));
3811 connection_close_immediate(conn
);
3813 * This can bypass normal channel checking since we did
3814 * connection_or_notify_error() above.
3816 connection_mark_for_close_internal(conn
);
3819 return 0; /* no change, see if next time is better */
3822 /* The connection is successful. */
3823 if (connection_finished_connecting(conn
)<0)
3827 max_to_write
= force
? (ssize_t
)conn
->outbuf_flushlen
3828 : connection_bucket_write_limit(conn
, now
);
3830 if (connection_speaks_cells(conn
) &&
3831 conn
->state
> OR_CONN_STATE_PROXY_HANDSHAKING
) {
3832 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
3833 size_t initial_size
;
3834 if (conn
->state
== OR_CONN_STATE_TLS_HANDSHAKING
||
3835 conn
->state
== OR_CONN_STATE_TLS_CLIENT_RENEGOTIATING
) {
3836 connection_stop_writing(conn
);
3837 if (connection_tls_continue_handshake(or_conn
) < 0) {
3838 /* Don't flush; connection is dead. */
3839 connection_or_notify_error(or_conn
,
3840 END_OR_CONN_REASON_MISC
,
3841 "TLS error in connection_tls_"
3842 "continue_handshake()");
3843 connection_close_immediate(conn
);
3845 * This can bypass normal channel checking since we did
3846 * connection_or_notify_error() above.
3848 connection_mark_for_close_internal(conn
);
3852 } else if (conn
->state
== OR_CONN_STATE_TLS_SERVER_RENEGOTIATING
) {
3853 return connection_handle_read(conn
);
3856 /* else open, or closing */
3857 initial_size
= buf_datalen(conn
->outbuf
);
3858 result
= flush_buf_tls(or_conn
->tls
, conn
->outbuf
,
3859 max_to_write
, &conn
->outbuf_flushlen
);
3861 /* If we just flushed the last bytes, tell the channel on the
3862 * or_conn to check if it needs to geoip_change_dirreq_state() */
3863 /* XXXX move this to flushed_some or finished_flushing -NM */
3864 if (buf_datalen(conn
->outbuf
) == 0 && or_conn
->chan
)
3865 channel_notify_flushed(TLS_CHAN_TO_BASE(or_conn
->chan
));
3868 CASE_TOR_TLS_ERROR_ANY
:
3870 log_info(LD_NET
, result
!= TOR_TLS_CLOSE
?
3871 "tls error. breaking.":"TLS connection closed on flush");
3872 /* Don't flush; connection is dead. */
3873 connection_or_notify_error(or_conn
,
3874 END_OR_CONN_REASON_MISC
,
3875 result
!= TOR_TLS_CLOSE
?
3876 "TLS error in during flush" :
3877 "TLS closed during flush");
3878 connection_close_immediate(conn
);
3880 * This can bypass normal channel checking since we did
3881 * connection_or_notify_error() above.
3883 connection_mark_for_close_internal(conn
);
3885 case TOR_TLS_WANTWRITE
:
3886 log_debug(LD_NET
,"wanted write.");
3887 /* we're already writing */
3888 dont_stop_writing
= 1;
3890 case TOR_TLS_WANTREAD
:
3891 /* Make sure to avoid a loop if the receive buckets are empty. */
3892 log_debug(LD_NET
,"wanted read.");
3893 if (!connection_is_reading(conn
)) {
3894 connection_stop_writing(conn
);
3895 conn
->write_blocked_on_bw
= 1;
3896 /* we'll start reading again when we get more tokens in our
3897 * read bucket; then we'll start writing again too.
3900 /* else no problem, we're already reading */
3902 /* case TOR_TLS_DONE:
3903 * for TOR_TLS_DONE, fall through to check if the flushlen
3904 * is empty, so we can stop writing.
3908 tor_tls_get_n_raw_bytes(or_conn
->tls
, &n_read
, &n_written
);
3909 log_debug(LD_GENERAL
, "After TLS write of %d: %ld read, %ld written",
3910 result
, (long)n_read
, (long)n_written
);
3911 or_conn
->bytes_xmitted
+= result
;
3912 or_conn
->bytes_xmitted_by_tls
+= n_written
;
3913 /* So we notice bytes were written even on error */
3914 /* XXXX This cast is safe since we can never write INT_MAX bytes in a
3915 * single set of TLS operations. But it looks kinda ugly. If we refactor
3916 * the *_buf_tls functions, we should make them return ssize_t or size_t
3918 result
= (int)(initial_size
-buf_datalen(conn
->outbuf
));
3920 CONN_LOG_PROTECT(conn
,
3921 result
= flush_buf(conn
->s
, conn
->outbuf
,
3922 max_to_write
, &conn
->outbuf_flushlen
));
3924 if (CONN_IS_EDGE(conn
))
3925 connection_edge_end_errno(TO_EDGE_CONN(conn
));
3926 if (conn
->type
== CONN_TYPE_AP
) {
3927 /* writing failed; we couldn't send a SOCKS reply if we wanted to */
3928 TO_ENTRY_CONN(conn
)->socks_request
->has_finished
= 1;
3931 connection_close_immediate(conn
); /* Don't flush; connection is dead. */
3932 connection_mark_for_close(conn
);
3935 n_written
= (size_t) result
;
3938 if (n_written
&& conn
->type
== CONN_TYPE_AP
) {
3939 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
3940 circuit_t
*circ
= circuit_get_by_edge_conn(edge_conn
);
3941 origin_circuit_t
*ocirc
;
3943 /* Check for overflow: */
3944 if (PREDICT_LIKELY(UINT32_MAX
- edge_conn
->n_written
> n_written
))
3945 edge_conn
->n_written
+= (int)n_written
;
3947 edge_conn
->n_written
= UINT32_MAX
;
3949 if (circ
&& CIRCUIT_IS_ORIGIN(circ
)) {
3950 ocirc
= TO_ORIGIN_CIRCUIT(circ
);
3951 if (PREDICT_LIKELY(UINT32_MAX
- ocirc
->n_written_circ_bw
> n_written
))
3952 ocirc
->n_written_circ_bw
+= (int)n_written
;
3954 ocirc
->n_written_circ_bw
= UINT32_MAX
;
3958 /* If CONN_BW events are enabled, update conn->n_written_conn_bw for
3959 * OR/DIR/EXIT connections, checking for overflow. */
3960 if (n_written
&& get_options()->TestingEnableConnBwEvent
&&
3961 (conn
->type
== CONN_TYPE_OR
||
3962 conn
->type
== CONN_TYPE_DIR
||
3963 conn
->type
== CONN_TYPE_EXIT
)) {
3964 if (PREDICT_LIKELY(UINT32_MAX
- conn
->n_written_conn_bw
> n_written
))
3965 conn
->n_written_conn_bw
+= (int)n_written
;
3967 conn
->n_written_conn_bw
= UINT32_MAX
;
3970 connection_buckets_decrement(conn
, approx_time(), n_read
, n_written
);
3973 /* If we wrote any bytes from our buffer, then call the appropriate
3975 if (connection_flushed_some(conn
) < 0) {
3976 if (connection_speaks_cells(conn
)) {
3977 connection_or_notify_error(TO_OR_CONN(conn
),
3978 END_OR_CONN_REASON_MISC
,
3979 "Got error back from "
3980 "connection_flushed_some()");
3984 * This can bypass normal channel checking since we did
3985 * connection_or_notify_error() above.
3987 connection_mark_for_close_internal(conn
);
3991 if (!connection_wants_to_flush(conn
) &&
3992 !dont_stop_writing
) { /* it's done flushing */
3993 if (connection_finished_flushing(conn
) < 0) {
3994 /* already marked */
4000 /* Call even if result is 0, since the global write bucket may
4001 * have reached 0 on a different conn, and this connection needs to
4002 * know to stop writing. */
4003 connection_consider_empty_write_buckets(conn
);
4004 if (n_read
> 0 && connection_is_reading(conn
))
4005 connection_consider_empty_read_buckets(conn
);
4010 /* DOCDOC connection_handle_write */
4012 connection_handle_write(connection_t
*conn
, int force
)
4015 tor_gettimeofday_cache_clear();
4016 conn
->in_connection_handle_write
= 1;
4017 res
= connection_handle_write_impl(conn
, force
);
4018 conn
->in_connection_handle_write
= 0;
4023 * Try to flush data that's waiting for a write on <b>conn</b>. Return
4024 * -1 on failure, 0 on success.
4026 * Don't use this function for regular writing; the buffers
4027 * system should be good enough at scheduling writes there. Instead, this
4028 * function is for cases when we're about to exit or something and we want
4029 * to report it right away.
4032 connection_flush(connection_t
*conn
)
4034 return connection_handle_write(conn
, 1);
4037 /** Append <b>len</b> bytes of <b>string</b> onto <b>conn</b>'s
4038 * outbuf, and ask it to start writing.
4040 * If <b>zlib</b> is nonzero, this is a directory connection that should get
4041 * its contents compressed or decompressed as they're written. If zlib is
4042 * negative, this is the last data to be compressed, and the connection's zlib
4043 * state should be flushed.
4046 connection_write_to_buf_impl_
,(const char *string
, size_t len
,
4047 connection_t
*conn
, int zlib
))
4049 /* XXXX This function really needs to return -1 on failure. */
4052 if (!len
&& !(zlib
<0))
4054 /* if it's marked for close, only allow write if we mean to flush it */
4055 if (conn
->marked_for_close
&& !conn
->hold_open_until_flushed
)
4058 old_datalen
= buf_datalen(conn
->outbuf
);
4060 dir_connection_t
*dir_conn
= TO_DIR_CONN(conn
);
4061 int done
= zlib
< 0;
4062 CONN_LOG_PROTECT(conn
, r
= write_to_buf_compress(conn
->outbuf
,
4063 dir_conn
->compress_state
,
4064 string
, len
, done
));
4066 CONN_LOG_PROTECT(conn
, r
= write_to_buf(string
, len
, conn
->outbuf
));
4069 if (CONN_IS_EDGE(conn
)) {
4070 /* if it failed, it means we have our package/delivery windows set
4071 wrong compared to our max outbuf size. close the whole circuit. */
4073 "write_to_buf failed. Closing circuit (fd %d).", (int)conn
->s
);
4074 circuit_mark_for_close(circuit_get_by_edge_conn(TO_EDGE_CONN(conn
)),
4075 END_CIRC_REASON_INTERNAL
);
4076 } else if (conn
->type
== CONN_TYPE_OR
) {
4077 or_connection_t
*orconn
= TO_OR_CONN(conn
);
4079 "write_to_buf failed on an orconn; notifying of error "
4080 "(fd %d)", (int)(conn
->s
));
4081 connection_or_close_for_error(orconn
, 0);
4084 "write_to_buf failed. Closing connection (fd %d).",
4086 connection_mark_for_close(conn
);
4091 /* If we receive optimistic data in the EXIT_CONN_STATE_RESOLVING
4092 * state, we don't want to try to write it right away, since
4093 * conn->write_event won't be set yet. Otherwise, write data from
4094 * this conn as the socket is available. */
4095 if (conn
->write_event
) {
4096 connection_start_writing(conn
);
4099 conn
->outbuf_flushlen
+= buf_datalen(conn
->outbuf
) - old_datalen
;
4101 conn
->outbuf_flushlen
+= len
;
4105 #define CONN_GET_ALL_TEMPLATE(var, test) \
4107 smartlist_t *conns = get_connection_array(); \
4108 smartlist_t *ret_conns = smartlist_new(); \
4109 SMARTLIST_FOREACH_BEGIN(conns, connection_t *, var) { \
4110 if (var && (test) && !var->marked_for_close) \
4111 smartlist_add(ret_conns, var); \
4112 } SMARTLIST_FOREACH_END(var); \
4116 /* Return a list of connections that aren't close and matches the given state.
4117 * The returned list can be empty and must be freed using smartlist_free().
4118 * The caller does NOT have owernship of the objects in the list so it must
4119 * not free them nor reference them as they can disapear. */
4121 connection_list_by_type_state(int type
, int state
)
4123 CONN_GET_ALL_TEMPLATE(conn
, (conn
->type
== type
&& conn
->state
== state
));
4126 /** Return a connection_t * from get_connection_array() that satisfies test on
4127 * var, and that is not marked for close. */
4128 #define CONN_GET_TEMPLATE(var, test) \
4130 smartlist_t *conns = get_connection_array(); \
4131 SMARTLIST_FOREACH(conns, connection_t *, var, \
4133 if (var && (test) && !var->marked_for_close) \
4139 /** Return a connection with given type, address, port, and purpose;
4140 * or NULL if no such connection exists (or if all such connections are marked
4142 MOCK_IMPL(connection_t
*,
4143 connection_get_by_type_addr_port_purpose
,(int type
,
4144 const tor_addr_t
*addr
, uint16_t port
,
4147 CONN_GET_TEMPLATE(conn
,
4148 (conn
->type
== type
&&
4149 tor_addr_eq(&conn
->addr
, addr
) &&
4150 conn
->port
== port
&&
4151 conn
->purpose
== purpose
));
4154 /** Return the stream with id <b>id</b> if it is not already marked for
4158 connection_get_by_global_id(uint64_t id
)
4160 CONN_GET_TEMPLATE(conn
, conn
->global_identifier
== id
);
4163 /** Return a connection of type <b>type</b> that is not marked for close.
4166 connection_get_by_type(int type
)
4168 CONN_GET_TEMPLATE(conn
, conn
->type
== type
);
4171 /** Return a connection of type <b>type</b> that is in state <b>state</b>,
4172 * and that is not marked for close.
4175 connection_get_by_type_state(int type
, int state
)
4177 CONN_GET_TEMPLATE(conn
, conn
->type
== type
&& conn
->state
== state
);
4180 /** Return a connection of type <b>type</b> that has rendquery equal
4181 * to <b>rendquery</b>, and that is not marked for close. If state
4182 * is non-zero, conn must be of that state too.
4185 connection_get_by_type_state_rendquery(int type
, int state
,
4186 const char *rendquery
)
4188 tor_assert(type
== CONN_TYPE_DIR
||
4189 type
== CONN_TYPE_AP
|| type
== CONN_TYPE_EXIT
);
4190 tor_assert(rendquery
);
4192 CONN_GET_TEMPLATE(conn
,
4193 (conn
->type
== type
&&
4194 (!state
|| state
== conn
->state
)) &&
4196 (type
== CONN_TYPE_DIR
&&
4197 TO_DIR_CONN(conn
)->rend_data
&&
4198 !rend_cmp_service_ids(rendquery
,
4199 rend_data_get_address(TO_DIR_CONN(conn
)->rend_data
)))
4201 (CONN_IS_EDGE(conn
) &&
4202 TO_EDGE_CONN(conn
)->rend_data
&&
4203 !rend_cmp_service_ids(rendquery
,
4204 rend_data_get_address(TO_EDGE_CONN(conn
)->rend_data
)))
4208 /** Return a new smartlist of dir_connection_t * from get_connection_array()
4209 * that satisfy conn_test on connection_t *conn_var, and dirconn_test on
4210 * dir_connection_t *dirconn_var. conn_var must be of CONN_TYPE_DIR and not
4211 * marked for close to be included in the list. */
4212 #define DIR_CONN_LIST_TEMPLATE(conn_var, conn_test, \
4213 dirconn_var, dirconn_test) \
4215 smartlist_t *conns = get_connection_array(); \
4216 smartlist_t *dir_conns = smartlist_new(); \
4217 SMARTLIST_FOREACH_BEGIN(conns, connection_t *, conn_var) { \
4218 if (conn_var && (conn_test) \
4219 && conn_var->type == CONN_TYPE_DIR \
4220 && !conn_var->marked_for_close) { \
4221 dir_connection_t *dirconn_var = TO_DIR_CONN(conn_var); \
4222 if (dirconn_var && (dirconn_test)) { \
4223 smartlist_add(dir_conns, dirconn_var); \
4226 } SMARTLIST_FOREACH_END(conn_var); \
4230 /** Return a list of directory connections that are fetching the item
4231 * described by <b>purpose</b>/<b>resource</b>. If there are none,
4232 * return an empty list. This list must be freed using smartlist_free,
4233 * but the pointers in it must not be freed.
4234 * Note that this list should not be cached, as the pointers in it can be
4235 * freed if their connections close. */
4237 connection_dir_list_by_purpose_and_resource(
4239 const char *resource
)
4241 DIR_CONN_LIST_TEMPLATE(conn
,
4242 conn
->purpose
== purpose
,
4244 0 == strcmp_opt(resource
,
4245 dirconn
->requested_resource
));
4248 /** Return a list of directory connections that are fetching the item
4249 * described by <b>purpose</b>/<b>resource</b>/<b>state</b>. If there are
4250 * none, return an empty list. This list must be freed using smartlist_free,
4251 * but the pointers in it must not be freed.
4252 * Note that this list should not be cached, as the pointers in it can be
4253 * freed if their connections close. */
4255 connection_dir_list_by_purpose_resource_and_state(
4257 const char *resource
,
4260 DIR_CONN_LIST_TEMPLATE(conn
,
4261 conn
->purpose
== purpose
&& conn
->state
== state
,
4263 0 == strcmp_opt(resource
,
4264 dirconn
->requested_resource
));
4267 #undef DIR_CONN_LIST_TEMPLATE
4269 /** Return an arbitrary active OR connection that isn't <b>this_conn</b>.
4271 * We use this to guess if we should tell the controller that we
4272 * didn't manage to connect to any of our bridges. */
4273 static connection_t
*
4274 connection_get_another_active_or_conn(const or_connection_t
*this_conn
)
4276 CONN_GET_TEMPLATE(conn
,
4277 conn
!= TO_CONN(this_conn
) && conn
->type
== CONN_TYPE_OR
);
4280 /** Return 1 if there are any active OR connections apart from
4283 * We use this to guess if we should tell the controller that we
4284 * didn't manage to connect to any of our bridges. */
4286 any_other_active_or_conns(const or_connection_t
*this_conn
)
4288 connection_t
*conn
= connection_get_another_active_or_conn(this_conn
);
4290 log_debug(LD_DIR
, "%s: Found an OR connection: %s",
4291 __func__
, conn
->address
);
4298 #undef CONN_GET_TEMPLATE
4300 /** Return 1 if <b>conn</b> is a listener conn, else return 0. */
4302 connection_is_listener(connection_t
*conn
)
4304 if (conn
->type
== CONN_TYPE_OR_LISTENER
||
4305 conn
->type
== CONN_TYPE_EXT_OR_LISTENER
||
4306 conn
->type
== CONN_TYPE_AP_LISTENER
||
4307 conn
->type
== CONN_TYPE_AP_TRANS_LISTENER
||
4308 conn
->type
== CONN_TYPE_AP_DNS_LISTENER
||
4309 conn
->type
== CONN_TYPE_AP_NATD_LISTENER
||
4310 conn
->type
== CONN_TYPE_DIR_LISTENER
||
4311 conn
->type
== CONN_TYPE_CONTROL_LISTENER
)
4316 /** Return 1 if <b>conn</b> is in state "open" and is not marked
4317 * for close, else return 0.
4320 connection_state_is_open(connection_t
*conn
)
4324 if (conn
->marked_for_close
)
4327 if ((conn
->type
== CONN_TYPE_OR
&& conn
->state
== OR_CONN_STATE_OPEN
) ||
4328 (conn
->type
== CONN_TYPE_EXT_OR
) ||
4329 (conn
->type
== CONN_TYPE_AP
&& conn
->state
== AP_CONN_STATE_OPEN
) ||
4330 (conn
->type
== CONN_TYPE_EXIT
&& conn
->state
== EXIT_CONN_STATE_OPEN
) ||
4331 (conn
->type
== CONN_TYPE_CONTROL
&&
4332 conn
->state
== CONTROL_CONN_STATE_OPEN
))
4338 /** Return 1 if conn is in 'connecting' state, else return 0. */
4340 connection_state_is_connecting(connection_t
*conn
)
4344 if (conn
->marked_for_close
)
4349 return conn
->state
== OR_CONN_STATE_CONNECTING
;
4350 case CONN_TYPE_EXIT
:
4351 return conn
->state
== EXIT_CONN_STATE_CONNECTING
;
4353 return conn
->state
== DIR_CONN_STATE_CONNECTING
;
4359 /** Allocates a base64'ed authenticator for use in http or https
4360 * auth, based on the input string <b>authenticator</b>. Returns it
4361 * if success, else returns NULL. */
4363 alloc_http_authenticator(const char *authenticator
)
4365 /* an authenticator in Basic authentication
4366 * is just the string "username:password" */
4367 const size_t authenticator_length
= strlen(authenticator
);
4368 const size_t base64_authenticator_length
=
4369 base64_encode_size(authenticator_length
, 0) + 1;
4370 char *base64_authenticator
= tor_malloc(base64_authenticator_length
);
4371 if (base64_encode(base64_authenticator
, base64_authenticator_length
,
4372 authenticator
, authenticator_length
, 0) < 0) {
4373 tor_free(base64_authenticator
); /* free and set to null */
4375 return base64_authenticator
;
4378 /** Given a socket handle, check whether the local address (sockname) of the
4379 * socket is one that we've connected from before. If so, double-check
4380 * whether our address has changed and we need to generate keys. If we do,
4384 client_check_address_changed(tor_socket_t sock
)
4386 struct sockaddr_storage out_sockaddr
;
4387 socklen_t out_addr_len
= (socklen_t
) sizeof(out_sockaddr
);
4388 tor_addr_t out_addr
, iface_addr
;
4389 tor_addr_t
**last_interface_ip_ptr
;
4392 if (!outgoing_addrs
)
4393 outgoing_addrs
= smartlist_new();
4395 if (getsockname(sock
, (struct sockaddr
*)&out_sockaddr
, &out_addr_len
)<0) {
4396 int e
= tor_socket_errno(sock
);
4397 log_warn(LD_NET
, "getsockname() to check for address change failed: %s",
4398 tor_socket_strerror(e
));
4401 tor_addr_from_sockaddr(&out_addr
, (struct sockaddr
*)&out_sockaddr
, NULL
);
4402 family
= tor_addr_family(&out_addr
);
4404 if (family
== AF_INET
)
4405 last_interface_ip_ptr
= &last_interface_ipv4
;
4406 else if (family
== AF_INET6
)
4407 last_interface_ip_ptr
= &last_interface_ipv6
;
4411 if (! *last_interface_ip_ptr
) {
4412 tor_addr_t
*a
= tor_malloc_zero(sizeof(tor_addr_t
));
4413 if (get_interface_address6(LOG_INFO
, family
, a
)==0) {
4414 *last_interface_ip_ptr
= a
;
4420 /* If we've used this address previously, we're okay. */
4421 SMARTLIST_FOREACH(outgoing_addrs
, const tor_addr_t
*, a_ptr
,
4422 if (tor_addr_eq(a_ptr
, &out_addr
))
4426 /* Uh-oh. We haven't connected from this address before. Has the interface
4427 * address changed? */
4428 if (get_interface_address6(LOG_INFO
, family
, &iface_addr
)<0)
4431 if (tor_addr_eq(&iface_addr
, *last_interface_ip_ptr
)) {
4432 /* Nope, it hasn't changed. Add this address to the list. */
4433 smartlist_add(outgoing_addrs
, tor_memdup(&out_addr
, sizeof(tor_addr_t
)));
4435 /* The interface changed. We're a client, so we need to regenerate our
4436 * keys. First, reset the state. */
4437 log_notice(LD_NET
, "Our IP address has changed. Rotating keys...");
4438 tor_addr_copy(*last_interface_ip_ptr
, &iface_addr
);
4439 SMARTLIST_FOREACH(outgoing_addrs
, tor_addr_t
*, a_ptr
, tor_free(a_ptr
));
4440 smartlist_clear(outgoing_addrs
);
4441 smartlist_add(outgoing_addrs
, tor_memdup(&out_addr
, sizeof(tor_addr_t
)));
4442 /* We'll need to resolve ourselves again. */
4443 reset_last_resolved_addr();
4444 /* Okay, now change our keys. */
4445 ip_address_changed(1);
4449 /** Some systems have limited system buffers for recv and xmit on
4450 * sockets allocated in a virtual server or similar environment. For a Tor
4451 * server this can produce the "Error creating network socket: No buffer
4452 * space available" error once all available TCP buffer space is consumed.
4453 * This method will attempt to constrain the buffers allocated for the socket
4454 * to the desired size to stay below system TCP buffer limits.
4457 set_constrained_socket_buffers(tor_socket_t sock
, int size
)
4459 void *sz
= (void*)&size
;
4460 socklen_t sz_sz
= (socklen_t
) sizeof(size
);
4461 if (setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
, sz
, sz_sz
) < 0) {
4462 int e
= tor_socket_errno(sock
);
4463 log_warn(LD_NET
, "setsockopt() to constrain send "
4464 "buffer to %d bytes failed: %s", size
, tor_socket_strerror(e
));
4466 if (setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
, sz
, sz_sz
) < 0) {
4467 int e
= tor_socket_errno(sock
);
4468 log_warn(LD_NET
, "setsockopt() to constrain recv "
4469 "buffer to %d bytes failed: %s", size
, tor_socket_strerror(e
));
4473 /** Process new bytes that have arrived on conn-\>inbuf.
4475 * This function just passes conn to the connection-specific
4476 * connection_*_process_inbuf() function. It also passes in
4477 * package_partial if wanted.
4480 connection_process_inbuf(connection_t
*conn
, int package_partial
)
4484 switch (conn
->type
) {
4486 return connection_or_process_inbuf(TO_OR_CONN(conn
));
4487 case CONN_TYPE_EXT_OR
:
4488 return connection_ext_or_process_inbuf(TO_OR_CONN(conn
));
4489 case CONN_TYPE_EXIT
:
4491 return connection_edge_process_inbuf(TO_EDGE_CONN(conn
),
4494 return connection_dir_process_inbuf(TO_DIR_CONN(conn
));
4495 case CONN_TYPE_CONTROL
:
4496 return connection_control_process_inbuf(TO_CONTROL_CONN(conn
));
4498 log_err(LD_BUG
,"got unexpected conn type %d.", conn
->type
);
4499 tor_fragile_assert();
4504 /** Called whenever we've written data on a connection. */
4506 connection_flushed_some(connection_t
*conn
)
4509 tor_assert(!conn
->in_flushed_some
);
4510 conn
->in_flushed_some
= 1;
4511 if (conn
->type
== CONN_TYPE_DIR
&&
4512 conn
->state
== DIR_CONN_STATE_SERVER_WRITING
) {
4513 r
= connection_dirserv_flushed_some(TO_DIR_CONN(conn
));
4514 } else if (conn
->type
== CONN_TYPE_OR
) {
4515 r
= connection_or_flushed_some(TO_OR_CONN(conn
));
4516 } else if (CONN_IS_EDGE(conn
)) {
4517 r
= connection_edge_flushed_some(TO_EDGE_CONN(conn
));
4519 conn
->in_flushed_some
= 0;
4523 /** We just finished flushing bytes to the appropriately low network layer,
4524 * and there are no more bytes remaining in conn-\>outbuf or
4525 * conn-\>tls to be flushed.
4527 * This function just passes conn to the connection-specific
4528 * connection_*_finished_flushing() function.
4531 connection_finished_flushing(connection_t
*conn
)
4535 /* If the connection is closed, don't try to do anything more here. */
4536 if (CONN_IS_CLOSED(conn
))
4539 // log_fn(LOG_DEBUG,"entered. Socket %u.", conn->s);
4541 connection_stop_writing(conn
);
4543 switch (conn
->type
) {
4545 return connection_or_finished_flushing(TO_OR_CONN(conn
));
4546 case CONN_TYPE_EXT_OR
:
4547 return connection_ext_or_finished_flushing(TO_OR_CONN(conn
));
4549 case CONN_TYPE_EXIT
:
4550 return connection_edge_finished_flushing(TO_EDGE_CONN(conn
));
4552 return connection_dir_finished_flushing(TO_DIR_CONN(conn
));
4553 case CONN_TYPE_CONTROL
:
4554 return connection_control_finished_flushing(TO_CONTROL_CONN(conn
));
4556 log_err(LD_BUG
,"got unexpected conn type %d.", conn
->type
);
4557 tor_fragile_assert();
4562 /** Called when our attempt to connect() to another server has just
4565 * This function just passes conn to the connection-specific
4566 * connection_*_finished_connecting() function.
4569 connection_finished_connecting(connection_t
*conn
)
4573 if (!server_mode(get_options())) {
4574 /* See whether getsockname() says our address changed. We need to do this
4575 * now that the connection has finished, because getsockname() on Windows
4576 * won't work until then. */
4577 client_check_address_changed(conn
->s
);
4583 return connection_or_finished_connecting(TO_OR_CONN(conn
));
4584 case CONN_TYPE_EXIT
:
4585 return connection_edge_finished_connecting(TO_EDGE_CONN(conn
));
4587 return connection_dir_finished_connecting(TO_DIR_CONN(conn
));
4589 log_err(LD_BUG
,"got unexpected conn type %d.", conn
->type
);
4590 tor_fragile_assert();
4595 /** Callback: invoked when a connection reaches an EOF event. */
4597 connection_reached_eof(connection_t
*conn
)
4599 switch (conn
->type
) {
4601 case CONN_TYPE_EXT_OR
:
4602 return connection_or_reached_eof(TO_OR_CONN(conn
));
4604 case CONN_TYPE_EXIT
:
4605 return connection_edge_reached_eof(TO_EDGE_CONN(conn
));
4607 return connection_dir_reached_eof(TO_DIR_CONN(conn
));
4608 case CONN_TYPE_CONTROL
:
4609 return connection_control_reached_eof(TO_CONTROL_CONN(conn
));
4611 log_err(LD_BUG
,"got unexpected conn type %d.", conn
->type
);
4612 tor_fragile_assert();
4617 /** Comparator for the two-orconn case in OOS victim sort */
4619 oos_victim_comparator_for_orconns(or_connection_t
*a
, or_connection_t
*b
)
4621 int a_circs
, b_circs
;
4622 /* Fewer circuits == higher priority for OOS kill, sort earlier */
4624 a_circs
= connection_or_get_num_circuits(a
);
4625 b_circs
= connection_or_get_num_circuits(b
);
4627 if (a_circs
< b_circs
) return 1;
4628 else if (a_circs
> b_circs
) return -1;
4632 /** Sort comparator for OOS victims; better targets sort before worse
4635 oos_victim_comparator(const void **a_v
, const void **b_v
)
4637 connection_t
*a
= NULL
, *b
= NULL
;
4639 /* Get connection pointers out */
4641 a
= (connection_t
*)(*a_v
);
4642 b
= (connection_t
*)(*b_v
);
4644 tor_assert(a
!= NULL
);
4645 tor_assert(b
!= NULL
);
4648 * We always prefer orconns as victims currently; we won't even see
4649 * these non-orconn cases, but if we do, sort them after orconns.
4651 if (a
->type
== CONN_TYPE_OR
&& b
->type
== CONN_TYPE_OR
) {
4652 return oos_victim_comparator_for_orconns(TO_OR_CONN(a
), TO_OR_CONN(b
));
4655 * One isn't an orconn; if one is, it goes first. We currently have no
4656 * opinions about cases where neither is an orconn.
4658 if (a
->type
== CONN_TYPE_OR
) return -1;
4659 else if (b
->type
== CONN_TYPE_OR
) return 1;
4664 /** Pick n victim connections for the OOS handler and return them in a
4667 MOCK_IMPL(STATIC smartlist_t
*,
4668 pick_oos_victims
, (int n
))
4670 smartlist_t
*eligible
= NULL
, *victims
= NULL
;
4672 int conn_counts_by_type
[CONN_TYPE_MAX_
+ 1], i
;
4675 * Big damn assumption (someone improve this someday!):
4677 * Socket exhaustion normally happens on high-volume relays, and so
4678 * most of the connections involved are orconns. We should pick victims
4679 * by assembling a list of all orconns, and sorting them in order of
4680 * how much 'damage' by some metric we'd be doing by dropping them.
4682 * If we move on from orconns, we should probably think about incoming
4683 * directory connections next, or exit connections. Things we should
4684 * probably never kill are controller connections and listeners.
4686 * This function will count how many connections of different types
4687 * exist and log it for purposes of gathering data on typical OOS
4688 * situations to guide future improvements.
4691 /* First, get the connection array */
4692 conns
= get_connection_array();
4694 * Iterate it and pick out eligible connection types, and log some stats
4697 eligible
= smartlist_new();
4698 memset(conn_counts_by_type
, 0, sizeof(conn_counts_by_type
));
4699 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, c
) {
4700 /* Bump the counter */
4701 tor_assert(c
->type
<= CONN_TYPE_MAX_
);
4702 ++(conn_counts_by_type
[c
->type
]);
4704 /* Skip anything without a socket we can free */
4705 if (!(SOCKET_OK(c
->s
))) {
4709 /* Skip anything we would count as moribund */
4710 if (connection_is_moribund(c
)) {
4716 /* We've got an orconn, it's eligible to be OOSed */
4717 smartlist_add(eligible
, c
);
4720 /* We don't know what to do with it, ignore it */
4723 } SMARTLIST_FOREACH_END(c
);
4725 /* Log some stats */
4726 if (smartlist_len(conns
) > 0) {
4727 /* At least one counter must be non-zero */
4728 log_info(LD_NET
, "Some stats on conn types seen during OOS follow");
4729 for (i
= CONN_TYPE_MIN_
; i
<= CONN_TYPE_MAX_
; ++i
) {
4730 /* Did we see any? */
4731 if (conn_counts_by_type
[i
] > 0) {
4732 log_info(LD_NET
, "%s: %d conns",
4733 conn_type_to_string(i
),
4734 conn_counts_by_type
[i
]);
4737 log_info(LD_NET
, "Done with OOS conn type stats");
4740 /* Did we find more eligible targets than we want to kill? */
4741 if (smartlist_len(eligible
) > n
) {
4742 /* Sort the list in order of target preference */
4743 smartlist_sort(eligible
, oos_victim_comparator
);
4744 /* Pick first n as victims */
4745 victims
= smartlist_new();
4746 for (i
= 0; i
< n
; ++i
) {
4747 smartlist_add(victims
, smartlist_get(eligible
, i
));
4749 /* Free the original list */
4750 smartlist_free(eligible
);
4752 /* No, we can just call them all victims */
4759 /** Kill a list of connections for the OOS handler. */
4760 MOCK_IMPL(STATIC
void,
4761 kill_conn_list_for_oos
, (smartlist_t
*conns
))
4765 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, c
) {
4766 /* Make sure the channel layer gets told about orconns */
4767 if (c
->type
== CONN_TYPE_OR
) {
4768 connection_or_close_for_error(TO_OR_CONN(c
), 1);
4770 connection_mark_for_close(c
);
4772 } SMARTLIST_FOREACH_END(c
);
4775 "OOS handler marked %d connections",
4776 smartlist_len(conns
));
4779 /** Out-of-Sockets handler; n_socks is the current number of open
4780 * sockets, and failed is non-zero if a socket exhaustion related
4781 * error immediately preceded this call. This is where to do
4782 * circuit-killing heuristics as needed.
4785 connection_check_oos(int n_socks
, int failed
)
4787 int target_n_socks
= 0, moribund_socks
, socks_to_kill
;
4790 /* Early exit: is OOS checking disabled? */
4791 if (get_options()->DisableOOSCheck
) {
4795 /* Sanity-check args */
4796 tor_assert(n_socks
>= 0);
4799 * Make some log noise; keep it at debug level since this gets a chance
4800 * to run on every connection attempt.
4803 "Running the OOS handler (%d open sockets, %s)",
4804 n_socks
, (failed
!= 0) ? "exhaustion seen" : "no exhaustion");
4807 * Check if we're really handling an OOS condition, and if so decide how
4808 * many sockets we want to get down to. Be sure we check if the threshold
4809 * is distinct from zero first; it's possible for this to be called a few
4810 * times before we've finished reading the config.
4812 if (n_socks
>= get_options()->ConnLimit_high_thresh
&&
4813 get_options()->ConnLimit_high_thresh
!= 0 &&
4814 get_options()->ConnLimit_
!= 0) {
4815 /* Try to get down to the low threshold */
4816 target_n_socks
= get_options()->ConnLimit_low_thresh
;
4818 "Current number of sockets %d is greater than configured "
4819 "limit %d; OOS handler trying to get down to %d",
4820 n_socks
, get_options()->ConnLimit_high_thresh
,
4822 } else if (failed
) {
4824 * If we're not at the limit but we hit a socket exhaustion error, try to
4825 * drop some (but not as aggressively as ConnLimit_low_threshold, which is
4826 * 3/4 of ConnLimit_)
4828 target_n_socks
= (n_socks
* 9) / 10;
4830 "We saw socket exhaustion at %d open sockets; OOS handler "
4831 "trying to get down to %d",
4832 n_socks
, target_n_socks
);
4835 if (target_n_socks
> 0) {
4839 * Count moribund sockets; it's be important that anything we decide
4840 * to get rid of here but don't immediately close get counted as moribund
4841 * on subsequent invocations so we don't try to kill too many things if
4842 * connection_check_oos() gets called multiple times.
4844 moribund_socks
= connection_count_moribund();
4846 if (moribund_socks
< n_socks
- target_n_socks
) {
4847 socks_to_kill
= n_socks
- target_n_socks
- moribund_socks
;
4849 conns
= pick_oos_victims(socks_to_kill
);
4851 kill_conn_list_for_oos(conns
);
4853 "OOS handler killed %d conns", smartlist_len(conns
));
4854 smartlist_free(conns
);
4856 log_notice(LD_NET
, "OOS handler failed to pick any victim conns");
4860 "Not killing any sockets for OOS because there are %d "
4861 "already moribund, and we only want to eliminate %d",
4862 moribund_socks
, n_socks
- target_n_socks
);
4867 /** Log how many bytes are used by buffers of different kinds and sizes. */
4869 connection_dump_buffer_mem_stats(int severity
)
4871 uint64_t used_by_type
[CONN_TYPE_MAX_
+1];
4872 uint64_t alloc_by_type
[CONN_TYPE_MAX_
+1];
4873 int n_conns_by_type
[CONN_TYPE_MAX_
+1];
4874 uint64_t total_alloc
= 0;
4875 uint64_t total_used
= 0;
4877 smartlist_t
*conns
= get_connection_array();
4879 memset(used_by_type
, 0, sizeof(used_by_type
));
4880 memset(alloc_by_type
, 0, sizeof(alloc_by_type
));
4881 memset(n_conns_by_type
, 0, sizeof(n_conns_by_type
));
4883 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, c
) {
4885 ++n_conns_by_type
[tp
];
4887 used_by_type
[tp
] += buf_datalen(c
->inbuf
);
4888 alloc_by_type
[tp
] += buf_allocation(c
->inbuf
);
4891 used_by_type
[tp
] += buf_datalen(c
->outbuf
);
4892 alloc_by_type
[tp
] += buf_allocation(c
->outbuf
);
4894 } SMARTLIST_FOREACH_END(c
);
4895 for (i
=0; i
<= CONN_TYPE_MAX_
; ++i
) {
4896 total_used
+= used_by_type
[i
];
4897 total_alloc
+= alloc_by_type
[i
];
4900 tor_log(severity
, LD_GENERAL
,
4901 "In buffers for %d connections: "U64_FORMAT
" used/"U64_FORMAT
" allocated",
4902 smartlist_len(conns
),
4903 U64_PRINTF_ARG(total_used
), U64_PRINTF_ARG(total_alloc
));
4904 for (i
=CONN_TYPE_MIN_
; i
<= CONN_TYPE_MAX_
; ++i
) {
4905 if (!n_conns_by_type
[i
])
4907 tor_log(severity
, LD_GENERAL
,
4908 " For %d %s connections: "U64_FORMAT
" used/"U64_FORMAT
" allocated",
4909 n_conns_by_type
[i
], conn_type_to_string(i
),
4910 U64_PRINTF_ARG(used_by_type
[i
]), U64_PRINTF_ARG(alloc_by_type
[i
]));
4914 /** Verify that connection <b>conn</b> has all of its invariants
4915 * correct. Trigger an assert if anything is invalid.
4918 assert_connection_ok(connection_t
*conn
, time_t now
)
4920 (void) now
; /* XXXX unused. */
4922 tor_assert(conn
->type
>= CONN_TYPE_MIN_
);
4923 tor_assert(conn
->type
<= CONN_TYPE_MAX_
);
4925 switch (conn
->type
) {
4927 case CONN_TYPE_EXT_OR
:
4928 tor_assert(conn
->magic
== OR_CONNECTION_MAGIC
);
4931 tor_assert(conn
->magic
== ENTRY_CONNECTION_MAGIC
);
4933 case CONN_TYPE_EXIT
:
4934 tor_assert(conn
->magic
== EDGE_CONNECTION_MAGIC
);
4937 tor_assert(conn
->magic
== DIR_CONNECTION_MAGIC
);
4939 case CONN_TYPE_CONTROL
:
4940 tor_assert(conn
->magic
== CONTROL_CONNECTION_MAGIC
);
4942 CASE_ANY_LISTENER_TYPE
:
4943 tor_assert(conn
->magic
== LISTENER_CONNECTION_MAGIC
);
4946 tor_assert(conn
->magic
== BASE_CONNECTION_MAGIC
);
4950 if (conn
->linked_conn
) {
4951 tor_assert(conn
->linked_conn
->linked_conn
== conn
);
4952 tor_assert(conn
->linked
);
4955 tor_assert(!SOCKET_OK(conn
->s
));
4957 if (conn
->outbuf_flushlen
> 0) {
4958 /* With optimistic data, we may have queued data in
4959 * EXIT_CONN_STATE_RESOLVING while the conn is not yet marked to writing.
4961 tor_assert((conn
->type
== CONN_TYPE_EXIT
&&
4962 conn
->state
== EXIT_CONN_STATE_RESOLVING
) ||
4963 connection_is_writing(conn
) ||
4964 conn
->write_blocked_on_bw
||
4965 (CONN_IS_EDGE(conn
) &&
4966 TO_EDGE_CONN(conn
)->edge_blocked_on_circ
));
4969 if (conn
->hold_open_until_flushed
)
4970 tor_assert(conn
->marked_for_close
);
4972 /* XXXX check: read_blocked_on_bw, write_blocked_on_bw, s, conn_array_index,
4973 * marked_for_close. */
4977 assert_buf_ok(conn
->inbuf
);
4979 assert_buf_ok(conn
->outbuf
);
4981 if (conn
->type
== CONN_TYPE_OR
) {
4982 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
4983 if (conn
->state
== OR_CONN_STATE_OPEN
) {
4984 /* tor_assert(conn->bandwidth > 0); */
4985 /* the above isn't necessarily true: if we just did a TLS
4986 * handshake but we didn't recognize the other peer, or it
4987 * gave a bad cert/etc, then we won't have assigned bandwidth,
4988 * yet it will be open. -RD
4990 // tor_assert(conn->read_bucket >= 0);
4992 // tor_assert(conn->addr && conn->port);
4993 tor_assert(conn
->address
);
4994 if (conn
->state
> OR_CONN_STATE_PROXY_HANDSHAKING
)
4995 tor_assert(or_conn
->tls
);
4998 if (CONN_IS_EDGE(conn
)) {
4999 /* XXX unchecked: package window, deliver window. */
5000 if (conn
->type
== CONN_TYPE_AP
) {
5001 entry_connection_t
*entry_conn
= TO_ENTRY_CONN(conn
);
5002 if (entry_conn
->chosen_exit_optional
|| entry_conn
->chosen_exit_retries
)
5003 tor_assert(entry_conn
->chosen_exit_name
);
5005 tor_assert(entry_conn
->socks_request
);
5006 if (conn
->state
== AP_CONN_STATE_OPEN
) {
5007 tor_assert(entry_conn
->socks_request
->has_finished
);
5008 if (!conn
->marked_for_close
) {
5009 tor_assert(ENTRY_TO_EDGE_CONN(entry_conn
)->cpath_layer
);
5010 assert_cpath_layer_ok(ENTRY_TO_EDGE_CONN(entry_conn
)->cpath_layer
);
5014 if (conn
->type
== CONN_TYPE_EXIT
) {
5015 tor_assert(conn
->purpose
== EXIT_PURPOSE_CONNECT
||
5016 conn
->purpose
== EXIT_PURPOSE_RESOLVE
);
5018 } else if (conn
->type
== CONN_TYPE_DIR
) {
5020 /* Purpose is only used for dir and exit types currently */
5021 tor_assert(!conn
->purpose
);
5026 CASE_ANY_LISTENER_TYPE
:
5027 tor_assert(conn
->state
== LISTENER_STATE_READY
);
5030 tor_assert(conn
->state
>= OR_CONN_STATE_MIN_
);
5031 tor_assert(conn
->state
<= OR_CONN_STATE_MAX_
);
5033 case CONN_TYPE_EXT_OR
:
5034 tor_assert(conn
->state
>= EXT_OR_CONN_STATE_MIN_
);
5035 tor_assert(conn
->state
<= EXT_OR_CONN_STATE_MAX_
);
5037 case CONN_TYPE_EXIT
:
5038 tor_assert(conn
->state
>= EXIT_CONN_STATE_MIN_
);
5039 tor_assert(conn
->state
<= EXIT_CONN_STATE_MAX_
);
5040 tor_assert(conn
->purpose
>= EXIT_PURPOSE_MIN_
);
5041 tor_assert(conn
->purpose
<= EXIT_PURPOSE_MAX_
);
5044 tor_assert(conn
->state
>= AP_CONN_STATE_MIN_
);
5045 tor_assert(conn
->state
<= AP_CONN_STATE_MAX_
);
5046 tor_assert(TO_ENTRY_CONN(conn
)->socks_request
);
5049 tor_assert(conn
->state
>= DIR_CONN_STATE_MIN_
);
5050 tor_assert(conn
->state
<= DIR_CONN_STATE_MAX_
);
5051 tor_assert(conn
->purpose
>= DIR_PURPOSE_MIN_
);
5052 tor_assert(conn
->purpose
<= DIR_PURPOSE_MAX_
);
5054 case CONN_TYPE_CONTROL
:
5055 tor_assert(conn
->state
>= CONTROL_CONN_STATE_MIN_
);
5056 tor_assert(conn
->state
<= CONTROL_CONN_STATE_MAX_
);
5063 /** Fills <b>addr</b> and <b>port</b> with the details of the global
5064 * proxy server we are using.
5065 * <b>conn</b> contains the connection we are using the proxy for.
5067 * Return 0 on success, -1 on failure.
5070 get_proxy_addrport(tor_addr_t
*addr
, uint16_t *port
, int *proxy_type
,
5071 const connection_t
*conn
)
5073 const or_options_t
*options
= get_options();
5075 /* Client Transport Plugins can use another proxy, but that should be hidden
5076 * from the rest of tor (as the plugin is responsible for dealing with the
5077 * proxy), check it first, then check the rest of the proxy types to allow
5078 * the config to have unused ClientTransportPlugin entries.
5080 if (options
->ClientTransportPlugin
) {
5081 const transport_t
*transport
= NULL
;
5083 r
= get_transport_by_bridge_addrport(&conn
->addr
, conn
->port
, &transport
);
5086 if (transport
) { /* transport found */
5087 tor_addr_copy(addr
, &transport
->addr
);
5088 *port
= transport
->port
;
5089 *proxy_type
= transport
->socks_version
;
5093 /* Unused ClientTransportPlugin. */
5096 if (options
->HTTPSProxy
) {
5097 tor_addr_copy(addr
, &options
->HTTPSProxyAddr
);
5098 *port
= options
->HTTPSProxyPort
;
5099 *proxy_type
= PROXY_CONNECT
;
5101 } else if (options
->Socks4Proxy
) {
5102 tor_addr_copy(addr
, &options
->Socks4ProxyAddr
);
5103 *port
= options
->Socks4ProxyPort
;
5104 *proxy_type
= PROXY_SOCKS4
;
5106 } else if (options
->Socks5Proxy
) {
5107 tor_addr_copy(addr
, &options
->Socks5ProxyAddr
);
5108 *port
= options
->Socks5ProxyPort
;
5109 *proxy_type
= PROXY_SOCKS5
;
5113 tor_addr_make_unspec(addr
);
5115 *proxy_type
= PROXY_NONE
;
5119 /** Log a failed connection to a proxy server.
5120 * <b>conn</b> is the connection we use the proxy server for. */
5122 log_failed_proxy_connection(connection_t
*conn
)
5124 tor_addr_t proxy_addr
;
5125 uint16_t proxy_port
;
5128 if (get_proxy_addrport(&proxy_addr
, &proxy_port
, &proxy_type
, conn
) != 0)
5129 return; /* if we have no proxy set up, leave this function. */
5132 "The connection to the %s proxy server at %s just failed. "
5133 "Make sure that the proxy server is up and running.",
5134 proxy_type_to_string(proxy_type
),
5135 fmt_addrport(&proxy_addr
, proxy_port
));
5138 /** Return string representation of <b>proxy_type</b>. */
5140 proxy_type_to_string(int proxy_type
)
5142 switch (proxy_type
) {
5143 case PROXY_CONNECT
: return "HTTP";
5144 case PROXY_SOCKS4
: return "SOCKS4";
5145 case PROXY_SOCKS5
: return "SOCKS5";
5146 case PROXY_PLUGGABLE
: return "pluggable transports SOCKS";
5147 case PROXY_NONE
: return "NULL";
5148 default: tor_assert(0);
5150 return NULL
; /*Unreached*/
5153 /** Call connection_free_() on every connection in our array, and release all
5154 * storage held by connection.c.
5156 * Don't do the checks in connection_free(), because they will
5160 connection_free_all(void)
5162 smartlist_t
*conns
= get_connection_array();
5164 /* We don't want to log any messages to controllers. */
5165 SMARTLIST_FOREACH(conns
, connection_t
*, conn
,
5166 if (conn
->type
== CONN_TYPE_CONTROL
)
5167 TO_CONTROL_CONN(conn
)->event_mask
= 0);
5169 control_update_global_event_mask();
5171 /* Unlink everything from the identity map. */
5172 connection_or_clear_identity_map();
5173 connection_or_clear_ext_or_id_map();
5175 /* Clear out our list of broken connections */
5176 clear_broken_connection_map(0);
5178 SMARTLIST_FOREACH(conns
, connection_t
*, conn
, connection_free_(conn
));
5180 if (outgoing_addrs
) {
5181 SMARTLIST_FOREACH(outgoing_addrs
, tor_addr_t
*, addr
, tor_free(addr
));
5182 smartlist_free(outgoing_addrs
);
5183 outgoing_addrs
= NULL
;
5186 tor_free(last_interface_ipv4
);
5187 tor_free(last_interface_ipv6
);
5190 /** Log a warning, and possibly emit a control event, that <b>received</b> came
5191 * at a skewed time. <b>trusted</b> indicates that the <b>source</b> was one
5192 * that we had more faith in and therefore the warning level should have higher
5196 clock_skew_warning(const connection_t
*conn
, long apparent_skew
, int trusted
,
5197 log_domain_mask_t domain
, const char *received
,
5201 char *ext_source
= NULL
;
5202 format_time_interval(dbuf
, sizeof(dbuf
), apparent_skew
);
5204 tor_asprintf(&ext_source
, "%s:%s:%d", source
, conn
->address
, conn
->port
);
5206 ext_source
= tor_strdup(source
);
5207 log_fn(trusted
? LOG_WARN
: LOG_INFO
, domain
,
5208 "Received %s with skewed time (%s): "
5209 "It seems that our clock is %s by %s, or that theirs is %s%s. "
5210 "Tor requires an accurate clock to work: please check your time, "
5211 "timezone, and date settings.", received
, ext_source
,
5212 apparent_skew
> 0 ? "ahead" : "behind", dbuf
,
5213 apparent_skew
> 0 ? "behind" : "ahead",
5214 (!conn
|| trusted
) ? "" : ", or they are sending us the wrong time");
5216 control_event_general_status(LOG_WARN
, "CLOCK_SKEW SKEW=%ld SOURCE=%s",
5217 apparent_skew
, ext_source
);
5218 tor_free(ext_source
);