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) 2007-2021, 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_buf_add(). 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
58 #include "core/or/or.h"
59 #include "feature/client/bridges.h"
60 #include "lib/buf/buffers.h"
61 #include "lib/tls/buffers_tls.h"
62 #include "lib/err/backtrace.h"
65 * Define this so we get channel internal functions, since we're implementing
66 * part of a subclass (channel_tls_t).
68 #define CHANNEL_OBJECT_PRIVATE
69 #include "app/config/config.h"
70 #include "app/config/resolve_addr.h"
71 #include "core/mainloop/connection.h"
72 #include "core/mainloop/mainloop.h"
73 #include "core/mainloop/netstatus.h"
74 #include "core/or/channel.h"
75 #include "core/or/channeltls.h"
76 #include "core/or/circuitbuild.h"
77 #include "core/or/circuitlist.h"
78 #include "core/or/circuituse.h"
79 #include "core/or/connection_edge.h"
80 #include "core/or/connection_or.h"
81 #include "core/or/dos.h"
82 #include "core/or/policies.h"
83 #include "core/or/reasons.h"
84 #include "core/or/relay.h"
85 #include "core/or/status.h"
86 #include "core/or/crypt_path.h"
87 #include "core/proto/proto_haproxy.h"
88 #include "core/proto/proto_http.h"
89 #include "core/proto/proto_socks.h"
90 #include "feature/client/dnsserv.h"
91 #include "feature/client/entrynodes.h"
92 #include "feature/client/transports.h"
93 #include "feature/control/control.h"
94 #include "feature/control/control_events.h"
95 #include "feature/dirauth/authmode.h"
96 #include "feature/dirauth/dirauth_config.h"
97 #include "feature/dircache/dirserv.h"
98 #include "feature/dircommon/directory.h"
99 #include "feature/hibernate/hibernate.h"
100 #include "feature/hs/hs_common.h"
101 #include "feature/hs/hs_ident.h"
102 #include "feature/hs/hs_metrics.h"
103 #include "feature/metrics/metrics.h"
104 #include "feature/nodelist/nodelist.h"
105 #include "feature/nodelist/routerlist.h"
106 #include "feature/relay/dns.h"
107 #include "feature/relay/ext_orport.h"
108 #include "feature/relay/routermode.h"
109 #include "feature/rend/rendcommon.h"
110 #include "feature/stats/connstats.h"
111 #include "feature/stats/rephist.h"
112 #include "feature/stats/bwhist.h"
113 #include "lib/crypt_ops/crypto_util.h"
114 #include "lib/crypt_ops/crypto_format.h"
115 #include "lib/geoip/geoip.h"
117 #include "lib/cc/ctassert.h"
118 #include "lib/sandbox/sandbox.h"
119 #include "lib/net/buffers_net.h"
120 #include "lib/net/address.h"
121 #include "lib/tls/tortls.h"
122 #include "lib/evloop/compat_libevent.h"
123 #include "lib/compress/compress.h"
132 #ifdef HAVE_SYS_STAT_H
133 #include <sys/stat.h>
137 #include <sys/socket.h>
141 #include "feature/dircommon/dir_connection_st.h"
142 #include "feature/control/control_connection_st.h"
143 #include "core/or/entry_connection_st.h"
144 #include "core/or/listener_connection_st.h"
145 #include "core/or/or_connection_st.h"
146 #include "core/or/port_cfg_st.h"
147 #include "feature/nodelist/routerinfo_st.h"
148 #include "core/or/socks_request_st.h"
150 #include "core/or/congestion_control_flow.h"
153 * On Windows and Linux we cannot reliably bind() a socket to an
154 * address and port if: 1) There's already a socket bound to wildcard
155 * address (0.0.0.0 or ::) with the same port; 2) We try to bind()
156 * to wildcard address and there's another socket bound to a
157 * specific address and the same port.
159 * To address this problem on these two platforms we implement a
161 * 1) Checks if first attempt to bind() a new socket failed with
163 * 2) If so, it will close the appropriate old listener connection and
164 * 3) Attempts bind()'ing the new listener socket again.
166 * Just to be safe, we are enabling listener rebind code on all platforms,
167 * to account for unexpected cases where it may be needed.
169 #define ENABLE_LISTENER_REBIND
171 static connection_t
*connection_listener_new(
172 const struct sockaddr
*listensockaddr
,
173 socklen_t listensocklen
, int type
,
175 const port_cfg_t
*portcfg
,
177 static connection_t
*connection_listener_new_for_port(
178 const port_cfg_t
*port
,
179 int *defer
, int *addr_in_use
);
180 static void connection_init(time_t now
, connection_t
*conn
, int type
,
182 static int connection_handle_listener_read(connection_t
*conn
, int new_type
);
183 static int connection_finished_flushing(connection_t
*conn
);
184 static int connection_flushed_some(connection_t
*conn
);
185 static int connection_finished_connecting(connection_t
*conn
);
186 static int connection_reached_eof(connection_t
*conn
);
187 static int connection_buf_read_from_socket(connection_t
*conn
,
188 ssize_t
*max_to_read
,
190 static int connection_process_inbuf(connection_t
*conn
, int package_partial
);
191 static void client_check_address_changed(tor_socket_t sock
);
192 static void set_constrained_socket_buffers(tor_socket_t sock
, int size
);
194 static const char *connection_proxy_state_to_string(int state
);
195 static int connection_read_https_proxy_response(connection_t
*conn
);
196 static void connection_send_socks5_connect(connection_t
*conn
);
197 static const char *proxy_type_to_string(int proxy_type
);
198 static int conn_get_proxy_type(const connection_t
*conn
);
199 const tor_addr_t
*conn_get_outbound_address(sa_family_t family
,
200 const or_options_t
*options
, unsigned int conn_type
);
201 static void reenable_blocked_connection_init(const or_options_t
*options
);
202 static void reenable_blocked_connection_schedule(void);
204 /** The last addresses that our network interface seemed to have been
205 * binding to. We use this as one way to detect when our IP changes.
207 * XXXX+ We should really use the entire list of interfaces here.
209 static tor_addr_t
*last_interface_ipv4
= NULL
;
210 /* DOCDOC last_interface_ipv6 */
211 static tor_addr_t
*last_interface_ipv6
= NULL
;
212 /** A list of tor_addr_t for addresses we've used in outgoing connections.
213 * Used to detect IP address changes. */
214 static smartlist_t
*outgoing_addrs
= NULL
;
216 #define CASE_ANY_LISTENER_TYPE \
217 case CONN_TYPE_OR_LISTENER: \
218 case CONN_TYPE_EXT_OR_LISTENER: \
219 case CONN_TYPE_AP_LISTENER: \
220 case CONN_TYPE_DIR_LISTENER: \
221 case CONN_TYPE_CONTROL_LISTENER: \
222 case CONN_TYPE_AP_TRANS_LISTENER: \
223 case CONN_TYPE_AP_NATD_LISTENER: \
224 case CONN_TYPE_AP_DNS_LISTENER: \
225 case CONN_TYPE_AP_HTTP_CONNECT_LISTENER: \
226 case CONN_TYPE_METRICS_LISTENER
228 /**************************************************************/
231 * Cast a `connection_t *` to a `listener_connection_t *`.
233 * Exit with an assertion failure if the input is not a
234 * `listener_connection_t`.
236 listener_connection_t
*
237 TO_LISTENER_CONN(connection_t
*c
)
239 tor_assert(c
->magic
== LISTENER_CONNECTION_MAGIC
);
240 return DOWNCAST(listener_connection_t
, c
);
244 * Cast a `const connection_t *` to a `const listener_connection_t *`.
246 * Exit with an assertion failure if the input is not a
247 * `listener_connection_t`.
249 const listener_connection_t
*
250 CONST_TO_LISTENER_CONN(const connection_t
*c
)
252 return TO_LISTENER_CONN((connection_t
*)c
);
256 connection_get_inbuf_len(const connection_t
*conn
)
258 return conn
->inbuf
? buf_datalen(conn
->inbuf
) : 0;
262 connection_get_outbuf_len(const connection_t
*conn
)
264 return conn
->outbuf
? buf_datalen(conn
->outbuf
) : 0;
268 * Return the human-readable name for the connection type <b>type</b>
271 conn_type_to_string(int type
)
275 case CONN_TYPE_OR_LISTENER
: return "OR listener";
276 case CONN_TYPE_OR
: return "OR";
277 case CONN_TYPE_EXIT
: return "Exit";
278 case CONN_TYPE_AP_LISTENER
: return "Socks listener";
279 case CONN_TYPE_AP_TRANS_LISTENER
:
280 return "Transparent pf/netfilter listener";
281 case CONN_TYPE_AP_NATD_LISTENER
: return "Transparent natd listener";
282 case CONN_TYPE_AP_DNS_LISTENER
: return "DNS listener";
283 case CONN_TYPE_AP
: return "Socks";
284 case CONN_TYPE_DIR_LISTENER
: return "Directory listener";
285 case CONN_TYPE_DIR
: return "Directory";
286 case CONN_TYPE_CONTROL_LISTENER
: return "Control listener";
287 case CONN_TYPE_CONTROL
: return "Control";
288 case CONN_TYPE_EXT_OR
: return "Extended OR";
289 case CONN_TYPE_EXT_OR_LISTENER
: return "Extended OR listener";
290 case CONN_TYPE_AP_HTTP_CONNECT_LISTENER
: return "HTTP tunnel listener";
291 case CONN_TYPE_METRICS_LISTENER
: return "Metrics listener";
292 case CONN_TYPE_METRICS
: return "Metrics";
294 log_warn(LD_BUG
, "unknown connection type %d", type
);
295 tor_snprintf(buf
, sizeof(buf
), "unknown [%d]", type
);
301 * Return the human-readable name for the connection state <b>state</b>
302 * for the connection type <b>type</b>
305 conn_state_to_string(int type
, int state
)
309 CASE_ANY_LISTENER_TYPE
:
310 if (state
== LISTENER_STATE_READY
)
315 case OR_CONN_STATE_CONNECTING
: return "connect()ing";
316 case OR_CONN_STATE_PROXY_HANDSHAKING
: return "handshaking (proxy)";
317 case OR_CONN_STATE_TLS_HANDSHAKING
: return "handshaking (TLS)";
318 case OR_CONN_STATE_TLS_CLIENT_RENEGOTIATING
:
319 return "renegotiating (TLS, v2 handshake)";
320 case OR_CONN_STATE_TLS_SERVER_RENEGOTIATING
:
321 return "waiting for renegotiation or V3 handshake";
322 case OR_CONN_STATE_OR_HANDSHAKING_V2
:
323 return "handshaking (Tor, v2 handshake)";
324 case OR_CONN_STATE_OR_HANDSHAKING_V3
:
325 return "handshaking (Tor, v3 handshake)";
326 case OR_CONN_STATE_OPEN
: return "open";
329 case CONN_TYPE_EXT_OR
:
331 case EXT_OR_CONN_STATE_AUTH_WAIT_AUTH_TYPE
:
332 return "waiting for authentication type";
333 case EXT_OR_CONN_STATE_AUTH_WAIT_CLIENT_NONCE
:
334 return "waiting for client nonce";
335 case EXT_OR_CONN_STATE_AUTH_WAIT_CLIENT_HASH
:
336 return "waiting for client hash";
337 case EXT_OR_CONN_STATE_OPEN
: return "open";
338 case EXT_OR_CONN_STATE_FLUSHING
: return "flushing final OKAY";
343 case EXIT_CONN_STATE_RESOLVING
: return "waiting for dest info";
344 case EXIT_CONN_STATE_CONNECTING
: return "connecting";
345 case EXIT_CONN_STATE_OPEN
: return "open";
346 case EXIT_CONN_STATE_RESOLVEFAILED
: return "resolve failed";
351 case AP_CONN_STATE_SOCKS_WAIT
: return "waiting for socks info";
352 case AP_CONN_STATE_NATD_WAIT
: return "waiting for natd dest info";
353 case AP_CONN_STATE_RENDDESC_WAIT
: return "waiting for rendezvous desc";
354 case AP_CONN_STATE_CONTROLLER_WAIT
: return "waiting for controller";
355 case AP_CONN_STATE_CIRCUIT_WAIT
: return "waiting for circuit";
356 case AP_CONN_STATE_CONNECT_WAIT
: return "waiting for connect response";
357 case AP_CONN_STATE_RESOLVE_WAIT
: return "waiting for resolve response";
358 case AP_CONN_STATE_OPEN
: return "open";
363 case DIR_CONN_STATE_CONNECTING
: return "connecting";
364 case DIR_CONN_STATE_CLIENT_SENDING
: return "client sending";
365 case DIR_CONN_STATE_CLIENT_READING
: return "client reading";
366 case DIR_CONN_STATE_CLIENT_FINISHED
: return "client finished";
367 case DIR_CONN_STATE_SERVER_COMMAND_WAIT
: return "waiting for command";
368 case DIR_CONN_STATE_SERVER_WRITING
: return "writing";
371 case CONN_TYPE_CONTROL
:
373 case CONTROL_CONN_STATE_OPEN
: return "open (protocol v1)";
374 case CONTROL_CONN_STATE_NEEDAUTH
:
375 return "waiting for authentication (protocol v1)";
381 return "uninitialized";
384 log_warn(LD_BUG
, "unknown connection state %d (type %d)", state
, type
);
385 tor_snprintf(buf
, sizeof(buf
),
386 "unknown state [%d] on unknown [%s] connection",
387 state
, conn_type_to_string(type
));
388 tor_assert_nonfatal_unreached_once();
393 * Helper: describe the peer or address of connection @a conn in a
394 * human-readable manner.
396 * Returns a pointer to a static buffer; future calls to
397 * connection_describe_peer_internal() will invalidate this buffer.
399 * If <b>include_preposition</b> is true, include a preposition before the
402 * Nobody should parse the output of this function; it can and will change in
403 * future versions of tor.
406 connection_describe_peer_internal(const connection_t
*conn
,
407 bool include_preposition
)
413 static char peer_buf
[256];
414 const tor_addr_t
*addr
= &conn
->addr
;
415 const char *address
= NULL
;
421 /* First, figure out the preposition to use */
422 switch (conn
->type
) {
423 CASE_ANY_LISTENER_TYPE
:
429 case CONN_TYPE_CONTROL
:
431 case CONN_TYPE_EXT_OR
:
439 /* Now figure out the address. */
440 if (conn
->socket_family
== AF_UNIX
) {
441 /* For unix sockets, we always use the `address` string. */
442 address
= conn
->address
? conn
->address
: "unix socket";
443 } else if (conn
->type
== CONN_TYPE_OR
) {
444 /* For OR connections, we have a lot to do. */
445 const or_connection_t
*or_conn
= CONST_TO_OR_CONN(conn
);
446 /* We report the IDs we're talking to... */
447 if (fast_digest_is_zero(or_conn
->identity_digest
)) {
448 // This could be a client, so scrub it. No identity to report.
451 const ed25519_public_key_t
*ed_id
=
452 connection_or_get_alleged_ed25519_id(or_conn
);
453 char ed_id_buf
[ED25519_BASE64_LEN
+1];
454 char rsa_id_buf
[HEX_DIGEST_LEN
+1];
456 ed25519_public_to_base64(ed_id_buf
, ed_id
);
458 strlcpy(ed_id_buf
, "<none>", sizeof(ed_id_buf
));
460 base16_encode(rsa_id_buf
, sizeof(rsa_id_buf
),
461 or_conn
->identity_digest
, DIGEST_LEN
);
462 tor_snprintf(extra_buf
, sizeof(extra_buf
),
463 " ID=%s RSA_ID=%s", ed_id_buf
, rsa_id_buf
);
465 if (! scrub
&& (! tor_addr_eq(addr
, &or_conn
->canonical_orport
.addr
) ||
466 conn
->port
!= or_conn
->canonical_orport
.port
)) {
467 /* We report canonical address, if it's different */
468 char canonical_addr_buf
[TOR_ADDR_BUF_LEN
];
469 if (tor_addr_to_str(canonical_addr_buf
, &or_conn
->canonical_orport
.addr
,
470 sizeof(canonical_addr_buf
), 1)) {
471 tor_snprintf(extra_buf
+strlen(extra_buf
),
472 sizeof(extra_buf
)-strlen(extra_buf
),
473 " canonical_addr=%s:%"PRIu16
,
475 or_conn
->canonical_orport
.port
);
478 } else if (conn
->type
== CONN_TYPE_EXIT
) {
479 scrub
= true; /* This is a client's request; scrub it with SafeLogging. */
480 if (tor_addr_is_null(addr
)) {
481 address
= conn
->address
;
482 strlcpy(extra_buf
, " (DNS lookup pending)", sizeof(extra_buf
));
486 char addr_buf
[TOR_ADDR_BUF_LEN
];
487 if (address
== NULL
) {
488 if (tor_addr_family(addr
) == 0) {
491 address
= tor_addr_to_str(addr_buf
, addr
, sizeof(addr_buf
), 1);
493 address
= "<can't format!>";
494 tor_assert_nonfatal_unreached_once();
501 if (scrub
&& get_options()->SafeLogging_
!= SAFELOG_SCRUB_NONE
) {
502 address
= "[scrubbed]";
504 /* Only set the port if we're not scrubbing the address. */
505 if (conn
->port
!= 0) {
506 tor_snprintf(portbuf
, sizeof(portbuf
), ":%d", conn
->port
);
510 const char *sp
= include_preposition
? " " : "";
511 if (! include_preposition
)
514 tor_snprintf(peer_buf
, sizeof(peer_buf
),
515 "%s%s%s%s%s", prep
, sp
, address
, portbuf
, extra_buf
);
520 * Describe the peer or address of connection @a conn in a
521 * human-readable manner.
523 * Returns a pointer to a static buffer; future calls to
524 * connection_describe_peer() or connection_describe() will invalidate this
527 * Nobody should parse the output of this function; it can and will change in
528 * future versions of tor.
531 connection_describe_peer(const connection_t
*conn
)
533 return connection_describe_peer_internal(conn
, false);
537 * Describe a connection for logging purposes.
539 * Returns a pointer to a static buffer; future calls to connection_describe()
540 * will invalidate this buffer.
542 * Nobody should parse the output of this function; it can and will change in
543 * future versions of tor.
546 connection_describe(const connection_t
*conn
)
549 return "null connection";
551 static char desc_buf
[256];
552 const char *peer
= connection_describe_peer_internal(conn
, true);
553 tor_snprintf(desc_buf
, sizeof(desc_buf
),
554 "%s connection (%s) %s",
555 conn_type_to_string(conn
->type
),
556 conn_state_to_string(conn
->type
, conn
->state
),
561 /** Allocate and return a new dir_connection_t, initialized as by
562 * connection_init(). */
564 dir_connection_new(int socket_family
)
566 dir_connection_t
*dir_conn
= tor_malloc_zero(sizeof(dir_connection_t
));
567 connection_init(time(NULL
), TO_CONN(dir_conn
), CONN_TYPE_DIR
, socket_family
);
571 /** Allocate and return a new or_connection_t, initialized as by
574 * Initialize active_circuit_pqueue.
576 * Set active_circuit_pqueue_last_recalibrated to current cell_ewma tick.
579 or_connection_new(int type
, int socket_family
)
581 or_connection_t
*or_conn
= tor_malloc_zero(sizeof(or_connection_t
));
582 time_t now
= time(NULL
);
583 tor_assert(type
== CONN_TYPE_OR
|| type
== CONN_TYPE_EXT_OR
);
584 connection_init(now
, TO_CONN(or_conn
), type
, socket_family
);
586 tor_addr_make_unspec(&or_conn
->canonical_orport
.addr
);
587 connection_or_set_canonical(or_conn
, 0);
589 if (type
== CONN_TYPE_EXT_OR
) {
590 /* If we aren't told an address for this connection, we should
591 * presume it isn't local, and should be rate-limited. */
592 TO_CONN(or_conn
)->always_rate_limit_as_remote
= 1;
593 connection_or_set_ext_or_identifier(or_conn
);
599 /** Allocate and return a new entry_connection_t, initialized as by
602 * Allocate space to store the socks_request.
605 entry_connection_new(int type
, int socket_family
)
607 entry_connection_t
*entry_conn
= tor_malloc_zero(sizeof(entry_connection_t
));
608 tor_assert(type
== CONN_TYPE_AP
);
609 connection_init(time(NULL
), ENTRY_TO_CONN(entry_conn
), type
, socket_family
);
610 entry_conn
->socks_request
= socks_request_new();
611 /* If this is coming from a listener, we'll set it up based on the listener
612 * in a little while. Otherwise, we're doing this as a linked connection
613 * of some kind, and we should set it up here based on the socket family */
614 if (socket_family
== AF_INET
)
615 entry_conn
->entry_cfg
.ipv4_traffic
= 1;
616 else if (socket_family
== AF_INET6
)
617 entry_conn
->entry_cfg
.ipv6_traffic
= 1;
619 /* Initialize the read token bucket to the maximum value which is the same as
620 * no rate limiting. */
621 token_bucket_rw_init(&ENTRY_TO_EDGE_CONN(entry_conn
)->bucket
, INT32_MAX
,
622 INT32_MAX
, monotime_coarse_get_stamp());
626 /** Allocate and return a new edge_connection_t, initialized as by
627 * connection_init(). */
629 edge_connection_new(int type
, int socket_family
)
631 edge_connection_t
*edge_conn
= tor_malloc_zero(sizeof(edge_connection_t
));
632 tor_assert(type
== CONN_TYPE_EXIT
);
633 connection_init(time(NULL
), TO_CONN(edge_conn
), type
, socket_family
);
634 /* Initialize the read token bucket to the maximum value which is the same as
635 * no rate limiting. */
636 token_bucket_rw_init(&edge_conn
->bucket
, INT32_MAX
, INT32_MAX
,
637 monotime_coarse_get_stamp());
641 /** Allocate and return a new control_connection_t, initialized as by
642 * connection_init(). */
643 control_connection_t
*
644 control_connection_new(int socket_family
)
646 control_connection_t
*control_conn
=
647 tor_malloc_zero(sizeof(control_connection_t
));
648 connection_init(time(NULL
),
649 TO_CONN(control_conn
), CONN_TYPE_CONTROL
, socket_family
);
653 /** Allocate and return a new listener_connection_t, initialized as by
654 * connection_init(). */
655 listener_connection_t
*
656 listener_connection_new(int type
, int socket_family
)
658 listener_connection_t
*listener_conn
=
659 tor_malloc_zero(sizeof(listener_connection_t
));
660 connection_init(time(NULL
), TO_CONN(listener_conn
), type
, socket_family
);
661 return listener_conn
;
664 /** Allocate, initialize, and return a new connection_t subtype of <b>type</b>
665 * to make or receive connections of address family <b>socket_family</b>. The
666 * type should be one of the CONN_TYPE_* constants. */
668 connection_new(int type
, int socket_family
)
672 case CONN_TYPE_EXT_OR
:
673 return TO_CONN(or_connection_new(type
, socket_family
));
676 return TO_CONN(edge_connection_new(type
, socket_family
));
679 return ENTRY_TO_CONN(entry_connection_new(type
, socket_family
));
682 return TO_CONN(dir_connection_new(socket_family
));
684 case CONN_TYPE_CONTROL
:
685 return TO_CONN(control_connection_new(socket_family
));
687 CASE_ANY_LISTENER_TYPE
:
688 return TO_CONN(listener_connection_new(type
, socket_family
));
691 connection_t
*conn
= tor_malloc_zero(sizeof(connection_t
));
692 connection_init(time(NULL
), conn
, type
, socket_family
);
698 /** Initializes conn. (you must call connection_add() to link it into the main
701 * Set conn-\>magic to the correct value.
703 * Set conn-\>type to <b>type</b>. Set conn-\>s and conn-\>conn_array_index to
704 * -1 to signify they are not yet assigned.
706 * Initialize conn's timestamps to now.
709 connection_init(time_t now
, connection_t
*conn
, int type
, int socket_family
)
711 static uint64_t n_connections_allocated
= 1;
715 case CONN_TYPE_EXT_OR
:
716 conn
->magic
= OR_CONNECTION_MAGIC
;
719 conn
->magic
= EDGE_CONNECTION_MAGIC
;
722 conn
->magic
= ENTRY_CONNECTION_MAGIC
;
725 conn
->magic
= DIR_CONNECTION_MAGIC
;
727 case CONN_TYPE_CONTROL
:
728 conn
->magic
= CONTROL_CONNECTION_MAGIC
;
730 CASE_ANY_LISTENER_TYPE
:
731 conn
->magic
= LISTENER_CONNECTION_MAGIC
;
734 conn
->magic
= BASE_CONNECTION_MAGIC
;
738 conn
->s
= TOR_INVALID_SOCKET
; /* give it a default of 'not used' */
739 conn
->conn_array_index
= -1; /* also default to 'not used' */
740 conn
->global_identifier
= n_connections_allocated
++;
743 conn
->socket_family
= socket_family
;
744 if (!connection_is_listener(conn
)) {
745 /* listeners never use their buf */
746 conn
->inbuf
= buf_new();
747 conn
->outbuf
= buf_new();
750 conn
->timestamp_created
= now
;
751 conn
->timestamp_last_read_allowed
= now
;
752 conn
->timestamp_last_write_allowed
= now
;
755 /** Create a link between <b>conn_a</b> and <b>conn_b</b>. */
757 connection_link_connections(connection_t
*conn_a
, connection_t
*conn_b
)
759 tor_assert(! SOCKET_OK(conn_a
->s
));
760 tor_assert(! SOCKET_OK(conn_b
->s
));
764 conn_a
->linked_conn
= conn_b
;
765 conn_b
->linked_conn
= conn_a
;
768 /** Return true iff the provided connection listener type supports AF_UNIX
771 conn_listener_type_supports_af_unix(int type
)
773 /* For now only control ports or SOCKS ports can be Unix domain sockets
774 * and listeners at the same time */
776 case CONN_TYPE_CONTROL_LISTENER
:
777 case CONN_TYPE_AP_LISTENER
:
784 /** Deallocate memory used by <b>conn</b>. Deallocate its buffers if
785 * necessary, close its socket if necessary, and mark the directory as dirty
786 * if <b>conn</b> is an OR or OP connection.
789 connection_free_minimal(connection_t
*conn
)
796 switch (conn
->type
) {
798 case CONN_TYPE_EXT_OR
:
799 tor_assert(conn
->magic
== OR_CONNECTION_MAGIC
);
800 mem
= TO_OR_CONN(conn
);
801 memlen
= sizeof(or_connection_t
);
804 tor_assert(conn
->magic
== ENTRY_CONNECTION_MAGIC
);
805 mem
= TO_ENTRY_CONN(conn
);
806 memlen
= sizeof(entry_connection_t
);
809 tor_assert(conn
->magic
== EDGE_CONNECTION_MAGIC
);
810 mem
= TO_EDGE_CONN(conn
);
811 memlen
= sizeof(edge_connection_t
);
814 tor_assert(conn
->magic
== DIR_CONNECTION_MAGIC
);
815 mem
= TO_DIR_CONN(conn
);
816 memlen
= sizeof(dir_connection_t
);
818 case CONN_TYPE_CONTROL
:
819 tor_assert(conn
->magic
== CONTROL_CONNECTION_MAGIC
);
820 mem
= TO_CONTROL_CONN(conn
);
821 memlen
= sizeof(control_connection_t
);
823 CASE_ANY_LISTENER_TYPE
:
824 tor_assert(conn
->magic
== LISTENER_CONNECTION_MAGIC
);
825 mem
= TO_LISTENER_CONN(conn
);
826 memlen
= sizeof(listener_connection_t
);
829 tor_assert(conn
->magic
== BASE_CONNECTION_MAGIC
);
831 memlen
= sizeof(connection_t
);
836 log_info(LD_GENERAL
, "Freeing linked %s connection [%s] with %d "
837 "bytes on inbuf, %d on outbuf.",
838 conn_type_to_string(conn
->type
),
839 conn_state_to_string(conn
->type
, conn
->state
),
840 (int)connection_get_inbuf_len(conn
),
841 (int)connection_get_outbuf_len(conn
));
844 if (!connection_is_listener(conn
)) {
845 buf_free(conn
->inbuf
);
846 buf_free(conn
->outbuf
);
848 if (conn
->socket_family
== AF_UNIX
) {
849 /* For now only control and SOCKS ports can be Unix domain sockets
850 * and listeners at the same time */
851 tor_assert(conn_listener_type_supports_af_unix(conn
->type
));
853 if (unlink(conn
->address
) < 0 && errno
!= ENOENT
) {
854 log_warn(LD_NET
, "Could not unlink %s: %s", conn
->address
,
860 tor_str_wipe_and_free(conn
->address
);
862 if (connection_speaks_cells(conn
)) {
863 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
865 if (! SOCKET_OK(conn
->s
)) {
866 /* The socket has been closed by somebody else; we must tell the
867 * TLS object not to close it. */
868 tor_tls_release_socket(or_conn
->tls
);
870 /* The tor_tls_free() call below will close the socket; we must tell
871 * the code below not to close it a second time. */
872 tor_release_socket_ownership(conn
->s
);
873 conn
->s
= TOR_INVALID_SOCKET
;
875 tor_tls_free(or_conn
->tls
);
878 or_handshake_state_free(or_conn
->handshake_state
);
879 or_conn
->handshake_state
= NULL
;
880 tor_str_wipe_and_free(or_conn
->nickname
);
882 /* Owww, this shouldn't happen, but... */
883 channel_t
*base_chan
= TLS_CHAN_TO_BASE(or_conn
->chan
);
884 tor_assert(base_chan
);
886 "Freeing orconn at %p, saw channel %p with ID "
887 "%"PRIu64
" left un-NULLed",
889 base_chan
->global_identifier
);
890 if (!CHANNEL_FINISHED(base_chan
)) {
891 channel_close_for_error(base_chan
);
894 or_conn
->chan
->conn
= NULL
;
895 or_conn
->chan
= NULL
;
898 if (conn
->type
== CONN_TYPE_AP
) {
899 entry_connection_t
*entry_conn
= TO_ENTRY_CONN(conn
);
900 tor_str_wipe_and_free(entry_conn
->chosen_exit_name
);
901 tor_str_wipe_and_free(entry_conn
->original_dest_address
);
902 if (entry_conn
->socks_request
)
903 socks_request_free(entry_conn
->socks_request
);
904 if (entry_conn
->pending_optimistic_data
) {
905 buf_free(entry_conn
->pending_optimistic_data
);
907 if (entry_conn
->sending_optimistic_data
) {
908 buf_free(entry_conn
->sending_optimistic_data
);
911 if (CONN_IS_EDGE(conn
)) {
912 hs_ident_edge_conn_free(TO_EDGE_CONN(conn
)->hs_ident
);
914 if (conn
->type
== CONN_TYPE_CONTROL
) {
915 control_connection_t
*control_conn
= TO_CONTROL_CONN(conn
);
916 tor_free(control_conn
->safecookie_client_hash
);
917 tor_free(control_conn
->incoming_cmd
);
918 tor_free(control_conn
->current_cmd
);
919 if (control_conn
->ephemeral_onion_services
) {
920 SMARTLIST_FOREACH(control_conn
->ephemeral_onion_services
, char *, cp
, {
921 memwipe(cp
, 0, strlen(cp
));
924 smartlist_free(control_conn
->ephemeral_onion_services
);
928 /* Probably already freed by connection_free. */
929 tor_event_free(conn
->read_event
);
930 tor_event_free(conn
->write_event
);
931 conn
->read_event
= conn
->write_event
= NULL
;
933 if (conn
->type
== CONN_TYPE_DIR
) {
934 dir_connection_t
*dir_conn
= TO_DIR_CONN(conn
);
935 tor_free(dir_conn
->requested_resource
);
937 tor_compress_free(dir_conn
->compress_state
);
938 dir_conn_clear_spool(dir_conn
);
940 hs_ident_dir_conn_free(dir_conn
->hs_ident
);
941 if (dir_conn
->guard_state
) {
942 /* Cancel before freeing, if it's still there. */
943 entry_guard_cancel(&dir_conn
->guard_state
);
945 circuit_guard_state_free(dir_conn
->guard_state
);
948 if (SOCKET_OK(conn
->s
)) {
949 log_debug(LD_NET
,"closing fd %d.",(int)conn
->s
);
950 tor_close_socket(conn
->s
);
951 conn
->s
= TOR_INVALID_SOCKET
;
954 if (conn
->type
== CONN_TYPE_OR
&&
955 !tor_digest_is_zero(TO_OR_CONN(conn
)->identity_digest
)) {
956 log_warn(LD_BUG
, "called on OR conn with non-zeroed identity_digest");
957 connection_or_clear_identity(TO_OR_CONN(conn
));
959 if (conn
->type
== CONN_TYPE_OR
|| conn
->type
== CONN_TYPE_EXT_OR
) {
960 tor_free(TO_OR_CONN(conn
)->ext_or_conn_id
);
961 tor_free(TO_OR_CONN(conn
)->ext_or_auth_correct_client_hash
);
962 tor_free(TO_OR_CONN(conn
)->ext_or_transport
);
965 memwipe(mem
, 0xCC, memlen
); /* poison memory */
969 /** Make sure <b>conn</b> isn't in any of the global conn lists; then free it.
972 connection_free_
,(connection_t
*conn
))
976 tor_assert(!connection_is_on_closeable_list(conn
));
977 tor_assert(!connection_in_array(conn
));
978 if (BUG(conn
->linked_conn
)) {
979 conn
->linked_conn
->linked_conn
= NULL
;
980 if (! conn
->linked_conn
->marked_for_close
&&
981 conn
->linked_conn
->reading_from_linked_conn
)
982 connection_start_reading(conn
->linked_conn
);
983 conn
->linked_conn
= NULL
;
985 if (connection_speaks_cells(conn
)) {
986 if (!tor_digest_is_zero(TO_OR_CONN(conn
)->identity_digest
)) {
987 connection_or_clear_identity(TO_OR_CONN(conn
));
990 if (conn
->type
== CONN_TYPE_CONTROL
) {
991 connection_control_closed(TO_CONTROL_CONN(conn
));
995 if (conn
->type
== CONN_TYPE_AP
) {
996 connection_ap_warn_and_unmark_if_pending_circ(TO_ENTRY_CONN(conn
),
1001 /* Notify the circuit creation DoS mitigation subsystem that an OR client
1002 * connection has been closed. And only do that if we track it. */
1003 if (conn
->type
== CONN_TYPE_OR
) {
1004 dos_close_client_conn(TO_OR_CONN(conn
));
1007 connection_unregister_events(conn
);
1008 connection_free_minimal(conn
);
1012 * Called when we're about to finally unlink and free a connection:
1013 * perform necessary accounting and cleanup
1014 * - Directory conns that failed to fetch a rendezvous descriptor
1015 * need to inform pending rendezvous streams.
1016 * - OR conns need to call rep_hist_note_*() to record status.
1017 * - AP conns need to send a socks reject if necessary.
1018 * - Exit conns need to call connection_dns_remove() if necessary.
1019 * - AP and Exit conns need to send an end cell if they can.
1020 * - DNS conns need to fail any resolves that are pending on them.
1021 * - OR and edge connections need to be unlinked from circuits.
1024 connection_about_to_close_connection(connection_t
*conn
)
1026 tor_assert(conn
->marked_for_close
);
1028 switch (conn
->type
) {
1030 connection_dir_about_to_close(TO_DIR_CONN(conn
));
1033 case CONN_TYPE_EXT_OR
:
1034 connection_or_about_to_close(TO_OR_CONN(conn
));
1037 connection_ap_about_to_close(TO_ENTRY_CONN(conn
));
1039 case CONN_TYPE_EXIT
:
1040 connection_exit_about_to_close(TO_EDGE_CONN(conn
));
1045 /** Return true iff connection_close_immediate() has been called on this
1047 #define CONN_IS_CLOSED(c) \
1048 ((c)->linked ? ((c)->linked_conn_is_closed) : (! SOCKET_OK(c->s)))
1050 /** Close the underlying socket for <b>conn</b>, so we don't try to
1051 * flush it. Must be used in conjunction with (right before)
1052 * connection_mark_for_close().
1055 connection_close_immediate(connection_t
*conn
)
1057 assert_connection_ok(conn
,0);
1058 if (CONN_IS_CLOSED(conn
)) {
1059 log_err(LD_BUG
,"Attempt to close already-closed connection.");
1060 tor_fragile_assert();
1063 if (connection_get_outbuf_len(conn
)) {
1064 log_info(LD_NET
,"fd %d, type %s, state %s, %"TOR_PRIuSZ
" bytes on outbuf.",
1065 (int)conn
->s
, conn_type_to_string(conn
->type
),
1066 conn_state_to_string(conn
->type
, conn
->state
),
1067 buf_datalen(conn
->outbuf
));
1070 connection_unregister_events(conn
);
1072 /* Prevent the event from getting unblocked. */
1073 conn
->read_blocked_on_bw
= 0;
1074 conn
->write_blocked_on_bw
= 0;
1076 if (SOCKET_OK(conn
->s
))
1077 tor_close_socket(conn
->s
);
1078 conn
->s
= TOR_INVALID_SOCKET
;
1080 conn
->linked_conn_is_closed
= 1;
1082 buf_clear(conn
->outbuf
);
1085 /** Mark <b>conn</b> to be closed next time we loop through
1086 * conn_close_if_marked() in main.c. */
1088 connection_mark_for_close_(connection_t
*conn
, int line
, const char *file
)
1090 assert_connection_ok(conn
,0);
1092 tor_assert(line
< 1<<16); /* marked_for_close can only fit a uint16_t. */
1095 if (conn
->type
== CONN_TYPE_OR
) {
1097 * An or_connection should have been closed through one of the channel-
1098 * aware functions in connection_or.c. We'll assume this is an error
1099 * close and do that, and log a bug warning.
1101 log_warn(LD_CHANNEL
| LD_BUG
,
1102 "Something tried to close an or_connection_t without going "
1103 "through channels at %s:%d",
1105 connection_or_close_for_error(TO_OR_CONN(conn
), 0);
1107 /* Pass it down to the real function */
1108 connection_mark_for_close_internal_(conn
, line
, file
);
1112 /** Mark <b>conn</b> to be closed next time we loop through
1113 * conn_close_if_marked() in main.c.
1115 * This _internal version bypasses the CONN_TYPE_OR checks; this should be
1116 * called when you either are sure that if this is an or_connection_t the
1117 * controlling channel has been notified (e.g. with
1118 * connection_or_notify_error()), or you actually are the
1119 * connection_or_close_for_error() or connection_or_close_normally() function.
1120 * For all other cases, use connection_mark_and_flush() which checks for
1121 * or_connection_t properly, instead. See below.
1123 * We want to keep this function simple and quick, since it can be called from
1124 * quite deep in the call chain, and hence it should avoid having side-effects
1125 * that interfere with its callers view of the connection.
1128 connection_mark_for_close_internal_
, (connection_t
*conn
,
1129 int line
, const char *file
))
1131 assert_connection_ok(conn
,0);
1133 tor_assert(line
< 1<<16); /* marked_for_close can only fit a uint16_t. */
1136 if (conn
->marked_for_close
) {
1137 log_warn(LD_BUG
,"Duplicate call to connection_mark_for_close at %s:%d"
1138 " (first at %s:%d)", file
, line
, conn
->marked_for_close_file
,
1139 conn
->marked_for_close
);
1140 tor_fragile_assert();
1144 if (conn
->type
== CONN_TYPE_OR
) {
1146 * Bad news if this happens without telling the controlling channel; do
1147 * this so we can find things that call this wrongly when the asserts hit.
1149 log_debug(LD_CHANNEL
,
1150 "Calling connection_mark_for_close_internal_() on an OR conn "
1155 conn
->marked_for_close
= line
;
1156 conn
->marked_for_close_file
= file
;
1157 add_connection_to_closeable_list(conn
);
1159 /* in case we're going to be held-open-til-flushed, reset
1160 * the number of seconds since last successful write, so
1161 * we get our whole 15 seconds */
1162 conn
->timestamp_last_write_allowed
= time(NULL
);
1165 /** Find each connection that has hold_open_until_flushed set to
1166 * 1 but hasn't written in the past 15 seconds, and set
1167 * hold_open_until_flushed to 0. This means it will get cleaned
1168 * up in the next loop through close_if_marked() in main.c.
1171 connection_expire_held_open(void)
1174 smartlist_t
*conns
= get_connection_array();
1178 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, conn
) {
1179 /* If we've been holding the connection open, but we haven't written
1182 if (conn
->hold_open_until_flushed
) {
1183 tor_assert(conn
->marked_for_close
);
1184 if (now
- conn
->timestamp_last_write_allowed
>= 15) {
1186 if (conn
->type
== CONN_TYPE_EXIT
||
1187 (conn
->type
== CONN_TYPE_DIR
&&
1188 conn
->purpose
== DIR_PURPOSE_SERVER
))
1189 severity
= LOG_INFO
;
1191 severity
= LOG_NOTICE
;
1192 log_fn(severity
, LD_NET
,
1193 "Giving up on marked_for_close conn that's been flushing "
1194 "for 15s (fd %d, type %s, state %s).",
1195 (int)conn
->s
, conn_type_to_string(conn
->type
),
1196 conn_state_to_string(conn
->type
, conn
->state
));
1197 conn
->hold_open_until_flushed
= 0;
1200 } SMARTLIST_FOREACH_END(conn
);
1203 #if defined(HAVE_SYS_UN_H) || defined(RUNNING_DOXYGEN)
1204 /** Create an AF_UNIX listenaddr struct.
1205 * <b>listenaddress</b> provides the path to the Unix socket.
1207 * Eventually <b>listenaddress</b> will also optionally contain user, group,
1208 * and file permissions for the new socket. But not yet. XXX
1209 * Also, since we do not create the socket here the information doesn't help
1212 * If not NULL <b>readable_address</b> will contain a copy of the path part of
1213 * <b>listenaddress</b>.
1215 * The listenaddr struct has to be freed by the caller.
1217 static struct sockaddr_un
*
1218 create_unix_sockaddr(const char *listenaddress
, char **readable_address
,
1221 struct sockaddr_un
*sockaddr
= NULL
;
1223 sockaddr
= tor_malloc_zero(sizeof(struct sockaddr_un
));
1224 sockaddr
->sun_family
= AF_UNIX
;
1225 if (strlcpy(sockaddr
->sun_path
, listenaddress
, sizeof(sockaddr
->sun_path
))
1226 >= sizeof(sockaddr
->sun_path
)) {
1227 log_warn(LD_CONFIG
, "Unix socket path '%s' is too long to fit.",
1228 escaped(listenaddress
));
1233 if (readable_address
)
1234 *readable_address
= tor_strdup(listenaddress
);
1236 *len_out
= sizeof(struct sockaddr_un
);
1239 #else /* !(defined(HAVE_SYS_UN_H) || defined(RUNNING_DOXYGEN)) */
1240 static struct sockaddr
*
1241 create_unix_sockaddr(const char *listenaddress
, char **readable_address
,
1244 (void)listenaddress
;
1245 (void)readable_address
;
1246 log_fn(LOG_ERR
, LD_BUG
,
1247 "Unix domain sockets not supported, yet we tried to create one.");
1249 tor_fragile_assert();
1252 #endif /* defined(HAVE_SYS_UN_H) || defined(RUNNING_DOXYGEN) */
1255 * A socket failed from resource exhaustion.
1257 * Among other actions, warn that an accept or a connect has failed because
1258 * we're running out of TCP sockets we can use on current system. Rate-limit
1259 * these warnings so that we don't spam the log. */
1261 socket_failed_from_resource_exhaustion(void)
1263 /* When we get to this point we know that a socket could not be
1264 * established. However the kernel does not let us know whether the reason is
1265 * because we ran out of TCP source ports, or because we exhausted all the
1266 * FDs on this system, or for any other reason.
1268 * For this reason, we are going to use the following heuristic: If our
1269 * system supports a lot of sockets, we will assume that it's a problem of
1270 * TCP port exhaustion. Otherwise, if our system does not support many
1271 * sockets, we will assume that this is because of file descriptor
1274 if (get_max_sockets() > 65535) {
1275 /* TCP port exhaustion */
1276 rep_hist_note_tcp_exhaustion();
1278 /* File descriptor exhaustion */
1279 rep_hist_note_overload(OVERLOAD_FD_EXHAUSTED
);
1282 #define WARN_TOO_MANY_CONNS_INTERVAL (6*60*60)
1283 static ratelim_t last_warned
= RATELIM_INIT(WARN_TOO_MANY_CONNS_INTERVAL
);
1285 if ((m
= rate_limit_log(&last_warned
, approx_time()))) {
1286 int n_conns
= get_n_open_sockets();
1287 log_warn(LD_NET
,"Failing because we have %d connections already. Please "
1288 "read doc/TUNING for guidance.%s", n_conns
, m
);
1290 control_event_general_status(LOG_WARN
, "TOO_MANY_CONNECTIONS CURRENT=%d",
1295 #ifdef HAVE_SYS_UN_H
1297 #define UNIX_SOCKET_PURPOSE_CONTROL_SOCKET 0
1298 #define UNIX_SOCKET_PURPOSE_SOCKS_SOCKET 1
1300 /** Check if the purpose isn't one of the ones we know what to do with */
1303 is_valid_unix_socket_purpose(int purpose
)
1308 case UNIX_SOCKET_PURPOSE_CONTROL_SOCKET
:
1309 case UNIX_SOCKET_PURPOSE_SOCKS_SOCKET
:
1317 /** Return a string description of a unix socket purpose */
1319 unix_socket_purpose_to_string(int purpose
)
1321 const char *s
= "unknown-purpose socket";
1324 case UNIX_SOCKET_PURPOSE_CONTROL_SOCKET
:
1325 s
= "control socket";
1327 case UNIX_SOCKET_PURPOSE_SOCKS_SOCKET
:
1335 /** Check whether we should be willing to open an AF_UNIX socket in
1336 * <b>path</b>. Return 0 if we should go ahead and -1 if we shouldn't. */
1338 check_location_for_unix_socket(const or_options_t
*options
, const char *path
,
1339 int purpose
, const port_cfg_t
*port
)
1344 tor_assert(is_valid_unix_socket_purpose(purpose
));
1346 p
= tor_strdup(path
);
1347 cpd_check_t flags
= CPD_CHECK_MODE_ONLY
;
1348 if (get_parent_directory(p
)<0 || p
[0] != '/') {
1349 log_warn(LD_GENERAL
, "Bad unix socket address '%s'. Tor does not support "
1350 "relative paths for unix sockets.", path
);
1354 if (port
->is_world_writable
) {
1355 /* World-writable sockets can go anywhere. */
1360 if (port
->is_group_writable
) {
1361 flags
|= CPD_GROUP_OK
;
1364 if (port
->relax_dirmode_check
) {
1365 flags
|= CPD_RELAX_DIRMODE_CHECK
;
1368 if (check_private_dir(p
, flags
, options
->User
) < 0) {
1369 char *escpath
, *escdir
;
1370 escpath
= esc_for_log(path
);
1371 escdir
= esc_for_log(p
);
1372 log_warn(LD_GENERAL
, "Before Tor can create a %s in %s, the directory "
1373 "%s needs to exist, and to be accessible only by the user%s "
1374 "account that is running Tor. (On some Unix systems, anybody "
1375 "who can list a socket can connect to it, so Tor is being "
1377 unix_socket_purpose_to_string(purpose
), escpath
, escdir
,
1378 port
->is_group_writable
? " and group" : "");
1389 #endif /* defined(HAVE_SYS_UN_H) */
1391 /** Tell the TCP stack that it shouldn't wait for a long time after
1392 * <b>sock</b> has closed before reusing its port. Return 0 on success,
1395 make_socket_reuseable(tor_socket_t sock
)
1403 /* REUSEADDR on normal places means you can rebind to the port
1404 * right after somebody else has let it go. But REUSEADDR on win32
1405 * means you can bind to the port _even when somebody else
1406 * already has it bound_. So, don't do that on Win32. */
1407 if (setsockopt(sock
, SOL_SOCKET
, SO_REUSEADDR
, (void*) &one
,
1408 (socklen_t
)sizeof(one
)) == -1) {
1412 #endif /* defined(_WIN32) */
1416 /** Tell the Windows TCP stack to prevent other applications from receiving
1417 * traffic from tor's open ports. Return 0 on success, -1 on failure. */
1419 make_win32_socket_exclusive(tor_socket_t sock
)
1421 #ifdef SO_EXCLUSIVEADDRUSE
1424 /* Any socket that sets REUSEADDR on win32 can bind to a port _even when
1425 * somebody else already has it bound_, and _even if the original socket
1426 * didn't set REUSEADDR_. Use EXCLUSIVEADDRUSE to prevent this port-stealing
1428 if (setsockopt(sock
, SOL_SOCKET
, SO_EXCLUSIVEADDRUSE
, (void*) &one
,
1429 (socklen_t
)sizeof(one
))) {
1433 #else /* !defined(SO_EXCLUSIVEADDRUSE) */
1436 #endif /* defined(SO_EXCLUSIVEADDRUSE) */
1438 #endif /* defined(_WIN32) */
1440 /** Max backlog to pass to listen. We start at */
1441 static int listen_limit
= INT_MAX
;
1443 /* Listen on <b>fd</b> with appropriate backlog. Return as for listen. */
1445 tor_listen(tor_socket_t fd
)
1449 if ((r
= listen(fd
, listen_limit
)) < 0) {
1450 if (listen_limit
== SOMAXCONN
)
1452 if ((r
= listen(fd
, SOMAXCONN
)) == 0) {
1453 listen_limit
= SOMAXCONN
;
1454 log_warn(LD_NET
, "Setting listen backlog to INT_MAX connections "
1455 "didn't work, but SOMAXCONN did. Lowering backlog limit.");
1461 /** Bind a new non-blocking socket listening to the socket described
1462 * by <b>listensockaddr</b>.
1464 * <b>address</b> is only used for logging purposes and to add the information
1467 * Set <b>addr_in_use</b> to true in case socket binding fails with
1470 static connection_t
*
1471 connection_listener_new(const struct sockaddr
*listensockaddr
,
1473 int type
, const char *address
,
1474 const port_cfg_t
*port_cfg
,
1477 listener_connection_t
*lis_conn
;
1478 connection_t
*conn
= NULL
;
1479 tor_socket_t s
= TOR_INVALID_SOCKET
; /* the socket we're going to make */
1480 or_options_t
const *options
= get_options();
1481 (void) options
; /* Windows doesn't use this. */
1482 #if defined(HAVE_PWD_H) && defined(HAVE_SYS_UN_H)
1483 const struct passwd
*pw
= NULL
;
1485 uint16_t usePort
= 0, gotPort
= 0;
1486 int start_reading
= 0;
1487 static int global_next_session_group
= SESSION_GROUP_FIRST_AUTO
;
1494 if (listensockaddr
->sa_family
== AF_INET
||
1495 listensockaddr
->sa_family
== AF_INET6
) {
1496 int is_stream
= (type
!= CONN_TYPE_AP_DNS_LISTENER
);
1500 tor_addr_from_sockaddr(&addr
, listensockaddr
, &usePort
);
1501 log_notice(LD_NET
, "Opening %s on %s",
1502 conn_type_to_string(type
), fmt_addrport(&addr
, usePort
));
1504 s
= tor_open_socket_nonblocking(tor_addr_family(&addr
),
1505 is_stream
? SOCK_STREAM
: SOCK_DGRAM
,
1506 is_stream
? IPPROTO_TCP
: IPPROTO_UDP
);
1507 if (!SOCKET_OK(s
)) {
1508 int e
= tor_socket_errno(s
);
1509 if (ERRNO_IS_RESOURCE_LIMIT(e
)) {
1510 socket_failed_from_resource_exhaustion();
1512 * We'll call the OOS handler at the error exit, so set the
1513 * exhaustion flag for it.
1517 log_warn(LD_NET
, "Socket creation failed: %s",
1518 tor_socket_strerror(e
));
1523 if (make_socket_reuseable(s
) < 0) {
1524 log_warn(LD_NET
, "Error setting SO_REUSEADDR flag on %s: %s",
1525 conn_type_to_string(type
),
1526 tor_socket_strerror(errno
));
1530 if (make_win32_socket_exclusive(s
) < 0) {
1531 log_warn(LD_NET
, "Error setting SO_EXCLUSIVEADDRUSE flag on %s: %s",
1532 conn_type_to_string(type
),
1533 tor_socket_strerror(errno
));
1535 #endif /* defined(_WIN32) */
1537 #if defined(USE_TRANSPARENT) && defined(IP_TRANSPARENT)
1538 if (options
->TransProxyType_parsed
== TPT_TPROXY
&&
1539 type
== CONN_TYPE_AP_TRANS_LISTENER
) {
1541 if (setsockopt(s
, SOL_IP
, IP_TRANSPARENT
, (void*)&one
,
1542 (socklen_t
)sizeof(one
)) < 0) {
1543 const char *extra
= "";
1544 int e
= tor_socket_errno(s
);
1546 extra
= "TransTPROXY requires root privileges or similar"
1548 log_warn(LD_NET
, "Error setting IP_TRANSPARENT flag: %s.%s",
1549 tor_socket_strerror(e
), extra
);
1552 #endif /* defined(USE_TRANSPARENT) && defined(IP_TRANSPARENT) */
1555 if (listensockaddr
->sa_family
== AF_INET6
) {
1557 /* We need to set IPV6_V6ONLY so that this socket can't get used for
1558 * IPv4 connections. */
1559 if (setsockopt(s
,IPPROTO_IPV6
, IPV6_V6ONLY
,
1560 (void*)&one
, (socklen_t
)sizeof(one
)) < 0) {
1561 int e
= tor_socket_errno(s
);
1562 log_warn(LD_NET
, "Error setting IPV6_V6ONLY flag: %s",
1563 tor_socket_strerror(e
));
1564 /* Keep going; probably not harmful. */
1567 #endif /* defined(IPV6_V6ONLY) */
1569 if (bind(s
,listensockaddr
,socklen
) < 0) {
1570 const char *helpfulhint
= "";
1571 int e
= tor_socket_errno(s
);
1572 if (ERRNO_IS_EADDRINUSE(e
)) {
1573 helpfulhint
= ". Is Tor already running?";
1577 log_warn(LD_NET
, "Could not bind to %s:%u: %s%s", address
, usePort
,
1578 tor_socket_strerror(e
), helpfulhint
);
1583 if (tor_listen(s
) < 0) {
1584 log_warn(LD_NET
, "Could not listen on %s:%u: %s", address
, usePort
,
1585 tor_socket_strerror(tor_socket_errno(s
)));
1594 struct sockaddr_storage ss
;
1595 socklen_t ss_len
=sizeof(ss
);
1596 if (getsockname(s
, (struct sockaddr
*)&ss
, &ss_len
)<0) {
1597 log_warn(LD_NET
, "getsockname() couldn't learn address for %s: %s",
1598 conn_type_to_string(type
),
1599 tor_socket_strerror(tor_socket_errno(s
)));
1602 tor_addr_from_sockaddr(&addr2
, (struct sockaddr
*)&ss
, &gotPort
);
1604 #ifdef HAVE_SYS_UN_H
1606 * AF_UNIX generic setup stuff
1608 } else if (listensockaddr
->sa_family
== AF_UNIX
) {
1609 /* We want to start reading for both AF_UNIX cases */
1612 tor_assert(conn_listener_type_supports_af_unix(type
));
1614 if (check_location_for_unix_socket(options
, address
,
1615 (type
== CONN_TYPE_CONTROL_LISTENER
) ?
1616 UNIX_SOCKET_PURPOSE_CONTROL_SOCKET
:
1617 UNIX_SOCKET_PURPOSE_SOCKS_SOCKET
, port_cfg
) < 0) {
1621 log_notice(LD_NET
, "Opening %s on %s",
1622 conn_type_to_string(type
), address
);
1624 tor_addr_make_unspec(&addr
);
1626 if (unlink(address
) < 0 && errno
!= ENOENT
) {
1627 log_warn(LD_NET
, "Could not unlink %s: %s", address
,
1632 s
= tor_open_socket_nonblocking(AF_UNIX
, SOCK_STREAM
, 0);
1633 if (! SOCKET_OK(s
)) {
1634 int e
= tor_socket_errno(s
);
1635 if (ERRNO_IS_RESOURCE_LIMIT(e
)) {
1636 socket_failed_from_resource_exhaustion();
1638 * We'll call the OOS handler at the error exit, so set the
1639 * exhaustion flag for it.
1643 log_warn(LD_NET
,"Socket creation failed: %s.", strerror(e
));
1648 if (bind(s
, listensockaddr
,
1649 (socklen_t
)sizeof(struct sockaddr_un
)) == -1) {
1650 log_warn(LD_NET
,"Bind to %s failed: %s.", address
,
1651 tor_socket_strerror(tor_socket_errno(s
)));
1656 if (options
->User
) {
1657 pw
= tor_getpwnam(options
->User
);
1660 log_warn(LD_NET
,"Unable to chown() %s socket: user %s not found.",
1661 address
, options
->User
);
1663 } else if (fstat(s
, &st
) == 0 &&
1664 st
.st_uid
== pw
->pw_uid
&& st
.st_gid
== pw
->pw_gid
) {
1665 /* No change needed */
1666 } else if (chown(sandbox_intern_string(address
),
1667 pw
->pw_uid
, pw
->pw_gid
) < 0) {
1668 log_warn(LD_NET
,"Unable to chown() %s socket: %s.",
1669 address
, strerror(errno
));
1673 #endif /* defined(HAVE_PWD_H) */
1679 if (port_cfg
->is_world_writable
) {
1681 status
= "world-writable";
1682 } else if (port_cfg
->is_group_writable
) {
1684 status
= "group-writable";
1689 /* We need to use chmod; fchmod doesn't work on sockets on all
1691 if (fstat(s
, &st
) == 0 && (st
.st_mode
& 0777) == mode
) {
1692 /* no change needed */
1693 } else if (chmod(sandbox_intern_string(address
), mode
) < 0) {
1694 log_warn(LD_FS
,"Unable to make %s %s.", address
, status
);
1699 if (listen(s
, SOMAXCONN
) < 0) {
1700 log_warn(LD_NET
, "Could not listen on %s: %s", address
,
1701 tor_socket_strerror(tor_socket_errno(s
)));
1706 /* This code was introduced to help debug #28229. */
1708 socklen_t len
= sizeof(value
);
1710 if (!getsockopt(s
, SOL_SOCKET
, SO_ACCEPTCONN
, &value
, &len
)) {
1712 log_err(LD_NET
, "Could not listen on %s - "
1713 "getsockopt(.,SO_ACCEPTCONN,.) yields 0.", address
);
1717 #endif /* !defined(__APPLE__) */
1718 #endif /* defined(HAVE_SYS_UN_H) */
1720 log_err(LD_BUG
, "Got unexpected address family %d.",
1721 listensockaddr
->sa_family
);
1725 lis_conn
= listener_connection_new(type
, listensockaddr
->sa_family
);
1726 conn
= TO_CONN(lis_conn
);
1727 conn
->socket_family
= listensockaddr
->sa_family
;
1729 s
= TOR_INVALID_SOCKET
; /* Prevent double-close */
1730 conn
->address
= tor_strdup(address
);
1731 conn
->port
= gotPort
;
1732 tor_addr_copy(&conn
->addr
, &addr
);
1734 memcpy(&lis_conn
->entry_cfg
, &port_cfg
->entry_cfg
, sizeof(entry_port_cfg_t
));
1736 if (port_cfg
->entry_cfg
.isolation_flags
) {
1737 lis_conn
->entry_cfg
.isolation_flags
= port_cfg
->entry_cfg
.isolation_flags
;
1738 if (port_cfg
->entry_cfg
.session_group
>= 0) {
1739 lis_conn
->entry_cfg
.session_group
= port_cfg
->entry_cfg
.session_group
;
1741 /* This can wrap after around INT_MAX listeners are opened. But I don't
1742 * believe that matters, since you would need to open a ridiculous
1743 * number of listeners while keeping the early ones open before you ever
1744 * hit this. An OR with a dozen ports open, for example, would have to
1745 * close and re-open its listeners every second for 4 years nonstop.
1747 lis_conn
->entry_cfg
.session_group
= global_next_session_group
--;
1751 if (connection_add(conn
) < 0) { /* no space, forget it */
1752 log_warn(LD_NET
,"connection_add for listener failed. Giving up.");
1756 log_fn(usePort
==gotPort
? LOG_DEBUG
: LOG_NOTICE
, LD_NET
,
1757 "%s listening on port %u.",
1758 conn_type_to_string(type
), gotPort
);
1760 conn
->state
= LISTENER_STATE_READY
;
1761 if (start_reading
) {
1762 connection_start_reading(conn
);
1764 tor_assert(type
== CONN_TYPE_AP_DNS_LISTENER
);
1765 dnsserv_configure_listener(conn
);
1769 * Normal exit; call the OOS handler since connection count just changed;
1770 * the exhaustion flag will always be zero here though.
1772 connection_check_oos(get_n_open_sockets(), 0);
1774 log_notice(LD_NET
, "Opened %s", connection_describe(conn
));
1780 tor_close_socket(s
);
1782 connection_free(conn
);
1784 /* Call the OOS handler, indicate if we saw an exhaustion-related error */
1785 connection_check_oos(get_n_open_sockets(), exhaustion
);
1791 * Create a new listener connection for a given <b>port</b>. In case we
1792 * for a reason that is not an error condition, set <b>defer</b>
1793 * to true. If we cannot bind listening socket because address is already
1794 * in use, set <b>addr_in_use</b> to true.
1796 static connection_t
*
1797 connection_listener_new_for_port(const port_cfg_t
*port
,
1798 int *defer
, int *addr_in_use
)
1801 struct sockaddr
*listensockaddr
;
1802 socklen_t listensocklen
= 0;
1804 int real_port
= port
->port
== CFG_AUTO_PORT
? 0 : port
->port
;
1805 tor_assert(real_port
<= UINT16_MAX
);
1810 if (port
->server_cfg
.no_listen
) {
1817 /* We don't need to be root to create a UNIX socket, so defer until after
1819 const or_options_t
*options
= get_options();
1820 if (port
->is_unix_addr
&& !geteuid() && (options
->User
) &&
1821 strcmp(options
->User
, "root")) {
1826 #endif /* !defined(_WIN32) */
1828 if (port
->is_unix_addr
) {
1829 listensockaddr
= (struct sockaddr
*)
1830 create_unix_sockaddr(port
->unix_addr
,
1831 &address
, &listensocklen
);
1833 listensockaddr
= tor_malloc(sizeof(struct sockaddr_storage
));
1834 listensocklen
= tor_addr_to_sockaddr(&port
->addr
,
1837 sizeof(struct sockaddr_storage
));
1838 address
= tor_addr_to_str_dup(&port
->addr
);
1841 if (listensockaddr
) {
1842 conn
= connection_listener_new(listensockaddr
, listensocklen
,
1843 port
->type
, address
, port
,
1845 tor_free(listensockaddr
);
1854 /** Do basic sanity checking on a newly received socket. Return 0
1855 * if it looks ok, else return -1.
1857 * Notably, some TCP stacks can erroneously have accept() return successfully
1858 * with socklen 0, when the client sends an RST before the accept call (as
1859 * nmap does). We want to detect that, and not go on with the connection.
1862 check_sockaddr(const struct sockaddr
*sa
, int len
, int level
)
1866 if (sa
->sa_family
== AF_INET
) {
1867 struct sockaddr_in
*sin
=(struct sockaddr_in
*)sa
;
1868 if (len
!= sizeof(struct sockaddr_in
)) {
1869 log_fn(level
, LD_NET
, "Length of address not as expected: %d vs %d",
1870 len
,(int)sizeof(struct sockaddr_in
));
1873 if (sin
->sin_addr
.s_addr
== 0 || sin
->sin_port
== 0) {
1874 log_fn(level
, LD_NET
,
1875 "Address for new connection has address/port equal to zero.");
1878 } else if (sa
->sa_family
== AF_INET6
) {
1879 struct sockaddr_in6
*sin6
=(struct sockaddr_in6
*)sa
;
1880 if (len
!= sizeof(struct sockaddr_in6
)) {
1881 log_fn(level
, LD_NET
, "Length of address not as expected: %d vs %d",
1882 len
,(int)sizeof(struct sockaddr_in6
));
1885 if (fast_mem_is_zero((void*)sin6
->sin6_addr
.s6_addr
, 16) ||
1886 sin6
->sin6_port
== 0) {
1887 log_fn(level
, LD_NET
,
1888 "Address for new connection has address/port equal to zero.");
1891 } else if (sa
->sa_family
== AF_UNIX
) {
1899 /** Check whether the socket family from an accepted socket <b>got</b> is the
1900 * same as the one that <b>listener</b> is waiting for. If it isn't, log
1901 * a useful message and return -1. Else return 0.
1903 * This is annoying, but can apparently happen on some Darwins. */
1905 check_sockaddr_family_match(sa_family_t got
, connection_t
*listener
)
1907 if (got
!= listener
->socket_family
) {
1908 log_info(LD_BUG
, "A listener connection returned a socket with a "
1909 "mismatched family. %s for addr_family %d gave us a socket "
1910 "with address family %d. Dropping.",
1911 conn_type_to_string(listener
->type
),
1912 (int)listener
->socket_family
,
1919 /** The listener connection <b>conn</b> told poll() it wanted to read.
1920 * Call accept() on conn-\>s, and add the new connection if necessary.
1923 connection_handle_listener_read(connection_t
*conn
, int new_type
)
1925 tor_socket_t news
; /* the new socket */
1926 connection_t
*newconn
= 0;
1927 /* information about the remote peer when connecting to other routers */
1928 struct sockaddr_storage addrbuf
;
1929 struct sockaddr
*remote
= (struct sockaddr
*)&addrbuf
;
1930 /* length of the remote address. Must be whatever accept() needs. */
1931 socklen_t remotelen
= (socklen_t
)sizeof(addrbuf
);
1932 const or_options_t
*options
= get_options();
1934 tor_assert((size_t)remotelen
>= sizeof(struct sockaddr_in
));
1935 memset(&addrbuf
, 0, sizeof(addrbuf
));
1937 news
= tor_accept_socket_nonblocking(conn
->s
,remote
,&remotelen
);
1938 if (!SOCKET_OK(news
)) { /* accept() error */
1939 int e
= tor_socket_errno(conn
->s
);
1940 if (ERRNO_IS_ACCEPT_EAGAIN(e
)) {
1942 * they hung up before we could accept(). that's fine.
1944 * give the OOS handler a chance to run though
1946 connection_check_oos(get_n_open_sockets(), 0);
1948 } else if (ERRNO_IS_RESOURCE_LIMIT(e
)) {
1949 socket_failed_from_resource_exhaustion();
1950 /* Exhaustion; tell the OOS handler */
1951 connection_check_oos(get_n_open_sockets(), 1);
1954 /* else there was a real error. */
1955 log_warn(LD_NET
,"accept() failed: %s. Closing listener.",
1956 tor_socket_strerror(e
));
1957 connection_mark_for_close(conn
);
1958 /* Tell the OOS handler about this too */
1959 connection_check_oos(get_n_open_sockets(), 0);
1963 "Connection accepted on socket %d (child of fd %d).",
1964 (int)news
,(int)conn
->s
);
1966 /* We accepted a new conn; run OOS handler */
1967 connection_check_oos(get_n_open_sockets(), 0);
1969 if (make_socket_reuseable(news
) < 0) {
1970 if (tor_socket_errno(news
) == EINVAL
) {
1971 /* This can happen on OSX if we get a badly timed shutdown. */
1972 log_debug(LD_NET
, "make_socket_reuseable returned EINVAL");
1974 log_warn(LD_NET
, "Error setting SO_REUSEADDR flag on %s: %s",
1975 conn_type_to_string(new_type
),
1976 tor_socket_strerror(errno
));
1978 tor_close_socket(news
);
1982 if (options
->ConstrainedSockets
)
1983 set_constrained_socket_buffers(news
, (int)options
->ConstrainedSockSize
);
1985 if (check_sockaddr_family_match(remote
->sa_family
, conn
) < 0) {
1986 tor_close_socket(news
);
1990 if (conn
->socket_family
== AF_INET
|| conn
->socket_family
== AF_INET6
||
1991 (conn
->socket_family
== AF_UNIX
&& new_type
== CONN_TYPE_AP
)) {
1994 if (check_sockaddr(remote
, remotelen
, LOG_INFO
)<0) {
1996 "accept() returned a strange address; closing connection.");
1997 tor_close_socket(news
);
2001 tor_addr_from_sockaddr(&addr
, remote
, &port
);
2003 /* process entrance policies here, before we even create the connection */
2004 if (new_type
== CONN_TYPE_AP
) {
2005 /* check sockspolicy to see if we should accept it */
2006 if (socks_policy_permits_address(&addr
) == 0) {
2008 "Denying socks connection from untrusted address %s.",
2009 fmt_and_decorate_addr(&addr
));
2010 tor_close_socket(news
);
2014 if (new_type
== CONN_TYPE_DIR
) {
2015 /* check dirpolicy to see if we should accept it */
2016 if (dir_policy_permits_address(&addr
) == 0) {
2017 log_notice(LD_DIRSERV
,"Denying dir connection from address %s.",
2018 fmt_and_decorate_addr(&addr
));
2019 tor_close_socket(news
);
2023 if (new_type
== CONN_TYPE_OR
) {
2024 /* Assess with the connection DoS mitigation subsystem if this address
2025 * can open a new connection. */
2026 if (dos_conn_addr_get_defense_type(&addr
) == DOS_CONN_DEFENSE_CLOSE
) {
2027 tor_close_socket(news
);
2032 newconn
= connection_new(new_type
, conn
->socket_family
);
2035 /* remember the remote address */
2036 tor_addr_copy(&newconn
->addr
, &addr
);
2037 if (new_type
== CONN_TYPE_AP
&& conn
->socket_family
== AF_UNIX
) {
2039 newconn
->address
= tor_strdup(conn
->address
);
2041 newconn
->port
= port
;
2042 newconn
->address
= tor_addr_to_str_dup(&addr
);
2045 if (new_type
== CONN_TYPE_AP
&& conn
->socket_family
!= AF_UNIX
) {
2046 log_info(LD_NET
, "New SOCKS connection opened from %s.",
2047 fmt_and_decorate_addr(&addr
));
2049 if (new_type
== CONN_TYPE_AP
&& conn
->socket_family
== AF_UNIX
) {
2050 log_info(LD_NET
, "New SOCKS AF_UNIX connection opened");
2052 if (new_type
== CONN_TYPE_CONTROL
) {
2053 log_notice(LD_CONTROL
, "New control connection opened from %s.",
2054 fmt_and_decorate_addr(&addr
));
2056 if (new_type
== CONN_TYPE_METRICS
) {
2057 log_info(LD_CONTROL
, "New metrics connection opened from %s.",
2058 fmt_and_decorate_addr(&addr
));
2061 } else if (conn
->socket_family
== AF_UNIX
&& conn
->type
!= CONN_TYPE_AP
) {
2062 tor_assert(conn
->type
== CONN_TYPE_CONTROL_LISTENER
);
2063 tor_assert(new_type
== CONN_TYPE_CONTROL
);
2064 log_notice(LD_CONTROL
, "New control connection opened.");
2066 newconn
= connection_new(new_type
, conn
->socket_family
);
2069 /* remember the remote address -- do we have anything sane to put here? */
2070 tor_addr_make_unspec(&newconn
->addr
);
2072 newconn
->address
= tor_strdup(conn
->address
);
2077 if (connection_add(newconn
) < 0) { /* no space, forget it */
2078 connection_free(newconn
);
2079 return 0; /* no need to tear down the parent */
2082 if (connection_init_accepted_conn(newconn
, TO_LISTENER_CONN(conn
)) < 0) {
2083 if (! newconn
->marked_for_close
)
2084 connection_mark_for_close(newconn
);
2088 note_connection(true /* inbound */, conn
->socket_family
);
2093 /** Initialize states for newly accepted connection <b>conn</b>.
2095 * If conn is an OR, start the TLS handshake.
2097 * If conn is a transparent AP, get its original destination
2098 * and place it in circuit_wait.
2100 * The <b>listener</b> parameter is only used for AP connections.
2103 connection_init_accepted_conn(connection_t
*conn
,
2104 const listener_connection_t
*listener
)
2108 connection_start_reading(conn
);
2110 switch (conn
->type
) {
2111 case CONN_TYPE_EXT_OR
:
2112 /* Initiate Extended ORPort authentication. */
2113 return connection_ext_or_start_auth(TO_OR_CONN(conn
));
2115 connection_or_event_status(TO_OR_CONN(conn
), OR_CONN_EVENT_NEW
, 0);
2116 rv
= connection_tls_start_handshake(TO_OR_CONN(conn
), 1);
2118 connection_or_close_for_error(TO_OR_CONN(conn
), 0);
2123 memcpy(&TO_ENTRY_CONN(conn
)->entry_cfg
, &listener
->entry_cfg
,
2124 sizeof(entry_port_cfg_t
));
2125 TO_ENTRY_CONN(conn
)->nym_epoch
= get_signewnym_epoch();
2126 TO_ENTRY_CONN(conn
)->socks_request
->listener_type
= listener
->base_
.type
;
2128 /* Any incoming connection on an entry port counts as user activity. */
2129 note_user_activity(approx_time());
2131 switch (TO_CONN(listener
)->type
) {
2132 case CONN_TYPE_AP_LISTENER
:
2133 conn
->state
= AP_CONN_STATE_SOCKS_WAIT
;
2134 TO_ENTRY_CONN(conn
)->socks_request
->socks_prefer_no_auth
=
2135 listener
->entry_cfg
.socks_prefer_no_auth
;
2136 TO_ENTRY_CONN(conn
)->socks_request
->socks_use_extended_errors
=
2137 listener
->entry_cfg
.extended_socks5_codes
;
2139 case CONN_TYPE_AP_TRANS_LISTENER
:
2140 TO_ENTRY_CONN(conn
)->is_transparent_ap
= 1;
2141 /* XXXX028 -- is this correct still, with the addition of
2142 * pending_entry_connections ? */
2143 conn
->state
= AP_CONN_STATE_CIRCUIT_WAIT
;
2144 return connection_ap_process_transparent(TO_ENTRY_CONN(conn
));
2145 case CONN_TYPE_AP_NATD_LISTENER
:
2146 TO_ENTRY_CONN(conn
)->is_transparent_ap
= 1;
2147 conn
->state
= AP_CONN_STATE_NATD_WAIT
;
2149 case CONN_TYPE_AP_HTTP_CONNECT_LISTENER
:
2150 conn
->state
= AP_CONN_STATE_HTTP_CONNECT_WAIT
;
2154 conn
->purpose
= DIR_PURPOSE_SERVER
;
2155 conn
->state
= DIR_CONN_STATE_SERVER_COMMAND_WAIT
;
2157 case CONN_TYPE_CONTROL
:
2158 conn
->state
= CONTROL_CONN_STATE_NEEDAUTH
;
2164 /** Take conn, make a nonblocking socket; try to connect to
2165 * sa, binding to bindaddr if sa is not localhost. If fail, return -1 and if
2166 * applicable put your best guess about errno into *<b>socket_error</b>.
2167 * If connected return 1, if EAGAIN return 0.
2169 MOCK_IMPL(STATIC
int,
2170 connection_connect_sockaddr
,(connection_t
*conn
,
2171 const struct sockaddr
*sa
,
2173 const struct sockaddr
*bindaddr
,
2174 socklen_t bindaddr_len
,
2179 const or_options_t
*options
= get_options();
2183 tor_assert(socket_error
);
2185 if (net_is_completely_disabled()) {
2186 /* We should never even try to connect anyplace if the network is
2187 * completely shut off.
2189 * (We don't check net_is_disabled() here, since we still sometimes
2190 * want to open connections when we're in soft hibernation.)
2192 static ratelim_t disablenet_violated
= RATELIM_INIT(30*60);
2193 *socket_error
= SOCK_ERRNO(ENETUNREACH
);
2194 log_fn_ratelim(&disablenet_violated
, LOG_WARN
, LD_BUG
,
2195 "Tried to open a socket with DisableNetwork set.");
2196 tor_fragile_assert();
2200 const int protocol_family
= sa
->sa_family
;
2201 const int proto
= (sa
->sa_family
== AF_INET6
||
2202 sa
->sa_family
== AF_INET
) ? IPPROTO_TCP
: 0;
2204 s
= tor_open_socket_nonblocking(protocol_family
, SOCK_STREAM
, proto
);
2205 if (! SOCKET_OK(s
)) {
2207 * Early OOS handler calls; it matters if it's an exhaustion-related
2210 *socket_error
= tor_socket_errno(s
);
2211 if (ERRNO_IS_RESOURCE_LIMIT(*socket_error
)) {
2212 socket_failed_from_resource_exhaustion();
2213 connection_check_oos(get_n_open_sockets(), 1);
2215 log_warn(LD_NET
,"Error creating network socket: %s",
2216 tor_socket_strerror(*socket_error
));
2217 connection_check_oos(get_n_open_sockets(), 0);
2222 if (make_socket_reuseable(s
) < 0) {
2223 log_warn(LD_NET
, "Error setting SO_REUSEADDR flag on new connection: %s",
2224 tor_socket_strerror(errno
));
2227 if (bindaddr
&& bind(s
, bindaddr
, bindaddr_len
) < 0) {
2228 *socket_error
= tor_socket_errno(s
);
2229 if (ERRNO_IS_EADDRINUSE(*socket_error
)) {
2230 socket_failed_from_resource_exhaustion();
2231 connection_check_oos(get_n_open_sockets(), 1);
2233 log_warn(LD_NET
,"Error binding network socket: %s",
2234 tor_socket_strerror(*socket_error
));
2235 connection_check_oos(get_n_open_sockets(), 0);
2237 tor_close_socket(s
);
2242 * We've got the socket open and bound; give the OOS handler a chance to
2243 * check against configured maximum socket number, but tell it no exhaustion
2246 connection_check_oos(get_n_open_sockets(), 0);
2248 tor_assert(options
);
2249 if (options
->ConstrainedSockets
)
2250 set_constrained_socket_buffers(s
, (int)options
->ConstrainedSockSize
);
2252 if (connect(s
, sa
, sa_len
) < 0) {
2253 int e
= tor_socket_errno(s
);
2254 if (!ERRNO_IS_CONN_EINPROGRESS(e
)) {
2255 /* yuck. kill it. */
2258 "connect() to socket failed: %s",
2259 tor_socket_strerror(e
));
2260 tor_close_socket(s
);
2267 note_connection(false /* outbound */, conn
->socket_family
);
2269 /* it succeeded. we're connected. */
2270 log_fn(inprogress
? LOG_DEBUG
: LOG_INFO
, LD_NET
,
2271 "Connection to socket %s (sock "TOR_SOCKET_T_FORMAT
").",
2272 inprogress
? "in progress" : "established", s
);
2274 if (connection_add_connecting(conn
) < 0) {
2275 /* no space, forget it */
2276 *socket_error
= SOCK_ERRNO(ENOBUFS
);
2280 return inprogress
? 0 : 1;
2283 /* Log a message if connection attempt is made when IPv4 or IPv6 is disabled.
2284 * Log a less severe message if we couldn't conform to ClientPreferIPv6ORPort
2285 * or ClientPreferIPv6ORPort. */
2287 connection_connect_log_client_use_ip_version(const connection_t
*conn
)
2289 const or_options_t
*options
= get_options();
2291 /* Only clients care about ClientUseIPv4/6, bail out early on servers, and
2292 * on connections we don't care about */
2293 if (server_mode(options
) || !conn
|| conn
->type
== CONN_TYPE_EXIT
) {
2297 /* We're only prepared to log OR and DIR connections here */
2298 if (conn
->type
!= CONN_TYPE_OR
&& conn
->type
!= CONN_TYPE_DIR
) {
2302 const int must_ipv4
= !reachable_addr_use_ipv6(options
);
2303 const int must_ipv6
= (options
->ClientUseIPv4
== 0);
2304 const int pref_ipv6
= (conn
->type
== CONN_TYPE_OR
2305 ? reachable_addr_prefer_ipv6_orport(options
)
2306 : reachable_addr_prefer_ipv6_dirport(options
));
2307 tor_addr_t real_addr
;
2308 tor_addr_copy(&real_addr
, &conn
->addr
);
2310 /* Check if we broke a mandatory address family restriction */
2311 if ((must_ipv4
&& tor_addr_family(&real_addr
) == AF_INET6
)
2312 || (must_ipv6
&& tor_addr_family(&real_addr
) == AF_INET
)) {
2313 static int logged_backtrace
= 0;
2314 log_info(LD_BUG
, "Outgoing %s connection to %s violated ClientUseIPv%s 0.",
2315 conn
->type
== CONN_TYPE_OR
? "OR" : "Dir",
2316 fmt_addr(&real_addr
),
2317 options
->ClientUseIPv4
== 0 ? "4" : "6");
2318 if (!logged_backtrace
) {
2319 log_backtrace(LOG_INFO
, LD_BUG
, "Address came from");
2320 logged_backtrace
= 1;
2324 /* Bridges are allowed to break IPv4/IPv6 ORPort preferences to connect to
2325 * the node's configured address when ClientPreferIPv6ORPort is auto */
2326 if (options
->UseBridges
&& conn
->type
== CONN_TYPE_OR
2327 && options
->ClientPreferIPv6ORPort
== -1) {
2331 if (reachable_addr_use_ipv6(options
)) {
2332 log_info(LD_NET
, "Our outgoing connection is using IPv%d.",
2333 tor_addr_family(&real_addr
) == AF_INET6
? 6 : 4);
2336 /* Check if we couldn't satisfy an address family preference */
2337 if ((!pref_ipv6
&& tor_addr_family(&real_addr
) == AF_INET6
)
2338 || (pref_ipv6
&& tor_addr_family(&real_addr
) == AF_INET
)) {
2339 log_info(LD_NET
, "Outgoing connection to %s doesn't satisfy "
2340 "ClientPreferIPv6%sPort %d, with ClientUseIPv4 %d, and "
2341 "reachable_addr_use_ipv6 %d (ClientUseIPv6 %d and UseBridges "
2343 fmt_addr(&real_addr
),
2344 conn
->type
== CONN_TYPE_OR
? "OR" : "Dir",
2345 conn
->type
== CONN_TYPE_OR
? options
->ClientPreferIPv6ORPort
2346 : options
->ClientPreferIPv6DirPort
,
2347 options
->ClientUseIPv4
, reachable_addr_use_ipv6(options
),
2348 options
->ClientUseIPv6
, options
->UseBridges
);
2352 /** Retrieve the outbound address depending on the protocol (IPv4 or IPv6)
2353 * and the connection type (relay, exit, ...)
2354 * Return a socket address or NULL in case nothing is configured.
2357 conn_get_outbound_address(sa_family_t family
,
2358 const or_options_t
*options
, unsigned int conn_type
)
2360 const tor_addr_t
*ext_addr
= NULL
;
2374 // If an exit connection, use the exit address (if present)
2375 if (conn_type
== CONN_TYPE_EXIT
) {
2376 if (!tor_addr_is_null(
2377 &options
->OutboundBindAddresses
[OUTBOUND_ADDR_EXIT
][fam_index
])) {
2378 ext_addr
= &options
->OutboundBindAddresses
[OUTBOUND_ADDR_EXIT
]
2380 } else if (!tor_addr_is_null(
2381 &options
->OutboundBindAddresses
[OUTBOUND_ADDR_ANY
]
2383 ext_addr
= &options
->OutboundBindAddresses
[OUTBOUND_ADDR_ANY
]
2386 } else { // All non-exit connections
2387 if (!tor_addr_is_null(
2388 &options
->OutboundBindAddresses
[OUTBOUND_ADDR_OR
][fam_index
])) {
2389 ext_addr
= &options
->OutboundBindAddresses
[OUTBOUND_ADDR_OR
]
2391 } else if (!tor_addr_is_null(
2392 &options
->OutboundBindAddresses
[OUTBOUND_ADDR_ANY
]
2394 ext_addr
= &options
->OutboundBindAddresses
[OUTBOUND_ADDR_ANY
]
2401 /** Take conn, make a nonblocking socket; try to connect to
2402 * addr:port (port arrives in *host order*). If fail, return -1 and if
2403 * applicable put your best guess about errno into *<b>socket_error</b>.
2404 * Else assign s to conn-\>s: if connected return 1, if EAGAIN return 0.
2406 * addr:port can be different to conn->addr:conn->port if connecting through
2409 * address is used to make the logs useful.
2411 * On success, add conn to the list of polled connections.
2414 connection_connect(connection_t
*conn
, const char *address
,
2415 const tor_addr_t
*addr
, uint16_t port
, int *socket_error
)
2417 struct sockaddr_storage addrbuf
;
2418 struct sockaddr_storage bind_addr_ss
;
2419 struct sockaddr
*bind_addr
= NULL
;
2420 struct sockaddr
*dest_addr
;
2421 int dest_addr_len
, bind_addr_len
= 0;
2423 /* Log if we didn't stick to ClientUseIPv4/6 or ClientPreferIPv6OR/DirPort
2425 connection_connect_log_client_use_ip_version(conn
);
2427 if (!tor_addr_is_loopback(addr
)) {
2428 const tor_addr_t
*ext_addr
= NULL
;
2429 ext_addr
= conn_get_outbound_address(tor_addr_family(addr
), get_options(),
2432 memset(&bind_addr_ss
, 0, sizeof(bind_addr_ss
));
2433 bind_addr_len
= tor_addr_to_sockaddr(ext_addr
, 0,
2434 (struct sockaddr
*) &bind_addr_ss
,
2435 sizeof(bind_addr_ss
));
2436 if (bind_addr_len
== 0) {
2438 "Error converting OutboundBindAddress %s into sockaddr. "
2439 "Ignoring.", fmt_and_decorate_addr(ext_addr
));
2441 bind_addr
= (struct sockaddr
*)&bind_addr_ss
;
2446 memset(&addrbuf
,0,sizeof(addrbuf
));
2447 dest_addr
= (struct sockaddr
*) &addrbuf
;
2448 dest_addr_len
= tor_addr_to_sockaddr(addr
, port
, dest_addr
, sizeof(addrbuf
));
2449 tor_assert(dest_addr_len
> 0);
2451 log_debug(LD_NET
, "Connecting to %s:%u.",
2452 escaped_safe_str_client(address
), port
);
2454 return connection_connect_sockaddr(conn
, dest_addr
, dest_addr_len
,
2455 bind_addr
, bind_addr_len
, socket_error
);
2458 #ifdef HAVE_SYS_UN_H
2460 /** Take conn, make a nonblocking socket; try to connect to
2461 * an AF_UNIX socket at socket_path. If fail, return -1 and if applicable
2462 * put your best guess about errno into *<b>socket_error</b>. Else assign s
2463 * to conn-\>s: if connected return 1, if EAGAIN return 0.
2465 * On success, add conn to the list of polled connections.
2468 connection_connect_unix(connection_t
*conn
, const char *socket_path
,
2471 struct sockaddr_un dest_addr
;
2473 tor_assert(socket_path
);
2475 /* Check that we'll be able to fit it into dest_addr later */
2476 if (strlen(socket_path
) + 1 > sizeof(dest_addr
.sun_path
)) {
2478 "Path %s is too long for an AF_UNIX socket\n",
2479 escaped_safe_str_client(socket_path
));
2480 *socket_error
= SOCK_ERRNO(ENAMETOOLONG
);
2484 memset(&dest_addr
, 0, sizeof(dest_addr
));
2485 dest_addr
.sun_family
= AF_UNIX
;
2486 strlcpy(dest_addr
.sun_path
, socket_path
, sizeof(dest_addr
.sun_path
));
2489 "Connecting to AF_UNIX socket at %s.",
2490 escaped_safe_str_client(socket_path
));
2492 return connection_connect_sockaddr(conn
,
2493 (struct sockaddr
*)&dest_addr
, sizeof(dest_addr
),
2494 NULL
, 0, socket_error
);
2497 #endif /* defined(HAVE_SYS_UN_H) */
2499 /** Convert state number to string representation for logging purposes.
2502 connection_proxy_state_to_string(int state
)
2504 static const char *unknown
= "???";
2505 static const char *states
[] = {
2508 "PROXY_HTTPS_WANT_CONNECT_OK",
2509 "PROXY_SOCKS4_WANT_CONNECT_OK",
2510 "PROXY_SOCKS5_WANT_AUTH_METHOD_NONE",
2511 "PROXY_SOCKS5_WANT_AUTH_METHOD_RFC1929",
2512 "PROXY_SOCKS5_WANT_AUTH_RFC1929_OK",
2513 "PROXY_SOCKS5_WANT_CONNECT_OK",
2514 "PROXY_HAPROXY_WAIT_FOR_FLUSH",
2518 CTASSERT(ARRAY_LENGTH(states
) == PROXY_CONNECTED
+1);
2520 if (state
< PROXY_NONE
|| state
> PROXY_CONNECTED
)
2523 return states
[state
];
2526 /** Returns the proxy type used by tor for a single connection, for
2527 * logging or high-level purposes. Don't use it to fill the
2528 * <b>proxy_type</b> field of or_connection_t; use the actual proxy
2529 * protocol instead.*/
2531 conn_get_proxy_type(const connection_t
*conn
)
2533 const or_options_t
*options
= get_options();
2535 if (options
->ClientTransportPlugin
) {
2536 /* If we have plugins configured *and* this addr/port is a known bridge
2537 * with a transport, then we should be PROXY_PLUGGABLE. */
2538 const transport_t
*transport
= NULL
;
2540 r
= get_transport_by_bridge_addrport(&conn
->addr
, conn
->port
, &transport
);
2541 if (r
== 0 && transport
)
2542 return PROXY_PLUGGABLE
;
2545 /* In all other cases, we're using a global proxy. */
2546 if (options
->HTTPSProxy
)
2547 return PROXY_CONNECT
;
2548 else if (options
->Socks4Proxy
)
2549 return PROXY_SOCKS4
;
2550 else if (options
->Socks5Proxy
)
2551 return PROXY_SOCKS5
;
2552 else if (options
->TCPProxy
) {
2553 /* The only supported protocol in TCPProxy is haproxy. */
2554 tor_assert(options
->TCPProxyProtocol
== TCP_PROXY_PROTOCOL_HAPROXY
);
2555 return PROXY_HAPROXY
;
2560 /* One byte for the version, one for the command, two for the
2561 port, and four for the addr... and, one more for the
2563 #define SOCKS4_STANDARD_BUFFER_SIZE (1 + 1 + 2 + 4 + 1)
2565 /** Write a proxy request of https to conn for conn->addr:conn->port,
2566 * authenticating with the auth details given in the configuration
2569 * Returns -1 if conn->addr is incompatible with the proxy protocol, and
2573 connection_https_proxy_connect(connection_t
*conn
)
2577 const or_options_t
*options
= get_options();
2579 char *base64_authenticator
= NULL
;
2580 const char *authenticator
= options
->HTTPSProxyAuthenticator
;
2582 /* Send HTTP CONNECT and authentication (if available) in
2585 if (authenticator
) {
2586 base64_authenticator
= alloc_http_authenticator(authenticator
);
2587 if (!base64_authenticator
)
2588 log_warn(LD_OR
, "Encoding https authenticator failed");
2591 if (base64_authenticator
) {
2592 const char *addrport
= fmt_addrport(&conn
->addr
, conn
->port
);
2593 tor_snprintf(buf
, sizeof(buf
), "CONNECT %s HTTP/1.1\r\n"
2595 "Proxy-Authorization: Basic %s\r\n\r\n",
2598 base64_authenticator
);
2599 tor_free(base64_authenticator
);
2601 tor_snprintf(buf
, sizeof(buf
), "CONNECT %s HTTP/1.0\r\n\r\n",
2602 fmt_addrport(&conn
->addr
, conn
->port
));
2605 connection_buf_add(buf
, strlen(buf
), conn
);
2606 conn
->proxy_state
= PROXY_HTTPS_WANT_CONNECT_OK
;
2611 /** Write a proxy request of socks4 to conn for conn->addr:conn->port.
2613 * Returns -1 if conn->addr is incompatible with the proxy protocol, and
2617 connection_socks4_proxy_connect(connection_t
*conn
)
2624 size_t buf_size
= 0;
2625 char *socks_args_string
= NULL
;
2627 /* Send a SOCKS4 connect request */
2629 if (tor_addr_family(&conn
->addr
) != AF_INET
) {
2630 log_warn(LD_NET
, "SOCKS4 client is incompatible with IPv6");
2634 { /* If we are here because we are trying to connect to a
2635 pluggable transport proxy, check if we have any SOCKS
2636 arguments to transmit. If we do, compress all arguments to
2637 a single string in 'socks_args_string': */
2639 if (conn_get_proxy_type(conn
) == PROXY_PLUGGABLE
) {
2641 pt_get_socks_args_for_proxy_addrport(&conn
->addr
, conn
->port
);
2642 if (socks_args_string
)
2643 log_debug(LD_NET
, "Sending out '%s' as our SOCKS argument string.",
2648 { /* Figure out the buffer size we need for the SOCKS message: */
2650 buf_size
= SOCKS4_STANDARD_BUFFER_SIZE
;
2652 /* If we have a SOCKS argument string, consider its size when
2653 calculating the buffer size: */
2654 if (socks_args_string
)
2655 buf_size
+= strlen(socks_args_string
);
2658 buf
= tor_malloc_zero(buf_size
);
2660 ip4addr
= tor_addr_to_ipv4n(&conn
->addr
);
2661 portn
= htons(conn
->port
);
2663 buf
[0] = 4; /* version */
2664 buf
[1] = SOCKS_COMMAND_CONNECT
; /* command */
2665 memcpy(buf
+ 2, &portn
, 2); /* port */
2666 memcpy(buf
+ 4, &ip4addr
, 4); /* addr */
2668 /* Next packet field is the userid. If we have pluggable
2669 transport SOCKS arguments, we have to embed them
2670 there. Otherwise, we use an empty userid. */
2671 if (socks_args_string
) { /* place the SOCKS args string: */
2672 tor_assert(strlen(socks_args_string
) > 0);
2673 tor_assert(buf_size
>=
2674 SOCKS4_STANDARD_BUFFER_SIZE
+ strlen(socks_args_string
));
2675 strlcpy((char *)buf
+ 8, socks_args_string
, buf_size
- 8);
2676 tor_free(socks_args_string
);
2678 buf
[8] = 0; /* no userid */
2681 connection_buf_add((char *)buf
, buf_size
, conn
);
2684 conn
->proxy_state
= PROXY_SOCKS4_WANT_CONNECT_OK
;
2688 /** Write a proxy request of socks5 to conn for conn->addr:conn->port,
2689 * authenticating with the auth details given in the configuration
2692 * Returns -1 if conn->addr is incompatible with the proxy protocol, and
2696 connection_socks5_proxy_connect(connection_t
*conn
)
2700 const or_options_t
*options
= get_options();
2701 unsigned char buf
[4]; /* fields: vers, num methods, method list */
2703 /* Send a SOCKS5 greeting (connect request must wait) */
2705 buf
[0] = 5; /* version */
2707 /* We have to use SOCKS5 authentication, if we have a
2708 Socks5ProxyUsername or if we want to pass arguments to our
2709 pluggable transport proxy: */
2710 if ((options
->Socks5ProxyUsername
) ||
2711 (conn_get_proxy_type(conn
) == PROXY_PLUGGABLE
&&
2712 (get_socks_args_by_bridge_addrport(&conn
->addr
, conn
->port
)))) {
2713 /* number of auth methods */
2715 buf
[2] = 0x00; /* no authentication */
2716 buf
[3] = 0x02; /* rfc1929 Username/Passwd auth */
2717 conn
->proxy_state
= PROXY_SOCKS5_WANT_AUTH_METHOD_RFC1929
;
2720 buf
[2] = 0x00; /* no authentication */
2721 conn
->proxy_state
= PROXY_SOCKS5_WANT_AUTH_METHOD_NONE
;
2724 connection_buf_add((char *)buf
, 2 + buf
[1], conn
);
2728 /** Write a proxy request of haproxy to conn for conn->addr:conn->port.
2730 * Returns -1 if conn->addr is incompatible with the proxy protocol, and
2734 connection_haproxy_proxy_connect(connection_t
*conn
)
2737 tor_addr_port_t
*addr_port
= tor_addr_port_new(&conn
->addr
, conn
->port
);
2738 char *buf
= haproxy_format_proxy_header_line(addr_port
);
2745 connection_buf_add(buf
, strlen(buf
), conn
);
2746 /* In haproxy, we don't have to wait for the response, but we wait for ack.
2747 * So we can set the state to be PROXY_HAPROXY_WAIT_FOR_FLUSH. */
2748 conn
->proxy_state
= PROXY_HAPROXY_WAIT_FOR_FLUSH
;
2753 tor_free(addr_port
);
2757 /** Write a proxy request of <b>type</b> (socks4, socks5, https, haproxy)
2758 * to conn for conn->addr:conn->port, authenticating with the auth details
2759 * given in the configuration (if available). SOCKS 5 and HTTP CONNECT
2760 * proxies support authentication.
2762 * Returns -1 if conn->addr is incompatible with the proxy protocol, and
2765 * Use connection_read_proxy_handshake() to complete the handshake.
2768 connection_proxy_connect(connection_t
*conn
, int type
)
2776 ret
= connection_https_proxy_connect(conn
);
2780 ret
= connection_socks4_proxy_connect(conn
);
2784 ret
= connection_socks5_proxy_connect(conn
);
2788 ret
= connection_haproxy_proxy_connect(conn
);
2792 log_err(LD_BUG
, "Invalid proxy protocol, %d", type
);
2793 tor_fragile_assert();
2799 log_debug(LD_NET
, "set state %s",
2800 connection_proxy_state_to_string(conn
->proxy_state
));
2806 /** Read conn's inbuf. If the http response from the proxy is all
2807 * here, make sure it's good news, then return 1. If it's bad news,
2808 * return -1. Else return 0 and hope for better luck next time.
2811 connection_read_https_proxy_response(connection_t
*conn
)
2818 switch (fetch_from_buf_http(conn
->inbuf
,
2819 &headers
, MAX_HEADERS_SIZE
,
2820 NULL
, NULL
, 10000, 0)) {
2821 case -1: /* overflow */
2822 log_warn(LD_PROTOCOL
,
2823 "Your https proxy sent back an oversized response. Closing.");
2826 log_info(LD_NET
,"https proxy response not all here yet. Waiting.");
2828 /* case 1, fall through */
2831 if (parse_http_response(headers
, &status_code
, &date_header
,
2832 NULL
, &reason
) < 0) {
2834 "Unparseable headers from proxy (%s). Closing.",
2835 connection_describe(conn
));
2840 if (!reason
) reason
= tor_strdup("[no reason given]");
2842 if (status_code
== 200) {
2844 "HTTPS connect for %s successful! (200 %s) Starting TLS.",
2845 connection_describe(conn
), escaped(reason
));
2849 /* else, bad news on the status code */
2850 switch (status_code
) {
2853 "The https proxy refused to allow connection to %s "
2854 "(status code %d, %s). Closing.",
2855 conn
->address
, status_code
, escaped(reason
));
2859 "The https proxy sent back an unexpected status code %d (%s). "
2861 status_code
, escaped(reason
));
2868 /** Send SOCKS5 CONNECT command to <b>conn</b>, copying <b>conn->addr</b>
2869 * and <b>conn->port</b> into the request.
2872 connection_send_socks5_connect(connection_t
*conn
)
2874 unsigned char buf
[1024];
2876 uint16_t port
= htons(conn
->port
);
2878 buf
[0] = 5; /* version */
2879 buf
[1] = SOCKS_COMMAND_CONNECT
; /* command */
2880 buf
[2] = 0; /* reserved */
2882 if (tor_addr_family(&conn
->addr
) == AF_INET
) {
2883 uint32_t addr
= tor_addr_to_ipv4n(&conn
->addr
);
2887 memcpy(buf
+ 4, &addr
, 4);
2888 memcpy(buf
+ 8, &port
, 2);
2889 } else { /* AF_INET6 */
2892 memcpy(buf
+ 4, tor_addr_to_in6_addr8(&conn
->addr
), 16);
2893 memcpy(buf
+ 20, &port
, 2);
2896 connection_buf_add((char *)buf
, reqsize
, conn
);
2898 conn
->proxy_state
= PROXY_SOCKS5_WANT_CONNECT_OK
;
2901 /** Wrapper around fetch_from_buf_socks_client: see that functions
2902 * for documentation of its behavior. */
2904 connection_fetch_from_buf_socks_client(connection_t
*conn
,
2905 int state
, char **reason
)
2907 return fetch_from_buf_socks_client(conn
->inbuf
, state
, reason
);
2910 /** Call this from connection_*_process_inbuf() to advance the proxy
2913 * No matter what proxy protocol is used, if this function returns 1, the
2914 * handshake is complete, and the data remaining on inbuf may contain the
2915 * start of the communication with the requested server.
2917 * Returns 0 if the current buffer contains an incomplete response, and -1
2921 connection_read_proxy_handshake(connection_t
*conn
)
2924 char *reason
= NULL
;
2926 log_debug(LD_NET
, "enter state %s",
2927 connection_proxy_state_to_string(conn
->proxy_state
));
2929 switch (conn
->proxy_state
) {
2930 case PROXY_HTTPS_WANT_CONNECT_OK
:
2931 ret
= connection_read_https_proxy_response(conn
);
2933 conn
->proxy_state
= PROXY_CONNECTED
;
2936 case PROXY_SOCKS4_WANT_CONNECT_OK
:
2937 ret
= connection_fetch_from_buf_socks_client(conn
,
2941 conn
->proxy_state
= PROXY_CONNECTED
;
2944 case PROXY_SOCKS5_WANT_AUTH_METHOD_NONE
:
2945 ret
= connection_fetch_from_buf_socks_client(conn
,
2948 /* no auth needed, do connect */
2950 connection_send_socks5_connect(conn
);
2955 case PROXY_SOCKS5_WANT_AUTH_METHOD_RFC1929
:
2956 ret
= connection_fetch_from_buf_socks_client(conn
,
2960 /* send auth if needed, otherwise do connect */
2962 connection_send_socks5_connect(conn
);
2964 } else if (ret
== 2) {
2965 unsigned char buf
[1024];
2966 size_t reqsize
, usize
, psize
;
2967 const char *user
, *pass
;
2968 char *socks_args_string
= NULL
;
2970 if (conn_get_proxy_type(conn
) == PROXY_PLUGGABLE
) {
2972 pt_get_socks_args_for_proxy_addrport(&conn
->addr
, conn
->port
);
2973 if (!socks_args_string
) {
2974 log_warn(LD_NET
, "Could not create SOCKS args string for PT.");
2979 log_debug(LD_NET
, "PT SOCKS5 arguments: %s", socks_args_string
);
2980 tor_assert(strlen(socks_args_string
) > 0);
2981 tor_assert(strlen(socks_args_string
) <= MAX_SOCKS5_AUTH_SIZE_TOTAL
);
2983 if (strlen(socks_args_string
) > MAX_SOCKS5_AUTH_FIELD_SIZE
) {
2984 user
= socks_args_string
;
2985 usize
= MAX_SOCKS5_AUTH_FIELD_SIZE
;
2986 pass
= socks_args_string
+ MAX_SOCKS5_AUTH_FIELD_SIZE
;
2987 psize
= strlen(socks_args_string
) - MAX_SOCKS5_AUTH_FIELD_SIZE
;
2989 user
= socks_args_string
;
2990 usize
= strlen(socks_args_string
);
2994 } else if (get_options()->Socks5ProxyUsername
) {
2995 user
= get_options()->Socks5ProxyUsername
;
2996 pass
= get_options()->Socks5ProxyPassword
;
2997 tor_assert(user
&& pass
);
2998 usize
= strlen(user
);
2999 psize
= strlen(pass
);
3001 log_err(LD_BUG
, "We entered %s for no reason!", __func__
);
3002 tor_fragile_assert();
3007 /* Username and password lengths should have been checked
3008 above and during torrc parsing. */
3009 tor_assert(usize
<= MAX_SOCKS5_AUTH_FIELD_SIZE
&&
3010 psize
<= MAX_SOCKS5_AUTH_FIELD_SIZE
);
3011 reqsize
= 3 + usize
+ psize
;
3013 buf
[0] = 1; /* negotiation version */
3015 memcpy(buf
+ 2, user
, usize
);
3016 buf
[2 + usize
] = psize
;
3017 memcpy(buf
+ 3 + usize
, pass
, psize
);
3019 if (socks_args_string
)
3020 tor_free(socks_args_string
);
3022 connection_buf_add((char *)buf
, reqsize
, conn
);
3024 conn
->proxy_state
= PROXY_SOCKS5_WANT_AUTH_RFC1929_OK
;
3029 case PROXY_SOCKS5_WANT_AUTH_RFC1929_OK
:
3030 ret
= connection_fetch_from_buf_socks_client(conn
,
3033 /* send the connect request */
3035 connection_send_socks5_connect(conn
);
3040 case PROXY_SOCKS5_WANT_CONNECT_OK
:
3041 ret
= connection_fetch_from_buf_socks_client(conn
,
3045 conn
->proxy_state
= PROXY_CONNECTED
;
3049 log_err(LD_BUG
, "Invalid proxy_state for reading, %d",
3051 tor_fragile_assert();
3056 log_debug(LD_NET
, "leaving state %s",
3057 connection_proxy_state_to_string(conn
->proxy_state
));
3061 log_warn(LD_NET
, "Proxy Client: unable to connect %s (%s)",
3062 connection_describe(conn
), escaped(reason
));
3065 log_warn(LD_NET
, "Proxy Client: unable to connect %s",
3066 connection_describe(conn
));
3068 } else if (ret
== 1) {
3069 log_info(LD_NET
, "Proxy Client: %s successful",
3070 connection_describe(conn
));
3076 /** Given a list of listener connections in <b>old_conns</b>, and list of
3077 * port_cfg_t entries in <b>ports</b>, open a new listener for every port in
3078 * <b>ports</b> that does not already have a listener in <b>old_conns</b>.
3080 * Remove from <b>old_conns</b> every connection that has a corresponding
3081 * entry in <b>ports</b>. Add to <b>new_conns</b> new every connection we
3082 * launch. If we may need to perform socket rebind when creating new
3083 * listener that replaces old one, create a <b>listener_replacement_t</b>
3084 * struct for affected pair and add it to <b>replacements</b>.
3086 * If <b>control_listeners_only</b> is true, then we only open control
3087 * listeners, and we do not remove any noncontrol listeners from
3090 * Return 0 on success, -1 on failure.
3093 retry_listener_ports(smartlist_t
*old_conns
,
3094 const smartlist_t
*ports
,
3095 smartlist_t
*new_conns
,
3096 smartlist_t
*replacements
,
3097 int control_listeners_only
)
3099 #ifndef ENABLE_LISTENER_REBIND
3103 smartlist_t
*launch
= smartlist_new();
3106 if (control_listeners_only
) {
3107 SMARTLIST_FOREACH(ports
, port_cfg_t
*, p
, {
3108 if (p
->type
== CONN_TYPE_CONTROL_LISTENER
)
3109 smartlist_add(launch
, p
);
3112 smartlist_add_all(launch
, ports
);
3115 /* Iterate through old_conns, comparing it to launch: remove from both lists
3116 * each pair of elements that corresponds to the same port. */
3117 SMARTLIST_FOREACH_BEGIN(old_conns
, connection_t
*, conn
) {
3118 const port_cfg_t
*found_port
= NULL
;
3120 /* Okay, so this is a listener. Is it configured? */
3121 /* That is, is it either: 1) exact match - address and port
3122 * pair match exactly between old listener and new port; or 2)
3123 * wildcard match - port matches exactly, but *one* of the
3124 * addresses is wildcard (0.0.0.0 or ::)?
3126 SMARTLIST_FOREACH_BEGIN(launch
, const port_cfg_t
*, wanted
) {
3127 if (conn
->type
!= wanted
->type
)
3129 if ((conn
->socket_family
!= AF_UNIX
&& wanted
->is_unix_addr
) ||
3130 (conn
->socket_family
== AF_UNIX
&& ! wanted
->is_unix_addr
))
3133 if (wanted
->server_cfg
.no_listen
)
3134 continue; /* We don't want to open a listener for this one */
3136 if (wanted
->is_unix_addr
) {
3137 if (conn
->socket_family
== AF_UNIX
&&
3138 !strcmp(wanted
->unix_addr
, conn
->address
)) {
3139 found_port
= wanted
;
3143 /* Numeric values of old and new port match exactly. */
3144 const int port_matches_exact
= (wanted
->port
== conn
->port
);
3145 /* Ports match semantically - either their specific values
3146 match exactly, or new port is 'auto'.
3148 const int port_matches
= (wanted
->port
== CFG_AUTO_PORT
||
3149 port_matches_exact
);
3151 if (port_matches
&& tor_addr_eq(&wanted
->addr
, &conn
->addr
)) {
3152 found_port
= wanted
;
3155 #ifdef ENABLE_LISTENER_REBIND
3156 /* Rebinding may be needed if all of the following are true:
3157 * 1) Address family is the same in old and new listeners.
3158 * 2) Port number matches exactly (numeric value is the same).
3159 * 3) *One* of listeners (either old one or new one) has a
3160 * wildcard IP address (0.0.0.0 or [::]).
3162 * These are the exact conditions for a first bind() syscall
3163 * to fail with EADDRINUSE.
3165 const int may_need_rebind
=
3166 tor_addr_family(&wanted
->addr
) == tor_addr_family(&conn
->addr
) &&
3167 port_matches_exact
&& bool_neq(tor_addr_is_null(&wanted
->addr
),
3168 tor_addr_is_null(&conn
->addr
));
3169 if (replacements
&& may_need_rebind
) {
3170 listener_replacement_t
*replacement
=
3171 tor_malloc(sizeof(listener_replacement_t
));
3173 replacement
->old_conn
= conn
;
3174 replacement
->new_port
= wanted
;
3175 smartlist_add(replacements
, replacement
);
3177 SMARTLIST_DEL_CURRENT(launch
, wanted
);
3178 SMARTLIST_DEL_CURRENT(old_conns
, conn
);
3181 #endif /* defined(ENABLE_LISTENER_REBIND) */
3183 } SMARTLIST_FOREACH_END(wanted
);
3186 /* This listener is already running; we don't need to launch it. */
3187 //log_debug(LD_NET, "Already have %s on %s:%d",
3188 // conn_type_to_string(found_port->type), conn->address, conn->port);
3189 smartlist_remove(launch
, found_port
);
3190 /* And we can remove the connection from old_conns too. */
3191 SMARTLIST_DEL_CURRENT(old_conns
, conn
);
3193 } SMARTLIST_FOREACH_END(conn
);
3195 /* Now open all the listeners that are configured but not opened. */
3196 SMARTLIST_FOREACH_BEGIN(launch
, const port_cfg_t
*, port
) {
3198 connection_t
*conn
= connection_listener_new_for_port(port
, &skip
, NULL
);
3200 if (conn
&& new_conns
)
3201 smartlist_add(new_conns
, conn
);
3204 } SMARTLIST_FOREACH_END(port
);
3206 smartlist_free(launch
);
3211 /** Launch listeners for each port you should have open. Only launch
3212 * listeners who are not already open, and only close listeners we no longer
3215 * Add all new connections to <b>new_conns</b>.
3217 * If <b>close_all_noncontrol</b> is true, then we only open control
3218 * listeners, and we close all other listeners.
3221 retry_all_listeners(smartlist_t
*new_conns
, int close_all_noncontrol
)
3223 smartlist_t
*listeners
= smartlist_new();
3224 smartlist_t
*replacements
= smartlist_new();
3225 const or_options_t
*options
= get_options();
3227 const uint16_t old_or_port
= routerconf_find_or_port(options
, AF_INET
);
3228 const uint16_t old_or_port_ipv6
=
3229 routerconf_find_or_port(options
,AF_INET6
);
3230 const uint16_t old_dir_port
= routerconf_find_dir_port(options
, 0);
3232 SMARTLIST_FOREACH_BEGIN(get_connection_array(), connection_t
*, conn
) {
3233 if (connection_is_listener(conn
) && !conn
->marked_for_close
)
3234 smartlist_add(listeners
, conn
);
3235 } SMARTLIST_FOREACH_END(conn
);
3237 if (retry_listener_ports(listeners
,
3238 get_configured_ports(),
3241 close_all_noncontrol
) < 0)
3244 #ifdef ENABLE_LISTENER_REBIND
3245 if (smartlist_len(replacements
))
3246 log_debug(LD_NET
, "%d replacements - starting rebinding loop.",
3247 smartlist_len(replacements
));
3249 SMARTLIST_FOREACH_BEGIN(replacements
, listener_replacement_t
*, r
) {
3250 int addr_in_use
= 0;
3253 tor_assert(r
->new_port
);
3254 tor_assert(r
->old_conn
);
3256 connection_t
*new_conn
=
3257 connection_listener_new_for_port(r
->new_port
, &skip
, &addr_in_use
);
3258 connection_t
*old_conn
= r
->old_conn
;
3261 log_debug(LD_NET
, "Skipping creating new listener for %s",
3262 connection_describe(old_conn
));
3266 connection_close_immediate(old_conn
);
3267 connection_mark_for_close(old_conn
);
3270 new_conn
= connection_listener_new_for_port(r
->new_port
,
3271 &skip
, &addr_in_use
);
3274 /* There are many reasons why we can't open a new listener port so in case
3275 * we hit those, bail early so tor can stop. */
3277 log_warn(LD_NET
, "Unable to create listener port: %s:%d",
3278 fmt_and_decorate_addr(&r
->new_port
->addr
), r
->new_port
->port
);
3283 smartlist_add(new_conns
, new_conn
);
3285 char *old_desc
= tor_strdup(connection_describe(old_conn
));
3286 log_notice(LD_NET
, "Closed no-longer-configured %s "
3288 old_desc
, connection_describe(new_conn
));
3290 } SMARTLIST_FOREACH_END(r
);
3291 #endif /* defined(ENABLE_LISTENER_REBIND) */
3293 /* Any members that were still in 'listeners' don't correspond to
3294 * any configured port. Kill 'em. */
3295 SMARTLIST_FOREACH_BEGIN(listeners
, connection_t
*, conn
) {
3296 log_notice(LD_NET
, "Closing no-longer-configured %s on %s:%d",
3297 conn_type_to_string(conn
->type
),
3298 fmt_and_decorate_addr(&conn
->addr
), conn
->port
);
3299 connection_close_immediate(conn
);
3300 connection_mark_for_close(conn
);
3301 } SMARTLIST_FOREACH_END(conn
);
3303 smartlist_free(listeners
);
3304 /* Cleanup any remaining listener replacement. */
3305 SMARTLIST_FOREACH(replacements
, listener_replacement_t
*, r
, tor_free(r
));
3306 smartlist_free(replacements
);
3308 if (old_or_port
!= routerconf_find_or_port(options
, AF_INET
) ||
3309 old_or_port_ipv6
!= routerconf_find_or_port(options
, AF_INET6
) ||
3310 old_dir_port
!= routerconf_find_dir_port(options
, 0)) {
3311 /* Our chosen ORPort or DirPort is not what it used to be: the
3312 * descriptor we had (if any) should be regenerated. (We won't
3313 * automatically notice this because of changes in the option,
3314 * since the value could be "auto".) */
3315 mark_my_descriptor_dirty("Chosen Or/DirPort changed");
3321 /** Mark every listener of type other than CONTROL_LISTENER to be closed. */
3323 connection_mark_all_noncontrol_listeners(void)
3325 SMARTLIST_FOREACH_BEGIN(get_connection_array(), connection_t
*, conn
) {
3326 if (conn
->marked_for_close
)
3328 if (conn
->type
== CONN_TYPE_CONTROL_LISTENER
)
3330 if (connection_is_listener(conn
))
3331 connection_mark_for_close(conn
);
3332 } SMARTLIST_FOREACH_END(conn
);
3335 /** Mark every external connection not used for controllers for close. */
3337 connection_mark_all_noncontrol_connections(void)
3339 SMARTLIST_FOREACH_BEGIN(get_connection_array(), connection_t
*, conn
) {
3340 if (conn
->marked_for_close
)
3342 switch (conn
->type
) {
3343 case CONN_TYPE_CONTROL_LISTENER
:
3344 case CONN_TYPE_CONTROL
:
3347 connection_mark_unattached_ap(TO_ENTRY_CONN(conn
),
3348 END_STREAM_REASON_HIBERNATING
);
3352 or_connection_t
*orconn
= TO_OR_CONN(conn
);
3354 connection_or_close_normally(orconn
, 0);
3357 * There should have been one, but mark for close and hope
3360 connection_mark_for_close(conn
);
3365 connection_mark_for_close(conn
);
3368 } SMARTLIST_FOREACH_END(conn
);
3371 /** Return 1 if we should apply rate limiting to <b>conn</b>, and 0
3373 * Right now this just checks if it's an internal IP address or an
3374 * internal connection. We also should, but don't, check if the connection
3375 * uses pluggable transports, since we should then limit it even if it
3376 * comes from an internal IP address. */
3378 connection_is_rate_limited(const connection_t
*conn
)
3380 const or_options_t
*options
= get_options();
3382 return 0; /* Internal connection */
3383 else if (! options
->CountPrivateBandwidth
&&
3384 ! conn
->always_rate_limit_as_remote
&&
3385 (tor_addr_family(&conn
->addr
) == AF_UNSPEC
|| /* no address */
3386 tor_addr_family(&conn
->addr
) == AF_UNIX
|| /* no address */
3387 tor_addr_is_internal(&conn
->addr
, 0)))
3388 return 0; /* Internal address */
3393 /** When was either global write bucket last empty? If this was recent, then
3394 * we're probably low on bandwidth, and we should be stingy with our bandwidth
3396 static time_t write_buckets_last_empty_at
= -100;
3398 /** How many seconds of no active local circuits will make the
3399 * connection revert to the "relayed" bandwidth class? */
3400 #define CLIENT_IDLE_TIME_FOR_PRIORITY 30
3402 /** Return 1 if <b>conn</b> should use tokens from the "relayed"
3403 * bandwidth rates, else 0. Currently, only OR conns with bandwidth
3404 * class 1, and directory conns that are serving data out, count.
3407 connection_counts_as_relayed_traffic(connection_t
*conn
, time_t now
)
3409 if (conn
->type
== CONN_TYPE_OR
&&
3410 connection_or_client_used(TO_OR_CONN(conn
)) +
3411 CLIENT_IDLE_TIME_FOR_PRIORITY
< now
)
3413 if (conn
->type
== CONN_TYPE_DIR
&& DIR_CONN_IS_SERVER(conn
))
3418 /** Helper function to decide how many bytes out of <b>global_bucket</b>
3419 * we're willing to use for this transaction. <b>base</b> is the size
3420 * of a cell on the network; <b>priority</b> says whether we should
3421 * write many of them or just a few; and <b>conn_bucket</b> (if
3422 * non-negative) provides an upper limit for our answer. */
3424 connection_bucket_get_share(int base
, int priority
,
3425 ssize_t global_bucket_val
, ssize_t conn_bucket
)
3428 ssize_t num_bytes_high
= (priority
? 32 : 16) * base
;
3429 ssize_t num_bytes_low
= (priority
? 4 : 2) * base
;
3431 /* Do a rudimentary limiting so one circuit can't hog a connection.
3432 * Pick at most 32 cells, at least 4 cells if possible, and if we're in
3433 * the middle pick 1/8 of the available bandwidth. */
3434 at_most
= global_bucket_val
/ 8;
3435 at_most
-= (at_most
% base
); /* round down */
3436 if (at_most
> num_bytes_high
) /* 16 KB, or 8 KB for low-priority */
3437 at_most
= num_bytes_high
;
3438 else if (at_most
< num_bytes_low
) /* 2 KB, or 1 KB for low-priority */
3439 at_most
= num_bytes_low
;
3441 if (at_most
> global_bucket_val
)
3442 at_most
= global_bucket_val
;
3444 if (conn_bucket
>= 0 && at_most
> conn_bucket
)
3445 at_most
= conn_bucket
;
3452 /** How many bytes at most can we read onto this connection? */
3454 connection_bucket_read_limit(connection_t
*conn
, time_t now
)
3456 int base
= RELAY_PAYLOAD_SIZE
;
3457 int priority
= conn
->type
!= CONN_TYPE_DIR
;
3458 ssize_t conn_bucket
= -1;
3459 size_t global_bucket_val
= token_bucket_rw_get_read(&global_bucket
);
3460 if (global_bucket_val
== 0) {
3461 /* We reached our global read limit: count this as an overload.
3463 * The token bucket is always initialized (see connection_bucket_init() and
3464 * options_validate_relay_bandwidth()) and hence we can assume that if the
3465 * token ever hits zero, it's a limit that got popped and not the bucket
3466 * being uninitialized.
3468 rep_hist_note_overload(OVERLOAD_READ
);
3471 if (connection_speaks_cells(conn
)) {
3472 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
3473 if (conn
->state
== OR_CONN_STATE_OPEN
)
3474 conn_bucket
= token_bucket_rw_get_read(&or_conn
->bucket
);
3475 base
= get_cell_network_size(or_conn
->wide_circ_ids
);
3478 /* Edge connection have their own read bucket due to flow control being able
3479 * to set a rate limit for them. However, for exit connections, we still need
3480 * to honor the global bucket as well. */
3481 if (CONN_IS_EDGE(conn
)) {
3482 const edge_connection_t
*edge_conn
= CONST_TO_EDGE_CONN(conn
);
3483 conn_bucket
= token_bucket_rw_get_read(&edge_conn
->bucket
);
3484 if (conn
->type
== CONN_TYPE_EXIT
) {
3485 /* Decide between our limit and the global one. */
3491 if (!connection_is_rate_limited(conn
)) {
3492 /* be willing to read on local conns even if our buckets are empty */
3493 return conn_bucket
>=0 ? conn_bucket
: 1<<14;
3496 if (connection_counts_as_relayed_traffic(conn
, now
)) {
3497 size_t relayed
= token_bucket_rw_get_read(&global_relayed_bucket
);
3498 global_bucket_val
= MIN(global_bucket_val
, relayed
);
3502 return connection_bucket_get_share(base
, priority
,
3503 global_bucket_val
, conn_bucket
);
3506 /** How many bytes at most can we write onto this connection? */
3508 connection_bucket_write_limit(connection_t
*conn
, time_t now
)
3510 int base
= RELAY_PAYLOAD_SIZE
;
3511 int priority
= conn
->type
!= CONN_TYPE_DIR
;
3512 size_t conn_bucket
= buf_datalen(conn
->outbuf
);
3513 size_t global_bucket_val
= token_bucket_rw_get_write(&global_bucket
);
3514 if (global_bucket_val
== 0) {
3515 /* We reached our global write limit: We should count this as an overload.
3516 * See above function for more information */
3517 rep_hist_note_overload(OVERLOAD_WRITE
);
3520 if (!connection_is_rate_limited(conn
)) {
3521 /* be willing to write to local conns even if our buckets are empty */
3525 if (connection_speaks_cells(conn
)) {
3526 /* use the per-conn write limit if it's lower */
3527 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
3528 if (conn
->state
== OR_CONN_STATE_OPEN
)
3529 conn_bucket
= MIN(conn_bucket
,
3530 token_bucket_rw_get_write(&or_conn
->bucket
));
3531 base
= get_cell_network_size(or_conn
->wide_circ_ids
);
3534 if (connection_counts_as_relayed_traffic(conn
, now
)) {
3535 size_t relayed
= token_bucket_rw_get_write(&global_relayed_bucket
);
3536 global_bucket_val
= MIN(global_bucket_val
, relayed
);
3539 return connection_bucket_get_share(base
, priority
,
3540 global_bucket_val
, conn_bucket
);
3543 /** Return true iff the global write buckets are low enough that we
3544 * shouldn't send <b>attempt</b> bytes of low-priority directory stuff
3545 * out to <b>conn</b>.
3547 * If we are a directory authority, always answer dir requests thus true is
3550 * Note: There are a lot of parameters we could use here:
3551 * - global_relayed_write_bucket. Low is bad.
3552 * - global_write_bucket. Low is bad.
3553 * - bandwidthrate. Low is bad.
3554 * - bandwidthburst. Not a big factor?
3555 * - attempt. High is bad.
3556 * - total bytes queued on outbufs. High is bad. But I'm wary of
3557 * using this, since a few slow-flushing queues will pump up the
3558 * number without meaning what we meant to mean. What we really
3559 * mean is "total directory bytes added to outbufs recently", but
3560 * that's harder to quantify and harder to keep track of.
3563 connection_dir_is_global_write_low(const connection_t
*conn
, size_t attempt
)
3565 size_t smaller_bucket
=
3566 MIN(token_bucket_rw_get_write(&global_bucket
),
3567 token_bucket_rw_get_write(&global_relayed_bucket
));
3569 /* Special case for authorities (directory only). */
3570 if (authdir_mode_v3(get_options())) {
3571 /* Are we configured to possibly reject requests under load? */
3572 if (!dirauth_should_reject_requests_under_load()) {
3573 /* Answer request no matter what. */
3576 /* Always answer requests from a known relay which includes the other
3577 * authorities. The following looks up the addresses for relays that we
3578 * have their descriptor _and_ any configured trusted directories. */
3579 if (nodelist_probably_contains_address(&conn
->addr
)) {
3584 if (!connection_is_rate_limited(conn
))
3585 return false; /* local conns don't get limited */
3587 if (smaller_bucket
< attempt
)
3588 return true; /* not enough space. */
3591 const time_t diff
= approx_time() - write_buckets_last_empty_at
;
3593 return true; /* we're already hitting our limits, no more please */
3598 /** When did we last tell the accounting subsystem about transmitted
3600 static time_t last_recorded_accounting_at
= 0;
3602 /** Helper: adjusts our bandwidth history and informs the controller as
3603 * appropriate, given that we have just read <b>num_read</b> bytes and written
3604 * <b>num_written</b> bytes on <b>conn</b>. */
3606 record_num_bytes_transferred_impl(connection_t
*conn
,
3607 time_t now
, size_t num_read
, size_t num_written
)
3609 /* Count bytes of answering direct and tunneled directory requests */
3610 if (conn
->type
== CONN_TYPE_DIR
&& conn
->purpose
== DIR_PURPOSE_SERVER
) {
3612 bwhist_note_dir_bytes_read(num_read
, now
);
3613 if (num_written
> 0)
3614 bwhist_note_dir_bytes_written(num_written
, now
);
3617 /* Linked connections and internal IPs aren't counted for statistics or
3619 * - counting linked connections would double-count BEGINDIR bytes, because
3620 * they are sent as Dir bytes on the linked connection, and OR bytes on
3621 * the OR connection;
3622 * - relays and clients don't connect to internal IPs, unless specifically
3623 * configured to do so. If they are configured that way, we don't count
3626 if (!connection_is_rate_limited(conn
))
3629 const bool is_ipv6
= (conn
->socket_family
== AF_INET6
);
3630 if (conn
->type
== CONN_TYPE_OR
)
3631 conn_stats_note_or_conn_bytes(conn
->global_identifier
, num_read
,
3632 num_written
, now
, is_ipv6
);
3635 bwhist_note_bytes_read(num_read
, now
, is_ipv6
);
3637 if (num_written
> 0) {
3638 bwhist_note_bytes_written(num_written
, now
, is_ipv6
);
3640 if (conn
->type
== CONN_TYPE_EXIT
)
3641 rep_hist_note_exit_bytes(conn
->port
, num_written
, num_read
);
3643 /* Remember these bytes towards statistics. */
3644 stats_increment_bytes_read_and_written(num_read
, num_written
);
3646 /* Remember these bytes towards accounting. */
3647 if (accounting_is_enabled(get_options())) {
3648 if (now
> last_recorded_accounting_at
&& last_recorded_accounting_at
) {
3649 accounting_add_bytes(num_read
, num_written
,
3650 (int)(now
- last_recorded_accounting_at
));
3652 accounting_add_bytes(num_read
, num_written
, 0);
3654 last_recorded_accounting_at
= now
;
3658 /** We just read <b>num_read</b> and wrote <b>num_written</b> bytes
3659 * onto <b>conn</b>. Decrement buckets appropriately. */
3661 connection_buckets_decrement(connection_t
*conn
, time_t now
,
3662 size_t num_read
, size_t num_written
)
3664 if (num_written
>= INT_MAX
|| num_read
>= INT_MAX
) {
3665 log_err(LD_BUG
, "Value out of range. num_read=%lu, num_written=%lu, "
3666 "connection type=%s, state=%s",
3667 (unsigned long)num_read
, (unsigned long)num_written
,
3668 conn_type_to_string(conn
->type
),
3669 conn_state_to_string(conn
->type
, conn
->state
));
3670 tor_assert_nonfatal_unreached();
3671 if (num_written
>= INT_MAX
)
3673 if (num_read
>= INT_MAX
)
3677 record_num_bytes_transferred_impl(conn
, now
, num_read
, num_written
);
3679 /* Edge connection need to decrement the read side of the bucket used by our
3680 * congestion control. */
3681 if (CONN_IS_EDGE(conn
) && num_read
> 0) {
3682 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
3683 token_bucket_rw_dec(&edge_conn
->bucket
, num_read
, 0);
3686 if (!connection_is_rate_limited(conn
))
3687 return; /* local IPs are free */
3690 if (connection_counts_as_relayed_traffic(conn
, now
)) {
3691 flags
= token_bucket_rw_dec(&global_relayed_bucket
, num_read
, num_written
);
3693 flags
|= token_bucket_rw_dec(&global_bucket
, num_read
, num_written
);
3695 if (flags
& TB_WRITE
) {
3696 write_buckets_last_empty_at
= now
;
3698 if (connection_speaks_cells(conn
) && conn
->state
== OR_CONN_STATE_OPEN
) {
3699 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
3700 token_bucket_rw_dec(&or_conn
->bucket
, num_read
, num_written
);
3705 * Mark <b>conn</b> as needing to stop reading because bandwidth has been
3706 * exhausted. If <b>is_global_bw</b>, it is closing because global bandwidth
3707 * limit has been exhausted. Otherwise, it is closing because its own
3708 * bandwidth limit has been exhausted.
3711 connection_read_bw_exhausted(connection_t
*conn
, bool is_global_bw
)
3714 conn
->read_blocked_on_bw
= 1;
3715 connection_stop_reading(conn
);
3716 reenable_blocked_connection_schedule();
3720 * Mark <b>conn</b> as needing to stop reading because write bandwidth has
3721 * been exhausted. If <b>is_global_bw</b>, it is closing because global
3722 * bandwidth limit has been exhausted. Otherwise, it is closing because its
3723 * own bandwidth limit has been exhausted.
3726 connection_write_bw_exhausted(connection_t
*conn
, bool is_global_bw
)
3729 conn
->write_blocked_on_bw
= 1;
3730 connection_stop_writing(conn
);
3731 reenable_blocked_connection_schedule();
3734 /** If we have exhausted our global buckets, or the buckets for conn,
3737 connection_consider_empty_read_buckets(connection_t
*conn
)
3742 if (CONN_IS_EDGE(conn
) &&
3743 token_bucket_rw_get_read(&TO_EDGE_CONN(conn
)->bucket
) <= 0) {
3744 reason
= "edge connection read bucket exhausted. Pausing.";
3746 } else if (!connection_is_rate_limited(conn
)) {
3747 return; /* Always okay. */
3748 } else if (token_bucket_rw_get_read(&global_bucket
) <= 0) {
3749 reason
= "global read bucket exhausted. Pausing.";
3750 } else if (connection_counts_as_relayed_traffic(conn
, approx_time()) &&
3751 token_bucket_rw_get_read(&global_relayed_bucket
) <= 0) {
3752 reason
= "global relayed read bucket exhausted. Pausing.";
3753 } else if (connection_speaks_cells(conn
) &&
3754 conn
->state
== OR_CONN_STATE_OPEN
&&
3755 token_bucket_rw_get_read(&TO_OR_CONN(conn
)->bucket
) <= 0) {
3756 reason
= "connection read bucket exhausted. Pausing.";
3759 return; /* all good, no need to stop it */
3762 LOG_FN_CONN(conn
, (LOG_DEBUG
, LD_NET
, "%s", reason
));
3763 connection_read_bw_exhausted(conn
, is_global
);
3766 /** If we have exhausted our global buckets, or the buckets for conn,
3769 connection_consider_empty_write_buckets(connection_t
*conn
)
3773 if (!connection_is_rate_limited(conn
))
3774 return; /* Always okay. */
3776 bool is_global
= true;
3777 if (token_bucket_rw_get_write(&global_bucket
) <= 0) {
3778 reason
= "global write bucket exhausted. Pausing.";
3779 } else if (connection_counts_as_relayed_traffic(conn
, approx_time()) &&
3780 token_bucket_rw_get_write(&global_relayed_bucket
) <= 0) {
3781 reason
= "global relayed write bucket exhausted. Pausing.";
3782 } else if (connection_speaks_cells(conn
) &&
3783 conn
->state
== OR_CONN_STATE_OPEN
&&
3784 token_bucket_rw_get_write(&TO_OR_CONN(conn
)->bucket
) <= 0) {
3785 reason
= "connection write bucket exhausted. Pausing.";
3788 return; /* all good, no need to stop it */
3790 LOG_FN_CONN(conn
, (LOG_DEBUG
, LD_NET
, "%s", reason
));
3791 connection_write_bw_exhausted(conn
, is_global
);
3794 /** Initialize the global buckets to the values configured in the
3797 connection_bucket_init(void)
3799 const or_options_t
*options
= get_options();
3800 const uint32_t now_ts
= monotime_coarse_get_stamp();
3801 token_bucket_rw_init(&global_bucket
,
3802 (int32_t)options
->BandwidthRate
,
3803 (int32_t)options
->BandwidthBurst
,
3805 if (options
->RelayBandwidthRate
) {
3806 token_bucket_rw_init(&global_relayed_bucket
,
3807 (int32_t)options
->RelayBandwidthRate
,
3808 (int32_t)options
->RelayBandwidthBurst
,
3811 token_bucket_rw_init(&global_relayed_bucket
,
3812 (int32_t)options
->BandwidthRate
,
3813 (int32_t)options
->BandwidthBurst
,
3817 reenable_blocked_connection_init(options
);
3820 /** Update the global connection bucket settings to a new value. */
3822 connection_bucket_adjust(const or_options_t
*options
)
3824 token_bucket_rw_adjust(&global_bucket
,
3825 (int32_t)options
->BandwidthRate
,
3826 (int32_t)options
->BandwidthBurst
);
3827 if (options
->RelayBandwidthRate
) {
3828 token_bucket_rw_adjust(&global_relayed_bucket
,
3829 (int32_t)options
->RelayBandwidthRate
,
3830 (int32_t)options
->RelayBandwidthBurst
);
3832 token_bucket_rw_adjust(&global_relayed_bucket
,
3833 (int32_t)options
->BandwidthRate
,
3834 (int32_t)options
->BandwidthBurst
);
3839 * Cached value of the last coarse-timestamp when we refilled the
3842 static uint32_t last_refilled_global_buckets_ts
=0;
3844 * Refill the token buckets for a single connection <b>conn</b>, and the
3845 * global token buckets as appropriate. Requires that <b>now_ts</b> is
3846 * the time in coarse timestamp units.
3849 connection_bucket_refill_single(connection_t
*conn
, uint32_t now_ts
)
3851 /* Note that we only check for equality here: the underlying
3852 * token bucket functions can handle moving backwards in time if they
3854 if (now_ts
!= last_refilled_global_buckets_ts
) {
3855 token_bucket_rw_refill(&global_bucket
, now_ts
);
3856 token_bucket_rw_refill(&global_relayed_bucket
, now_ts
);
3857 last_refilled_global_buckets_ts
= now_ts
;
3860 if (connection_speaks_cells(conn
) && conn
->state
== OR_CONN_STATE_OPEN
) {
3861 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
3862 token_bucket_rw_refill(&or_conn
->bucket
, now_ts
);
3865 if (CONN_IS_EDGE(conn
)) {
3866 token_bucket_rw_refill(&TO_EDGE_CONN(conn
)->bucket
, now_ts
);
3871 * Event to re-enable all connections that were previously blocked on read or
3874 static mainloop_event_t
*reenable_blocked_connections_ev
= NULL
;
3876 /** True iff reenable_blocked_connections_ev is currently scheduled. */
3877 static int reenable_blocked_connections_is_scheduled
= 0;
3879 /** Delay after which to run reenable_blocked_connections_ev. */
3880 static struct timeval reenable_blocked_connections_delay
;
3883 * Re-enable all connections that were previously blocked on read or write.
3884 * This event is scheduled after enough time has elapsed to be sure
3885 * that the buckets will refill when the connections have something to do.
3888 reenable_blocked_connections_cb(mainloop_event_t
*ev
, void *arg
)
3892 SMARTLIST_FOREACH_BEGIN(get_connection_array(), connection_t
*, conn
) {
3893 if (conn
->read_blocked_on_bw
== 1) {
3894 connection_start_reading(conn
);
3895 conn
->read_blocked_on_bw
= 0;
3897 if (conn
->write_blocked_on_bw
== 1) {
3898 connection_start_writing(conn
);
3899 conn
->write_blocked_on_bw
= 0;
3901 } SMARTLIST_FOREACH_END(conn
);
3903 reenable_blocked_connections_is_scheduled
= 0;
3907 * Initialize the mainloop event that we use to wake up connections that
3908 * find themselves blocked on bandwidth.
3911 reenable_blocked_connection_init(const or_options_t
*options
)
3913 if (! reenable_blocked_connections_ev
) {
3914 reenable_blocked_connections_ev
=
3915 mainloop_event_new(reenable_blocked_connections_cb
, NULL
);
3916 reenable_blocked_connections_is_scheduled
= 0;
3918 time_t sec
= options
->TokenBucketRefillInterval
/ 1000;
3919 int msec
= (options
->TokenBucketRefillInterval
% 1000);
3920 reenable_blocked_connections_delay
.tv_sec
= sec
;
3921 reenable_blocked_connections_delay
.tv_usec
= msec
* 1000;
3925 * Called when we have blocked a connection for being low on bandwidth:
3926 * schedule an event to reenable such connections, if it is not already
3930 reenable_blocked_connection_schedule(void)
3932 if (reenable_blocked_connections_is_scheduled
)
3934 if (BUG(reenable_blocked_connections_ev
== NULL
)) {
3935 reenable_blocked_connection_init(get_options());
3937 mainloop_event_schedule(reenable_blocked_connections_ev
,
3938 &reenable_blocked_connections_delay
);
3939 reenable_blocked_connections_is_scheduled
= 1;
3942 /** Read bytes from conn-\>s and process them.
3944 * It calls connection_buf_read_from_socket() to bring in any new bytes,
3945 * and then calls connection_process_inbuf() to process them.
3947 * Mark the connection and return -1 if you want to close it, else
3951 connection_handle_read_impl(connection_t
*conn
)
3953 ssize_t max_to_read
=-1, try_to_read
;
3954 size_t before
, n_read
= 0;
3955 int socket_error
= 0;
3957 if (conn
->marked_for_close
)
3958 return 0; /* do nothing */
3960 conn
->timestamp_last_read_allowed
= approx_time();
3962 connection_bucket_refill_single(conn
, monotime_coarse_get_stamp());
3964 switch (conn
->type
) {
3965 case CONN_TYPE_OR_LISTENER
:
3966 return connection_handle_listener_read(conn
, CONN_TYPE_OR
);
3967 case CONN_TYPE_EXT_OR_LISTENER
:
3968 return connection_handle_listener_read(conn
, CONN_TYPE_EXT_OR
);
3969 case CONN_TYPE_AP_LISTENER
:
3970 case CONN_TYPE_AP_TRANS_LISTENER
:
3971 case CONN_TYPE_AP_NATD_LISTENER
:
3972 case CONN_TYPE_AP_HTTP_CONNECT_LISTENER
:
3973 return connection_handle_listener_read(conn
, CONN_TYPE_AP
);
3974 case CONN_TYPE_DIR_LISTENER
:
3975 return connection_handle_listener_read(conn
, CONN_TYPE_DIR
);
3976 case CONN_TYPE_CONTROL_LISTENER
:
3977 return connection_handle_listener_read(conn
, CONN_TYPE_CONTROL
);
3978 case CONN_TYPE_METRICS_LISTENER
:
3979 return connection_handle_listener_read(conn
, CONN_TYPE_METRICS
);
3980 case CONN_TYPE_AP_DNS_LISTENER
:
3981 /* This should never happen; eventdns.c handles the reads here. */
3982 tor_fragile_assert();
3987 try_to_read
= max_to_read
;
3988 tor_assert(!conn
->marked_for_close
);
3990 before
= buf_datalen(conn
->inbuf
);
3991 if (connection_buf_read_from_socket(conn
, &max_to_read
, &socket_error
) < 0) {
3992 /* There's a read error; kill the connection.*/
3993 if (conn
->type
== CONN_TYPE_OR
) {
3994 connection_or_notify_error(TO_OR_CONN(conn
),
3996 errno_to_orconn_end_reason(socket_error
) :
3997 END_OR_CONN_REASON_CONNRESET
,
3999 tor_socket_strerror(socket_error
) :
4000 "(unknown, errno was 0)");
4002 if (CONN_IS_EDGE(conn
)) {
4003 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
4004 connection_edge_end_errno(edge_conn
);
4005 if (conn
->type
== CONN_TYPE_AP
&& TO_ENTRY_CONN(conn
)->socks_request
) {
4006 /* broken, don't send a socks reply back */
4007 TO_ENTRY_CONN(conn
)->socks_request
->has_finished
= 1;
4010 connection_close_immediate(conn
); /* Don't flush; connection is dead. */
4012 * This can bypass normal channel checking since we did
4013 * connection_or_notify_error() above.
4015 connection_mark_for_close_internal(conn
);
4018 n_read
+= buf_datalen(conn
->inbuf
) - before
;
4019 if (CONN_IS_EDGE(conn
) && try_to_read
!= max_to_read
) {
4020 /* instruct it not to try to package partial cells. */
4021 if (connection_process_inbuf(conn
, 0) < 0) {
4024 if (!conn
->marked_for_close
&&
4025 connection_is_reading(conn
) &&
4026 !conn
->inbuf_reached_eof
&&
4028 goto loop_again
; /* try reading again, in case more is here now */
4030 /* one last try, packaging partial cells and all. */
4031 if (!conn
->marked_for_close
&&
4032 connection_process_inbuf(conn
, 1) < 0) {
4035 if (conn
->linked_conn
) {
4036 /* The other side's handle_write() will never actually get called, so
4037 * we need to invoke the appropriate callbacks ourself. */
4038 connection_t
*linked
= conn
->linked_conn
;
4041 /* Probably a no-op, since linked conns typically don't count for
4042 * bandwidth rate limiting. But do it anyway so we can keep stats
4043 * accurately. Note that since we read the bytes from conn, and
4044 * we're writing the bytes onto the linked connection, we count
4045 * these as <i>written</i> bytes. */
4046 connection_buckets_decrement(linked
, approx_time(), 0, n_read
);
4048 if (connection_flushed_some(linked
) < 0)
4049 connection_mark_for_close(linked
);
4050 if (!connection_wants_to_flush(linked
))
4051 connection_finished_flushing(linked
);
4054 if (!buf_datalen(linked
->outbuf
) && conn
->active_on_link
)
4055 connection_stop_reading_from_linked_conn(conn
);
4057 /* If we hit the EOF, call connection_reached_eof(). */
4058 if (!conn
->marked_for_close
&&
4059 conn
->inbuf_reached_eof
&&
4060 connection_reached_eof(conn
) < 0) {
4066 /* DOCDOC connection_handle_read */
4068 connection_handle_read(connection_t
*conn
)
4071 update_current_time(time(NULL
));
4072 res
= connection_handle_read_impl(conn
);
4076 /** Pull in new bytes from conn-\>s or conn-\>linked_conn onto conn-\>inbuf,
4077 * either directly or via TLS. Reduce the token buckets by the number of bytes
4080 * If *max_to_read is -1, then decide it ourselves, else go with the
4081 * value passed to us. When returning, if it's changed, subtract the
4082 * number of bytes we read from *max_to_read.
4084 * Return -1 if we want to break conn, else return 0.
4087 connection_buf_read_from_socket(connection_t
*conn
, ssize_t
*max_to_read
,
4091 ssize_t at_most
= *max_to_read
;
4092 size_t slack_in_buf
, more_to_read
;
4093 size_t n_read
= 0, n_written
= 0;
4095 if (at_most
== -1) { /* we need to initialize it */
4096 /* how many bytes are we allowed to read? */
4097 at_most
= connection_bucket_read_limit(conn
, approx_time());
4100 /* Do not allow inbuf to grow past BUF_MAX_LEN. */
4101 const ssize_t maximum
= BUF_MAX_LEN
- buf_datalen(conn
->inbuf
);
4102 if (at_most
> maximum
) {
4106 slack_in_buf
= buf_slack(conn
->inbuf
);
4108 if ((size_t)at_most
> slack_in_buf
&& slack_in_buf
>= 1024) {
4109 more_to_read
= at_most
- slack_in_buf
;
4110 at_most
= slack_in_buf
;
4115 if (connection_speaks_cells(conn
) &&
4116 conn
->state
> OR_CONN_STATE_PROXY_HANDSHAKING
) {
4118 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
4119 size_t initial_size
;
4120 if (conn
->state
== OR_CONN_STATE_TLS_HANDSHAKING
||
4121 conn
->state
== OR_CONN_STATE_TLS_CLIENT_RENEGOTIATING
) {
4122 /* continue handshaking even if global token bucket is empty */
4123 return connection_tls_continue_handshake(or_conn
);
4127 "%d: starting, inbuf_datalen %ld (%d pending in tls object)."
4129 (int)conn
->s
,(long)buf_datalen(conn
->inbuf
),
4130 tor_tls_get_pending_bytes(or_conn
->tls
), (long)at_most
);
4132 initial_size
= buf_datalen(conn
->inbuf
);
4133 /* else open, or closing */
4134 result
= buf_read_from_tls(conn
->inbuf
, or_conn
->tls
, at_most
);
4135 if (TOR_TLS_IS_ERROR(result
) || result
== TOR_TLS_CLOSE
)
4136 or_conn
->tls_error
= result
;
4138 or_conn
->tls_error
= 0;
4142 case TOR_TLS_ERROR_IO
:
4143 log_debug(LD_NET
,"TLS %s closed %son read. Closing.",
4144 connection_describe(conn
),
4145 result
== TOR_TLS_CLOSE
? "cleanly " : "");
4147 CASE_TOR_TLS_ERROR_ANY_NONIO
:
4148 log_debug(LD_NET
,"tls error [%s] from %s. Breaking.",
4149 tor_tls_err_to_string(result
),
4150 connection_describe(conn
));
4152 case TOR_TLS_WANTWRITE
:
4153 connection_start_writing(conn
);
4155 case TOR_TLS_WANTREAD
:
4156 if (conn
->in_connection_handle_write
) {
4157 /* We've been invoked from connection_handle_write, because we're
4158 * waiting for a TLS renegotiation, the renegotiation started, and
4159 * SSL_read returned WANTWRITE. But now SSL_read is saying WANTREAD
4160 * again. Stop waiting for write events now, or else we'll
4161 * busy-loop until data arrives for us to read.
4162 * XXX: remove this when v2 handshakes support is dropped. */
4163 connection_stop_writing(conn
);
4164 if (!connection_is_reading(conn
))
4165 connection_start_reading(conn
);
4167 /* we're already reading, one hopes */
4169 case TOR_TLS_DONE
: /* no data read, so nothing to process */
4170 break; /* so we call bucket_decrement below */
4174 pending
= tor_tls_get_pending_bytes(or_conn
->tls
);
4176 /* If we have any pending bytes, we read them now. This *can*
4177 * take us over our read allotment, but really we shouldn't be
4178 * believing that SSL bytes are the same as TCP bytes anyway. */
4179 int r2
= buf_read_from_tls(conn
->inbuf
, or_conn
->tls
, pending
);
4181 log_warn(LD_BUG
, "apparently, reading pending bytes can fail.");
4185 result
= (int)(buf_datalen(conn
->inbuf
)-initial_size
);
4186 tor_tls_get_n_raw_bytes(or_conn
->tls
, &n_read
, &n_written
);
4187 log_debug(LD_GENERAL
, "After TLS read of %d: %ld read, %ld written",
4188 result
, (long)n_read
, (long)n_written
);
4189 } else if (conn
->linked
) {
4190 if (conn
->linked_conn
) {
4191 result
= (int) buf_move_all(conn
->inbuf
, conn
->linked_conn
->outbuf
);
4195 //log_notice(LD_GENERAL, "Moved %d bytes on an internal link!", result);
4196 /* If the other side has disappeared, or if it's been marked for close and
4197 * we flushed its outbuf, then we should set our inbuf_reached_eof. */
4198 if (!conn
->linked_conn
||
4199 (conn
->linked_conn
->marked_for_close
&&
4200 buf_datalen(conn
->linked_conn
->outbuf
) == 0))
4201 conn
->inbuf_reached_eof
= 1;
4203 n_read
= (size_t) result
;
4205 /* !connection_speaks_cells, !conn->linked_conn. */
4206 int reached_eof
= 0;
4207 CONN_LOG_PROTECT(conn
,
4208 result
= buf_read_from_socket(conn
->inbuf
, conn
->s
,
4213 conn
->inbuf_reached_eof
= 1;
4215 // log_fn(LOG_DEBUG,"read_to_buf returned %d.",read_result);
4219 n_read
= (size_t) result
;
4223 /* change *max_to_read */
4224 *max_to_read
= at_most
- n_read
;
4226 /* Onion service application connection. Note read bytes for metrics. */
4227 if (CONN_IS_EDGE(conn
) && TO_EDGE_CONN(conn
)->hs_ident
) {
4228 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
4229 hs_metrics_app_read_bytes(&edge_conn
->hs_ident
->identity_pk
,
4230 edge_conn
->hs_ident
->orig_virtual_port
,
4234 /* Update edge_conn->n_read */
4235 if (conn
->type
== CONN_TYPE_AP
) {
4236 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
4238 /* Check for overflow: */
4239 if (PREDICT_LIKELY(UINT32_MAX
- edge_conn
->n_read
> n_read
))
4240 edge_conn
->n_read
+= (int)n_read
;
4242 edge_conn
->n_read
= UINT32_MAX
;
4245 /* If CONN_BW events are enabled, update conn->n_read_conn_bw for
4246 * OR/DIR/EXIT connections, checking for overflow. */
4247 if (get_options()->TestingEnableConnBwEvent
&&
4248 (conn
->type
== CONN_TYPE_OR
||
4249 conn
->type
== CONN_TYPE_DIR
||
4250 conn
->type
== CONN_TYPE_EXIT
)) {
4251 if (PREDICT_LIKELY(UINT32_MAX
- conn
->n_read_conn_bw
> n_read
))
4252 conn
->n_read_conn_bw
+= (int)n_read
;
4254 conn
->n_read_conn_bw
= UINT32_MAX
;
4258 connection_buckets_decrement(conn
, approx_time(), n_read
, n_written
);
4260 if (more_to_read
&& result
== at_most
) {
4261 slack_in_buf
= buf_slack(conn
->inbuf
);
4262 at_most
= more_to_read
;
4266 /* Call even if result is 0, since the global read bucket may
4267 * have reached 0 on a different conn, and this connection needs to
4268 * know to stop reading. */
4269 connection_consider_empty_read_buckets(conn
);
4270 if (n_written
> 0 && connection_is_writing(conn
))
4271 connection_consider_empty_write_buckets(conn
);
4276 /** A pass-through to fetch_from_buf. */
4278 connection_buf_get_bytes(char *string
, size_t len
, connection_t
*conn
)
4280 return buf_get_bytes(conn
->inbuf
, string
, len
);
4283 /** As buf_get_line(), but read from a connection's input buffer. */
4285 connection_buf_get_line(connection_t
*conn
, char *data
,
4288 return buf_get_line(conn
->inbuf
, data
, data_len
);
4291 /** As fetch_from_buf_http, but fetches from a connection's input buffer_t as
4294 connection_fetch_from_buf_http(connection_t
*conn
,
4295 char **headers_out
, size_t max_headerlen
,
4296 char **body_out
, size_t *body_used
,
4297 size_t max_bodylen
, int force_complete
)
4299 return fetch_from_buf_http(conn
->inbuf
, headers_out
, max_headerlen
,
4300 body_out
, body_used
, max_bodylen
, force_complete
);
4303 /** Return true if this connection has data to flush. */
4305 connection_wants_to_flush(connection_t
*conn
)
4307 return connection_get_outbuf_len(conn
) > 0;
4310 /** Are there too many bytes on edge connection <b>conn</b>'s outbuf to
4311 * send back a relay-level sendme yet? Return 1 if so, 0 if not. Used by
4312 * connection_edge_consider_sending_sendme().
4315 connection_outbuf_too_full(connection_t
*conn
)
4317 return connection_get_outbuf_len(conn
) > 10*CELL_PAYLOAD_SIZE
;
4321 * On Windows Vista and Windows 7, tune the send buffer size according to a
4324 * This should help fix slow upload rates.
4327 update_send_buffer_size(tor_socket_t sock
)
4330 /* We only do this on Vista and 7, because earlier versions of Windows
4331 * don't have the SIO_IDEAL_SEND_BACKLOG_QUERY functionality, and on
4332 * later versions it isn't necessary. */
4333 static int isVistaOr7
= -1;
4334 if (isVistaOr7
== -1) {
4336 OSVERSIONINFO osvi
= { 0 };
4337 osvi
.dwOSVersionInfoSize
= sizeof(OSVERSIONINFO
);
4338 GetVersionEx(&osvi
);
4339 if (osvi
.dwMajorVersion
== 6 && osvi
.dwMinorVersion
< 2)
4344 if (get_options()->ConstrainedSockets
)
4347 DWORD bytesReturned
= 0;
4348 if (!WSAIoctl(sock
, SIO_IDEAL_SEND_BACKLOG_QUERY
, NULL
, 0,
4349 &isb
, sizeof(isb
), &bytesReturned
, NULL
, NULL
)) {
4350 setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
, (const char*)&isb
, sizeof(isb
));
4352 #else /* !defined(_WIN32) */
4354 #endif /* defined(_WIN32) */
4357 /** Try to flush more bytes onto <b>conn</b>-\>s.
4359 * This function is called in connection_handle_write(), which gets
4360 * called from conn_write_callback() in main.c when libevent tells us
4361 * that <b>conn</b> wants to write.
4363 * Update <b>conn</b>-\>timestamp_last_write_allowed to now, and call flush_buf
4364 * or flush_buf_tls appropriately. If it succeeds and there are no more
4365 * more bytes on <b>conn</b>-\>outbuf, then call connection_finished_flushing
4368 * If <b>force</b>, then write as many bytes as possible, ignoring bandwidth
4369 * limits. (Used for flushing messages to controller connections on fatal
4372 * Mark the connection and return -1 if you want to close it, else
4376 connection_handle_write_impl(connection_t
*conn
, int force
)
4379 socklen_t len
=(socklen_t
)sizeof(e
);
4381 ssize_t max_to_write
;
4382 time_t now
= approx_time();
4383 size_t n_read
= 0, n_written
= 0;
4384 int dont_stop_writing
= 0;
4386 tor_assert(!connection_is_listener(conn
));
4388 if (conn
->marked_for_close
|| !SOCKET_OK(conn
->s
))
4389 return 0; /* do nothing */
4391 if (conn
->in_flushed_some
) {
4392 log_warn(LD_BUG
, "called recursively from inside conn->in_flushed_some");
4396 conn
->timestamp_last_write_allowed
= now
;
4398 connection_bucket_refill_single(conn
, monotime_coarse_get_stamp());
4400 /* Sometimes, "writable" means "connected". */
4401 if (connection_state_is_connecting(conn
)) {
4402 if (getsockopt(conn
->s
, SOL_SOCKET
, SO_ERROR
, (void*)&e
, &len
) < 0) {
4403 log_warn(LD_BUG
, "getsockopt() syscall failed");
4404 if (conn
->type
== CONN_TYPE_OR
) {
4405 or_connection_t
*orconn
= TO_OR_CONN(conn
);
4406 connection_or_close_for_error(orconn
, 0);
4408 if (CONN_IS_EDGE(conn
)) {
4409 connection_edge_end_errno(TO_EDGE_CONN(conn
));
4411 connection_mark_for_close(conn
);
4416 /* some sort of error, but maybe just inprogress still */
4417 if (!ERRNO_IS_CONN_EINPROGRESS(e
)) {
4418 log_info(LD_NET
,"in-progress connect failed. Removing. (%s)",
4419 tor_socket_strerror(e
));
4420 if (CONN_IS_EDGE(conn
))
4421 connection_edge_end_errno(TO_EDGE_CONN(conn
));
4422 if (conn
->type
== CONN_TYPE_OR
)
4423 connection_or_notify_error(TO_OR_CONN(conn
),
4424 errno_to_orconn_end_reason(e
),
4425 tor_socket_strerror(e
));
4427 connection_close_immediate(conn
);
4429 * This can bypass normal channel checking since we did
4430 * connection_or_notify_error() above.
4432 connection_mark_for_close_internal(conn
);
4435 return 0; /* no change, see if next time is better */
4438 /* The connection is successful. */
4439 if (connection_finished_connecting(conn
)<0)
4443 max_to_write
= force
? (ssize_t
)buf_datalen(conn
->outbuf
)
4444 : connection_bucket_write_limit(conn
, now
);
4446 if (connection_speaks_cells(conn
) &&
4447 conn
->state
> OR_CONN_STATE_PROXY_HANDSHAKING
) {
4448 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
4449 size_t initial_size
;
4450 if (conn
->state
== OR_CONN_STATE_TLS_HANDSHAKING
||
4451 conn
->state
== OR_CONN_STATE_TLS_CLIENT_RENEGOTIATING
) {
4452 connection_stop_writing(conn
);
4453 if (connection_tls_continue_handshake(or_conn
) < 0) {
4454 /* Don't flush; connection is dead. */
4455 connection_or_notify_error(or_conn
,
4456 END_OR_CONN_REASON_MISC
,
4457 "TLS error in connection_tls_"
4458 "continue_handshake()");
4459 connection_close_immediate(conn
);
4461 * This can bypass normal channel checking since we did
4462 * connection_or_notify_error() above.
4464 connection_mark_for_close_internal(conn
);
4468 } else if (conn
->state
== OR_CONN_STATE_TLS_SERVER_RENEGOTIATING
) {
4469 return connection_handle_read(conn
);
4472 /* else open, or closing */
4473 initial_size
= buf_datalen(conn
->outbuf
);
4474 result
= buf_flush_to_tls(conn
->outbuf
, or_conn
->tls
,
4478 update_send_buffer_size(conn
->s
);
4480 /* If we just flushed the last bytes, tell the channel on the
4481 * or_conn to check if it needs to geoip_change_dirreq_state() */
4482 /* XXXX move this to flushed_some or finished_flushing -NM */
4483 if (buf_datalen(conn
->outbuf
) == 0 && or_conn
->chan
)
4484 channel_notify_flushed(TLS_CHAN_TO_BASE(or_conn
->chan
));
4487 CASE_TOR_TLS_ERROR_ANY
:
4489 or_conn
->tls_error
= result
;
4490 log_info(LD_NET
, result
!= TOR_TLS_CLOSE
?
4491 "tls error. breaking.":"TLS connection closed on flush");
4492 /* Don't flush; connection is dead. */
4493 connection_or_notify_error(or_conn
,
4494 END_OR_CONN_REASON_MISC
,
4495 result
!= TOR_TLS_CLOSE
?
4496 "TLS error in during flush" :
4497 "TLS closed during flush");
4498 connection_close_immediate(conn
);
4500 * This can bypass normal channel checking since we did
4501 * connection_or_notify_error() above.
4503 connection_mark_for_close_internal(conn
);
4505 case TOR_TLS_WANTWRITE
:
4506 log_debug(LD_NET
,"wanted write.");
4507 /* we're already writing */
4508 dont_stop_writing
= 1;
4510 case TOR_TLS_WANTREAD
:
4511 /* Make sure to avoid a loop if the receive buckets are empty. */
4512 log_debug(LD_NET
,"wanted read.");
4513 if (!connection_is_reading(conn
)) {
4514 connection_write_bw_exhausted(conn
, true);
4515 /* we'll start reading again when we get more tokens in our
4516 * read bucket; then we'll start writing again too.
4519 /* else no problem, we're already reading */
4521 /* case TOR_TLS_DONE:
4522 * for TOR_TLS_DONE, fall through to check if the flushlen
4523 * is empty, so we can stop writing.
4527 tor_tls_get_n_raw_bytes(or_conn
->tls
, &n_read
, &n_written
);
4528 log_debug(LD_GENERAL
, "After TLS write of %d: %ld read, %ld written",
4529 result
, (long)n_read
, (long)n_written
);
4530 or_conn
->bytes_xmitted
+= result
;
4531 or_conn
->bytes_xmitted_by_tls
+= n_written
;
4532 /* So we notice bytes were written even on error */
4533 /* XXXX This cast is safe since we can never write INT_MAX bytes in a
4534 * single set of TLS operations. But it looks kinda ugly. If we refactor
4535 * the *_buf_tls functions, we should make them return ssize_t or size_t
4537 result
= (int)(initial_size
-buf_datalen(conn
->outbuf
));
4539 CONN_LOG_PROTECT(conn
,
4540 result
= buf_flush_to_socket(conn
->outbuf
, conn
->s
,
4543 if (CONN_IS_EDGE(conn
))
4544 connection_edge_end_errno(TO_EDGE_CONN(conn
));
4545 if (conn
->type
== CONN_TYPE_AP
) {
4546 /* writing failed; we couldn't send a SOCKS reply if we wanted to */
4547 TO_ENTRY_CONN(conn
)->socks_request
->has_finished
= 1;
4550 connection_close_immediate(conn
); /* Don't flush; connection is dead. */
4551 connection_mark_for_close(conn
);
4554 update_send_buffer_size(conn
->s
);
4555 n_written
= (size_t) result
;
4558 if (n_written
&& conn
->type
== CONN_TYPE_AP
) {
4559 edge_connection_t
*edge_conn
= TO_EDGE_CONN(conn
);
4561 /* Check for overflow: */
4562 if (PREDICT_LIKELY(UINT32_MAX
- edge_conn
->n_written
> n_written
))
4563 edge_conn
->n_written
+= (int)n_written
;
4565 edge_conn
->n_written
= UINT32_MAX
;
4568 /* If CONN_BW events are enabled, update conn->n_written_conn_bw for
4569 * OR/DIR/EXIT connections, checking for overflow. */
4570 if (n_written
&& get_options()->TestingEnableConnBwEvent
&&
4571 (conn
->type
== CONN_TYPE_OR
||
4572 conn
->type
== CONN_TYPE_DIR
||
4573 conn
->type
== CONN_TYPE_EXIT
)) {
4574 if (PREDICT_LIKELY(UINT32_MAX
- conn
->n_written_conn_bw
> n_written
))
4575 conn
->n_written_conn_bw
+= (int)n_written
;
4577 conn
->n_written_conn_bw
= UINT32_MAX
;
4580 connection_buckets_decrement(conn
, approx_time(), n_read
, n_written
);
4583 /* If we wrote any bytes from our buffer, then call the appropriate
4585 if (connection_flushed_some(conn
) < 0) {
4586 if (connection_speaks_cells(conn
)) {
4587 connection_or_notify_error(TO_OR_CONN(conn
),
4588 END_OR_CONN_REASON_MISC
,
4589 "Got error back from "
4590 "connection_flushed_some()");
4594 * This can bypass normal channel checking since we did
4595 * connection_or_notify_error() above.
4597 connection_mark_for_close_internal(conn
);
4601 if (!connection_wants_to_flush(conn
) &&
4602 !dont_stop_writing
) { /* it's done flushing */
4603 if (connection_finished_flushing(conn
) < 0) {
4604 /* already marked */
4610 /* Call even if result is 0, since the global write bucket may
4611 * have reached 0 on a different conn, and this connection needs to
4612 * know to stop writing. */
4613 connection_consider_empty_write_buckets(conn
);
4614 if (n_read
> 0 && connection_is_reading(conn
))
4615 connection_consider_empty_read_buckets(conn
);
4618 /* If this is an edge connection with congestion control, check to see
4619 * if it is time to send an xon */
4620 if (conn_uses_flow_control(conn
)) {
4621 flow_control_decide_xon(TO_EDGE_CONN(conn
), n_written
);
4630 /* DOCDOC connection_handle_write */
4632 connection_handle_write(connection_t
*conn
, int force
)
4635 update_current_time(time(NULL
));
4636 /* connection_handle_write_impl() might call connection_handle_read()
4637 * if we're in the middle of a v2 handshake, in which case it needs this
4639 conn
->in_connection_handle_write
= 1;
4640 res
= connection_handle_write_impl(conn
, force
);
4641 conn
->in_connection_handle_write
= 0;
4646 * Try to flush data that's waiting for a write on <b>conn</b>. Return
4647 * -1 on failure, 0 on success.
4649 * Don't use this function for regular writing; the buffers
4650 * system should be good enough at scheduling writes there. Instead, this
4651 * function is for cases when we're about to exit or something and we want
4652 * to report it right away.
4655 connection_flush(connection_t
*conn
)
4657 return connection_handle_write(conn
, 1);
4660 /** Helper for connection_write_to_buf_impl and connection_write_buf_to_buf:
4662 * Return true iff it is okay to queue bytes on <b>conn</b>'s outbuf for
4666 connection_may_write_to_buf(connection_t
*conn
)
4668 /* if it's marked for close, only allow write if we mean to flush it */
4669 if (conn
->marked_for_close
&& !conn
->hold_open_until_flushed
)
4675 /** Helper for connection_write_to_buf_impl and connection_write_buf_to_buf:
4677 * Called when an attempt to add bytes on <b>conn</b>'s outbuf has failed;
4678 * mark the connection and warn as appropriate.
4681 connection_write_to_buf_failed(connection_t
*conn
)
4683 if (CONN_IS_EDGE(conn
)) {
4684 /* if it failed, it means we have our package/delivery windows set
4685 wrong compared to our max outbuf size. close the whole circuit. */
4687 "write_to_buf failed. Closing circuit (fd %d).", (int)conn
->s
);
4688 circuit_mark_for_close(circuit_get_by_edge_conn(TO_EDGE_CONN(conn
)),
4689 END_CIRC_REASON_INTERNAL
);
4690 } else if (conn
->type
== CONN_TYPE_OR
) {
4691 or_connection_t
*orconn
= TO_OR_CONN(conn
);
4693 "write_to_buf failed on an orconn; notifying of error "
4694 "(fd %d)", (int)(conn
->s
));
4695 connection_or_close_for_error(orconn
, 0);
4698 "write_to_buf failed. Closing connection (fd %d).",
4700 connection_mark_for_close(conn
);
4704 /** Helper for connection_write_to_buf_impl and connection_write_buf_to_buf:
4706 * Called when an attempt to add bytes on <b>conn</b>'s outbuf has succeeded:
4707 * start writing if appropriate.
4710 connection_write_to_buf_commit(connection_t
*conn
)
4712 /* If we receive optimistic data in the EXIT_CONN_STATE_RESOLVING
4713 * state, we don't want to try to write it right away, since
4714 * conn->write_event won't be set yet. Otherwise, write data from
4715 * this conn as the socket is available. */
4716 if (conn
->write_event
) {
4717 connection_start_writing(conn
);
4721 /** Append <b>len</b> bytes of <b>string</b> onto <b>conn</b>'s
4722 * outbuf, and ask it to start writing.
4724 * If <b>zlib</b> is nonzero, this is a directory connection that should get
4725 * its contents compressed or decompressed as they're written. If zlib is
4726 * negative, this is the last data to be compressed, and the connection's zlib
4727 * state should be flushed.
4730 connection_write_to_buf_impl_
,(const char *string
, size_t len
,
4731 connection_t
*conn
, int zlib
))
4733 /* XXXX This function really needs to return -1 on failure. */
4735 if (!len
&& !(zlib
<0))
4738 if (!connection_may_write_to_buf(conn
))
4742 dir_connection_t
*dir_conn
= TO_DIR_CONN(conn
);
4743 int done
= zlib
< 0;
4744 CONN_LOG_PROTECT(conn
, r
= buf_add_compress(conn
->outbuf
,
4745 dir_conn
->compress_state
,
4746 string
, len
, done
));
4748 CONN_LOG_PROTECT(conn
, r
= buf_add(conn
->outbuf
, string
, len
));
4751 connection_write_to_buf_failed(conn
);
4754 connection_write_to_buf_commit(conn
);
4758 * Write a <b>string</b> (of size <b>len</b> to directory connection
4759 * <b>dir_conn</b>. Apply compression if connection is configured to use
4760 * it and finalize it if <b>done</b> is true.
4763 connection_dir_buf_add(const char *string
, size_t len
,
4764 dir_connection_t
*dir_conn
, int done
)
4766 if (dir_conn
->compress_state
!= NULL
) {
4767 connection_buf_add_compress(string
, len
, dir_conn
, done
);
4771 connection_buf_add(string
, len
, TO_CONN(dir_conn
));
4775 connection_buf_add_compress(const char *string
, size_t len
,
4776 dir_connection_t
*conn
, int done
)
4778 connection_write_to_buf_impl_(string
, len
, TO_CONN(conn
), done
? -1 : 1);
4782 * Add all bytes from <b>buf</b> to <b>conn</b>'s outbuf, draining them
4783 * from <b>buf</b>. (If the connection is marked and will soon be closed,
4784 * nothing is drained.)
4787 connection_buf_add_buf(connection_t
*conn
, buf_t
*buf
)
4791 size_t len
= buf_datalen(buf
);
4795 if (!connection_may_write_to_buf(conn
))
4798 buf_move_all(conn
->outbuf
, buf
);
4799 connection_write_to_buf_commit(conn
);
4802 #define CONN_GET_ALL_TEMPLATE(var, test) \
4804 smartlist_t *conns = get_connection_array(); \
4805 smartlist_t *ret_conns = smartlist_new(); \
4806 SMARTLIST_FOREACH_BEGIN(conns, connection_t *, var) { \
4807 if (var && (test) && !var->marked_for_close) \
4808 smartlist_add(ret_conns, var); \
4809 } SMARTLIST_FOREACH_END(var); \
4813 /* Return a list of connections that aren't close and matches the given type
4814 * and state. The returned list can be empty and must be freed using
4815 * smartlist_free(). The caller does NOT have ownership of the objects in the
4816 * list so it must not free them nor reference them as they can disappear. */
4818 connection_list_by_type_state(int type
, int state
)
4820 CONN_GET_ALL_TEMPLATE(conn
, (conn
->type
== type
&& conn
->state
== state
));
4823 /* Return a list of connections that aren't close and matches the given type
4824 * and purpose. The returned list can be empty and must be freed using
4825 * smartlist_free(). The caller does NOT have ownership of the objects in the
4826 * list so it must not free them nor reference them as they can disappear. */
4828 connection_list_by_type_purpose(int type
, int purpose
)
4830 CONN_GET_ALL_TEMPLATE(conn
,
4831 (conn
->type
== type
&& conn
->purpose
== purpose
));
4834 /** Return a connection_t * from get_connection_array() that satisfies test on
4835 * var, and that is not marked for close. */
4836 #define CONN_GET_TEMPLATE(var, test) \
4838 smartlist_t *conns = get_connection_array(); \
4839 SMARTLIST_FOREACH(conns, connection_t *, var, \
4841 if (var && (test) && !var->marked_for_close) \
4847 /** Return a connection with given type, address, port, and purpose;
4848 * or NULL if no such connection exists (or if all such connections are marked
4850 MOCK_IMPL(connection_t
*,
4851 connection_get_by_type_addr_port_purpose
,(int type
,
4852 const tor_addr_t
*addr
, uint16_t port
,
4855 CONN_GET_TEMPLATE(conn
,
4856 (conn
->type
== type
&&
4857 tor_addr_eq(&conn
->addr
, addr
) &&
4858 conn
->port
== port
&&
4859 conn
->purpose
== purpose
));
4862 /** Return the stream with id <b>id</b> if it is not already marked for
4866 connection_get_by_global_id(uint64_t id
)
4868 CONN_GET_TEMPLATE(conn
, conn
->global_identifier
== id
);
4871 /** Return a connection of type <b>type</b> that is not marked for close.
4874 connection_get_by_type(int type
)
4876 CONN_GET_TEMPLATE(conn
, conn
->type
== type
);
4879 /** Return a connection of type <b>type</b> that is in state <b>state</b>,
4880 * and that is not marked for close.
4883 connection_get_by_type_state(int type
, int state
)
4885 CONN_GET_TEMPLATE(conn
, conn
->type
== type
&& conn
->state
== state
);
4889 * Return a connection of type <b>type</b> that is not an internally linked
4890 * connection, and is not marked for close.
4892 MOCK_IMPL(connection_t
*,
4893 connection_get_by_type_nonlinked
,(int type
))
4895 CONN_GET_TEMPLATE(conn
, conn
->type
== type
&& !conn
->linked
);
4898 /** Return a new smartlist of dir_connection_t * from get_connection_array()
4899 * that satisfy conn_test on connection_t *conn_var, and dirconn_test on
4900 * dir_connection_t *dirconn_var. conn_var must be of CONN_TYPE_DIR and not
4901 * marked for close to be included in the list. */
4902 #define DIR_CONN_LIST_TEMPLATE(conn_var, conn_test, \
4903 dirconn_var, dirconn_test) \
4905 smartlist_t *conns = get_connection_array(); \
4906 smartlist_t *dir_conns = smartlist_new(); \
4907 SMARTLIST_FOREACH_BEGIN(conns, connection_t *, conn_var) { \
4908 if (conn_var && (conn_test) \
4909 && conn_var->type == CONN_TYPE_DIR \
4910 && !conn_var->marked_for_close) { \
4911 dir_connection_t *dirconn_var = TO_DIR_CONN(conn_var); \
4912 if (dirconn_var && (dirconn_test)) { \
4913 smartlist_add(dir_conns, dirconn_var); \
4916 } SMARTLIST_FOREACH_END(conn_var); \
4920 /** Return a list of directory connections that are fetching the item
4921 * described by <b>purpose</b>/<b>resource</b>. If there are none,
4922 * return an empty list. This list must be freed using smartlist_free,
4923 * but the pointers in it must not be freed.
4924 * Note that this list should not be cached, as the pointers in it can be
4925 * freed if their connections close. */
4927 connection_dir_list_by_purpose_and_resource(
4929 const char *resource
)
4931 DIR_CONN_LIST_TEMPLATE(conn
,
4932 conn
->purpose
== purpose
,
4934 0 == strcmp_opt(resource
,
4935 dirconn
->requested_resource
));
4938 /** Return a list of directory connections that are fetching the item
4939 * described by <b>purpose</b>/<b>resource</b>/<b>state</b>. If there are
4940 * none, return an empty list. This list must be freed using smartlist_free,
4941 * but the pointers in it must not be freed.
4942 * Note that this list should not be cached, as the pointers in it can be
4943 * freed if their connections close. */
4945 connection_dir_list_by_purpose_resource_and_state(
4947 const char *resource
,
4950 DIR_CONN_LIST_TEMPLATE(conn
,
4951 conn
->purpose
== purpose
&& conn
->state
== state
,
4953 0 == strcmp_opt(resource
,
4954 dirconn
->requested_resource
));
4957 #undef DIR_CONN_LIST_TEMPLATE
4959 /** Return an arbitrary active OR connection that isn't <b>this_conn</b>.
4961 * We use this to guess if we should tell the controller that we
4962 * didn't manage to connect to any of our bridges. */
4963 static connection_t
*
4964 connection_get_another_active_or_conn(const or_connection_t
*this_conn
)
4966 CONN_GET_TEMPLATE(conn
,
4967 conn
!= TO_CONN(this_conn
) && conn
->type
== CONN_TYPE_OR
);
4970 /** Return 1 if there are any active OR connections apart from
4973 * We use this to guess if we should tell the controller that we
4974 * didn't manage to connect to any of our bridges. */
4976 any_other_active_or_conns(const or_connection_t
*this_conn
)
4978 connection_t
*conn
= connection_get_another_active_or_conn(this_conn
);
4980 log_debug(LD_DIR
, "%s: Found an OR connection: %s",
4981 __func__
, connection_describe(conn
));
4988 #undef CONN_GET_TEMPLATE
4990 /** Return 1 if <b>conn</b> is a listener conn, else return 0. */
4992 connection_is_listener(connection_t
*conn
)
4994 if (conn
->type
== CONN_TYPE_OR_LISTENER
||
4995 conn
->type
== CONN_TYPE_EXT_OR_LISTENER
||
4996 conn
->type
== CONN_TYPE_AP_LISTENER
||
4997 conn
->type
== CONN_TYPE_AP_TRANS_LISTENER
||
4998 conn
->type
== CONN_TYPE_AP_DNS_LISTENER
||
4999 conn
->type
== CONN_TYPE_AP_NATD_LISTENER
||
5000 conn
->type
== CONN_TYPE_AP_HTTP_CONNECT_LISTENER
||
5001 conn
->type
== CONN_TYPE_DIR_LISTENER
||
5002 conn
->type
== CONN_TYPE_METRICS_LISTENER
||
5003 conn
->type
== CONN_TYPE_CONTROL_LISTENER
)
5008 /** Return 1 if <b>conn</b> is in state "open" and is not marked
5009 * for close, else return 0.
5012 connection_state_is_open(connection_t
*conn
)
5016 if (conn
->marked_for_close
)
5019 if ((conn
->type
== CONN_TYPE_OR
&& conn
->state
== OR_CONN_STATE_OPEN
) ||
5020 (conn
->type
== CONN_TYPE_EXT_OR
) ||
5021 (conn
->type
== CONN_TYPE_AP
&& conn
->state
== AP_CONN_STATE_OPEN
) ||
5022 (conn
->type
== CONN_TYPE_EXIT
&& conn
->state
== EXIT_CONN_STATE_OPEN
) ||
5023 (conn
->type
== CONN_TYPE_CONTROL
&&
5024 conn
->state
== CONTROL_CONN_STATE_OPEN
))
5030 /** Return 1 if conn is in 'connecting' state, else return 0. */
5032 connection_state_is_connecting(connection_t
*conn
)
5036 if (conn
->marked_for_close
)
5041 return conn
->state
== OR_CONN_STATE_CONNECTING
;
5042 case CONN_TYPE_EXIT
:
5043 return conn
->state
== EXIT_CONN_STATE_CONNECTING
;
5045 return conn
->state
== DIR_CONN_STATE_CONNECTING
;
5051 /** Allocates a base64'ed authenticator for use in http or https
5052 * auth, based on the input string <b>authenticator</b>. Returns it
5053 * if success, else returns NULL. */
5055 alloc_http_authenticator(const char *authenticator
)
5057 /* an authenticator in Basic authentication
5058 * is just the string "username:password" */
5059 const size_t authenticator_length
= strlen(authenticator
);
5060 const size_t base64_authenticator_length
=
5061 base64_encode_size(authenticator_length
, 0) + 1;
5062 char *base64_authenticator
= tor_malloc(base64_authenticator_length
);
5063 if (base64_encode(base64_authenticator
, base64_authenticator_length
,
5064 authenticator
, authenticator_length
, 0) < 0) {
5065 tor_free(base64_authenticator
); /* free and set to null */
5067 return base64_authenticator
;
5070 /** Given a socket handle, check whether the local address (sockname) of the
5071 * socket is one that we've connected from before. If so, double-check
5072 * whether our address has changed and we need to generate keys. If we do,
5076 client_check_address_changed(tor_socket_t sock
)
5078 tor_addr_t out_addr
, iface_addr
;
5079 tor_addr_t
**last_interface_ip_ptr
;
5082 if (!outgoing_addrs
)
5083 outgoing_addrs
= smartlist_new();
5085 if (tor_addr_from_getsockname(&out_addr
, sock
) < 0) {
5086 int e
= tor_socket_errno(sock
);
5087 log_warn(LD_NET
, "getsockname() to check for address change failed: %s",
5088 tor_socket_strerror(e
));
5091 family
= tor_addr_family(&out_addr
);
5093 if (family
== AF_INET
)
5094 last_interface_ip_ptr
= &last_interface_ipv4
;
5095 else if (family
== AF_INET6
)
5096 last_interface_ip_ptr
= &last_interface_ipv6
;
5100 if (! *last_interface_ip_ptr
) {
5101 tor_addr_t
*a
= tor_malloc_zero(sizeof(tor_addr_t
));
5102 if (get_interface_address6(LOG_INFO
, family
, a
)==0) {
5103 *last_interface_ip_ptr
= a
;
5109 /* If we've used this address previously, we're okay. */
5110 SMARTLIST_FOREACH(outgoing_addrs
, const tor_addr_t
*, a_ptr
,
5111 if (tor_addr_eq(a_ptr
, &out_addr
))
5115 /* Uh-oh. We haven't connected from this address before. Has the interface
5116 * address changed? */
5117 if (get_interface_address6(LOG_INFO
, family
, &iface_addr
)<0)
5120 if (tor_addr_eq(&iface_addr
, *last_interface_ip_ptr
)) {
5121 /* Nope, it hasn't changed. Add this address to the list. */
5122 smartlist_add(outgoing_addrs
, tor_memdup(&out_addr
, sizeof(tor_addr_t
)));
5124 /* The interface changed. We're a client, so we need to regenerate our
5125 * keys. First, reset the state. */
5126 log_notice(LD_NET
, "Our IP address has changed. Rotating keys...");
5127 tor_addr_copy(*last_interface_ip_ptr
, &iface_addr
);
5128 SMARTLIST_FOREACH(outgoing_addrs
, tor_addr_t
*, a_ptr
, tor_free(a_ptr
));
5129 smartlist_clear(outgoing_addrs
);
5130 smartlist_add(outgoing_addrs
, tor_memdup(&out_addr
, sizeof(tor_addr_t
)));
5131 /* We'll need to resolve ourselves again. */
5132 resolved_addr_reset_last(AF_INET
);
5133 /* Okay, now change our keys. */
5134 ip_address_changed(1);
5138 /** Some systems have limited system buffers for recv and xmit on
5139 * sockets allocated in a virtual server or similar environment. For a Tor
5140 * server this can produce the "Error creating network socket: No buffer
5141 * space available" error once all available TCP buffer space is consumed.
5142 * This method will attempt to constrain the buffers allocated for the socket
5143 * to the desired size to stay below system TCP buffer limits.
5146 set_constrained_socket_buffers(tor_socket_t sock
, int size
)
5148 void *sz
= (void*)&size
;
5149 socklen_t sz_sz
= (socklen_t
) sizeof(size
);
5150 if (setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
, sz
, sz_sz
) < 0) {
5151 int e
= tor_socket_errno(sock
);
5152 log_warn(LD_NET
, "setsockopt() to constrain send "
5153 "buffer to %d bytes failed: %s", size
, tor_socket_strerror(e
));
5155 if (setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
, sz
, sz_sz
) < 0) {
5156 int e
= tor_socket_errno(sock
);
5157 log_warn(LD_NET
, "setsockopt() to constrain recv "
5158 "buffer to %d bytes failed: %s", size
, tor_socket_strerror(e
));
5162 /** Process new bytes that have arrived on conn-\>inbuf.
5164 * This function just passes conn to the connection-specific
5165 * connection_*_process_inbuf() function. It also passes in
5166 * package_partial if wanted.
5169 connection_process_inbuf(connection_t
*conn
, int package_partial
)
5173 switch (conn
->type
) {
5175 return connection_or_process_inbuf(TO_OR_CONN(conn
));
5176 case CONN_TYPE_EXT_OR
:
5177 return connection_ext_or_process_inbuf(TO_OR_CONN(conn
));
5178 case CONN_TYPE_EXIT
:
5180 return connection_edge_process_inbuf(TO_EDGE_CONN(conn
),
5183 return connection_dir_process_inbuf(TO_DIR_CONN(conn
));
5184 case CONN_TYPE_CONTROL
:
5185 return connection_control_process_inbuf(TO_CONTROL_CONN(conn
));
5186 case CONN_TYPE_METRICS
:
5187 return metrics_connection_process_inbuf(conn
);
5189 log_err(LD_BUG
,"got unexpected conn type %d.", conn
->type
);
5190 tor_fragile_assert();
5195 /** Called whenever we've written data on a connection. */
5197 connection_flushed_some(connection_t
*conn
)
5200 tor_assert(!conn
->in_flushed_some
);
5201 conn
->in_flushed_some
= 1;
5202 if (conn
->type
== CONN_TYPE_DIR
&&
5203 conn
->state
== DIR_CONN_STATE_SERVER_WRITING
) {
5204 r
= connection_dirserv_flushed_some(TO_DIR_CONN(conn
));
5205 } else if (conn
->type
== CONN_TYPE_OR
) {
5206 r
= connection_or_flushed_some(TO_OR_CONN(conn
));
5207 } else if (CONN_IS_EDGE(conn
)) {
5208 r
= connection_edge_flushed_some(TO_EDGE_CONN(conn
));
5210 conn
->in_flushed_some
= 0;
5214 /** We just finished flushing bytes to the appropriately low network layer,
5215 * and there are no more bytes remaining in conn-\>outbuf or
5216 * conn-\>tls to be flushed.
5218 * This function just passes conn to the connection-specific
5219 * connection_*_finished_flushing() function.
5222 connection_finished_flushing(connection_t
*conn
)
5226 /* If the connection is closed, don't try to do anything more here. */
5227 if (CONN_IS_CLOSED(conn
))
5230 // log_fn(LOG_DEBUG,"entered. Socket %u.", conn->s);
5232 connection_stop_writing(conn
);
5234 switch (conn
->type
) {
5236 return connection_or_finished_flushing(TO_OR_CONN(conn
));
5237 case CONN_TYPE_EXT_OR
:
5238 return connection_ext_or_finished_flushing(TO_OR_CONN(conn
));
5240 case CONN_TYPE_EXIT
:
5241 return connection_edge_finished_flushing(TO_EDGE_CONN(conn
));
5243 return connection_dir_finished_flushing(TO_DIR_CONN(conn
));
5244 case CONN_TYPE_CONTROL
:
5245 return connection_control_finished_flushing(TO_CONTROL_CONN(conn
));
5246 case CONN_TYPE_METRICS
:
5247 return metrics_connection_finished_flushing(conn
);
5249 log_err(LD_BUG
,"got unexpected conn type %d.", conn
->type
);
5250 tor_fragile_assert();
5255 /** Called when our attempt to connect() to a server has just succeeded.
5257 * This function checks if the interface address has changed (clients only),
5258 * and then passes conn to the connection-specific
5259 * connection_*_finished_connecting() function.
5262 connection_finished_connecting(connection_t
*conn
)
5266 if (!server_mode(get_options())) {
5267 /* See whether getsockname() says our address changed. We need to do this
5268 * now that the connection has finished, because getsockname() on Windows
5269 * won't work until then. */
5270 client_check_address_changed(conn
->s
);
5276 return connection_or_finished_connecting(TO_OR_CONN(conn
));
5277 case CONN_TYPE_EXIT
:
5278 return connection_edge_finished_connecting(TO_EDGE_CONN(conn
));
5280 return connection_dir_finished_connecting(TO_DIR_CONN(conn
));
5282 log_err(LD_BUG
,"got unexpected conn type %d.", conn
->type
);
5283 tor_fragile_assert();
5288 /** Callback: invoked when a connection reaches an EOF event. */
5290 connection_reached_eof(connection_t
*conn
)
5292 switch (conn
->type
) {
5294 case CONN_TYPE_EXT_OR
:
5295 return connection_or_reached_eof(TO_OR_CONN(conn
));
5297 case CONN_TYPE_EXIT
:
5298 return connection_edge_reached_eof(TO_EDGE_CONN(conn
));
5300 return connection_dir_reached_eof(TO_DIR_CONN(conn
));
5301 case CONN_TYPE_CONTROL
:
5302 return connection_control_reached_eof(TO_CONTROL_CONN(conn
));
5303 case CONN_TYPE_METRICS
:
5304 return metrics_connection_reached_eof(conn
);
5306 log_err(LD_BUG
,"got unexpected conn type %d.", conn
->type
);
5307 tor_fragile_assert();
5312 /** Comparator for the two-orconn case in OOS victim sort */
5314 oos_victim_comparator_for_orconns(or_connection_t
*a
, or_connection_t
*b
)
5316 int a_circs
, b_circs
;
5317 /* Fewer circuits == higher priority for OOS kill, sort earlier */
5319 a_circs
= connection_or_get_num_circuits(a
);
5320 b_circs
= connection_or_get_num_circuits(b
);
5322 if (a_circs
< b_circs
) return 1;
5323 else if (a_circs
> b_circs
) return -1;
5327 /** Sort comparator for OOS victims; better targets sort before worse
5330 oos_victim_comparator(const void **a_v
, const void **b_v
)
5332 connection_t
*a
= NULL
, *b
= NULL
;
5334 /* Get connection pointers out */
5336 a
= (connection_t
*)(*a_v
);
5337 b
= (connection_t
*)(*b_v
);
5339 tor_assert(a
!= NULL
);
5340 tor_assert(b
!= NULL
);
5343 * We always prefer orconns as victims currently; we won't even see
5344 * these non-orconn cases, but if we do, sort them after orconns.
5346 if (a
->type
== CONN_TYPE_OR
&& b
->type
== CONN_TYPE_OR
) {
5347 return oos_victim_comparator_for_orconns(TO_OR_CONN(a
), TO_OR_CONN(b
));
5350 * One isn't an orconn; if one is, it goes first. We currently have no
5351 * opinions about cases where neither is an orconn.
5353 if (a
->type
== CONN_TYPE_OR
) return -1;
5354 else if (b
->type
== CONN_TYPE_OR
) return 1;
5359 /** Pick n victim connections for the OOS handler and return them in a
5362 MOCK_IMPL(STATIC smartlist_t
*,
5363 pick_oos_victims
, (int n
))
5365 smartlist_t
*eligible
= NULL
, *victims
= NULL
;
5367 int conn_counts_by_type
[CONN_TYPE_MAX_
+ 1], i
;
5370 * Big damn assumption (someone improve this someday!):
5372 * Socket exhaustion normally happens on high-volume relays, and so
5373 * most of the connections involved are orconns. We should pick victims
5374 * by assembling a list of all orconns, and sorting them in order of
5375 * how much 'damage' by some metric we'd be doing by dropping them.
5377 * If we move on from orconns, we should probably think about incoming
5378 * directory connections next, or exit connections. Things we should
5379 * probably never kill are controller connections and listeners.
5381 * This function will count how many connections of different types
5382 * exist and log it for purposes of gathering data on typical OOS
5383 * situations to guide future improvements.
5386 /* First, get the connection array */
5387 conns
= get_connection_array();
5389 * Iterate it and pick out eligible connection types, and log some stats
5392 eligible
= smartlist_new();
5393 memset(conn_counts_by_type
, 0, sizeof(conn_counts_by_type
));
5394 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, c
) {
5395 /* Bump the counter */
5396 tor_assert(c
->type
<= CONN_TYPE_MAX_
);
5397 ++(conn_counts_by_type
[c
->type
]);
5399 /* Skip anything without a socket we can free */
5400 if (!(SOCKET_OK(c
->s
))) {
5404 /* Skip anything we would count as moribund */
5405 if (connection_is_moribund(c
)) {
5411 /* We've got an orconn, it's eligible to be OOSed */
5412 smartlist_add(eligible
, c
);
5415 /* We don't know what to do with it, ignore it */
5418 } SMARTLIST_FOREACH_END(c
);
5420 /* Log some stats */
5421 if (smartlist_len(conns
) > 0) {
5422 /* At least one counter must be non-zero */
5423 log_info(LD_NET
, "Some stats on conn types seen during OOS follow");
5424 for (i
= CONN_TYPE_MIN_
; i
<= CONN_TYPE_MAX_
; ++i
) {
5425 /* Did we see any? */
5426 if (conn_counts_by_type
[i
] > 0) {
5427 log_info(LD_NET
, "%s: %d conns",
5428 conn_type_to_string(i
),
5429 conn_counts_by_type
[i
]);
5432 log_info(LD_NET
, "Done with OOS conn type stats");
5435 /* Did we find more eligible targets than we want to kill? */
5436 if (smartlist_len(eligible
) > n
) {
5437 /* Sort the list in order of target preference */
5438 smartlist_sort(eligible
, oos_victim_comparator
);
5439 /* Pick first n as victims */
5440 victims
= smartlist_new();
5441 for (i
= 0; i
< n
; ++i
) {
5442 smartlist_add(victims
, smartlist_get(eligible
, i
));
5444 /* Free the original list */
5445 smartlist_free(eligible
);
5447 /* No, we can just call them all victims */
5454 /** Kill a list of connections for the OOS handler. */
5455 MOCK_IMPL(STATIC
void,
5456 kill_conn_list_for_oos
, (smartlist_t
*conns
))
5460 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, c
) {
5461 /* Make sure the channel layer gets told about orconns */
5462 if (c
->type
== CONN_TYPE_OR
) {
5463 connection_or_close_for_error(TO_OR_CONN(c
), 1);
5465 connection_mark_for_close(c
);
5467 } SMARTLIST_FOREACH_END(c
);
5470 "OOS handler marked %d connections",
5471 smartlist_len(conns
));
5474 /** Check if a connection is on the way out so the OOS handler doesn't try
5475 * to kill more than it needs. */
5477 connection_is_moribund(connection_t
*conn
)
5480 (conn
->conn_array_index
< 0 ||
5481 conn
->marked_for_close
)) {
5488 /** Out-of-Sockets handler; n_socks is the current number of open
5489 * sockets, and failed is non-zero if a socket exhaustion related
5490 * error immediately preceded this call. This is where to do
5491 * circuit-killing heuristics as needed.
5494 connection_check_oos(int n_socks
, int failed
)
5496 int target_n_socks
= 0, moribund_socks
, socks_to_kill
;
5499 /* Early exit: is OOS checking disabled? */
5500 if (get_options()->DisableOOSCheck
) {
5504 /* Sanity-check args */
5505 tor_assert(n_socks
>= 0);
5508 * Make some log noise; keep it at debug level since this gets a chance
5509 * to run on every connection attempt.
5512 "Running the OOS handler (%d open sockets, %s)",
5513 n_socks
, (failed
!= 0) ? "exhaustion seen" : "no exhaustion");
5516 * Check if we're really handling an OOS condition, and if so decide how
5517 * many sockets we want to get down to. Be sure we check if the threshold
5518 * is distinct from zero first; it's possible for this to be called a few
5519 * times before we've finished reading the config.
5521 if (n_socks
>= get_options()->ConnLimit_high_thresh
&&
5522 get_options()->ConnLimit_high_thresh
!= 0 &&
5523 get_options()->ConnLimit_
!= 0) {
5524 /* Try to get down to the low threshold */
5525 target_n_socks
= get_options()->ConnLimit_low_thresh
;
5527 "Current number of sockets %d is greater than configured "
5528 "limit %d; OOS handler trying to get down to %d",
5529 n_socks
, get_options()->ConnLimit_high_thresh
,
5531 } else if (failed
) {
5533 * If we're not at the limit but we hit a socket exhaustion error, try to
5534 * drop some (but not as aggressively as ConnLimit_low_threshold, which is
5535 * 3/4 of ConnLimit_)
5537 target_n_socks
= (n_socks
* 9) / 10;
5539 "We saw socket exhaustion at %d open sockets; OOS handler "
5540 "trying to get down to %d",
5541 n_socks
, target_n_socks
);
5544 if (target_n_socks
> 0) {
5548 * Count moribund sockets; it's be important that anything we decide
5549 * to get rid of here but don't immediately close get counted as moribund
5550 * on subsequent invocations so we don't try to kill too many things if
5551 * connection_check_oos() gets called multiple times.
5553 moribund_socks
= connection_count_moribund();
5555 if (moribund_socks
< n_socks
- target_n_socks
) {
5556 socks_to_kill
= n_socks
- target_n_socks
- moribund_socks
;
5558 conns
= pick_oos_victims(socks_to_kill
);
5560 kill_conn_list_for_oos(conns
);
5562 "OOS handler killed %d conns", smartlist_len(conns
));
5563 smartlist_free(conns
);
5565 log_notice(LD_NET
, "OOS handler failed to pick any victim conns");
5569 "Not killing any sockets for OOS because there are %d "
5570 "already moribund, and we only want to eliminate %d",
5571 moribund_socks
, n_socks
- target_n_socks
);
5576 /** Log how many bytes are used by buffers of different kinds and sizes. */
5578 connection_dump_buffer_mem_stats(int severity
)
5580 uint64_t used_by_type
[CONN_TYPE_MAX_
+1];
5581 uint64_t alloc_by_type
[CONN_TYPE_MAX_
+1];
5582 int n_conns_by_type
[CONN_TYPE_MAX_
+1];
5583 uint64_t total_alloc
= 0;
5584 uint64_t total_used
= 0;
5586 smartlist_t
*conns
= get_connection_array();
5588 memset(used_by_type
, 0, sizeof(used_by_type
));
5589 memset(alloc_by_type
, 0, sizeof(alloc_by_type
));
5590 memset(n_conns_by_type
, 0, sizeof(n_conns_by_type
));
5592 SMARTLIST_FOREACH_BEGIN(conns
, connection_t
*, c
) {
5594 ++n_conns_by_type
[tp
];
5596 used_by_type
[tp
] += buf_datalen(c
->inbuf
);
5597 alloc_by_type
[tp
] += buf_allocation(c
->inbuf
);
5600 used_by_type
[tp
] += buf_datalen(c
->outbuf
);
5601 alloc_by_type
[tp
] += buf_allocation(c
->outbuf
);
5603 } SMARTLIST_FOREACH_END(c
);
5604 for (i
=0; i
<= CONN_TYPE_MAX_
; ++i
) {
5605 total_used
+= used_by_type
[i
];
5606 total_alloc
+= alloc_by_type
[i
];
5609 tor_log(severity
, LD_GENERAL
,
5610 "In buffers for %d connections: %"PRIu64
" used/%"PRIu64
" allocated",
5611 smartlist_len(conns
),
5612 (total_used
), (total_alloc
));
5613 for (i
=CONN_TYPE_MIN_
; i
<= CONN_TYPE_MAX_
; ++i
) {
5614 if (!n_conns_by_type
[i
])
5616 tor_log(severity
, LD_GENERAL
,
5617 " For %d %s connections: %"PRIu64
" used/%"PRIu64
" allocated",
5618 n_conns_by_type
[i
], conn_type_to_string(i
),
5619 (used_by_type
[i
]), (alloc_by_type
[i
]));
5623 /** Verify that connection <b>conn</b> has all of its invariants
5624 * correct. Trigger an assert if anything is invalid.
5627 assert_connection_ok(connection_t
*conn
, time_t now
)
5629 (void) now
; /* XXXX unused. */
5631 tor_assert(conn
->type
>= CONN_TYPE_MIN_
);
5632 tor_assert(conn
->type
<= CONN_TYPE_MAX_
);
5634 switch (conn
->type
) {
5636 case CONN_TYPE_EXT_OR
:
5637 tor_assert(conn
->magic
== OR_CONNECTION_MAGIC
);
5640 tor_assert(conn
->magic
== ENTRY_CONNECTION_MAGIC
);
5642 case CONN_TYPE_EXIT
:
5643 tor_assert(conn
->magic
== EDGE_CONNECTION_MAGIC
);
5646 tor_assert(conn
->magic
== DIR_CONNECTION_MAGIC
);
5648 case CONN_TYPE_CONTROL
:
5649 tor_assert(conn
->magic
== CONTROL_CONNECTION_MAGIC
);
5651 CASE_ANY_LISTENER_TYPE
:
5652 tor_assert(conn
->magic
== LISTENER_CONNECTION_MAGIC
);
5655 tor_assert(conn
->magic
== BASE_CONNECTION_MAGIC
);
5659 if (conn
->linked_conn
) {
5660 tor_assert(conn
->linked_conn
->linked_conn
== conn
);
5661 tor_assert(conn
->linked
);
5664 tor_assert(!SOCKET_OK(conn
->s
));
5666 if (conn
->hold_open_until_flushed
)
5667 tor_assert(conn
->marked_for_close
);
5669 /* XXXX check: read_blocked_on_bw, write_blocked_on_bw, s, conn_array_index,
5670 * marked_for_close. */
5674 buf_assert_ok(conn
->inbuf
);
5676 buf_assert_ok(conn
->outbuf
);
5678 if (conn
->type
== CONN_TYPE_OR
) {
5679 or_connection_t
*or_conn
= TO_OR_CONN(conn
);
5680 if (conn
->state
== OR_CONN_STATE_OPEN
) {
5681 /* tor_assert(conn->bandwidth > 0); */
5682 /* the above isn't necessarily true: if we just did a TLS
5683 * handshake but we didn't recognize the other peer, or it
5684 * gave a bad cert/etc, then we won't have assigned bandwidth,
5685 * yet it will be open. -RD
5687 // tor_assert(conn->read_bucket >= 0);
5689 // tor_assert(conn->addr && conn->port);
5690 tor_assert(conn
->address
);
5691 if (conn
->state
> OR_CONN_STATE_PROXY_HANDSHAKING
)
5692 tor_assert(or_conn
->tls
);
5695 if (CONN_IS_EDGE(conn
)) {
5696 /* XXX unchecked: package window, deliver window. */
5697 if (conn
->type
== CONN_TYPE_AP
) {
5698 entry_connection_t
*entry_conn
= TO_ENTRY_CONN(conn
);
5699 if (entry_conn
->chosen_exit_optional
|| entry_conn
->chosen_exit_retries
)
5700 tor_assert(entry_conn
->chosen_exit_name
);
5702 tor_assert(entry_conn
->socks_request
);
5703 if (conn
->state
== AP_CONN_STATE_OPEN
) {
5704 tor_assert(entry_conn
->socks_request
->has_finished
);
5705 if (!conn
->marked_for_close
) {
5706 tor_assert(ENTRY_TO_EDGE_CONN(entry_conn
)->cpath_layer
);
5707 cpath_assert_layer_ok(ENTRY_TO_EDGE_CONN(entry_conn
)->cpath_layer
);
5711 if (conn
->type
== CONN_TYPE_EXIT
) {
5712 tor_assert(conn
->purpose
== EXIT_PURPOSE_CONNECT
||
5713 conn
->purpose
== EXIT_PURPOSE_RESOLVE
);
5715 } else if (conn
->type
== CONN_TYPE_DIR
) {
5717 /* Purpose is only used for dir and exit types currently */
5718 tor_assert(!conn
->purpose
);
5723 CASE_ANY_LISTENER_TYPE
:
5724 tor_assert(conn
->state
== LISTENER_STATE_READY
);
5727 tor_assert(conn
->state
>= OR_CONN_STATE_MIN_
);
5728 tor_assert(conn
->state
<= OR_CONN_STATE_MAX_
);
5730 case CONN_TYPE_EXT_OR
:
5731 tor_assert(conn
->state
>= EXT_OR_CONN_STATE_MIN_
);
5732 tor_assert(conn
->state
<= EXT_OR_CONN_STATE_MAX_
);
5734 case CONN_TYPE_EXIT
:
5735 tor_assert(conn
->state
>= EXIT_CONN_STATE_MIN_
);
5736 tor_assert(conn
->state
<= EXIT_CONN_STATE_MAX_
);
5737 tor_assert(conn
->purpose
>= EXIT_PURPOSE_MIN_
);
5738 tor_assert(conn
->purpose
<= EXIT_PURPOSE_MAX_
);
5741 tor_assert(conn
->state
>= AP_CONN_STATE_MIN_
);
5742 tor_assert(conn
->state
<= AP_CONN_STATE_MAX_
);
5743 tor_assert(TO_ENTRY_CONN(conn
)->socks_request
);
5746 tor_assert(conn
->state
>= DIR_CONN_STATE_MIN_
);
5747 tor_assert(conn
->state
<= DIR_CONN_STATE_MAX_
);
5748 tor_assert(conn
->purpose
>= DIR_PURPOSE_MIN_
);
5749 tor_assert(conn
->purpose
<= DIR_PURPOSE_MAX_
);
5751 case CONN_TYPE_CONTROL
:
5752 tor_assert(conn
->state
>= CONTROL_CONN_STATE_MIN_
);
5753 tor_assert(conn
->state
<= CONTROL_CONN_STATE_MAX_
);
5755 case CONN_TYPE_METRICS
:
5763 /** Fills <b>addr</b> and <b>port</b> with the details of the global
5764 * proxy server we are using. Store a 1 to the int pointed to by
5765 * <b>is_put_out</b> if the connection is using a pluggable
5766 * transport; store 0 otherwise. <b>conn</b> contains the connection
5767 * we are using the proxy for.
5769 * Return 0 on success, -1 on failure.
5772 get_proxy_addrport(tor_addr_t
*addr
, uint16_t *port
, int *proxy_type
,
5773 int *is_pt_out
, const connection_t
*conn
)
5775 const or_options_t
*options
= get_options();
5778 /* Client Transport Plugins can use another proxy, but that should be hidden
5779 * from the rest of tor (as the plugin is responsible for dealing with the
5780 * proxy), check it first, then check the rest of the proxy types to allow
5781 * the config to have unused ClientTransportPlugin entries.
5783 if (options
->ClientTransportPlugin
) {
5784 const transport_t
*transport
= NULL
;
5786 r
= get_transport_by_bridge_addrport(&conn
->addr
, conn
->port
, &transport
);
5789 if (transport
) { /* transport found */
5790 tor_addr_copy(addr
, &transport
->addr
);
5791 *port
= transport
->port
;
5792 *proxy_type
= transport
->socks_version
;
5797 /* Unused ClientTransportPlugin. */
5800 if (options
->HTTPSProxy
) {
5801 tor_addr_copy(addr
, &options
->HTTPSProxyAddr
);
5802 *port
= options
->HTTPSProxyPort
;
5803 *proxy_type
= PROXY_CONNECT
;
5805 } else if (options
->Socks4Proxy
) {
5806 tor_addr_copy(addr
, &options
->Socks4ProxyAddr
);
5807 *port
= options
->Socks4ProxyPort
;
5808 *proxy_type
= PROXY_SOCKS4
;
5810 } else if (options
->Socks5Proxy
) {
5811 tor_addr_copy(addr
, &options
->Socks5ProxyAddr
);
5812 *port
= options
->Socks5ProxyPort
;
5813 *proxy_type
= PROXY_SOCKS5
;
5815 } else if (options
->TCPProxy
) {
5816 tor_addr_copy(addr
, &options
->TCPProxyAddr
);
5817 *port
= options
->TCPProxyPort
;
5818 /* The only supported protocol in TCPProxy is haproxy. */
5819 tor_assert(options
->TCPProxyProtocol
== TCP_PROXY_PROTOCOL_HAPROXY
);
5820 *proxy_type
= PROXY_HAPROXY
;
5824 tor_addr_make_unspec(addr
);
5826 *proxy_type
= PROXY_NONE
;
5830 /** Log a failed connection to a proxy server.
5831 * <b>conn</b> is the connection we use the proxy server for. */
5833 log_failed_proxy_connection(connection_t
*conn
)
5835 tor_addr_t proxy_addr
;
5836 uint16_t proxy_port
;
5837 int proxy_type
, is_pt
;
5839 if (get_proxy_addrport(&proxy_addr
, &proxy_port
, &proxy_type
, &is_pt
,
5841 return; /* if we have no proxy set up, leave this function. */
5845 "The connection to the %s proxy server at %s just failed. "
5846 "Make sure that the proxy server is up and running.",
5847 proxy_type_to_string(proxy_type
),
5848 fmt_addrport(&proxy_addr
, proxy_port
));
5851 /** Return string representation of <b>proxy_type</b>. */
5853 proxy_type_to_string(int proxy_type
)
5855 switch (proxy_type
) {
5856 case PROXY_CONNECT
: return "HTTP";
5857 case PROXY_SOCKS4
: return "SOCKS4";
5858 case PROXY_SOCKS5
: return "SOCKS5";
5859 case PROXY_HAPROXY
: return "HAPROXY";
5860 case PROXY_PLUGGABLE
: return "pluggable transports SOCKS";
5861 case PROXY_NONE
: return "NULL";
5862 default: tor_assert(0);
5864 return NULL
; /*Unreached*/
5867 /** Call connection_free_minimal() on every connection in our array, and
5868 * release all storage held by connection.c.
5870 * Don't do the checks in connection_free(), because they will
5874 connection_free_all(void)
5876 smartlist_t
*conns
= get_connection_array();
5878 /* We don't want to log any messages to controllers. */
5879 SMARTLIST_FOREACH(conns
, connection_t
*, conn
,
5880 if (conn
->type
== CONN_TYPE_CONTROL
)
5881 TO_CONTROL_CONN(conn
)->event_mask
= 0);
5883 control_update_global_event_mask();
5885 /* Unlink everything from the identity map. */
5886 connection_or_clear_identity_map();
5888 /* Clear out our list of broken connections */
5889 clear_broken_connection_map(0);
5891 SMARTLIST_FOREACH(conns
, connection_t
*, conn
,
5892 connection_free_minimal(conn
));
5894 if (outgoing_addrs
) {
5895 SMARTLIST_FOREACH(outgoing_addrs
, tor_addr_t
*, addr
, tor_free(addr
));
5896 smartlist_free(outgoing_addrs
);
5897 outgoing_addrs
= NULL
;
5900 tor_free(last_interface_ipv4
);
5901 tor_free(last_interface_ipv6
);
5902 last_recorded_accounting_at
= 0;
5904 mainloop_event_free(reenable_blocked_connections_ev
);
5905 reenable_blocked_connections_is_scheduled
= 0;
5906 memset(&reenable_blocked_connections_delay
, 0, sizeof(struct timeval
));
5909 /** Log a warning, and possibly emit a control event, that <b>received</b> came
5910 * at a skewed time. <b>trusted</b> indicates that the <b>source</b> was one
5911 * that we had more faith in and therefore the warning level should have higher
5915 clock_skew_warning
, (const connection_t
*conn
, long apparent_skew
, int trusted
,
5916 log_domain_mask_t domain
, const char *received
,
5917 const char *source
))
5920 char *ext_source
= NULL
, *warn
= NULL
;
5921 format_time_interval(dbuf
, sizeof(dbuf
), apparent_skew
);
5923 tor_asprintf(&ext_source
, "%s:%s:%d", source
,
5924 fmt_and_decorate_addr(&conn
->addr
), conn
->port
);
5926 ext_source
= tor_strdup(source
);
5927 log_fn(trusted
? LOG_WARN
: LOG_INFO
, domain
,
5928 "Received %s with skewed time (%s): "
5929 "It seems that our clock is %s by %s, or that theirs is %s%s. "
5930 "Tor requires an accurate clock to work: please check your time, "
5931 "timezone, and date settings.", received
, ext_source
,
5932 apparent_skew
> 0 ? "ahead" : "behind", dbuf
,
5933 apparent_skew
> 0 ? "behind" : "ahead",
5934 (!conn
|| trusted
) ? "" : ", or they are sending us the wrong time");
5936 control_event_general_status(LOG_WARN
, "CLOCK_SKEW SKEW=%ld SOURCE=%s",
5937 apparent_skew
, ext_source
);
5938 tor_asprintf(&warn
, "Clock skew %ld in %s from %s", apparent_skew
,
5940 control_event_bootstrap_problem(warn
, "CLOCK_SKEW", conn
, 1);
5943 tor_free(ext_source
);