tor/src/or/circuituse.c

1944 lines
72 KiB
C
Raw Normal View History

2006-02-09 06:46:49 +01:00
/* Copyright (c) 2001 Matej Pfajfar.
* Copyright (c) 2001-2004, Roger Dingledine.
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
2011-01-03 17:50:39 +01:00
* Copyright (c) 2007-2011, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file circuituse.c
* \brief Launch the right sort of circuits and attach streams to them.
**/
#include "or.h"
2010-07-22 01:21:00 +02:00
#include "circuitbuild.h"
2010-07-22 09:46:23 +02:00
#include "circuitlist.h"
2010-07-22 10:03:40 +02:00
#include "circuituse.h"
2010-07-22 10:22:51 +02:00
#include "config.h"
2010-07-22 10:32:52 +02:00
#include "connection.h"
2010-07-22 10:43:02 +02:00
#include "connection_edge.h"
2010-07-22 11:35:09 +02:00
#include "control.h"
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
#include "nodelist.h"
#include "networkstatus.h"
2010-07-23 20:51:25 +02:00
#include "policies.h"
2010-07-22 00:13:51 +02:00
#include "rendclient.h"
2010-07-21 17:52:54 +02:00
#include "rendcommon.h"
2010-07-22 00:30:17 +02:00
#include "rendservice.h"
2010-07-23 22:57:20 +02:00
#include "rephist.h"
2010-07-21 16:17:10 +02:00
#include "router.h"
2010-07-21 17:08:11 +02:00
#include "routerlist.h"
/********* START VARIABLES **********/
extern circuit_t *global_circuitlist; /* from circuitlist.c */
/********* END VARIABLES ************/
static void circuit_expire_old_circuits_clientside(void);
static void circuit_increment_failure_count(void);
/** Return 1 if <b>circ</b> could be returned by circuit_get_best().
* Else return 0.
*/
static int
2011-07-06 22:30:02 +02:00
circuit_is_acceptable(const origin_circuit_t *origin_circ,
const entry_connection_t *conn,
int must_be_open, uint8_t purpose,
int need_uptime, int need_internal,
time_t now)
{
2011-07-06 22:30:02 +02:00
const circuit_t *circ = TO_CIRCUIT(origin_circ);
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
const node_t *exitnode;
cpath_build_state_t *build_state;
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
tor_assert(circ);
tor_assert(conn);
tor_assert(conn->socks_request);
if (must_be_open && (circ->state != CIRCUIT_STATE_OPEN || !circ->n_conn))
return 0; /* ignore non-open circs */
if (circ->marked_for_close)
return 0;
/* if this circ isn't our purpose, skip. */
if (purpose == CIRCUIT_PURPOSE_C_REND_JOINED && !must_be_open) {
if (circ->purpose != CIRCUIT_PURPOSE_C_ESTABLISH_REND &&
circ->purpose != CIRCUIT_PURPOSE_C_REND_READY &&
circ->purpose != CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED &&
circ->purpose != CIRCUIT_PURPOSE_C_REND_JOINED)
return 0;
} else if (purpose == CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT &&
!must_be_open) {
if (circ->purpose != CIRCUIT_PURPOSE_C_INTRODUCING &&
circ->purpose != CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT)
return 0;
} else {
if (purpose != circ->purpose)
return 0;
}
if (purpose == CIRCUIT_PURPOSE_C_GENERAL ||
purpose == CIRCUIT_PURPOSE_C_REND_JOINED)
if (circ->timestamp_dirty &&
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
circ->timestamp_dirty+get_options()->MaxCircuitDirtiness <= now)
return 0;
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/* decide if this circ is suitable for this conn */
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/* for rend circs, circ->cpath->prev is not the last router in the
* circuit, it's the magical extra bob hop. so just check the nickname
* of the one we meant to finish at.
*/
2011-07-06 22:30:02 +02:00
build_state = origin_circ->build_state;
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
exitnode = build_state_get_exit_node(build_state);
if (need_uptime && !build_state->need_uptime)
return 0;
if (need_internal != build_state->is_internal)
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
return 0;
if (purpose == CIRCUIT_PURPOSE_C_GENERAL) {
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
if (!exitnode && !build_state->onehop_tunnel) {
log_debug(LD_CIRC,"Not considering circuit with unknown router.");
return 0; /* this circuit is screwed and doesn't know it yet,
* or is a rendezvous circuit. */
}
if (build_state->onehop_tunnel) {
if (!conn->want_onehop) {
log_debug(LD_CIRC,"Skipping one-hop circuit.");
return 0;
}
tor_assert(conn->chosen_exit_name);
if (build_state->chosen_exit) {
char digest[DIGEST_LEN];
if (hexdigest_to_digest(conn->chosen_exit_name, digest) < 0)
return 0; /* broken digest, we don't want it */
if (tor_memneq(digest, build_state->chosen_exit->identity_digest,
DIGEST_LEN))
return 0; /* this is a circuit to somewhere else */
if (tor_digest_is_zero(digest)) {
/* we don't know the digest; have to compare addr:port */
tor_addr_t addr;
int r = tor_addr_parse(&addr, conn->socks_request->address);
if (r < 0 ||
!tor_addr_eq(&build_state->chosen_exit->addr, &addr) ||
build_state->chosen_exit->port != conn->socks_request->port)
return 0;
}
}
} else {
if (conn->want_onehop) {
/* don't use three-hop circuits -- that could hurt our anonymity. */
return 0;
}
}
if (exitnode && !connection_ap_can_use_exit(conn, exitnode)) {
/* can't exit from this router */
return 0;
}
} else { /* not general */
const edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(conn);
if ((edge_conn->rend_data && !origin_circ->rend_data) ||
(!edge_conn->rend_data && origin_circ->rend_data) ||
(edge_conn->rend_data && origin_circ->rend_data &&
rend_cmp_service_ids(edge_conn->rend_data->onion_address,
2011-07-06 22:30:02 +02:00
origin_circ->rend_data->onion_address))) {
/* this circ is not for this conn */
return 0;
}
}
if (!connection_edge_compatible_with_circuit(conn, origin_circ)) {
/* conn needs to be isolated from other conns that have already used
* origin_circ */
return 0;
}
return 1;
}
/** Return 1 if circuit <b>a</b> is better than circuit <b>b</b> for
* <b>conn</b>, and return 0 otherwise. Used by circuit_get_best.
*/
static int
2011-07-06 22:30:02 +02:00
circuit_is_better(const origin_circuit_t *oa, const origin_circuit_t *ob,
const entry_connection_t *conn)
{
2011-07-06 22:30:02 +02:00
const circuit_t *a = TO_CIRCUIT(oa);
const circuit_t *b = TO_CIRCUIT(ob);
const uint8_t purpose = ENTRY_TO_CONN(conn)->purpose;
int a_bits, b_bits;
2011-07-06 22:30:02 +02:00
switch (purpose) {
case CIRCUIT_PURPOSE_C_GENERAL:
/* if it's used but less dirty it's best;
* else if it's more recently created it's best
*/
if (b->timestamp_dirty) {
if (a->timestamp_dirty &&
a->timestamp_dirty > b->timestamp_dirty)
return 1;
} else {
if (a->timestamp_dirty ||
timercmp(&a->timestamp_created, &b->timestamp_created, >))
return 1;
2011-07-06 22:30:02 +02:00
if (ob->build_state->is_internal)
/* XXX023 what the heck is this internal thing doing here. I
* think we can get rid of it. circuit_is_acceptable() already
* makes sure that is_internal is exactly what we need it to
* be. -RD */
return 1;
}
break;
case CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT:
/* the closer it is to ack_wait the better it is */
if (a->purpose > b->purpose)
return 1;
break;
case CIRCUIT_PURPOSE_C_REND_JOINED:
/* the closer it is to rend_joined the better it is */
if (a->purpose > b->purpose)
return 1;
break;
}
/* XXXX023 Maybe this check should get a higher priority to avoid
* using up circuits too rapidly. */
a_bits = connection_edge_update_circuit_isolation(conn,
(origin_circuit_t*)oa, 1);
b_bits = connection_edge_update_circuit_isolation(conn,
(origin_circuit_t*)ob, 1);
/* if x_bits < 0, then we have not used x for anything; better not to dirty
* a connection if we can help it. */
if (a_bits < 0) {
return 0;
} else if (b_bits < 0) {
return 1;
}
a_bits &= ~ oa->isolation_flags_mixed;
a_bits &= ~ ob->isolation_flags_mixed;
if (n_bits_set_u8(a_bits) < n_bits_set_u8(b_bits)) {
/* The fewer new restrictions we need to make on a circuit for stream
* isolation, the better. */
return 1;
}
return 0;
}
/** Find the best circ that conn can use, preferably one which is
* dirty. Circ must not be too old.
*
* Conn must be defined.
*
* If must_be_open, ignore circs not in CIRCUIT_STATE_OPEN.
*
* circ_purpose specifies what sort of circuit we must have.
* It can be C_GENERAL, C_INTRODUCE_ACK_WAIT, or C_REND_JOINED.
*
* If it's REND_JOINED and must_be_open==0, then return the closest
* rendezvous-purposed circuit that you can find.
*
* If it's INTRODUCE_ACK_WAIT and must_be_open==0, then return the
* closest introduce-purposed circuit that you can find.
*/
static origin_circuit_t *
circuit_get_best(const entry_connection_t *conn,
2011-07-06 22:30:02 +02:00
int must_be_open, uint8_t purpose,
int need_uptime, int need_internal)
{
2011-07-06 22:30:02 +02:00
circuit_t *circ;
origin_circuit_t *best=NULL;
struct timeval now;
int intro_going_on_but_too_old = 0;
tor_assert(conn);
tor_assert(purpose == CIRCUIT_PURPOSE_C_GENERAL ||
purpose == CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT ||
purpose == CIRCUIT_PURPOSE_C_REND_JOINED);
tor_gettimeofday(&now);
for (circ=global_circuitlist;circ;circ = circ->next) {
2011-07-06 22:30:02 +02:00
origin_circuit_t *origin_circ;
if (!CIRCUIT_IS_ORIGIN(circ))
continue;
origin_circ = TO_ORIGIN_CIRCUIT(circ);
if (!circuit_is_acceptable(origin_circ,conn,must_be_open,purpose,
need_uptime,need_internal,now.tv_sec))
continue;
if (purpose == CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT &&
!must_be_open && circ->state != CIRCUIT_STATE_OPEN &&
tv_mdiff(&now, &circ->timestamp_created) > circ_times.timeout_ms) {
intro_going_on_but_too_old = 1;
continue;
}
/* now this is an acceptable circ to hand back. but that doesn't
* mean it's the *best* circ to hand back. try to decide.
*/
if (!best || circuit_is_better(origin_circ,best,conn))
2011-07-06 22:30:02 +02:00
best = origin_circ;
}
if (!best && intro_going_on_but_too_old)
log_info(LD_REND|LD_CIRC, "There is an intro circuit being created "
"right now, but it has already taken quite a while. Starting "
"one in parallel.");
2011-07-06 22:30:02 +02:00
return best;
}
/** Return the number of not-yet-open general-purpose origin circuits. */
static int
count_pending_general_client_circuits(void)
{
const circuit_t *circ;
int count = 0;
for (circ = global_circuitlist; circ; circ = circ->next) {
if (circ->marked_for_close ||
circ->state == CIRCUIT_STATE_OPEN ||
circ->purpose != CIRCUIT_PURPOSE_C_GENERAL ||
!CIRCUIT_IS_ORIGIN(circ))
continue;
++count;
}
return count;
}
#if 0
/** Check whether, according to the policies in <b>options</b>, the
* circuit <b>circ</b> makes sense. */
/* XXXX currently only checks Exclude{Exit}Nodes; it should check more.
* Also, it doesn't have the right definition of an exit circuit. Also,
* it's never called. */
int
circuit_conforms_to_options(const origin_circuit_t *circ,
const or_options_t *options)
{
const crypt_path_t *cpath, *cpath_next = NULL;
/* first check if it includes any excluded nodes */
for (cpath = circ->cpath; cpath_next != circ->cpath; cpath = cpath_next) {
cpath_next = cpath->next;
if (routerset_contains_extendinfo(options->ExcludeNodes,
cpath->extend_info))
return 0;
}
/* then consider the final hop */
if (routerset_contains_extendinfo(options->ExcludeExitNodes,
circ->cpath->prev->extend_info))
return 0;
return 1;
}
#endif
/** Close all circuits that start at us, aren't open, and were born
* at least CircuitBuildTimeout seconds ago.
*/
void
circuit_expire_building(void)
{
circuit_t *victim, *next_circ = global_circuitlist;
/* circ_times.timeout_ms and circ_times.close_ms are from
* circuit_build_times_get_initial_timeout() if we haven't computed
* custom timeouts yet */
struct timeval general_cutoff, begindir_cutoff, fourhop_cutoff,
cannibalize_cutoff, close_cutoff, extremely_old_cutoff;
struct timeval now;
cpath_build_state_t *build_state;
tor_gettimeofday(&now);
#define SET_CUTOFF(target, msec) do { \
long ms = tor_lround(msec); \
struct timeval diff; \
diff.tv_sec = ms / 1000; \
2011-04-07 05:30:49 +02:00
diff.tv_usec = (int)((ms % 1000) * 1000); \
timersub(&now, &diff, &target); \
} while (0)
SET_CUTOFF(general_cutoff, circ_times.timeout_ms);
SET_CUTOFF(begindir_cutoff, circ_times.timeout_ms / 2.0);
SET_CUTOFF(fourhop_cutoff, circ_times.timeout_ms * (4/3.0));
SET_CUTOFF(cannibalize_cutoff, circ_times.timeout_ms / 2.0);
SET_CUTOFF(close_cutoff, circ_times.close_ms);
SET_CUTOFF(extremely_old_cutoff, circ_times.close_ms*2 + 1000);
while (next_circ) {
struct timeval cutoff;
victim = next_circ;
next_circ = next_circ->next;
if (!CIRCUIT_IS_ORIGIN(victim) || /* didn't originate here */
victim->marked_for_close) /* don't mess with marked circs */
continue;
build_state = TO_ORIGIN_CIRCUIT(victim)->build_state;
if (build_state && build_state->onehop_tunnel)
cutoff = begindir_cutoff;
else if (build_state && build_state->desired_path_len == 4
&& !TO_ORIGIN_CIRCUIT(victim)->has_opened)
cutoff = fourhop_cutoff;
else if (TO_ORIGIN_CIRCUIT(victim)->has_opened)
cutoff = cannibalize_cutoff;
else if (victim->purpose == CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT)
cutoff = close_cutoff;
else
cutoff = general_cutoff;
if (timercmp(&victim->timestamp_created, &cutoff, >))
continue; /* it's still young, leave it alone */
#if 0
/* some debug logs, to help track bugs */
if (victim->purpose >= CIRCUIT_PURPOSE_C_INTRODUCING &&
victim->purpose <= CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED) {
if (!victim->timestamp_dirty)
2009-05-27 23:55:51 +02:00
log_fn(LOG_DEBUG,"Considering %sopen purpose %d to %s (circid %d)."
"(clean).",
victim->state == CIRCUIT_STATE_OPEN ? "" : "non",
2004-07-03 01:40:03 +02:00
victim->purpose, victim->build_state->chosen_exit_name,
victim->n_circ_id);
else
2009-05-27 23:55:51 +02:00
log_fn(LOG_DEBUG,"Considering %sopen purpose %d to %s (circid %d). "
"%d secs since dirty.",
victim->state == CIRCUIT_STATE_OPEN ? "" : "non",
2004-07-03 01:40:03 +02:00
victim->purpose, victim->build_state->chosen_exit_name,
victim->n_circ_id,
(int)(now - victim->timestamp_dirty));
}
#endif
/* if circ is !open, or if it's open but purpose is a non-finished
* intro or rend, then mark it for close */
if (victim->state == CIRCUIT_STATE_OPEN) {
switch (victim->purpose) {
default: /* most open circuits can be left alone. */
continue; /* yes, continue inside a switch refers to the nearest
* enclosing loop. C is smart. */
case CIRCUIT_PURPOSE_C_ESTABLISH_REND:
case CIRCUIT_PURPOSE_C_INTRODUCING:
case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO:
break; /* too old, need to die */
case CIRCUIT_PURPOSE_C_REND_READY:
/* it's a rend_ready circ -- has it already picked a query? */
/* c_rend_ready circs measure age since timestamp_dirty,
* because that's set when they switch purposes
*/
if (TO_ORIGIN_CIRCUIT(victim)->rend_data ||
victim->timestamp_dirty > cutoff.tv_sec)
continue;
break;
case CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED:
case CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT:
/* rend and intro circs become dirty each time they
* make an introduction attempt. so timestamp_dirty
* will reflect the time since the last attempt.
*/
if (victim->timestamp_dirty > cutoff.tv_sec)
continue;
break;
}
} else { /* circuit not open, consider recording failure as timeout */
int first_hop_succeeded = TO_ORIGIN_CIRCUIT(victim)->cpath &&
TO_ORIGIN_CIRCUIT(victim)->cpath->state == CPATH_STATE_OPEN;
if (TO_ORIGIN_CIRCUIT(victim)->p_streams != NULL) {
log_warn(LD_BUG, "Circuit %d (purpose %d, %s) has timed out, "
"yet has attached streams!",
TO_ORIGIN_CIRCUIT(victim)->global_identifier,
victim->purpose,
circuit_purpose_to_string(victim->purpose));
tor_fragile_assert();
continue;
}
if (circuit_timeout_want_to_count_circ(TO_ORIGIN_CIRCUIT(victim)) &&
circuit_build_times_enough_to_compute(&circ_times)) {
/* Circuits are allowed to last longer for measurement.
* Switch their purpose and wait. */
if (victim->purpose != CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT) {
control_event_circuit_status(TO_ORIGIN_CIRCUIT(victim),
CIRC_EVENT_FAILED,
END_CIRC_REASON_TIMEOUT);
victim->purpose = CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT;
/* Record this failure to check for too many timeouts
* in a row. This function does not record a time value yet
* (we do that later); it only counts the fact that we did
* have a timeout. */
circuit_build_times_count_timeout(&circ_times,
first_hop_succeeded);
continue;
}
/*
* If the circuit build time is much greater than we would have cut
* it off at, we probably had a suspend event along this codepath,
* and we should discard the value.
*/
if (timercmp(&victim->timestamp_created, &extremely_old_cutoff, <)) {
log_notice(LD_CIRC,
"Extremely large value for circuit build timeout: %lds. "
"Assuming clock jump. Purpose %d (%s)",
(long)(now.tv_sec - victim->timestamp_created.tv_sec),
victim->purpose,
circuit_purpose_to_string(victim->purpose));
} else if (circuit_build_times_count_close(&circ_times,
first_hop_succeeded,
victim->timestamp_created.tv_sec)) {
circuit_build_times_set_timeout(&circ_times);
}
}
}
if (victim->n_conn)
log_info(LD_CIRC,"Abandoning circ %s:%d:%d (state %d:%s, purpose %d)",
victim->n_conn->_base.address, victim->n_conn->_base.port,
victim->n_circ_id,
victim->state, circuit_state_to_string(victim->state),
victim->purpose);
else
log_info(LD_CIRC,"Abandoning circ %d (state %d:%s, purpose %d)",
victim->n_circ_id, victim->state,
circuit_state_to_string(victim->state), victim->purpose);
circuit_log_path(LOG_INFO,LD_CIRC,TO_ORIGIN_CIRCUIT(victim));
if (victim->purpose == CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT)
circuit_mark_for_close(victim, END_CIRC_REASON_MEASUREMENT_EXPIRED);
else
circuit_mark_for_close(victim, END_CIRC_REASON_TIMEOUT);
}
}
/** Remove any elements in <b>needed_ports</b> that are handled by an
* open or in-progress circuit.
*/
void
circuit_remove_handled_ports(smartlist_t *needed_ports)
{
int i;
uint16_t *port;
for (i = 0; i < smartlist_len(needed_ports); ++i) {
port = smartlist_get(needed_ports, i);
tor_assert(*port);
if (circuit_stream_is_being_handled(NULL, *port,
MIN_CIRCUITS_HANDLING_STREAM)) {
// log_debug(LD_CIRC,"Port %d is already being handled; removing.", port);
smartlist_del(needed_ports, i--);
tor_free(port);
} else {
log_debug(LD_CIRC,"Port %d is not handled.", *port);
}
}
}
/** Return 1 if at least <b>min</b> general-purpose non-internal circuits
* will have an acceptable exit node for exit stream <b>conn</b> if it
* is defined, else for "*:port".
* Else return 0.
*/
int
circuit_stream_is_being_handled(entry_connection_t *conn,
uint16_t port, int min)
{
circuit_t *circ;
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
const node_t *exitnode;
int num=0;
time_t now = time(NULL);
int need_uptime = smartlist_string_num_isin(get_options()->LongLivedPorts,
conn ? conn->socks_request->port : port);
for (circ=global_circuitlist;circ;circ = circ->next) {
if (CIRCUIT_IS_ORIGIN(circ) &&
!circ->marked_for_close &&
circ->purpose == CIRCUIT_PURPOSE_C_GENERAL &&
(!circ->timestamp_dirty ||
circ->timestamp_dirty + get_options()->MaxCircuitDirtiness > now)) {
cpath_build_state_t *build_state = TO_ORIGIN_CIRCUIT(circ)->build_state;
if (build_state->is_internal || build_state->onehop_tunnel)
continue;
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
exitnode = build_state_get_exit_node(build_state);
if (exitnode && (!need_uptime || build_state->need_uptime)) {
int ok;
if (conn) {
ok = connection_ap_can_use_exit(conn, exitnode);
} else {
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
addr_policy_result_t r;
r = compare_tor_addr_to_node_policy(NULL, port, exitnode);
ok = r != ADDR_POLICY_REJECTED && r != ADDR_POLICY_PROBABLY_REJECTED;
}
if (ok) {
if (++num >= min)
return 1;
}
}
}
}
return 0;
}
/** Don't keep more than this many unused open circuits around. */
#define MAX_UNUSED_OPEN_CIRCUITS 14
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/** Figure out how many circuits we have open that are clean. Make
* sure it's enough for all the upcoming behaviors we predict we'll have.
* But put an upper bound on the total number of circuits.
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
*/
static void
circuit_predict_and_launch_new(void)
{
circuit_t *circ;
int num=0, num_internal=0, num_uptime_internal=0;
int hidserv_needs_uptime=0, hidserv_needs_capacity=1;
int port_needs_uptime=0, port_needs_capacity=1;
time_t now = time(NULL);
int flags = 0;
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/* First, count how many of each type of circuit we have already. */
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
for (circ=global_circuitlist;circ;circ = circ->next) {
cpath_build_state_t *build_state;
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
if (!CIRCUIT_IS_ORIGIN(circ))
continue;
if (circ->marked_for_close)
continue; /* don't mess with marked circs */
if (circ->timestamp_dirty)
continue; /* only count clean circs */
if (circ->purpose != CIRCUIT_PURPOSE_C_GENERAL)
continue; /* only pay attention to general-purpose circs */
build_state = TO_ORIGIN_CIRCUIT(circ)->build_state;
if (build_state->onehop_tunnel)
continue;
num++;
if (build_state->is_internal)
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
num_internal++;
if (build_state->need_uptime && build_state->is_internal)
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
num_uptime_internal++;
}
/* If that's enough, then stop now. */
if (num >= MAX_UNUSED_OPEN_CIRCUITS)
return; /* we already have many, making more probably will hurt */
/* Second, see if we need any more exit circuits. */
/* check if we know of a port that's been requested recently
* and no circuit is currently available that can handle it. */
if (!circuit_all_predicted_ports_handled(now, &port_needs_uptime,
&port_needs_capacity)) {
if (port_needs_uptime)
flags |= CIRCLAUNCH_NEED_UPTIME;
if (port_needs_capacity)
flags |= CIRCLAUNCH_NEED_CAPACITY;
log_info(LD_CIRC,
"Have %d clean circs (%d internal), need another exit circ.",
num, num_internal);
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
circuit_launch(CIRCUIT_PURPOSE_C_GENERAL, flags);
return;
}
/* Third, see if we need any more hidden service (server) circuits. */
if (num_rend_services() && num_uptime_internal < 3) {
flags = (CIRCLAUNCH_NEED_CAPACITY | CIRCLAUNCH_NEED_UPTIME |
CIRCLAUNCH_IS_INTERNAL);
log_info(LD_CIRC,
"Have %d clean circs (%d internal), need another internal "
"circ for my hidden service.",
num, num_internal);
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
circuit_launch(CIRCUIT_PURPOSE_C_GENERAL, flags);
return;
}
/* Fourth, see if we need any more hidden service (client) circuits. */
if (rep_hist_get_predicted_internal(now, &hidserv_needs_uptime,
&hidserv_needs_capacity) &&
((num_uptime_internal<2 && hidserv_needs_uptime) ||
num_internal<2)) {
if (hidserv_needs_uptime)
flags |= CIRCLAUNCH_NEED_UPTIME;
if (hidserv_needs_capacity)
flags |= CIRCLAUNCH_NEED_CAPACITY;
flags |= CIRCLAUNCH_IS_INTERNAL;
log_info(LD_CIRC,
"Have %d clean circs (%d uptime-internal, %d internal), need"
2009-05-27 23:55:51 +02:00
" another hidden service circ.",
num, num_uptime_internal, num_internal);
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
circuit_launch(CIRCUIT_PURPOSE_C_GENERAL, flags);
return;
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
}
/* Finally, check to see if we still need more circuits to learn
* a good build timeout. But if we're close to our max number we
* want, don't do another -- we want to leave a few slots open so
* we can still build circuits preemptively as needed. */
if (num < MAX_UNUSED_OPEN_CIRCUITS-2 &&
circuit_build_times_needs_circuits_now(&circ_times)) {
flags = CIRCLAUNCH_NEED_CAPACITY;
log_info(LD_CIRC,
"Have %d clean circs need another buildtime test circ.", num);
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
circuit_launch(CIRCUIT_PURPOSE_C_GENERAL, flags);
return;
}
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
}
/** Build a new test circuit every 5 minutes */
#define TESTING_CIRCUIT_INTERVAL 300
/** This function is called once a second, if router_have_min_dir_info() is
* true. Its job is to make sure all services we offer have enough circuits
* available. Some services just want enough circuits for current tasks,
* whereas others want a minimum set of idle circuits hanging around.
*/
void
circuit_build_needed_circs(time_t now)
{
static time_t time_to_new_circuit = 0;
const or_options_t *options = get_options();
/* launch a new circ for any pending streams that need one */
connection_ap_attach_pending();
/* make sure any hidden services have enough intro points */
rend_services_introduce();
if (time_to_new_circuit < now) {
circuit_reset_failure_count(1);
time_to_new_circuit = now + options->NewCircuitPeriod;
if (proxy_mode(get_options()))
addressmap_clean(now);
circuit_expire_old_circuits_clientside();
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
#if 0 /* disable for now, until predict-and-launch-new can cull leftovers */
circ = circuit_get_youngest_clean_open(CIRCUIT_PURPOSE_C_GENERAL);
if (get_options()->RunTesting &&
circ &&
circ->timestamp_created.tv_sec + TESTING_CIRCUIT_INTERVAL < now) {
log_fn(LOG_INFO,"Creating a new testing circuit.");
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
circuit_launch(CIRCUIT_PURPOSE_C_GENERAL, 0);
}
#endif
}
if (!options->DisablePredictedCircuits)
circuit_predict_and_launch_new();
}
/** If the stream <b>conn</b> is a member of any of the linked
* lists of <b>circ</b>, then remove it from the list.
*/
void
circuit_detach_stream(circuit_t *circ, edge_connection_t *conn)
{
edge_connection_t *prevconn;
tor_assert(circ);
tor_assert(conn);
if (conn->_base.type == CONN_TYPE_AP) {
entry_connection_t *entry_conn = EDGE_TO_ENTRY_CONN(conn);
entry_conn->may_use_optimistic_data = 0;
}
conn->cpath_layer = NULL; /* don't keep a stale pointer */
conn->on_circuit = NULL;
if (CIRCUIT_IS_ORIGIN(circ)) {
origin_circuit_t *origin_circ = TO_ORIGIN_CIRCUIT(circ);
if (conn == origin_circ->p_streams) {
origin_circ->p_streams = conn->next_stream;
return;
}
for (prevconn = origin_circ->p_streams;
prevconn && prevconn->next_stream && prevconn->next_stream != conn;
prevconn = prevconn->next_stream)
;
if (prevconn && prevconn->next_stream) {
prevconn->next_stream = conn->next_stream;
return;
}
} else {
or_circuit_t *or_circ = TO_OR_CIRCUIT(circ);
if (conn == or_circ->n_streams) {
or_circ->n_streams = conn->next_stream;
return;
}
if (conn == or_circ->resolving_streams) {
or_circ->resolving_streams = conn->next_stream;
return;
}
for (prevconn = or_circ->n_streams;
prevconn && prevconn->next_stream && prevconn->next_stream != conn;
prevconn = prevconn->next_stream)
;
if (prevconn && prevconn->next_stream) {
prevconn->next_stream = conn->next_stream;
return;
}
for (prevconn = or_circ->resolving_streams;
prevconn && prevconn->next_stream && prevconn->next_stream != conn;
prevconn = prevconn->next_stream)
;
if (prevconn && prevconn->next_stream) {
prevconn->next_stream = conn->next_stream;
return;
}
}
log_warn(LD_BUG,"Edge connection not in circuit's list.");
/* Don't give an error here; it's harmless. */
tor_fragile_assert();
}
/** If we haven't yet decided on a good timeout value for circuit
* building, we close idles circuits aggressively so we can get more
* data points. */
#define IDLE_TIMEOUT_WHILE_LEARNING (10*60)
/** Find each circuit that has been unused for too long, or dirty
* for too long and has no streams on it: mark it for close.
*/
static void
circuit_expire_old_circuits_clientside(void)
{
circuit_t *circ;
struct timeval cutoff, now;
tor_gettimeofday(&now);
cutoff = now;
if (circuit_build_times_needs_circuits(&circ_times)) {
/* Circuits should be shorter lived if we need more of them
* for learning a good build timeout */
cutoff.tv_sec -= IDLE_TIMEOUT_WHILE_LEARNING;
} else {
cutoff.tv_sec -= get_options()->CircuitIdleTimeout;
}
for (circ = global_circuitlist; circ; circ = circ->next) {
2010-08-10 00:50:49 +02:00
if (circ->marked_for_close || !CIRCUIT_IS_ORIGIN(circ))
continue;
/* If the circuit has been dirty for too long, and there are no streams
* on it, mark it for close.
*/
if (circ->timestamp_dirty &&
circ->timestamp_dirty + get_options()->MaxCircuitDirtiness <
now.tv_sec &&
!TO_ORIGIN_CIRCUIT(circ)->p_streams /* nothing attached */ ) {
log_debug(LD_CIRC, "Closing n_circ_id %d (dirty %ld sec ago, "
"purpose %d)",
circ->n_circ_id, (long)(now.tv_sec - circ->timestamp_dirty),
circ->purpose);
circuit_mark_for_close(circ, END_CIRC_REASON_FINISHED);
} else if (!circ->timestamp_dirty && circ->state == CIRCUIT_STATE_OPEN) {
if (timercmp(&circ->timestamp_created, &cutoff, <)) {
if (circ->purpose == CIRCUIT_PURPOSE_C_GENERAL ||
circ->purpose == CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT ||
circ->purpose == CIRCUIT_PURPOSE_S_ESTABLISH_INTRO ||
circ->purpose == CIRCUIT_PURPOSE_TESTING ||
(circ->purpose >= CIRCUIT_PURPOSE_C_INTRODUCING &&
circ->purpose <= CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED) ||
circ->purpose == CIRCUIT_PURPOSE_S_CONNECT_REND) {
log_debug(LD_CIRC,
"Closing circuit that has been unused for %ld msec.",
tv_mdiff(&circ->timestamp_created, &now));
circuit_mark_for_close(circ, END_CIRC_REASON_FINISHED);
} else if (!TO_ORIGIN_CIRCUIT(circ)->is_ancient) {
2010-08-14 10:22:08 +02:00
/* Server-side rend joined circuits can end up really old, because
* they are reused by clients for longer than normal. The client
2010-08-14 02:50:51 +02:00
* controls their lifespan. (They never become dirty, because
2010-08-14 10:22:08 +02:00
* connection_exit_begin_conn() never marks anything as dirty.)
* Similarly, server-side intro circuits last a long time. */
if (circ->purpose != CIRCUIT_PURPOSE_S_REND_JOINED &&
circ->purpose != CIRCUIT_PURPOSE_S_INTRO) {
log_notice(LD_CIRC,
"Ancient non-dirty circuit %d is still around after "
"%ld milliseconds. Purpose: %d (%s)",
TO_ORIGIN_CIRCUIT(circ)->global_identifier,
tv_mdiff(&circ->timestamp_created, &now),
circ->purpose,
circuit_purpose_to_string(circ->purpose));
TO_ORIGIN_CIRCUIT(circ)->is_ancient = 1;
}
}
}
}
}
}
2010-04-24 00:35:11 +02:00
/** How long do we wait before killing circuits with the properties
* described below?
*
* Probably we could choose a number here as low as 5 to 10 seconds,
* since these circs are used for begindir, and a) generally you either
* ask another begindir question right after or you don't for a long time,
* b) clients at least through 0.2.1.x choose from the whole set of
* directory mirrors at each choice, and c) re-establishing a one-hop
* circuit via create-fast is a light operation assuming the TLS conn is
* still there.
*
* I expect "b" to go away one day when we move to using directory
* guards, but I think "a" and "c" are good enough reasons that a low
* number is safe even then.
*/
#define IDLE_ONE_HOP_CIRC_TIMEOUT 60
/** Find each non-origin circuit that has been unused for too long,
* has no streams on it, used a create_fast, and ends here: mark it
* for close.
*/
void
circuit_expire_old_circuits_serverside(time_t now)
{
circuit_t *circ;
or_circuit_t *or_circ;
time_t cutoff = now - IDLE_ONE_HOP_CIRC_TIMEOUT;
for (circ = global_circuitlist; circ; circ = circ->next) {
if (circ->marked_for_close || CIRCUIT_IS_ORIGIN(circ))
continue;
or_circ = TO_OR_CIRCUIT(circ);
/* If the circuit has been idle for too long, and there are no streams
* on it, and it ends here, and it used a create_fast, mark it for close.
*/
if (or_circ->is_first_hop && !circ->n_conn &&
!or_circ->n_streams && !or_circ->resolving_streams &&
or_circ->p_conn &&
or_circ->p_conn->timestamp_last_added_nonpadding <= cutoff) {
log_info(LD_CIRC, "Closing circ_id %d (empty %d secs ago)",
or_circ->p_circ_id,
(int)(now - or_circ->p_conn->timestamp_last_added_nonpadding));
circuit_mark_for_close(circ, END_CIRC_REASON_FINISHED);
}
}
}
2007-02-24 08:50:38 +01:00
/** Number of testing circuits we want open before testing our bandwidth. */
#define NUM_PARALLEL_TESTING_CIRCS 4
2007-02-24 08:50:38 +01:00
/** True iff we've ever had enough testing circuits open to test our
* bandwidth. */
static int have_performed_bandwidth_test = 0;
/** Reset have_performed_bandwidth_test, so we'll start building
* testing circuits again so we can exercise our bandwidth. */
void
reset_bandwidth_test(void)
{
have_performed_bandwidth_test = 0;
}
/** Return 1 if we've already exercised our bandwidth, or if we
* have fewer than NUM_PARALLEL_TESTING_CIRCS testing circuits
* established or on the way. Else return 0.
*/
int
circuit_enough_testing_circs(void)
{
circuit_t *circ;
int num = 0;
if (have_performed_bandwidth_test)
return 1;
for (circ = global_circuitlist; circ; circ = circ->next) {
if (!circ->marked_for_close && CIRCUIT_IS_ORIGIN(circ) &&
circ->purpose == CIRCUIT_PURPOSE_TESTING &&
circ->state == CIRCUIT_STATE_OPEN)
num++;
}
return num >= NUM_PARALLEL_TESTING_CIRCS;
}
/** A testing circuit has completed. Take whatever stats we want.
* Noticing reachability is taken care of in onionskin_answer(),
* so there's no need to record anything here. But if we still want
* to do the bandwidth test, and we now have enough testing circuits
* open, do it.
*/
static void
circuit_testing_opened(origin_circuit_t *circ)
{
if (have_performed_bandwidth_test ||
!check_whether_orport_reachable()) {
/* either we've already done everything we want with testing circuits,
* or this testing circuit became open due to a fluke, e.g. we picked
* a last hop where we already had the connection open due to an
* outgoing local circuit. */
circuit_mark_for_close(TO_CIRCUIT(circ), END_CIRC_AT_ORIGIN);
} else if (circuit_enough_testing_circs()) {
router_perform_bandwidth_test(NUM_PARALLEL_TESTING_CIRCS, time(NULL));
have_performed_bandwidth_test = 1;
} else
consider_testing_reachability(1, 0);
}
/** A testing circuit has failed to build. Take whatever stats we want. */
static void
circuit_testing_failed(origin_circuit_t *circ, int at_last_hop)
{
if (server_mode(get_options()) && check_whether_orport_reachable())
return;
log_info(LD_GENERAL,
"Our testing circuit (to see if your ORPort is reachable) "
"has failed. I'll try again later.");
/* These aren't used yet. */
(void)circ;
(void)at_last_hop;
}
/** The circuit <b>circ</b> has just become open. Take the next
* step: for rendezvous circuits, we pass circ to the appropriate
* function in rendclient or rendservice. For general circuits, we
* call connection_ap_attach_pending, which looks for pending streams
* that could use circ.
*/
void
circuit_has_opened(origin_circuit_t *circ)
{
int can_try_clearing_isolation = 0, tried_clearing_isolation = 0;
control_event_circuit_status(circ, CIRC_EVENT_BUILT, 0);
/* Remember that this circuit has finished building. Now if we start
* it building again later (e.g. by extending it), we will know not
* to consider its build time. */
circ->has_opened = 1;
again:
switch (TO_CIRCUIT(circ)->purpose) {
case CIRCUIT_PURPOSE_C_ESTABLISH_REND:
rend_client_rendcirc_has_opened(circ);
2011-12-05 17:29:43 +01:00
/* XXXX We'd like to set something like "can_try_clearing_isolation"
* here, so that we can change the isolation of this circuit (and maybe
* its purpose too) if it turns out that we no longer have any streams
* that want to use it. But connection_ap_attach_pending() doesn't
* actually attach streams to a C_ESTABLISH_REND circuit-- streams
* don't get attached until the circuit is in C_REND_JOINED... so
* we can't clear isolation now.
*/
connection_ap_attach_pending();
break;
case CIRCUIT_PURPOSE_C_INTRODUCING:
rend_client_introcirc_has_opened(circ);
break;
case CIRCUIT_PURPOSE_C_GENERAL:
/* Tell any AP connections that have been waiting for a new
* circuit that one is ready. */
can_try_clearing_isolation = 1;
connection_ap_attach_pending();
break;
case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO:
/* at Bob, waiting for introductions */
rend_service_intro_has_opened(circ);
break;
case CIRCUIT_PURPOSE_S_CONNECT_REND:
/* at Bob, connecting to rend point */
rend_service_rendezvous_has_opened(circ);
break;
case CIRCUIT_PURPOSE_TESTING:
circuit_testing_opened(circ);
break;
/* default:
* This won't happen in normal operation, but might happen if the
* controller did it. Just let it slide. */
}
if (/* The circuit may have become non-open if it was cannibalized.*/
circ->_base.state == CIRCUIT_STATE_OPEN &&
/* Only if the purpose is clearable, and only if we haven't tried
* to clear isolation yet, do we try. */
can_try_clearing_isolation && !tried_clearing_isolation &&
/* If !isolation_values_set, there is nothing to clear. */
circ->isolation_values_set &&
/* It's not legal to clear a circuit's isolation info if it's ever had
* streams attached */
!circ->isolation_any_streams_attached) {
/* If we have any isolation information set on this circuit, and
* we didn't manage to attach any streams to it, then we can
* and should clear it and try again. */
circuit_clear_isolation(circ);
tried_clearing_isolation = 1;
goto again;
}
}
/** Called whenever a circuit could not be successfully built.
*/
void
circuit_build_failed(origin_circuit_t *circ)
{
/* we should examine circ and see if it failed because of
* the last hop or an earlier hop. then use this info below.
*/
int failed_at_last_hop = 0;
/* If the last hop isn't open, and the second-to-last is, we failed
* at the last hop. */
if (circ->cpath &&
circ->cpath->prev->state != CPATH_STATE_OPEN &&
circ->cpath->prev->prev->state == CPATH_STATE_OPEN) {
failed_at_last_hop = 1;
}
if (circ->cpath &&
circ->cpath->state != CPATH_STATE_OPEN) {
/* We failed at the first hop. If there's an OR connection
* to blame, blame it. Also, avoid this relay for a while, and
* fail any one-hop directory fetches destined for it. */
const char *n_conn_id = circ->cpath->extend_info->identity_digest;
int already_marked = 0;
if (circ->_base.n_conn) {
or_connection_t *n_conn = circ->_base.n_conn;
if (n_conn->is_bad_for_new_circs) {
/* We only want to blame this router when a fresh healthy
* connection fails. So don't mark this router as newly failed,
* since maybe this was just an old circuit attempt that's
* finally timing out now. Also, there's no need to blow away
* circuits/streams/etc, since the failure of an unhealthy conn
* doesn't tell us much about whether a healthy conn would
* succeed. */
already_marked = 1;
}
log_info(LD_OR,
"Our circuit failed to get a response from the first hop "
"(%s:%d). I'm going to try to rotate to a better connection.",
n_conn->_base.address, n_conn->_base.port);
n_conn->is_bad_for_new_circs = 1;
} else {
log_info(LD_OR,
"Our circuit died before the first hop with no connection");
}
if (n_conn_id && !already_marked) {
entry_guard_register_connect_status(n_conn_id, 0, 1, time(NULL));
/* if there are any one-hop streams waiting on this circuit, fail
* them now so they can retry elsewhere. */
connection_ap_fail_onehop(n_conn_id, circ->build_state);
}
}
switch (circ->_base.purpose) {
case CIRCUIT_PURPOSE_C_GENERAL:
/* If we never built the circuit, note it as a failure. */
circuit_increment_failure_count();
if (failed_at_last_hop) {
/* Make sure any streams that demand our last hop as their exit
* know that it's unlikely to happen. */
circuit_discard_optional_exit_enclaves(circ->cpath->prev->extend_info);
}
break;
case CIRCUIT_PURPOSE_TESTING:
circuit_testing_failed(circ, failed_at_last_hop);
break;
case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO:
/* at Bob, waiting for introductions */
if (circ->_base.state != CIRCUIT_STATE_OPEN) {
circuit_increment_failure_count();
}
/* no need to care here, because bob will rebuild intro
* points periodically. */
break;
case CIRCUIT_PURPOSE_C_INTRODUCING:
/* at Alice, connecting to intro point */
/* Don't increment failure count, since Bob may have picked
* the introduction point maliciously */
/* Alice will pick a new intro point when this one dies, if
* the stream in question still cares. No need to act here. */
break;
case CIRCUIT_PURPOSE_C_ESTABLISH_REND:
/* at Alice, waiting for Bob */
circuit_increment_failure_count();
/* Alice will pick a new rend point when this one dies, if
* the stream in question still cares. No need to act here. */
break;
case CIRCUIT_PURPOSE_S_CONNECT_REND:
/* at Bob, connecting to rend point */
/* Don't increment failure count, since Alice may have picked
* the rendezvous point maliciously */
log_info(LD_REND,
"Couldn't connect to Alice's chosen rend point %s "
"(%s hop failed).",
escaped(build_state_get_exit_nickname(circ->build_state)),
failed_at_last_hop?"last":"non-last");
rend_service_relaunch_rendezvous(circ);
break;
/* default:
* This won't happen in normal operation, but might happen if the
* controller did it. Just let it slide. */
}
}
/** Number of consecutive failures so far; should only be touched by
* circuit_launch_new and circuit_*_failure_count.
*/
static int n_circuit_failures = 0;
/** Before the last time we called circuit_reset_failure_count(), were
* there a lot of failures? */
static int did_circs_fail_last_period = 0;
/** Don't retry launching a new circuit if we try this many times with no
* success. */
#define MAX_CIRCUIT_FAILURES 5
/** Launch a new circuit; see circuit_launch_by_extend_info() for
* details on arguments. */
origin_circuit_t *
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
circuit_launch(uint8_t purpose, int flags)
2004-07-03 01:40:03 +02:00
{
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
return circuit_launch_by_extend_info(purpose, NULL, flags);
}
/** Launch a new circuit with purpose <b>purpose</b> and exit node
* <b>extend_info</b> (or NULL to select a random exit node). If flags
* contains CIRCLAUNCH_NEED_UPTIME, choose among routers with high uptime. If
* CIRCLAUNCH_NEED_CAPACITY is set, choose among routers with high bandwidth.
* If CIRCLAUNCH_IS_INTERNAL is true, the last hop need not be an exit node.
* If CIRCLAUNCH_ONEHOP_TUNNEL is set, the circuit will have only one hop.
* Return the newly allocated circuit on success, or NULL on failure. */
origin_circuit_t *
circuit_launch_by_extend_info(uint8_t purpose,
extend_info_t *extend_info,
int flags)
{
origin_circuit_t *circ;
2008-07-01 01:27:11 +02:00
int onehop_tunnel = (flags & CIRCLAUNCH_ONEHOP_TUNNEL) != 0;
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
if (!onehop_tunnel && !router_have_minimum_dir_info()) {
log_debug(LD_CIRC,"Haven't fetched enough directory info yet; canceling "
"circuit launch.");
return NULL;
}
if ((extend_info || purpose != CIRCUIT_PURPOSE_C_GENERAL) &&
purpose != CIRCUIT_PURPOSE_TESTING && !onehop_tunnel) {
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/* see if there are appropriate circs available to cannibalize. */
/* XXX if we're planning to add a hop, perhaps we want to look for
* internal circs rather than exit circs? -RD */
circ = circuit_find_to_cannibalize(purpose, extend_info, flags);
if (circ) {
log_info(LD_CIRC,"Cannibalizing circ '%s' for purpose %d (%s)",
build_state_get_exit_nickname(circ->build_state), purpose,
circuit_purpose_to_string(purpose));
circ->_base.purpose = purpose;
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/* reset the birth date of this circ, else expire_building
* will see it and think it's been trying to build since it
* began. */
tor_gettimeofday(&circ->_base.timestamp_created);
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
switch (purpose) {
case CIRCUIT_PURPOSE_C_ESTABLISH_REND:
case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO:
/* it's ready right now */
break;
case CIRCUIT_PURPOSE_C_INTRODUCING:
case CIRCUIT_PURPOSE_S_CONNECT_REND:
case CIRCUIT_PURPOSE_C_GENERAL:
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/* need to add a new hop */
tor_assert(extend_info);
if (circuit_extend_to_new_exit(circ, extend_info) < 0)
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
return NULL;
break;
default:
log_warn(LD_BUG,
"unexpected purpose %d when cannibalizing a circ.",
purpose);
tor_fragile_assert();
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
return NULL;
}
return circ;
}
}
if (did_circs_fail_last_period &&
n_circuit_failures > MAX_CIRCUIT_FAILURES) {
/* too many failed circs in a row. don't try. */
// log_fn(LOG_INFO,"%d failures so far, not trying.",n_circuit_failures);
return NULL;
}
/* try a circ. if it fails, circuit_mark_for_close will increment
* n_circuit_failures */
return circuit_establish_circuit(purpose, extend_info, flags);
2004-07-03 01:40:03 +02:00
}
/** Record another failure at opening a general circuit. When we have
* too many, we'll stop trying for the remainder of this minute.
*/
static void
circuit_increment_failure_count(void)
{
++n_circuit_failures;
log_debug(LD_CIRC,"n_circuit_failures now %d.",n_circuit_failures);
}
/** Reset the failure count for opening general circuits. This means
* we will try MAX_CIRCUIT_FAILURES times more (if necessary) before
* stopping again.
*/
void
circuit_reset_failure_count(int timeout)
{
if (timeout && n_circuit_failures > MAX_CIRCUIT_FAILURES)
did_circs_fail_last_period = 1;
else
did_circs_fail_last_period = 0;
n_circuit_failures = 0;
}
/** Find an open circ that we're happy to use for <b>conn</b> and return 1. If
* there isn't one, and there isn't one on the way, launch one and return
* 0. If it will never work, return -1.
*
* Write the found or in-progress or launched circ into *circp.
*/
static int
circuit_get_open_circ_or_launch(entry_connection_t *conn,
uint8_t desired_circuit_purpose,
origin_circuit_t **circp)
{
origin_circuit_t *circ;
int check_exit_policy;
int need_uptime, need_internal;
int want_onehop;
const or_options_t *options = get_options();
tor_assert(conn);
tor_assert(circp);
tor_assert(ENTRY_TO_CONN(conn)->state == AP_CONN_STATE_CIRCUIT_WAIT);
check_exit_policy =
conn->socks_request->command == SOCKS_COMMAND_CONNECT &&
!conn->use_begindir &&
!connection_edge_is_rendezvous_stream(ENTRY_TO_EDGE_CONN(conn));
want_onehop = conn->want_onehop;
need_uptime = !conn->want_onehop && !conn->use_begindir &&
smartlist_string_num_isin(options->LongLivedPorts,
conn->socks_request->port);
need_internal = desired_circuit_purpose != CIRCUIT_PURPOSE_C_GENERAL;
circ = circuit_get_best(conn, 1, desired_circuit_purpose,
need_uptime, need_internal);
if (circ) {
*circp = circ;
return 1; /* we're happy */
}
if (!want_onehop && !router_have_minimum_dir_info()) {
if (!connection_get_by_type(CONN_TYPE_DIR)) {
int severity = LOG_NOTICE;
2009-05-27 23:55:51 +02:00
/* FFFF if this is a tunneled directory fetch, don't yell
* as loudly. the user doesn't even know it's happening. */
if (entry_list_is_constrained(options) &&
entries_known_but_down(options)) {
log_fn(severity, LD_APP|LD_DIR,
"Application request when we haven't used client functionality "
"lately. Optimistically trying known %s again.",
options->UseBridges ? "bridges" : "entrynodes");
entries_retry_all(options);
} else if (!options->UseBridges || any_bridge_descriptors_known()) {
log_fn(severity, LD_APP|LD_DIR,
"Application request when we haven't used client functionality "
"lately. Optimistically trying directory fetches again.");
routerlist_retry_directory_downloads(time(NULL));
}
}
/* the stream will be dealt with when router_have_minimum_dir_info becomes
* 1, or when all directory attempts fail and directory_all_unreachable()
* kills it.
*/
return 0;
}
/* Do we need to check exit policy? */
if (check_exit_policy) {
if (!conn->chosen_exit_name) {
struct in_addr in;
tor_addr_t addr, *addrp=NULL;
if (tor_inet_aton(conn->socks_request->address, &in)) {
tor_addr_from_in(&addr, &in);
addrp = &addr;
}
if (router_exit_policy_all_nodes_reject(addrp,
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
conn->socks_request->port,
need_uptime)) {
log_notice(LD_APP,
2008-11-07 02:48:55 +01:00
"No Tor server allows exit to %s:%d. Rejecting.",
safe_str_client(conn->socks_request->address),
conn->socks_request->port);
return -1;
}
} else {
/* XXXX023 Duplicates checks in connection_ap_handshake_attach_circuit:
* refactor into a single function? */
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
const node_t *node = node_get_by_nickname(conn->chosen_exit_name, 1);
int opt = conn->chosen_exit_optional;
if (node && !connection_ap_can_use_exit(conn, node)) {
log_fn(opt ? LOG_INFO : LOG_WARN, LD_APP,
"Requested exit point '%s' is excluded or "
"would refuse request. %s.",
conn->chosen_exit_name, opt ? "Trying others" : "Closing");
if (opt) {
conn->chosen_exit_optional = 0;
tor_free(conn->chosen_exit_name);
/* Try again. */
return circuit_get_open_circ_or_launch(conn,
desired_circuit_purpose,
circp);
}
return -1;
}
}
}
/* is one already on the way? */
circ = circuit_get_best(conn, 0, desired_circuit_purpose,
need_uptime, need_internal);
if (circ)
log_debug(LD_CIRC, "one on the way!");
if (!circ) {
extend_info_t *extend_info=NULL;
uint8_t new_circ_purpose;
const int n_pending = count_pending_general_client_circuits();
if (n_pending >= options->MaxClientCircuitsPending) {
static ratelim_t delay_limit = RATELIM_INIT(10*60);
char *m;
if ((m = rate_limit_log(&delay_limit, approx_time()))) {
log_notice(LD_APP, "We'd like to launch a circuit to handle a "
"connection, but we already have %d general-purpose client "
"circuits pending. Waiting until some finish.",
n_pending);
tor_free(m);
}
return 0;
}
if (desired_circuit_purpose == CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT) {
/* need to pick an intro point */
rend_data_t *rend_data = ENTRY_TO_EDGE_CONN(conn)->rend_data;
tor_assert(rend_data);
extend_info = rend_client_get_random_intro(rend_data);
if (!extend_info) {
log_info(LD_REND,
2009-05-27 23:55:51 +02:00
"No intro points for '%s': re-fetching service descriptor.",
safe_str_client(rend_data->onion_address));
rend_client_refetch_v2_renddesc(rend_data);
ENTRY_TO_CONN(conn)->state = AP_CONN_STATE_RENDDESC_WAIT;
return 0;
}
log_info(LD_REND,"Chose %s as intro point for '%s'.",
extend_info_describe(extend_info),
safe_str_client(rend_data->onion_address));
}
/* If we have specified a particular exit node for our
* connection, then be sure to open a circuit to that exit node.
*/
2004-11-30 03:26:41 +01:00
if (desired_circuit_purpose == CIRCUIT_PURPOSE_C_GENERAL) {
if (conn->chosen_exit_name) {
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
const node_t *r;
int opt = conn->chosen_exit_optional;
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
r = node_get_by_nickname(conn->chosen_exit_name, 1);
if (r && node_has_descriptor(r)) {
/* We might want to connect to an IPv6 bridge for loading
descriptors so we use the preferred address rather than
the primary. */
extend_info = extend_info_from_node(r, conn->want_onehop ? 1 : 0);
} else {
log_debug(LD_DIR, "considering %d, %s",
want_onehop, conn->chosen_exit_name);
if (want_onehop && conn->chosen_exit_name[0] == '$') {
/* We're asking for a one-hop circuit to a router that
* we don't have a routerinfo about. Make up an extend_info. */
char digest[DIGEST_LEN];
char *hexdigest = conn->chosen_exit_name+1;
tor_addr_t addr;
if (strlen(hexdigest) < HEX_DIGEST_LEN ||
base16_decode(digest,DIGEST_LEN,hexdigest,HEX_DIGEST_LEN)<0) {
log_info(LD_DIR, "Broken exit digest on tunnel conn. Closing.");
return -1;
}
if (tor_addr_parse(&addr, conn->socks_request->address) < 0) {
log_info(LD_DIR, "Broken address %s on tunnel conn. Closing.",
escaped_safe_str_client(conn->socks_request->address));
return -1;
}
extend_info = extend_info_alloc(conn->chosen_exit_name+1,
digest, NULL, &addr,
conn->socks_request->port);
} else {
/* We will need an onion key for the router, and we
* don't have one. Refuse or relax requirements. */
log_fn(opt ? LOG_INFO : LOG_WARN, LD_APP,
"Requested exit point '%s' is not known. %s.",
conn->chosen_exit_name, opt ? "Trying others" : "Closing");
if (opt) {
conn->chosen_exit_optional = 0;
tor_free(conn->chosen_exit_name);
/* Try again with no requested exit */
return circuit_get_open_circ_or_launch(conn,
desired_circuit_purpose,
circp);
}
return -1;
}
}
}
}
if (desired_circuit_purpose == CIRCUIT_PURPOSE_C_REND_JOINED)
new_circ_purpose = CIRCUIT_PURPOSE_C_ESTABLISH_REND;
else if (desired_circuit_purpose == CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT)
new_circ_purpose = CIRCUIT_PURPOSE_C_INTRODUCING;
else
new_circ_purpose = desired_circuit_purpose;
if (options->Tor2webMode &&
(new_circ_purpose == CIRCUIT_PURPOSE_C_ESTABLISH_REND ||
new_circ_purpose == CIRCUIT_PURPOSE_C_INTRODUCING)) {
want_onehop = 1;
}
{
int flags = CIRCLAUNCH_NEED_CAPACITY;
if (want_onehop) flags |= CIRCLAUNCH_ONEHOP_TUNNEL;
if (need_uptime) flags |= CIRCLAUNCH_NEED_UPTIME;
if (need_internal) flags |= CIRCLAUNCH_IS_INTERNAL;
circ = circuit_launch_by_extend_info(new_circ_purpose, extend_info,
flags);
}
extend_info_free(extend_info);
if (desired_circuit_purpose == CIRCUIT_PURPOSE_C_GENERAL) {
/* We just caused a circuit to get built because of this stream.
* If this stream has caused a _lot_ of circuits to be built, that's
* a bad sign: we should tell the user. */
if (conn->num_circuits_launched < NUM_CIRCUITS_LAUNCHED_THRESHOLD &&
++conn->num_circuits_launched == NUM_CIRCUITS_LAUNCHED_THRESHOLD)
log_info(LD_CIRC, "The application request to %s:%d has launched "
"%d circuits without finding one it likes.",
escaped_safe_str_client(conn->socks_request->address),
conn->socks_request->port,
conn->num_circuits_launched);
} else {
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/* help predict this next time */
rep_hist_note_used_internal(time(NULL), need_uptime, 1);
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
if (circ) {
/* write the service_id into circ */
circ->rend_data = rend_data_dup(ENTRY_TO_EDGE_CONN(conn)->rend_data);
if (circ->_base.purpose == CIRCUIT_PURPOSE_C_ESTABLISH_REND &&
circ->_base.state == CIRCUIT_STATE_OPEN)
rend_client_rendcirc_has_opened(circ);
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
}
}
} /* endif (!circ) */
if (circ) {
/* Mark the circuit with the isolation fields for this connection.
* When the circuit arrives, we'll clear these flags: this is
* just some internal bookkeeping to make sure that we have
* launched enough circuits.
*/
connection_edge_update_circuit_isolation(conn, circ, 0);
} else {
log_info(LD_APP,
"No safe circuit (purpose %d) ready for edge "
"connection; delaying.",
desired_circuit_purpose);
}
*circp = circ;
return 0;
}
/** Return true iff <b>crypt_path</b> is one of the crypt_paths for
* <b>circ</b>. */
static int
cpath_is_on_circuit(origin_circuit_t *circ, crypt_path_t *crypt_path)
{
crypt_path_t *cpath, *cpath_next = NULL;
for (cpath = circ->cpath; cpath_next != circ->cpath; cpath = cpath_next) {
cpath_next = cpath->next;
if (crypt_path == cpath)
return 1;
}
return 0;
}
/** Return true iff client-side optimistic data is supported. */
static int
optimistic_data_enabled(void)
{
const or_options_t *options = get_options();
if (options->OptimisticData < 0) {
/* XXX023 consider having auto default to 1 rather than 0 before
* the 0.2.3 branch goes stable. See bug 3617. -RD */
const int32_t enabled =
networkstatus_get_param(NULL, "UseOptimisticData", 0, 0, 1);
return (int)enabled;
}
return options->OptimisticData;
}
/** Attach the AP stream <b>apconn</b> to circ's linked list of
* p_streams. Also set apconn's cpath_layer to <b>cpath</b>, or to the last
* hop in circ's cpath if <b>cpath</b> is NULL.
*/
static void
link_apconn_to_circ(entry_connection_t *apconn, origin_circuit_t *circ,
crypt_path_t *cpath)
{
const node_t *exitnode;
/* add it into the linked list of streams on this circuit */
log_debug(LD_APP|LD_CIRC, "attaching new conn to circ. n_circ_id %d.",
circ->_base.n_circ_id);
/* reset it, so we can measure circ timeouts */
ENTRY_TO_CONN(apconn)->timestamp_lastread = time(NULL);
ENTRY_TO_EDGE_CONN(apconn)->next_stream = circ->p_streams;
ENTRY_TO_EDGE_CONN(apconn)->on_circuit = TO_CIRCUIT(circ);
/* assert_connection_ok(conn, time(NULL)); */
circ->p_streams = ENTRY_TO_EDGE_CONN(apconn);
if (connection_edge_is_rendezvous_stream(ENTRY_TO_EDGE_CONN(apconn))) {
/* We are attaching a stream to a rendezvous circuit. That means
* that an attempt to connect to a hidden service just
* succeeded. Tell rendclient.c. */
rend_client_note_connection_attempt_ended(
ENTRY_TO_EDGE_CONN(apconn)->rend_data->onion_address);
}
if (cpath) { /* we were given one; use it */
tor_assert(cpath_is_on_circuit(circ, cpath));
} else {
/* use the last hop in the circuit */
tor_assert(circ->cpath);
tor_assert(circ->cpath->prev);
tor_assert(circ->cpath->prev->state == CPATH_STATE_OPEN);
cpath = circ->cpath->prev;
}
ENTRY_TO_EDGE_CONN(apconn)->cpath_layer = cpath;
circ->isolation_any_streams_attached = 1;
connection_edge_update_circuit_isolation(apconn, circ, 0);
/* See if we can use optimistic data on this circuit */
if (cpath->extend_info &&
(exitnode = node_get_by_id(cpath->extend_info->identity_digest)) &&
exitnode->rs) {
/* Okay; we know what exit node this is. */
if (optimistic_data_enabled() &&
circ->_base.purpose == CIRCUIT_PURPOSE_C_GENERAL &&
exitnode->rs->version_supports_optimistic_data)
apconn->may_use_optimistic_data = 1;
else
apconn->may_use_optimistic_data = 0;
log_info(LD_APP, "Looks like completed circuit to %s %s allow "
"optimistic data for connection to %s",
safe_str_client(node_describe(exitnode)),
apconn->may_use_optimistic_data ? "does" : "doesn't",
safe_str_client(apconn->socks_request->address));
}
}
/** Return true iff <b>address</b> is matched by one of the entries in
* TrackHostExits. */
int
hostname_in_track_host_exits(const or_options_t *options, const char *address)
{
if (!options->TrackHostExits)
return 0;
SMARTLIST_FOREACH_BEGIN(options->TrackHostExits, const char *, cp) {
if (cp[0] == '.') { /* match end */
if (cp[1] == '\0' ||
!strcasecmpend(address, cp) ||
!strcasecmp(address, &cp[1]))
return 1;
} else if (strcasecmp(cp, address) == 0) {
return 1;
}
} SMARTLIST_FOREACH_END(cp);
return 0;
}
/** If an exit wasn't explicitly specified for <b>conn</b>, consider saving
* the exit that we *did* choose for use by future connections to
* <b>conn</b>'s destination.
*/
static void
consider_recording_trackhost(const entry_connection_t *conn,
2011-07-06 22:30:02 +02:00
const origin_circuit_t *circ)
{
const or_options_t *options = get_options();
char *new_address = NULL;
char fp[HEX_DIGEST_LEN+1];
/* Search the addressmap for this conn's destination. */
/* If he's not in the address map.. */
if (!options->TrackHostExits ||
addressmap_have_mapping(conn->socks_request->address,
options->TrackHostExitsExpire))
return; /* nothing to track, or already mapped */
if (!hostname_in_track_host_exits(options, conn->socks_request->address) ||
!circ->build_state->chosen_exit)
return;
/* write down the fingerprint of the chosen exit, not the nickname,
* because the chosen exit might not be named. */
base16_encode(fp, sizeof(fp),
circ->build_state->chosen_exit->identity_digest, DIGEST_LEN);
/* Add this exit/hostname pair to the addressmap. */
tor_asprintf(&new_address, "%s.%s.exit",
conn->socks_request->address, fp);
addressmap_register(conn->socks_request->address, new_address,
time(NULL) + options->TrackHostExitsExpire,
ADDRMAPSRC_TRACKEXIT, 0, 0);
}
/** Attempt to attach the connection <b>conn</b> to <b>circ</b>, and send a
* begin or resolve cell as appropriate. Return values are as for
* connection_ap_handshake_attach_circuit. The stream will exit from the hop
* indicated by <b>cpath</b>, or from the last hop in circ's cpath if
* <b>cpath</b> is NULL. */
int
connection_ap_handshake_attach_chosen_circuit(entry_connection_t *conn,
origin_circuit_t *circ,
crypt_path_t *cpath)
{
connection_t *base_conn = ENTRY_TO_CONN(conn);
tor_assert(conn);
tor_assert(base_conn->state == AP_CONN_STATE_CIRCUIT_WAIT ||
base_conn->state == AP_CONN_STATE_CONTROLLER_WAIT);
tor_assert(conn->socks_request);
tor_assert(circ);
tor_assert(circ->_base.state == CIRCUIT_STATE_OPEN);
base_conn->state = AP_CONN_STATE_CIRCUIT_WAIT;
if (!circ->_base.timestamp_dirty)
circ->_base.timestamp_dirty = time(NULL);
link_apconn_to_circ(conn, circ, cpath);
tor_assert(conn->socks_request);
if (conn->socks_request->command == SOCKS_COMMAND_CONNECT) {
if (!conn->use_begindir)
consider_recording_trackhost(conn, circ);
if (connection_ap_handshake_send_begin(conn) < 0)
return -1;
} else {
if (connection_ap_handshake_send_resolve(conn) < 0)
return -1;
}
return 1;
}
/** Try to find a safe live circuit for CONN_TYPE_AP connection conn. If
* we don't find one: if conn cannot be handled by any known nodes,
* warn and return -1 (conn needs to die, and is maybe already marked);
* else launch new circuit (if necessary) and return 0.
* Otherwise, associate conn with a safe live circuit, do the
* right next step, and return 1.
*/
/* XXXX this function should mark for close whenever it returns -1;
* its callers shouldn't have to worry about that. */
int
connection_ap_handshake_attach_circuit(entry_connection_t *conn)
{
connection_t *base_conn = ENTRY_TO_CONN(conn);
int retval;
int conn_age;
int want_onehop;
tor_assert(conn);
tor_assert(base_conn->state == AP_CONN_STATE_CIRCUIT_WAIT);
tor_assert(conn->socks_request);
want_onehop = conn->want_onehop;
conn_age = (int)(time(NULL) - base_conn->timestamp_created);
if (conn_age >= get_options()->SocksTimeout) {
int severity = (tor_addr_is_null(&base_conn->addr) && !base_conn->port) ?
LOG_INFO : LOG_NOTICE;
log_fn(severity, LD_APP,
"Tried for %d seconds to get a connection to %s:%d. Giving up.",
conn_age, safe_str_client(conn->socks_request->address),
conn->socks_request->port);
return -1;
}
if (!connection_edge_is_rendezvous_stream(ENTRY_TO_EDGE_CONN(conn))) {
/* we're a general conn */
origin_circuit_t *circ=NULL;
if (conn->chosen_exit_name) {
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
const node_t *node = node_get_by_nickname(conn->chosen_exit_name, 1);
int opt = conn->chosen_exit_optional;
Initial conversion to use node_t throughout our codebase. A node_t is an abstraction over routerstatus_t, routerinfo_t, and microdesc_t. It should try to present a consistent interface to all of them. There should be a node_t for a server whenever there is * A routerinfo_t for it in the routerlist * A routerstatus_t in the current_consensus. (note that a microdesc_t alone isn't enough to make a node_t exist, since microdescriptors aren't usable on their own.) There are three ways to get a node_t right now: looking it up by ID, looking it up by nickname, and iterating over the whole list of microdescriptors. All (or nearly all) functions that are supposed to return "a router" -- especially those used in building connections and circuits -- should return a node_t, not a routerinfo_t or a routerstatus_t. A node_t should hold all the *mutable* flags about a node. This patch moves the is_foo flags from routerinfo_t into node_t. The flags in routerstatus_t remain, but they get set from the consensus and should not change. Some other highlights of this patch are: * Looking up routerinfo and routerstatus by nickname is now unified and based on the "look up a node by nickname" function. This tries to look only at the values from current consensus, and not get confused by the routerinfo_t->is_named flag, which could get set for other weird reasons. This changes the behavior of how authorities (when acting as clients) deal with nodes that have been listed by nickname. * I tried not to artificially increase the size of the diff here by moving functions around. As a result, some functions that now operate on nodes are now in the wrong file -- they should get moved to nodelist.c once this refactoring settles down. This moving should happen as part of a patch that moves functions AND NOTHING ELSE. * Some old code is now left around inside #if 0/1 blocks, and should get removed once I've verified that I don't want it sitting around to see how we used to do things. There are still some unimplemented functions: these are flagged with "UNIMPLEMENTED_NODELIST()." I'll work on filling in the implementation here, piece by piece. I wish this patch could have been smaller, but there did not seem to be any piece of it that was independent from the rest. Moving flags forces many functions that once returned routerinfo_t * to return node_t *, which forces their friends to change, and so on.
2010-09-29 21:00:41 +02:00
if (!node && !want_onehop) {
/* We ran into this warning when trying to extend a circuit to a
* hidden service directory for which we didn't have a router
* descriptor. See flyspray task 767 for more details. We should
* keep this in mind when deciding to use BEGIN_DIR cells for other
* directory requests as well. -KL*/
log_fn(opt ? LOG_INFO : LOG_WARN, LD_APP,
"Requested exit point '%s' is not known. %s.",
conn->chosen_exit_name, opt ? "Trying others" : "Closing");
if (opt) {
conn->chosen_exit_optional = 0;
tor_free(conn->chosen_exit_name);
return 0;
}
return -1;
}
if (node && !connection_ap_can_use_exit(conn, node)) {
log_fn(opt ? LOG_INFO : LOG_WARN, LD_APP,
"Requested exit point '%s' is excluded or "
"would refuse request. %s.",
conn->chosen_exit_name, opt ? "Trying others" : "Closing");
if (opt) {
conn->chosen_exit_optional = 0;
tor_free(conn->chosen_exit_name);
return 0;
}
return -1;
}
}
/* find the circuit that we should use, if there is one. */
retval = circuit_get_open_circ_or_launch(
conn, CIRCUIT_PURPOSE_C_GENERAL, &circ);
if (retval < 1) // XXX022 if we totally fail, this still returns 0 -RD
return retval;
log_debug(LD_APP|LD_CIRC,
"Attaching apconn to circ %d (stream %d sec old).",
circ->_base.n_circ_id, conn_age);
/* print the circ's path, so people can figure out which circs are
* sucking. */
circuit_log_path(LOG_INFO,LD_APP|LD_CIRC,circ);
/* We have found a suitable circuit for our conn. Hurray. */
return connection_ap_handshake_attach_chosen_circuit(conn, circ, NULL);
} else { /* we're a rendezvous conn */
origin_circuit_t *rendcirc=NULL, *introcirc=NULL;
tor_assert(!ENTRY_TO_EDGE_CONN(conn)->cpath_layer);
/* start by finding a rendezvous circuit for us */
retval = circuit_get_open_circ_or_launch(
conn, CIRCUIT_PURPOSE_C_REND_JOINED, &rendcirc);
if (retval < 0) return -1; /* failed */
if (retval > 0) {
tor_assert(rendcirc);
/* one is already established, attach */
log_info(LD_REND,
"rend joined circ %d already here. attaching. "
"(stream %d sec old)",
rendcirc->_base.n_circ_id, conn_age);
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
/* Mark rendezvous circuits as 'newly dirty' every time you use
* them, since the process of rebuilding a rendezvous circ is so
2009-05-27 23:55:51 +02:00
* expensive. There is a tradeoff between linkability and
Introduce a notion of 'internal' circs, which are chosen without regard to the exit policy of the last hop. Intro and rendezvous circs must be internal circs, to avoid leaking information. Resolve and connect streams can use internal circs if they want. New circuit pooling algorithm: make sure to have enough circs around to satisfy any predicted ports, and also make sure to have 2 internal circs around if we've required internal circs lately (with high uptime if we've seen that lately). Split NewCircuitPeriod config option into NewCircuitPeriod (30 secs), which describes how often we retry making new circuits if current ones are dirty, and MaxCircuitDirtiness (10 mins), which describes how long we're willing to make use of an already-dirty circuit. Once rendezvous circuits are established, keep using the same circuit as long as you attach a new stream to it at least every 10 minutes. (So web browsing doesn't require you to build new rend circs every 30 seconds.) Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND circ as necessary, if there are any completed ones lying around when we try to launch one. Re-instate the ifdef's to use version-0 style introduce cells, since there was yet another bug in handling version-1 style. We'll try switching over again after 0.0.9 is obsolete. Bugfix: when choosing an exit node for a new non-internal circ, don't take into account whether it'll be useful for any pending x.onion addresses -- it won't. Bugfix: we weren't actually publishing the hidden service descriptor when it became dirty. So we only published it every 20 minutes or so, which means when you first start your Tor, the hidden service will seem broken. svn:r3360
2005-01-17 19:13:09 +01:00
* feasibility, at this point.
*/
rendcirc->_base.timestamp_dirty = time(NULL);
link_apconn_to_circ(conn, rendcirc, NULL);
if (connection_ap_handshake_send_begin(conn) < 0)
return 0; /* already marked, let them fade away */
return 1;
}
if (rendcirc && (rendcirc->_base.purpose ==
CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED)) {
log_info(LD_REND,
"pending-join circ %d already here, with intro ack. "
"Stalling. (stream %d sec old)",
rendcirc->_base.n_circ_id, conn_age);
return 0;
}
/* it's on its way. find an intro circ. */
retval = circuit_get_open_circ_or_launch(
conn, CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT, &introcirc);
if (retval < 0) return -1; /* failed */
if (retval > 0) {
/* one has already sent the intro. keep waiting. */
circuit_t *c = NULL;
tor_assert(introcirc);
log_info(LD_REND, "Intro circ %d present and awaiting ack (rend %d). "
"Stalling. (stream %d sec old)",
introcirc->_base.n_circ_id,
rendcirc ? rendcirc->_base.n_circ_id : 0,
conn_age);
/* abort parallel intro circs, if any */
for (c = global_circuitlist; c; c = c->next) {
if (c->purpose == CIRCUIT_PURPOSE_C_INTRODUCING &&
!c->marked_for_close && CIRCUIT_IS_ORIGIN(c)) {
origin_circuit_t *oc = TO_ORIGIN_CIRCUIT(c);
if (oc->rend_data &&
!rend_cmp_service_ids(
ENTRY_TO_EDGE_CONN(conn)->rend_data->onion_address,
oc->rend_data->onion_address)) {
log_info(LD_REND|LD_CIRC, "Closing introduction circuit that we "
"built in parallel.");
circuit_mark_for_close(c, END_CIRC_REASON_TIMEOUT);
}
}
}
return 0;
}
/* now rendcirc and introcirc are each either undefined or not finished */
if (rendcirc && introcirc &&
rendcirc->_base.purpose == CIRCUIT_PURPOSE_C_REND_READY) {
log_info(LD_REND,
"ready rend circ %d already here (no intro-ack yet on "
"intro %d). (stream %d sec old)",
rendcirc->_base.n_circ_id,
introcirc->_base.n_circ_id, conn_age);
tor_assert(introcirc->_base.purpose == CIRCUIT_PURPOSE_C_INTRODUCING);
if (introcirc->_base.state == CIRCUIT_STATE_OPEN) {
log_info(LD_REND,"found open intro circ %d (rend %d); sending "
"introduction. (stream %d sec old)",
introcirc->_base.n_circ_id, rendcirc->_base.n_circ_id,
conn_age);
switch (rend_client_send_introduction(introcirc, rendcirc)) {
case 0: /* success */
rendcirc->_base.timestamp_dirty = time(NULL);
introcirc->_base.timestamp_dirty = time(NULL);
assert_circuit_ok(TO_CIRCUIT(rendcirc));
assert_circuit_ok(TO_CIRCUIT(introcirc));
return 0;
case -1: /* transient error */
return 0;
case -2: /* permanent error */
return -1;
default: /* oops */
tor_fragile_assert();
return -1;
}
}
}
log_info(LD_REND, "Intro (%d) and rend (%d) circs are not both ready. "
"Stalling conn. (%d sec old)",
introcirc ? introcirc->_base.n_circ_id : 0,
rendcirc ? rendcirc->_base.n_circ_id : 0, conn_age);
return 0;
}
}