Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
/* Copyright 2001,2002 Roger Dingledine, Matej Pfajfar. */
|
|
|
|
/* See LICENSE for licensing information */
|
|
|
|
/* $Id$ */
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
#include "or.h"
|
|
|
|
|
|
|
|
/********* START VARIABLES **********/
|
|
|
|
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
or_options_t options; /* command-line and config-file options */
|
2002-07-12 19:12:08 +02:00
|
|
|
int global_role;
|
2002-06-27 00:45:49 +02:00
|
|
|
|
2002-07-05 08:27:23 +02:00
|
|
|
static connection_t *connection_array[MAXCONNECTIONS] =
|
2002-06-27 00:45:49 +02:00
|
|
|
{ NULL };
|
|
|
|
|
2002-07-05 08:27:23 +02:00
|
|
|
static struct pollfd poll_array[MAXCONNECTIONS] =
|
2002-06-27 00:45:49 +02:00
|
|
|
{ [0 ... MAXCONNECTIONS-1] = { -1, 0, 0 } };
|
|
|
|
|
2002-07-05 08:27:23 +02:00
|
|
|
static int nfds=0; /* number of connections currently active */
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
/* private key */
|
2002-07-05 08:27:23 +02:00
|
|
|
static RSA *prkey = NULL;
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
/* router array */
|
2002-07-05 08:27:23 +02:00
|
|
|
static routerinfo_t **router_array = NULL;
|
|
|
|
static int rarray_len = 0;
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
/********* END VARIABLES ************/
|
|
|
|
|
2002-07-05 08:27:23 +02:00
|
|
|
/****************************************************************************
|
|
|
|
*
|
|
|
|
* This section contains accessors and other methods on the connection_array
|
|
|
|
* and poll_array variables (which are global within this file and unavailable
|
|
|
|
* outside it).
|
|
|
|
*
|
|
|
|
****************************************************************************/
|
|
|
|
|
2002-06-27 00:45:49 +02:00
|
|
|
int connection_add(connection_t *conn) {
|
|
|
|
|
|
|
|
if(nfds >= MAXCONNECTIONS-2) { /* 2, for some breathing room. should count the fenceposts. */
|
|
|
|
/* FIXME should use the 'max connections' option */
|
2002-07-16 04:12:58 +02:00
|
|
|
log(LOG_INFO,"connection_add(): failing because nfds is too high.");
|
2002-06-27 00:45:49 +02:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
conn->poll_index = nfds;
|
|
|
|
connection_set_poll_socket(conn);
|
|
|
|
connection_array[nfds] = conn;
|
|
|
|
|
|
|
|
/* zero these out here, because otherwise we'll inherit values from the previously freed one */
|
|
|
|
poll_array[nfds].events = 0;
|
|
|
|
poll_array[nfds].revents = 0;
|
|
|
|
|
|
|
|
nfds++;
|
|
|
|
|
2002-07-16 04:12:58 +02:00
|
|
|
log(LOG_INFO,"connection_add(): new conn type %d, socket %d, nfds %d.",conn->type, conn->s, nfds);
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
void connection_set_poll_socket(connection_t *conn) {
|
|
|
|
poll_array[conn->poll_index].fd = conn->s;
|
|
|
|
}
|
|
|
|
|
|
|
|
int connection_remove(connection_t *conn) {
|
|
|
|
int current_index;
|
|
|
|
|
|
|
|
assert(conn);
|
|
|
|
assert(nfds>0);
|
|
|
|
|
|
|
|
circuit_about_to_close_connection(conn); /* flush and send destroys for all circuits on this conn */
|
|
|
|
|
|
|
|
current_index = conn->poll_index;
|
|
|
|
if(current_index == nfds-1) { /* this is the end */
|
|
|
|
// connection_free(conn);
|
|
|
|
nfds--;
|
2002-07-16 04:12:58 +02:00
|
|
|
log(LOG_INFO,"connection_remove(): nfds now %d.",nfds);
|
2002-06-27 00:45:49 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* we replace this one with the one at the end, then free it */
|
|
|
|
nfds--;
|
|
|
|
poll_array[current_index].fd = poll_array[nfds].fd;
|
|
|
|
poll_array[current_index].events = poll_array[nfds].events;
|
|
|
|
poll_array[current_index].revents = poll_array[nfds].revents;
|
|
|
|
connection_array[current_index] = connection_array[nfds];
|
|
|
|
connection_array[current_index]->poll_index = current_index;
|
|
|
|
|
2002-07-16 04:12:58 +02:00
|
|
|
log(LOG_INFO,"connection_remove(): nfds now %d.",nfds);
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2002-07-08 10:59:15 +02:00
|
|
|
connection_t *connection_twin_get_by_addr_port(uint32_t addr, uint16_t port) {
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
/* Find a connection to the router described by addr and port,
|
|
|
|
* or alternately any router which knows its key.
|
|
|
|
* This connection *must* be in 'open' state.
|
|
|
|
* If not, return NULL.
|
|
|
|
*/
|
2002-07-08 10:59:15 +02:00
|
|
|
int i;
|
|
|
|
connection_t *conn;
|
|
|
|
|
|
|
|
/* first check if it's there exactly */
|
|
|
|
conn = connection_exact_get_by_addr_port(addr,port);
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
if(conn && connection_state_is_open(conn)) {
|
2002-07-08 10:59:15 +02:00
|
|
|
return conn;
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
}
|
2002-07-08 10:59:15 +02:00
|
|
|
|
|
|
|
/* now check if any of the other open connections are a twin for this one */
|
|
|
|
|
|
|
|
/* XXX */
|
|
|
|
|
|
|
|
/* guess not */
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
connection_t *connection_exact_get_by_addr_port(uint32_t addr, uint16_t port) {
|
2002-06-27 00:45:49 +02:00
|
|
|
int i;
|
|
|
|
connection_t *conn;
|
|
|
|
|
|
|
|
for(i=0;i<nfds;i++) {
|
|
|
|
conn = connection_array[i];
|
|
|
|
assert(conn);
|
|
|
|
if(conn->addr == addr && conn->port == port)
|
|
|
|
return conn;
|
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
connection_t *connection_get_by_type(int type) {
|
|
|
|
int i;
|
|
|
|
connection_t *conn;
|
|
|
|
|
|
|
|
for(i=0;i<nfds;i++) {
|
|
|
|
conn = connection_array[i];
|
|
|
|
if(conn->type == type)
|
|
|
|
return conn;
|
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2002-07-05 08:27:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/* the next 4 functions should move to routers.c once we get it
|
|
|
|
* cleaned up more. The router_array and rarray_len variables should
|
|
|
|
* move there too.
|
|
|
|
*/
|
|
|
|
|
2002-06-27 00:45:49 +02:00
|
|
|
routerinfo_t *router_get_by_addr_port(uint32_t addr, uint16_t port) {
|
|
|
|
int i;
|
|
|
|
routerinfo_t *router;
|
|
|
|
|
2002-07-05 08:27:23 +02:00
|
|
|
assert(router_array);
|
2002-06-27 00:45:49 +02:00
|
|
|
|
2002-07-05 08:27:23 +02:00
|
|
|
for(i=0;i<rarray_len;i++) {
|
2002-06-27 00:45:49 +02:00
|
|
|
router = router_array[i];
|
Integrated onion proxy into or/
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
2002-07-02 11:36:58 +02:00
|
|
|
if ((router->addr == addr) && (router->or_port == port))
|
2002-06-27 00:45:49 +02:00
|
|
|
return router;
|
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
Integrated onion proxy into or/
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
2002-07-02 11:36:58 +02:00
|
|
|
routerinfo_t *router_get_first_in_route(unsigned int *route, size_t routelen) {
|
|
|
|
return router_array[route[routelen-1]];
|
|
|
|
}
|
|
|
|
|
|
|
|
/* a wrapper around new_route. put all these in routers.c perhaps? */
|
|
|
|
unsigned int *router_new_route(size_t *rlen) {
|
2002-07-12 19:12:08 +02:00
|
|
|
return new_route(options.CoinWeight, router_array,rarray_len, rlen);
|
Integrated onion proxy into or/
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
2002-07-02 11:36:58 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* a wrapper around create_onion */
|
|
|
|
unsigned char *router_create_onion(unsigned int *route, size_t routelen, size_t *lenp, crypt_path_t **cpathp) {
|
|
|
|
return create_onion(router_array,rarray_len,route,routelen,lenp,cpathp);
|
|
|
|
}
|
|
|
|
|
2002-07-05 08:27:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Integrated onion proxy into or/
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
2002-07-02 11:36:58 +02:00
|
|
|
connection_t *connect_to_router_as_op(routerinfo_t *router) {
|
2002-07-12 19:12:08 +02:00
|
|
|
return connection_connect_to_router_as_op(router, prkey, options.ORPort);
|
Integrated onion proxy into or/
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
2002-07-02 11:36:58 +02:00
|
|
|
}
|
|
|
|
|
2002-06-27 00:45:49 +02:00
|
|
|
void connection_watch_events(connection_t *conn, short events) {
|
|
|
|
|
|
|
|
assert(conn && conn->poll_index < nfds);
|
|
|
|
|
|
|
|
poll_array[conn->poll_index].events = events;
|
|
|
|
}
|
|
|
|
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
void connection_stop_reading(connection_t *conn) {
|
|
|
|
|
|
|
|
assert(conn && conn->poll_index < nfds);
|
|
|
|
|
|
|
|
if(poll_array[conn->poll_index].events & POLLIN)
|
|
|
|
poll_array[conn->poll_index].events -= POLLIN;
|
|
|
|
}
|
|
|
|
|
|
|
|
void connection_start_reading(connection_t *conn) {
|
|
|
|
|
|
|
|
assert(conn && conn->poll_index < nfds);
|
|
|
|
|
|
|
|
poll_array[conn->poll_index].events |= POLLIN;
|
|
|
|
}
|
|
|
|
|
2002-06-27 00:45:49 +02:00
|
|
|
void check_conn_read(int i) {
|
|
|
|
int retval;
|
|
|
|
connection_t *conn;
|
|
|
|
|
|
|
|
if(poll_array[i].revents & POLLIN) { /* something to read */
|
|
|
|
|
|
|
|
conn = connection_array[i];
|
|
|
|
assert(conn);
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
// log(LOG_DEBUG,"check_conn_read(): socket %d has something to read.",conn->s);
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
if (conn->type == CONN_TYPE_OP_LISTENER) {
|
|
|
|
retval = connection_op_handle_listener_read(conn);
|
|
|
|
} else if (conn->type == CONN_TYPE_OR_LISTENER) {
|
|
|
|
retval = connection_or_handle_listener_read(conn);
|
Integrated onion proxy into or/
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
2002-07-02 11:36:58 +02:00
|
|
|
} else if (conn->type == CONN_TYPE_AP_LISTENER) {
|
|
|
|
retval = connection_ap_handle_listener_read(conn);
|
2002-06-27 00:45:49 +02:00
|
|
|
} else {
|
2002-06-30 09:37:49 +02:00
|
|
|
/* else it's an OP, OR, or exit */
|
2002-06-27 00:45:49 +02:00
|
|
|
retval = connection_read_to_buf(conn);
|
|
|
|
if (retval >= 0) { /* all still well */
|
|
|
|
retval = connection_process_inbuf(conn);
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
// log(LOG_DEBUG,"check_conn_read(): connection_process_inbuf returned %d.",retval);
|
|
|
|
if(retval >= 0 && !connection_state_is_open(conn) && conn->receiver_bucket == 0) {
|
|
|
|
log(LOG_DEBUG,"check_conn_read(): receiver bucket reached 0 before handshake finished. Closing.");
|
|
|
|
retval = -1;
|
|
|
|
}
|
2002-06-27 00:45:49 +02:00
|
|
|
}
|
|
|
|
}
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
|
2002-06-27 00:45:49 +02:00
|
|
|
if(retval < 0) { /* this connection is broken. remove it */
|
|
|
|
log(LOG_DEBUG,"check_conn_read(): Connection broken, removing.");
|
|
|
|
connection_remove(conn);
|
|
|
|
connection_free(conn);
|
|
|
|
if(i<nfds) { /* we just replaced the one at i with a new one.
|
|
|
|
process it too. */
|
|
|
|
check_conn_read(i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void check_conn_write(int i) {
|
|
|
|
int retval;
|
|
|
|
connection_t *conn;
|
|
|
|
|
|
|
|
if(poll_array[i].revents & POLLOUT) { /* something to write */
|
|
|
|
|
|
|
|
conn = connection_array[i];
|
|
|
|
log(LOG_DEBUG,"check_conn_write(): socket %d wants to write.",conn->s);
|
|
|
|
|
|
|
|
if(conn->type == CONN_TYPE_OP_LISTENER ||
|
|
|
|
conn->type == CONN_TYPE_OR_LISTENER) {
|
|
|
|
log(LOG_DEBUG,"check_conn_write(): Got a listener socket. Can't happen!");
|
|
|
|
retval = -1;
|
|
|
|
} else {
|
2002-06-30 09:37:49 +02:00
|
|
|
/* else it's an OP, OR, or exit */
|
2002-06-27 00:45:49 +02:00
|
|
|
retval = connection_flush_buf(conn); /* conns in CONNECTING state will fall through... */
|
|
|
|
if(retval == 0) { /* it's done flushing */
|
|
|
|
retval = connection_finished_flushing(conn); /* ...and get handled here. */
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if(retval < 0) { /* this connection is broken. remove it. */
|
|
|
|
log(LOG_DEBUG,"check_conn_write(): Connection broken, removing.");
|
|
|
|
connection_remove(conn);
|
|
|
|
connection_free(conn);
|
|
|
|
if(i<nfds) { /* we just replaced the one at i with a new one.
|
|
|
|
process it too. */
|
|
|
|
check_conn_read(i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void check_conn_marked(int i) {
|
|
|
|
connection_t *conn;
|
|
|
|
|
|
|
|
conn = connection_array[i];
|
|
|
|
assert(conn);
|
|
|
|
if(conn->marked_for_close) {
|
|
|
|
log(LOG_DEBUG,"check_conn_marked(): Cleaning up connection.");
|
Integrated onion proxy into or/
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
2002-07-02 11:36:58 +02:00
|
|
|
if(conn->s >= 0) { /* might be an incomplete exit connection */
|
|
|
|
/* FIXME there's got to be a better way to check for this -- and make other checks? */
|
|
|
|
connection_flush_buf(conn); /* flush it first */
|
|
|
|
}
|
2002-06-27 00:45:49 +02:00
|
|
|
connection_remove(conn);
|
|
|
|
connection_free(conn);
|
|
|
|
if(i<nfds) { /* we just replaced the one at i with a new one.
|
|
|
|
process it too. */
|
2002-07-05 08:27:23 +02:00
|
|
|
check_conn_marked(i);
|
2002-06-27 00:45:49 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
int prepare_for_poll(int *timeout) {
|
|
|
|
int i;
|
|
|
|
int need_to_refill_buckets = 0;
|
|
|
|
connection_t *conn = NULL;
|
|
|
|
connection_t *tmpconn;
|
|
|
|
struct timeval now, soonest;
|
|
|
|
static int current_second = 0; /* from previous calls to gettimeofday */
|
|
|
|
int ms_until_conn;
|
|
|
|
|
|
|
|
*timeout = -1; /* set it to never timeout, possibly overridden below */
|
|
|
|
|
|
|
|
/* first check if we need to refill buckets */
|
|
|
|
for(i=0;i<nfds;i++) {
|
|
|
|
if(connection_receiver_bucket_should_increase(connection_array[i])) {
|
|
|
|
need_to_refill_buckets = 1;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if(gettimeofday(&now,NULL) < 0)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
if(need_to_refill_buckets) {
|
|
|
|
if(now.tv_sec > current_second) { /* the second has already rolled over! */
|
|
|
|
log(LOG_DEBUG,"prepare_for_poll(): The second has rolled over, immediately refilling.");
|
|
|
|
increment_receiver_buckets();
|
|
|
|
current_second = now.tv_sec; /* remember which second it is, for next time */
|
|
|
|
}
|
|
|
|
*timeout = 1000 - (now.tv_usec / 1000); /* how many milliseconds til the next second? */
|
|
|
|
// log(LOG_DEBUG,"prepare_for_poll(): %d milliseconds til next second.",*timeout);
|
|
|
|
}
|
|
|
|
|
|
|
|
if(options.LinkPadding) {
|
|
|
|
/* now check which conn wants to speak soonest */
|
|
|
|
for(i=0;i<nfds;i++) {
|
|
|
|
tmpconn = connection_array[i];
|
|
|
|
if(tmpconn->type != CONN_TYPE_OR && tmpconn->type != CONN_TYPE_OP)
|
|
|
|
continue; /* this conn type doesn't send cells */
|
|
|
|
if(!connection_state_is_open(tmpconn))
|
|
|
|
continue; /* only conns in state 'open' have a valid send_timeval */
|
|
|
|
while(tv_cmp(&tmpconn->send_timeval,&now) <= 0) { /* send_timeval has already passed, let it send a cell */
|
2002-07-16 20:24:12 +02:00
|
|
|
// log(LOG_DEBUG,"prepare_for_poll(): doing backlogged connection_send_cell on socket %d (%d ms old)",tmpconn->s,
|
|
|
|
// (now.tv_sec - tmpconn->send_timeval.tv_sec)*1000 +
|
|
|
|
// (now.tv_usec - tmpconn->send_timeval.tv_usec)/1000
|
|
|
|
// );
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
connection_send_cell(tmpconn);
|
|
|
|
}
|
|
|
|
if(!conn || tv_cmp(&tmpconn->send_timeval, &soonest) < 0) { /* this is the best choice so far */
|
|
|
|
// log(LOG_DEBUG,"prepare_for_poll(): chose socket %d as best connection so far",tmpconn->s);
|
|
|
|
conn = tmpconn;
|
|
|
|
soonest.tv_sec = conn->send_timeval.tv_sec;
|
|
|
|
soonest.tv_usec = conn->send_timeval.tv_usec;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if(conn) { /* we might want to set *timeout sooner */
|
|
|
|
ms_until_conn = (soonest.tv_sec - now.tv_sec)*1000 +
|
|
|
|
(soonest.tv_usec - now.tv_usec)/1000;
|
|
|
|
// log(LOG_DEBUG,"prepare_for_poll(): conn %d times out in %d ms.",conn->s, ms_until_conn);
|
|
|
|
if(*timeout == -1 || ms_until_conn < *timeout) { /* use the new one */
|
|
|
|
// log(LOG_DEBUG,"prepare_for_poll(): conn %d soonest, in %d ms.",conn->s,ms_until_conn);
|
|
|
|
*timeout = ms_until_conn;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
void increment_receiver_buckets(void) {
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for(i=0;i<nfds;i++)
|
|
|
|
connection_increment_receiver_bucket(connection_array[i]);
|
|
|
|
}
|
|
|
|
|
2002-06-27 00:45:49 +02:00
|
|
|
int do_main_loop(void) {
|
|
|
|
int i;
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
int timeout;
|
|
|
|
int poll_result;
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
/* load the routers file */
|
2002-07-12 19:12:08 +02:00
|
|
|
router_array = getrouters(options.RouterFile,&rarray_len, options.ORPort);
|
2002-06-27 00:45:49 +02:00
|
|
|
if (!router_array)
|
|
|
|
{
|
|
|
|
log(LOG_ERR,"Error loading router list.");
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
return -1;
|
2002-06-27 00:45:49 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* load the private key */
|
2002-07-12 19:12:08 +02:00
|
|
|
prkey = load_prkey(options.PrivateKeyFile);
|
2002-06-27 00:45:49 +02:00
|
|
|
if (!prkey)
|
|
|
|
{
|
|
|
|
log(LOG_ERR,"Error loading private key.");
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
return -1;
|
2002-06-27 00:45:49 +02:00
|
|
|
}
|
|
|
|
log(LOG_DEBUG,"core : Loaded private key of size %u bytes.",RSA_size(prkey));
|
|
|
|
|
Integrated onion proxy into or/
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
2002-07-02 11:36:58 +02:00
|
|
|
/* start-up the necessary connections based on global_role. This is where we
|
|
|
|
* try to connect to all the other ORs, and start the listeners */
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
retry_all_connections(options.Role, router_array, rarray_len, prkey,
|
2002-07-12 19:12:08 +02:00
|
|
|
options.ORPort, options.OPPort, options.APPort);
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
for(;;) {
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
if(prepare_for_poll(&timeout) < 0) {
|
|
|
|
log(LOG_DEBUG,"do_main_loop(): prepare_for_poll failed, exiting.");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
/* now timeout is the value we'll hand to poll. It's either -1, meaning
|
|
|
|
* don't timeout, else it indicates the soonest event (either the
|
|
|
|
* one-second rollover for refilling receiver buckets, or the soonest
|
|
|
|
* conn that needs to send a cell)
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* if the timeout is less than 10, set it to 10 */
|
|
|
|
if(timeout >= 0 && timeout < 10)
|
|
|
|
timeout = 10;
|
|
|
|
|
|
|
|
/* poll until we have an event, or it's time to do something */
|
|
|
|
poll_result = poll(poll_array, nfds, timeout);
|
|
|
|
|
2002-07-16 20:24:12 +02:00
|
|
|
#if 0 /* let catch() handle things like ^c, and otherwise don't worry about it */
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
if(poll_result < 0) {
|
|
|
|
log(LOG_ERR,"do_main_loop(): poll failed.");
|
|
|
|
if(errno != EINTR) /* let the program survive things like ^z */
|
|
|
|
return -1;
|
|
|
|
}
|
2002-07-16 20:24:12 +02:00
|
|
|
#endif
|
2002-06-27 00:45:49 +02:00
|
|
|
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
if(poll_result > 0) { /* we have at least one connection to deal with */
|
|
|
|
/* do all the reads first, so we can detect closed sockets */
|
|
|
|
for(i=0;i<nfds;i++)
|
|
|
|
check_conn_read(i); /* this also blows away broken connections */
|
2002-06-27 00:45:49 +02:00
|
|
|
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
/* then do the writes */
|
|
|
|
for(i=0;i<nfds;i++)
|
|
|
|
check_conn_write(i);
|
2002-06-27 00:45:49 +02:00
|
|
|
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
/* any of the conns need to be closed now? */
|
|
|
|
for(i=0;i<nfds;i++)
|
|
|
|
check_conn_marked(i);
|
|
|
|
}
|
|
|
|
/* refilling buckets and sending cells happens at the beginning of the
|
|
|
|
* next iteration of the loop, inside prepare_for_poll()
|
|
|
|
*/
|
2002-06-27 00:45:49 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2002-06-30 09:37:49 +02:00
|
|
|
void catch () {
|
2002-06-27 00:45:49 +02:00
|
|
|
errno = 0; /* netcat does this. it looks fun. */
|
|
|
|
|
|
|
|
log(LOG_DEBUG,"Catching ^c, exiting cleanly.");
|
|
|
|
|
|
|
|
exit(0);
|
|
|
|
}
|
|
|
|
|
2002-06-30 09:37:49 +02:00
|
|
|
int main(int argc, char *argv[]) {
|
2002-06-27 00:45:49 +02:00
|
|
|
int retval = 0;
|
|
|
|
|
|
|
|
signal (SIGINT, catch); /* to catch ^c so we can exit cleanly */
|
|
|
|
|
2002-07-12 19:12:08 +02:00
|
|
|
if ( getoptions(argc,argv,&options) ) exit(1);
|
2002-07-12 20:14:17 +02:00
|
|
|
log(options.loglevel,NULL); /* assign logging severity level from options */
|
Implemented link padding and receiver token buckets
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
2002-07-16 03:12:15 +02:00
|
|
|
global_role = options.Role; /* assign global_role from options. FIX: remove from global namespace later. */
|
2002-06-27 00:45:49 +02:00
|
|
|
|
|
|
|
ERR_load_crypto_strings();
|
|
|
|
retval = do_main_loop();
|
|
|
|
ERR_free_strings();
|
|
|
|
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|