major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
/* Copyright 2003 Roger Dingledine. */
|
|
|
|
/* See LICENSE for licensing information */
|
|
|
|
/* $Id$ */
|
|
|
|
|
|
|
|
#include "or.h"
|
|
|
|
|
2003-02-14 08:53:55 +01:00
|
|
|
#define MAX_ADDRESSLEN 256
|
|
|
|
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
#define MAX_DNSSLAVES 50
|
|
|
|
#define MIN_DNSSLAVES 3 /* 1 for the tor process, 3 slaves */
|
|
|
|
|
|
|
|
struct slave_data_t {
|
|
|
|
int fd; /* socket to talk on */
|
|
|
|
int num_processed; /* number of times we've used this slave */
|
|
|
|
char busy; /* whether this slave currently has a task */
|
2003-02-14 08:53:55 +01:00
|
|
|
char question[MAX_ADDRESSLEN]; /* the hostname that we're resolving */
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
unsigned char question_len; /* how many bytes in question */
|
2003-02-14 08:53:55 +01:00
|
|
|
char answer[MAX_ADDRESSLEN]; /* the answer to the question */
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
unsigned char answer_len; /* how many bytes in answer */
|
|
|
|
};
|
|
|
|
|
|
|
|
struct slave_data_t slave_data[MAX_DNSSLAVES+1];
|
|
|
|
struct pollfd poll_data[MAX_DNSSLAVES+1];
|
|
|
|
|
|
|
|
static int dns_spawn_slave(void);
|
|
|
|
static int dns_read_block(int fd, char *string, unsigned char *len);
|
|
|
|
static int dns_write_block(int fd, char *string, unsigned char len);
|
|
|
|
static int dns_read_tor_question(int index);
|
|
|
|
static int dns_read_slave_response(int index);
|
|
|
|
static int dns_find_idle_slave(int max);
|
|
|
|
static int dns_assign_to_slave(int from, int to);
|
|
|
|
static int dns_master_to_tor(int from, int to);
|
|
|
|
static void dns_master_main(int fd);
|
2003-02-14 08:53:55 +01:00
|
|
|
static int dns_tor_to_master(connection_t *exitconn);
|
|
|
|
static int dns_found_answer(char *question, uint32_t answer, uint32_t valid);
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
|
|
|
|
int connection_dns_finished_flushing(connection_t *conn) {
|
|
|
|
|
|
|
|
assert(conn && conn->type == CONN_TYPE_DNSMASTER);
|
|
|
|
|
|
|
|
connection_stop_writing(conn);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int connection_dns_process_inbuf(connection_t *conn) {
|
|
|
|
unsigned char length;
|
2003-02-14 08:53:55 +01:00
|
|
|
char buf[MAX_ADDRESSLEN];
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
char *question;
|
2003-02-14 08:53:55 +01:00
|
|
|
uint32_t answer;
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
|
|
|
|
assert(conn && conn->type == CONN_TYPE_DNSMASTER);
|
|
|
|
assert(conn->state == DNSMASTER_STATE_OPEN);
|
|
|
|
|
|
|
|
if(conn->inbuf_reached_eof) {
|
|
|
|
log(LOG_ERR,"connection_dns_process_inbuf(): Read eof. No more dnsmaster!");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(conn->inbuf);
|
|
|
|
|
2003-02-06 09:00:49 +01:00
|
|
|
if(conn->inbuf_datalen <= 0)
|
|
|
|
return 0;
|
|
|
|
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
/* peek into the inbuf, so we can check if it's all here */
|
|
|
|
length = *conn->inbuf; /* warning: abstraction violation :( */
|
|
|
|
assert(length < 240);
|
|
|
|
|
|
|
|
if(conn->inbuf_datalen < 1+length) { /* entire answer available? */
|
|
|
|
log(LOG_INFO,"connection_dns_process_inbuf(): %d available, waiting for %d.", conn->inbuf_datalen, length+1);
|
|
|
|
return 0; /* not yet */
|
|
|
|
}
|
|
|
|
|
|
|
|
if(connection_fetch_from_buf(buf,1+length,conn) < 0) {
|
|
|
|
log(LOG_ERR,"connection_dns_process_inbuf(): Broken inbuf. No more dnsmaster!");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
question = buf+1;
|
|
|
|
log(LOG_DEBUG,"connection_dns_process_inbuf(): length %d, question '%s', strlen question %d", length, question, strlen(question));
|
|
|
|
assert(length == 4 + strlen(question) + 1);
|
|
|
|
|
2003-02-14 08:53:55 +01:00
|
|
|
answer = *(uint32_t *)(buf+1+length-4);
|
|
|
|
dns_found_answer(question, answer, (answer != 0));
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
return connection_process_inbuf(conn); /* process the remainder of the buffer */
|
|
|
|
}
|
|
|
|
|
|
|
|
/* return -1 if error, else the fd that can talk to the dns master */
|
|
|
|
int dns_master_start(void) {
|
|
|
|
connection_t *conn;
|
|
|
|
pid_t pid;
|
|
|
|
int fd[2];
|
|
|
|
|
|
|
|
if(socketpair(AF_UNIX, SOCK_STREAM, 0, fd) < 0) {
|
|
|
|
log(LOG_ERR,"dns_master_start(): socketpair failed.");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
pid = fork();
|
|
|
|
if(pid < 0) {
|
|
|
|
log(LOG_ERR,"dns_master_start(): fork failed.");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
if(pid == 0) { /* i'm the child */
|
|
|
|
log(LOG_DEBUG,"dns_master_start(): child says fd0 %d, fd1 %d.", fd[0], fd[1]);
|
|
|
|
close(fd[0]);
|
|
|
|
dns_master_main(fd[1]);
|
|
|
|
assert(0); /* never gets here */
|
|
|
|
}
|
|
|
|
|
|
|
|
/* i'm the parent */
|
|
|
|
|
|
|
|
close(fd[1]);
|
|
|
|
|
|
|
|
fcntl(fd[0], F_SETFL, O_NONBLOCK); /* set s to non-blocking */
|
|
|
|
|
|
|
|
conn = connection_new(CONN_TYPE_DNSMASTER);
|
|
|
|
if(!conn) {
|
|
|
|
log(LOG_INFO,"dns_master_start(): connection_new failed. Giving up.");
|
|
|
|
/* XXX tell the dnsmaster to die */
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
conn->s = fd[0];
|
|
|
|
conn->address = strdup("localhost");
|
|
|
|
conn->receiver_bucket = -1; /* edge connections don't do receiver buckets */
|
|
|
|
conn->bandwidth = -1;
|
|
|
|
|
|
|
|
if(connection_add(conn) < 0) { /* no space, forget it */
|
|
|
|
log(LOG_INFO,"dns_master_start(): connection_add failed. Giving up.");
|
|
|
|
connection_free(conn);
|
|
|
|
/* XXX tell the dnsmaster to die */
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
conn->state = DNSMASTER_STATE_OPEN;
|
|
|
|
connection_start_reading(conn);
|
|
|
|
log(LOG_INFO,"dns_master_start(): dns handler is spawned.");
|
|
|
|
return fd[0];
|
|
|
|
}
|
|
|
|
|
|
|
|
static void dns_slave_main(int fd) {
|
2003-02-14 08:53:55 +01:00
|
|
|
char question[MAX_ADDRESSLEN];
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
unsigned char question_len;
|
|
|
|
struct hostent *rent;
|
|
|
|
|
|
|
|
for(;;) {
|
|
|
|
if(dns_read_block(fd, question, &question_len) < 0) { /* the master wants us to die */
|
|
|
|
log(LOG_INFO,"dns_slave_main(): eof on read from master. Exiting.");
|
|
|
|
exit(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
rent = gethostbyname(question);
|
|
|
|
if (!rent) {
|
|
|
|
log(LOG_INFO,"dns_slave_main(): Could not resolve dest addr %s. Returning nulls.",question);
|
|
|
|
if(dns_write_block(fd, "\0\0\0\0", 4) < 0) {
|
|
|
|
log(LOG_INFO,"dns_slave_main(): writing to master failed. Exiting.");
|
|
|
|
exit(0);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if(dns_write_block(fd, rent->h_addr, rent->h_length) < 0) {
|
|
|
|
log(LOG_INFO,"dns_slave_main(): writing to master failed. Exiting.");
|
|
|
|
exit(0);
|
|
|
|
}
|
|
|
|
log(LOG_INFO,"dns_slave_main(): Answered question '%s'.",question);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dns_spawn_slave(void) {
|
|
|
|
pid_t pid;
|
|
|
|
int fd[2];
|
|
|
|
|
|
|
|
if(socketpair(AF_UNIX, SOCK_STREAM, 0, fd) < 0) {
|
|
|
|
perror("socketpair");
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
pid = fork();
|
|
|
|
if(pid < 0) {
|
|
|
|
perror("fork");
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
if(pid == 0) { /* i'm the child */
|
|
|
|
close(fd[0]);
|
|
|
|
dns_slave_main(fd[1]);
|
|
|
|
assert(0); /* never gets here */
|
|
|
|
}
|
|
|
|
|
|
|
|
/* i'm the parent */
|
|
|
|
log(LOG_INFO,"dns_spawn_slave(): just spawned a slave."); // XXX change to debug
|
|
|
|
close(fd[1]);
|
|
|
|
return fd[0];
|
|
|
|
}
|
|
|
|
|
|
|
|
/* read a first byte from fd, put it into *len. Then read *len
|
|
|
|
* bytes from fd and put it into string.
|
|
|
|
* Return -1 if eof or read error or bad len, else return 0.
|
|
|
|
*/
|
|
|
|
int dns_read_block(int fd, char *string, unsigned char *len) {
|
|
|
|
int read_result;
|
|
|
|
|
|
|
|
log(LOG_DEBUG,"dns_read_block(): Calling read to learn length (fd %d).", fd);
|
|
|
|
read_result = read(fd, len, 1);
|
|
|
|
log(LOG_DEBUG,"dns_read_block(): read finished, returned %d", read_result);
|
|
|
|
if (read_result < 0) {
|
|
|
|
log(LOG_INFO,"dns_read_block(): read len returned error");
|
|
|
|
return -1;
|
|
|
|
} else if (read_result == 0) {
|
|
|
|
log(LOG_INFO,"dns_read_block(): Encountered eof reading len");
|
|
|
|
return -1;
|
|
|
|
} else if (*len <= 0) {
|
|
|
|
log(LOG_INFO,"dns_read_block(): len not >0");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
log(LOG_DEBUG,"dns_read_block(): Calling read to get string, length %u.", *len);
|
|
|
|
read_result = read(fd, string, *len);
|
|
|
|
if (read_result < 0) {
|
|
|
|
log(LOG_INFO,"dns_read_block(): read string returned error");
|
|
|
|
return -1;
|
|
|
|
} else if (read_result == 0) {
|
|
|
|
log(LOG_INFO,"dns_read_block(): Encountered eof reading string");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
string[*len] = 0; /* null terminate it, just in case */
|
2003-02-06 09:00:49 +01:00
|
|
|
// log(LOG_INFO,"dns_read_block(): Read '%s', len %u.",string,*len);
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* write ("%c%s", string, len) onto fd */
|
|
|
|
static int dns_write_block(int fd, char *string, unsigned char len) {
|
|
|
|
int write_result;
|
|
|
|
int written=0;
|
|
|
|
char tmp[257];
|
|
|
|
|
|
|
|
assert(len <= 250);
|
|
|
|
tmp[0] = len;
|
|
|
|
memcpy(tmp+1, string, len);
|
|
|
|
log(LOG_DEBUG,"dns_write_block(): writing length %u, fd %d.", len, fd);
|
|
|
|
|
|
|
|
while(written < len+1) {
|
|
|
|
write_result = write(fd, tmp, len+1-written);
|
|
|
|
if (write_result < 0) {
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
written += write_result;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* pull in question. block until we've read everything.
|
|
|
|
* return -1 if eof. */
|
|
|
|
static int dns_read_tor_question(int index) {
|
|
|
|
|
|
|
|
log(LOG_DEBUG,"dns_read_tor_question(): Pulling question from tor");
|
|
|
|
if(dns_read_block(slave_data[index].fd,
|
|
|
|
slave_data[index].question,
|
|
|
|
&slave_data[index].question_len) < 0)
|
|
|
|
return -1;
|
|
|
|
|
2003-02-06 09:00:49 +01:00
|
|
|
// log(LOG_INFO,"dns_read_tor_question(): Read question '%s'",slave_data[index].question);
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* pull in answer. block until we've read it. return -1 if eof. */
|
|
|
|
static int dns_read_slave_response(int index) {
|
|
|
|
|
|
|
|
if(dns_read_block(slave_data[index].fd,
|
|
|
|
slave_data[index].answer,
|
|
|
|
&slave_data[index].answer_len) < 0)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dns_find_idle_slave(int max) {
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for(i=1;i<max;i++)
|
|
|
|
if(slave_data[i].busy == 0) {
|
|
|
|
log(LOG_DEBUG,"dns_find_idle_slave(): slave %d is chosen.",i);
|
|
|
|
return i;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(0); /* should never get here */
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dns_assign_to_slave(int from, int to) {
|
|
|
|
|
|
|
|
slave_data[to].question_len = slave_data[from].question_len;
|
|
|
|
memcpy(slave_data[to].question, slave_data[from].question, slave_data[from].question_len);
|
|
|
|
|
|
|
|
// slave_data[from].question_len = 0;
|
|
|
|
|
|
|
|
log(LOG_DEBUG,"dns_assign_to_slave(): from index %d to %d (writing fd %d)",from,to,slave_data[to].fd);
|
|
|
|
if(dns_write_block(slave_data[to].fd,
|
|
|
|
slave_data[to].question,
|
|
|
|
slave_data[to].question_len) < 0) {
|
|
|
|
log(LOG_INFO,"dns_assign_to_slave(): writing to slave failed.");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dns_master_to_tor(int from, int to) {
|
2003-02-14 08:53:55 +01:00
|
|
|
char tmp[MAX_ADDRESSLEN*2];
|
major overhaul: dns slave subsystem, topics
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
2003-01-26 10:02:24 +01:00
|
|
|
unsigned char len;
|
|
|
|
|
|
|
|
len = slave_data[from].question_len+1+slave_data[from].answer_len;
|
|
|
|
memcpy(tmp, slave_data[from].question, slave_data[from].question_len);
|
|
|
|
tmp[slave_data[from].question_len] = 0; /* null terminate it */
|
|
|
|
memcpy(tmp+1+slave_data[from].question_len, slave_data[from].answer, slave_data[from].answer_len);
|
|
|
|
|
|
|
|
log(LOG_DEBUG,"dns_master_to_tor(): question is '%s', length %d",slave_data[from].question,slave_data[from].question_len);
|
|
|
|
log(LOG_DEBUG,"dns_master_to_tor(): answer is %d %d %d %d",
|
|
|
|
slave_data[from].answer[0],
|
|
|
|
slave_data[from].answer[1],
|
|
|
|
slave_data[from].answer[2],
|
|
|
|
slave_data[from].answer[3]);
|
|
|
|
assert(slave_data[from].answer_len == 4);
|
|
|
|
if(dns_write_block(slave_data[to].fd, tmp, len) < 0) {
|
|
|
|
log(LOG_INFO,"dns_master_to_tor(): writing to tor failed.");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void dns_master_main(int fd) {
|
|
|
|
int nfds=1; /* the 0th index is the tor process, the rest are slaves */
|
|
|
|
int num_slaves_busy=0;
|
|
|
|
int num_slaves_needed = MIN_DNSSLAVES;
|
|
|
|
int poll_result, idle, i;
|
|
|
|
|
|
|
|
poll_data[0].fd = slave_data[0].fd = fd;
|
|
|
|
poll_data[0].events = POLLIN;
|
|
|
|
|
|
|
|
for(;;) { /* loop forever */
|
|
|
|
|
|
|
|
assert(num_slaves_needed < MAX_DNSSLAVES);
|
|
|
|
while(nfds-1 < num_slaves_needed) {
|
|
|
|
/* add another slave. */
|
|
|
|
|
|
|
|
i = nfds;
|
|
|
|
memset(&slave_data[i], 0, sizeof(struct slave_data_t));
|
|
|
|
memset(&poll_data[i], 0, sizeof(struct pollfd));
|
|
|
|
slave_data[i].fd = poll_data[i].fd = dns_spawn_slave();
|
|
|
|
poll_data[i].events = POLLIN; /* listen always, to prevent accidental deadlock */
|
|
|
|
nfds++;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* XXX later, decide on a timeout value, to catch wedged slaves */
|
|
|
|
|
|
|
|
poll_result = poll(poll_data, nfds, -1);
|
|
|
|
log(LOG_DEBUG,"dns_master_main(): Poll returned -- activity!");
|
|
|
|
for(i=0;i<nfds;i++) {
|
|
|
|
if(poll_data[i].revents & POLLIN) {
|
|
|
|
if(i==0) { /* note that we read only one question per poll loop */
|
|
|
|
if(dns_read_tor_question(i) >= 0) {
|
|
|
|
while(1) {
|
|
|
|
idle = dns_find_idle_slave(nfds);
|
|
|
|
if(dns_assign_to_slave(i, idle) >= 0)
|
|
|
|
break; /* successfully assigned to one */
|
|
|
|
/* XXX slave must die, recalc num slaves and num busy */
|
|
|
|
}
|
|
|
|
num_slaves_busy++;
|
|
|
|
} else { /* error */
|
|
|
|
log(LOG_INFO,"dns_master_main(): dns_read_tor_question failed. Master dying.");
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if(dns_read_slave_response(i) >= 0) {
|
|
|
|
if(dns_master_to_tor(i, 0) < 0) {
|
|
|
|
log(LOG_INFO,"dns_master_main(): dns_master_to_tor failed. Master dying.");
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
slave_data[i].busy = 0;
|
|
|
|
num_slaves_busy--;
|
|
|
|
poll_data[0].events = POLLIN; /* resume reading from tor if we'd stopped */
|
|
|
|
} else { /* error */
|
|
|
|
log(LOG_INFO,"dns_master_main(): dns_read_slave_response failed. Leaving slave stranded (FIXME)");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
log(LOG_DEBUG,"dns_master_main(): Finished looping over fd's.");
|
|
|
|
|
|
|
|
if(num_slaves_busy >= num_slaves_needed) {
|
|
|
|
if(num_slaves_needed == MAX_DNSSLAVES-1)
|
|
|
|
poll_data[0].events = 0; /* stop reading from tor */
|
|
|
|
else
|
|
|
|
num_slaves_needed++;
|
|
|
|
}
|
|
|
|
|
|
|
|
}
|
|
|
|
assert(0); /* should never get here */
|
|
|
|
}
|
|
|
|
|
2003-02-14 08:53:55 +01:00
|
|
|
|
|
|
|
|
|
|
|
#include "tree.h"
|
|
|
|
|
|
|
|
struct pending_connection_t {
|
|
|
|
struct connection_t *conn;
|
|
|
|
struct pending_connection_t *next;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct cached_resolve {
|
|
|
|
SPLAY_ENTRY(cached_resolve) node;
|
|
|
|
char question[MAX_ADDRESSLEN]; /* the hostname to be resolved */
|
|
|
|
uint32_t answer; /* in host order. I know I'm horrible for assuming ipv4 */
|
|
|
|
char state; /* 0 is pending; 1 means answer is valid; 2 means resolve failed */
|
|
|
|
#define CACHE_STATE_PENDING 0
|
|
|
|
#define CACHE_STATE_VALID 1
|
|
|
|
#define CACHE_STATE_FAILED 2
|
|
|
|
uint32_t expire; /* remove untouched items from cache after some time? */
|
|
|
|
struct pending_connection_t *pending_connections;
|
|
|
|
struct cached_resolve *next;
|
|
|
|
};
|
|
|
|
|
|
|
|
SPLAY_HEAD(cache_tree, cached_resolve) cache_root;
|
|
|
|
|
|
|
|
static int compare_cached_resolves(struct cached_resolve *a, struct cached_resolve *b) {
|
|
|
|
/* make this smarter one day? */
|
|
|
|
return strncasecmp(a->question, b->question, MAX_ADDRESSLEN);
|
|
|
|
}
|
|
|
|
|
|
|
|
SPLAY_PROTOTYPE(cache_tree, cached_resolve, node, compare_cached_resolves);
|
|
|
|
SPLAY_GENERATE(cache_tree, cached_resolve, node, compare_cached_resolves);
|
|
|
|
|
|
|
|
void init_cache_tree(void) {
|
|
|
|
SPLAY_INIT(&cache_root);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* see if the question 'exitconn->address' has been answered. if so,
|
|
|
|
* if resolve valid, put it into exitconn->addr and call
|
|
|
|
* connection_exit_connect directly. If resolve failed, return -1.
|
|
|
|
*
|
|
|
|
* Else, if seen before and pending, add conn to the pending list,
|
|
|
|
* and return 0.
|
|
|
|
*
|
|
|
|
* Else, if not seen before, add conn to pending list, hand to
|
|
|
|
* dns farm, and return 0.
|
|
|
|
*/
|
|
|
|
int dns_resolve(connection_t *exitconn) {
|
|
|
|
struct cached_resolve *new_resolve;
|
|
|
|
struct cached_resolve *resolve;
|
|
|
|
struct pending_connection_t *pending_connection;
|
|
|
|
|
|
|
|
new_resolve = malloc(sizeof(struct cached_resolve));
|
|
|
|
memset(new_resolve, 0, sizeof(struct cached_resolve));
|
|
|
|
strncpy(new_resolve->question, exitconn->address, MAX_ADDRESSLEN);
|
|
|
|
|
|
|
|
/* try adding it to the tree. if it's already there it will
|
|
|
|
* return it. */
|
|
|
|
resolve = SPLAY_INSERT(cache_tree, &cache_root, new_resolve);
|
|
|
|
if(resolve) { /* already there. free up new_resolve */
|
|
|
|
free(new_resolve);
|
|
|
|
switch(resolve->state) {
|
|
|
|
case CACHE_STATE_PENDING:
|
|
|
|
/* add us to the pending list */
|
|
|
|
pending_connection = malloc(sizeof(struct pending_connection_t));
|
|
|
|
pending_connection->conn = exitconn;
|
|
|
|
pending_connection->next = new_resolve->pending_connections;
|
|
|
|
new_resolve->pending_connections = pending_connection;
|
|
|
|
|
|
|
|
return dns_tor_to_master(exitconn);
|
|
|
|
case CACHE_STATE_VALID:
|
|
|
|
exitconn->addr = resolve->answer;
|
|
|
|
return connection_exit_connect(exitconn);
|
|
|
|
case CACHE_STATE_FAILED:
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
} else { /* this was newly added to the tree. ask the dns farm. */
|
|
|
|
new_resolve->state = CACHE_STATE_PENDING;
|
|
|
|
|
|
|
|
/* add us to the pending list */
|
|
|
|
pending_connection = malloc(sizeof(struct pending_connection_t));
|
|
|
|
pending_connection->conn = exitconn;
|
|
|
|
pending_connection->next = new_resolve->pending_connections;
|
|
|
|
new_resolve->pending_connections = pending_connection;
|
|
|
|
|
|
|
|
return dns_tor_to_master(exitconn);
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(0);
|
|
|
|
return 0; /* not reached; keep gcc happy */
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dns_tor_to_master(connection_t *exitconn) {
|
|
|
|
connection_t *dnsconn;
|
|
|
|
unsigned char len;
|
|
|
|
int do_dns_directly=0;
|
|
|
|
|
|
|
|
dnsconn = connection_get_by_type(CONN_TYPE_DNSMASTER);
|
|
|
|
|
|
|
|
if(!dnsconn) {
|
|
|
|
log(LOG_ERR,"dns_tor_to_master(): dns master nowhere to be found!");
|
|
|
|
}
|
|
|
|
|
|
|
|
if(!dnsconn || do_dns_directly) {
|
|
|
|
/* new version which does it all right here */
|
|
|
|
struct hostent *rent;
|
|
|
|
rent = gethostbyname(exitconn->address);
|
|
|
|
if (!rent) {
|
|
|
|
return dns_found_answer(exitconn->address, 0, 0);
|
|
|
|
}
|
|
|
|
return dns_found_answer(exitconn->address, *(uint32_t *)rent->h_addr, 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
len = strlen(exitconn->address);
|
|
|
|
if(connection_write_to_buf(&len, 1, dnsconn) < 0) {
|
|
|
|
log(LOG_DEBUG,"dns_tor_to_master(): Couldn't write length.");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(connection_write_to_buf(exitconn->address, len, dnsconn) < 0) {
|
|
|
|
log(LOG_DEBUG,"dns_tor_to_master(): Couldn't write address.");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
// log(LOG_DEBUG,"dns_tor_to_master(): submitted '%s'", address);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dns_found_answer(char *question, uint32_t answer, uint32_t valid) {
|
|
|
|
struct pending_connection_t *pend;
|
|
|
|
struct cached_resolve search;
|
|
|
|
struct cached_resolve *resolve;
|
|
|
|
|
|
|
|
strncpy(search.question, question, MAX_ADDRESSLEN);
|
|
|
|
|
|
|
|
resolve = SPLAY_FIND(cache_tree, &cache_root, &search);
|
|
|
|
if(!resolve) {
|
|
|
|
log(LOG_ERR,"dns_found_answer(): Answer to unasked question '%s'? Dropping.", question);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(resolve->state == CACHE_STATE_PENDING);
|
|
|
|
|
|
|
|
if(valid)
|
|
|
|
resolve->state = CACHE_STATE_VALID;
|
|
|
|
else
|
|
|
|
resolve->state = CACHE_STATE_FAILED;
|
|
|
|
|
|
|
|
while(resolve->pending_connections) {
|
|
|
|
pend = resolve->pending_connections;
|
|
|
|
pend->conn->addr = ntohl(answer);
|
|
|
|
if(resolve->state == CACHE_STATE_FAILED || connection_exit_connect(pend->conn) < 0) {
|
|
|
|
pend->conn->marked_for_close = 1;
|
|
|
|
}
|
|
|
|
resolve->pending_connections = pend->next;
|
|
|
|
free(pend);
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|