A bug? Socket broken due to resource leak in failed connect?


#1

It is really a triky problem.

Following code is used to illustrate the problem I encountered. Before
you execute it, please ensure to be offline, so it can be run very fast.
Please notice the line in tcpclient.rb “145.times do”, it is the key
issue.

Let me explain it.

Tcpserver.rb is a local tcp server simply sends a hello message whenever
a client connects and sends a request. Tcpclient.rb first makes a
connect to the local tcp server, then tries many times to connect to a
bad ip addresses which indeed can never be connected. So every try will
raise a exception " Unknown Error - connect(2)". This exception is
simply ignored and the client continues trying.

The wierd thing is , after certain number of tries, the established
connection between the tcp client and the local tcp server is broken.
So the line “from tcpclient.rb:19:in goodConn.puts” also raises a
exception . Even more,
what exception “goodConn.puts” raised is realted to the number of
tries to connect to the bad ip.

a. if times is less than 140 , it acts normally, nothing wrong, a hello
message is received from the local tcp server. I got this:

Exception raised: Unknown Error - connect(2)
Hello from local tcp server

b. If 140< times <145, I got this :

Exception raised: Unknown Error - connect(2)
tcpclient.rb:19:in write': Bad file descriptor (Errno::EBADF) from tcpclient.rb:19:inputs’
from tcpclient.rb:19:in `testsocket’
from tcpclient.rb:25

c. if times > 145, I got this:
Exception raised: Unknown Error - connect(2)
tcpclient.rb:19:in write': Invalid argument (Errno::EINVAL) from tcpclient.rb:19:inputs’
from tcpclient.rb:19:in `testsocket’
from tcpclient.rb:25

General speaking, it seems a resource leak happens every time a
“Unknown Error-connect(2)” exception raised when the tcpclient
failed to connnect to a bad ip address. So after substantial times
of tries, the leak accumulates and the socket is crushed. Is it a bug
of socket library??

Code is listed below, if you are insterested, please try it, run the
ftpserver.rb first, the run the tcpclient in another console. Be sure to
be offline when you try the code. Being offline will make it run fast,
otherwise it will take long time to run

tcpserver.rb:
require ‘socket’

server = TCPServer.new(‘localhost’,21)
while(session = server.accept)
puts “Client Request: #{session.gets}”
session.print “Hello from local tcp server”
session.close
end

tcpclient.rb:

require ‘socket’
def testsocket
goodAddr = “localhost”
goodConn = TCPSocket.new(goodAddr,21)

badAddr = “137.144.70.12” #no connection could established on this
address
145.times do
badconn = nil
begin
badconn = TCPSocket.new(badAddr,21)
rescue Exception => aException
puts “Exception raised: #{aException.to_s}”
ensure
badconn.close if @badconn and not @badconn.closed?
end
end

goodConn.puts(“hello”)
s = goodConn.gets
puts s
goodConn.close
end

testsocket

My OS is windows 98, ruby version is ruby
1.8.2 (2004-12-25) [i386-mswin32].


#2

Maybe this post is too long to read. Let me make it simple.

General speaking, I think it is likely a bug of ruby’s socket library.
It seems that a resource leak happens every time TCPSocket.new failed.

Do you agree it is a bug? Or I misunderstood something. Any opinion is
highly appreciated.


#3

Hi,

In message “Re: A bug? Socket broken due to resource leak in failed
conn”
on Wed, 17 May 2006 00:03:29 +0900, uncutstone wu
removed_email_address@domain.invalid writes:

|Maybe this post is too long to read. Let me make it simple.
|
|General speaking, I think it is likely a bug of ruby’s socket library.
|It seems that a resource leak happens every time TCPSocket.new failed.

I’m afraid that it’s a bug in Winsock library on Win98, since it
should never raise Unknown Error on any case.

						matz.

#4

Yukihiro M. wrote:

Hi,

In message “Re: A bug? Socket broken due to resource leak in failed
conn”
on Wed, 17 May 2006 00:03:29 +0900, uncutstone wu
removed_email_address@domain.invalid writes:

|Maybe this post is too long to read. Let me make it simple.
|
|General speaking, I think it is likely a bug of ruby’s socket library.
|It seems that a resource leak happens every time TCPSocket.new failed.

I’m afraid that it’s a bug in Winsock library on Win98, since it
should never raise Unknown Error on any case.

  					matz.

It’s my pleasure to get reply from you, the matz. :slight_smile:

Yes, I agree, it is very likely a win98 socket bug. I did think so
before I post this message.

I am writing a program which brute force scans a large group of IP
addresses to find ftp servers. Most of the ip addresses cannot be
connected. After cetain number of connection tries, the socket gets
broken.

How can I work around to this since it maybe a bug in the platform?

FYI, this program is part of a ftp search engine which I am developing
using ruby.


#5

uncutstone wu wrote:

I am writing a program which brute force scans a large group of IP
addresses to find ftp servers. Most of the ip addresses cannot be
connected. After cetain number of connection tries, the socket gets
broken.

How can I work around to this since it maybe a bug in the platform?

Maybe you could talk to the TCP socket at a lower protocol level than
connect, using raw IP packets. The source of nmap might be helpful.
(I’ve no idea whether win98 supports raw sockets, though…)