How to swindle hoster or the tale of how to defeat greed.

I remembered one interesting story.
One time we bought dedicated server, when we made out the order i specified partition layout in additional comment section (raid 1 + lvm), when i got access to the server i saw that OS have one partition on LVM spliced over both drives.

We wrote email, and asked to reinstall OS with specified layout. The hoster answered that OS installed automatically, if we need another layout, then we must pay additional 100$. After that i started to google about how i can reinstall OS remotely, i found few complicated solution, but did not used it.

Also i remembered that we have IP KVM, i logged in (it was HP KVM), found that i can connect ISO to virtual cd-drive, tried it, but without result.

While i tried to use virtual cd, i rebooted host few times, while it was rebooting i saw that they used boot on lan and have installed gPXE. Unfortunately i did not found debian installation in their net boot environment, but i have deal with gPXE before and remembered that i can boot anything what i want over network thru gPXE console. I found my old gPXE script, modify it to boot debian installer, and taadaaam!

% chainload

I installed OS as i want and save money.
It was interesting, when i started to install OS i observed that they reinstall OS with new partition layout. One part of first HDD was used for OS, another part and second HDD was used for LVM, idiots..

How to transfer data between hosts securely.

like a bossFrom time to time i faced with task how to transfer important data between servers securely (ie over ssl or something similar). I do not use passwords for remote access and do not have private keys on remote systems, so i can not use ssh for this purposes.

First i wanted to write about solution that i used few days ago (based on socat), but this solution is to complicated (later you will see why). I remembered that openssl can encrypt files with password and send result to STDOUT. While i was reading manual for openssl, i found that openssl can be used like netcat (s_server and s_client commands), unfortunately i did not found way how to use openssl for data transfer, because in session that you can establish openssl interprets some chars as commands (R for renegotiation for example), so if you want to use openssl client/server for data transfer, you need something like base64 encoding, but without control characters.

Solution with netcat and openssl:
First i created file for test (you can use data from STDIN, tar output for example, or transfer existing file):

client$ dd if=/dev/urandom of=/tmp/rand bs=1M count=10
10+0 records in
10+0 records out
10485760 bytes (10 MB) copied, 1.36697 seconds, 7.7 MB/s
client$ md5sum /tmp/rand
10fe36edbbd48cde844ad1a2a29a8e0f  /tmp/rand

Next, prepare server side:

server$ read pass
server$ nc -l -p 6667|openssl aes-256-cbc -d -k $pass > /tmp/rand

I used “read” to prevent save key into history file.
There “PasSwOrD” is your key, i use ssh to organize data transfer, so i did not worried that the traffic with key can be captured.
Next initiate transfer from client side:

client$ read pass
client$ cat /tmp/rand |openssl  aes-256-cbc -salt -k $pass|nc -w1 server.remote 6667

Check sum:

server$ md5sum /tmp/rand
10fe36edbbd48cde844ad1a2a29a8e0f  /tmp/rand

Yeah! Your see? I transfered mah file.
UPDATE: [2019-05-26] Nowaday openssl has broken backward compatibility, so when you try to decrypt file you cold get error message like `digital envelope routines:EVP_DecryptFinal_ex:bad decrypt:crypto/evp/evp_enc.c:535`, if so you need to add -md md5 or -md sha256 on both sides to openssl’s options.

Ok, next we will do it with socat.
First you need to generate client side and server side key and certificates, let’s do it on server:

server$ openssl genrsa -out server.key 2048
Generating RSA private key, 2048 bit long modulus
e is 65537 (0x10001)

Create certificate:

server$  openssl req -new -key server.key -x509 -days 108 -batch -out server.crt

Create pem file:

server$ cat server.* > ./server.pem

After that you will need to execute same commands on client side, but you will need to change filenames from “server” to “client”.
Next step is to exchange certificates between client and server (do it on both sides), they could be copy pasted:

server$ cat > ./client.crt << EOF
.... a lot of garbage ....

Now we ready to transfer file, prepare server:

server$ socat openssl-listen:4433,reuseaddr,cert=./server.pem,cafile=./client.crt STDIO > /tmp/rand


client$ socat STDIO openssl-connect:server.remote:4433,cert=./client.pem,cafile=./server.crt < /tmp/rand
client$ md5sum /tmp/rand
10fe36edbbd48cde844ad1a2a29a8e0f  /tmp/rand


As you can see, socat with TLS not a easy solution if you need just a transfer file, so i will recommend to use first solution. Also, in debian, you can use snakeoil key and cert, but it is your homework.

I found how to use openssl for data transfer, only one problem, they did not close socket after EOF, so you need to stop it by hands:
Prepare server (this time i used snakeoil cert):

server% sudo openssl s_server -quiet -accept 4343 -cert /etc/ssl/certs/ssl-cert-snakeoil.pem -key /etc/ssl/private/ssl-cert-snakeoil.key < /tmp/test

Run client:

client% md5sum /tmp/rnd
86246865b3932804979fdac48a99cebf  /tmp/rnd
client% openssl s_client -connect localhost:4343 -quiet > /tmp/rnd

After data transfered, hit ^C on server side and check:

server% md5sum /tmp/test
86246865b3932804979fdac48a99cebf  /tmp/test

Bug in munin

bug in munin traffic graphFew weeks i observed strange graphs for network usage produced by munin, i did not attach any importance to this. But few days ago when i seen again 600Mbit badwidth usage on host that had 10Mbit connection i remembered that before made some changes in munin.conf. I looked at config and found that changed directive ‘graph_period’ to ‘minute’ before(do not remember why). When i change it back to ‘second’, i got normal graphs.

A riddle from abyss

A week ago, when i worked at home computer i encountered with strange issue, xorg was stuck, i tried ctrl-alt-backspace and magic keys without result, so i rebooted host.
While computer is booted i saw few errors from different daemons that they can not found some files and finally when i tried to login i got fresh login screen after entered password, again and again. Superuser was disabled, so i booted into singlemod but only one errer that i saw was ‘Unable to cd to ‘/home/ivan’ for user ‘ivan'”. Considering the suddenness of what happened, i thought that my HDD broken, i booted from livecd and checked s.m.a.r.t., hdd surface for bad blocks and did not found anything. After that my paranoia  took over and i started to search rootkits or something that can broke my OS and nothing found again.

I configured network from single mode, start xsession with browser and started searching, suddenly i found article when someone said that run install script and that script changed permissions on ‘/’, i looked at my ‘/’ and observed 0644. For me is still a mystery how it happened.

Apt pinning note

Some days ago i got additional VDS and wanted to install php5-fpm from dotdeb, but after i added dotdeb repository, apt started to try upgrade mysql and nginx from dotdeb. Solution to stop that:

$ cat > /etc/apt/preferences.d/dotdeb << EOF
Package: *
Pin: origin ""
Pin-Priority: 50

Next time if i will want to install something from dotdeb, i can use -t switch for apt-get or aptitude.

Tor blacklist

Few month ago i was interested in how to block incoming traffic from Tor network. Tor network have finitely numbers of exit nodes, so the solution is to block traffic from this nodes. I see two solutions how to block exit nodes.  First use TorDNSEL service, in brief you can check that connection comes from Tor exit node or not by querying special domain name. It is useful if you want to do this check in your php script, for example. But i do not know how to protect  host completely by this method. Another way is to use exitlist. By this way you can get list of addresses. Both ways have advantages and disadvantages. I prefer last solution because, it allow to protect my hosts completely.  Cons of this method – near 700-800 rules into firewall, that can slow down your host.

Anyway, first that i done – simply script that fetch list of IP and print them to standard output. When i started to writing script i wanted to expand features in future. I want to generate ready firewall rules for example, but now this script can only print list of IP’s. I do not want to integrate firewall rules application in this script, because i think this job must be done by another script.
So, now this script only fetch and print IP addresses or exit with error code if can not do it:

#!/usr/bin/perl -w
use strict;
use LWP::Simple;
my $list = get("");
my $i;
my @ips;
if( ! defined( $list ) ) {
die( "Can not fetch addresses.\n" );
foreach $i (split( /\n/, $list )) {
push( @ips, $1 ) if( $i =~ m/((?:\d{1,3}\.){3}\d{1,3})/);
if( defined( $ARGV[0] ) &&  $ARGV[0]  eq "-ip" ) {
print( join( "\n", @ips ));
} else {
die( "Usage $0 -ip\n" );

Next that you need – apply rules on firewall. As i say before, when you apply one rule per IP you will get near 700-800 rules, it is reason why you need optimize sequence of your rules. It can be done in next manner:
1. Rule to accept packets flowing thru loopback.
2. Rule to accept packets of RELATED and ESTABLISHED connections.
3. Rule to drop packets from blacklisted sources.
4. Rule to accept packets from whitelisted sources.
5. Rule for fail2ban (optional).
6. Rule for Tor exit nodes.
7. Other rules (for example, accept connection on ssh, http, https ports).

This sequence allow  to make decision for most packets before they pass Tor filtration rules.
I organize Tor filtration rules in separate table called TorExitnodes, it allow to update only this table and do not touch other firewall rules. Also i create table TorBlockAndLog that contain LOG and DROP target, but it is optional, you can simply drop packets.

For updating firewall rules for exit nodes i made next bash scrip and place it in cront:

if /usr/bin/perl $SOLVE
/sbin/iptables -F TorExitnodes
/sbin/iptables -I TorExitnodes -j RETURN
for i in `/usr/bin/perl $SOLVE -ip`
/sbin/iptables -I TorExitnodes -s $i -j TorBlockAndLog
echo "Can not fetch Tor exit nodes" 1>&2
exit 1;

Where is variable “SOLVE” store path to previous perl script, as i say before, if you do not need special action for incoming connections from Tor, you can change “TorBlockAndLog” on “DROP”.

Thats all.


It is good idea to place whitelist before TorExitnodes, who can know, may be one fine day you found -s -j DROP into this table. =)

Port based routing

After i came on new work i found that can not send email thru SMTPS, because port 465 closed on router. At this point i already had configured VPN access on my home router, so i think that it is good idea to route SMTPS traffic thru VPN, let’s start.

For this purposes i needed iproute2 and iptables. First i created new route table and add default route:

$ echo "1    VPN" >> /etc/iproute2/rt_tables
$ ip route add default via src dev tun_vpn table VPN
$ ip route show table VPN
default via dev tun_vpn src

Where – ip of my router into VPN and tun_vpn – VPN interface.
After that i created rule, that route marked packets thru VPN route table:

$ ip rule add from all fwmark 0x16 lookup VPN
$ ip  ru sh
0:      from all lookup local
32765:  from all fwmark 0x16 lookup VPN
32766:  from all lookup main
32767:  from all lookup default

There is time to mark SMTPS packets:

$ iptables -t mangle -I PREROUTING -p tcp --dport 465 -j MARK --set-mark 0x16
$ iptables -t mangle -I OUTPUT -p tcp --dport 465 -j MARK --set-mark 0x16

Let’s check:

$ traceroute -n -T -p 993
traceroute to (, 30 hops max, 60 byte packets
1  0.217 ms  0.233 ms  0.201 ms
2  *  2.318 ms  2.377 ms  2.503 ms
3  *  1.411 ms  1.714 ms  1.947 ms
4  *  1.486 ms  1.733 ms  1.796 ms
5  * 12.762 ms  12.791 ms  12.836 ms
6  * 65.528 ms  61.534 ms  67.431 ms
7  66.606 ms  61.808 ms  60.219 ms
8  66.225 ms  61.190 ms  66.038 ms
9  60.271 ms  60.510 ms  60.956 ms
10  65.304 ms  65.610 ms  76.697 ms
11  66.954 ms  65.824 ms  61.563 ms
$ traceroute -n -T -p 465
traceroute to (, 30 hops max, 60 byte packets
1  26.088 ms  42.767 ms  42.748 ms
2  *  42.813 ms  42.799 ms  68.297 ms
3  *  42.668 ms  42.665 ms  42.619 ms
4  * 42.539 ms  42.521 ms  42.504 ms
5  *  42.522 ms  68.071 ms  68.039 ms
6  *  68.015 ms  76.070 ms  85.085 ms
7  * 136.618 ms  136.634 ms  136.555 ms
8  110.732 ms  110.744 ms  110.712 ms
9  136.549 ms  136.506 ms  136.463 ms
10  *  137.978 ms  137.887 ms
11  137.893 ms  137.846 ms  130.177 ms

In my situation i observed strange effect, although that i set src ip, my host trying to send packets with src ip of local ethernet interface, so i just add masquerade rule into iptables.

$ iptables -A POSTROUTING -o tun_vpn -J MASQUERADE

Another way to fix it, set route to local work network on router, but i too lazy to do it.