9:21 a.m.
is there any way to specify here more than one source-address
(the usual comma seperated way does not work in this context)
a complete ACCEPT before is no solution because it would bypass
any selective ACCEPT-rule
iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m recent --set
iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m recent
--update --seconds 1 --hitcount
75 -j REJECT --reject-with tcp-reset
12:46 p.m.
On Thu, May 03, 2012 at 04:21:20PM +0200, Reindl Harald wrote:
is there any way to specify here more than one source-address
(the usual comma seperated way does not work in this context)
a complete ACCEPT before is no solution because it would bypass
any selective ACCEPT-rule
iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m recent
--set
iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m recent
--update --seconds 1 --hitcount 75 -j REJECT --reject-with tcp-reset
Even when you use comma-separated addresses (allowed when not using
the '!' operator), iptables actually creates separate rules in
response to the command. I believe that's what you need to do in this
situation.
--
Paul W. Frields http://paul.frields.org/
gpg fingerprint: 3DA6 A0AC 6D58 FEC4 0233 5906 ACDB C937 BD11 3717
http://redhat.com/ - - - - http://pfrields.fedorapeople.org/
The open source story continues to grow: http://opensource.com
12:57 p.m.
Am 03.05.2012 19:46, schrieb Paul W. Frields:
On Thu, May 03, 2012 at 04:21:20PM +0200, Reindl Harald wrote:
> is there any way to specify here more than one source-address
> (the usual comma seperated way does not work in this context)
>
> a complete ACCEPT before is no solution because it would bypass
> any selective ACCEPT-rule
>
> iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m recent
--set
> iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m recent
--update --seconds 1 --hitcount 75 -j REJECT --reject-with tcp-reset
Even when you use comma-separated addresses (allowed when not using
the '!' operator), iptables actually creates separate rules in
response to the command. I believe that's what you need to do in this
situation
in theory yes
but practically the reject of this rule would be triggered
a secuity auditor from a customer is whining the he no longer
can make security-scans and it will get hard to arue that
we can not whitelist him in this case :-(
8:10 p.m.
On 2012/05/03 10:57, Reindl Harald wrote:
Am 03.05.2012 19:46, schrieb Paul W. Frields:
> On Thu, May 03, 2012 at 04:21:20PM +0200, Reindl Harald wrote:
>> is there any way to specify here more than one source-address
>> (the usual comma seperated way does not work in this context)
>>
>> a complete ACCEPT before is no solution because it would bypass
>> any selective ACCEPT-rule
>>
>> iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m
recent --set
>> iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m
recent --update --seconds 1 --hitcount 75 -j REJECT --reject-with tcp-reset
>
> Even when you use comma-separated addresses (allowed when not using
> the '!' operator), iptables actually creates separate rules in
> response to the command. I believe that's what you need to do in this
> situation
in theory yes
but practically the reject of this rule would be triggered
a secuity auditor from a customer is whining the he no longer
can make security-scans and it will get hard to arue that
we can not whitelist him in this case :-(
Ah, wait a minute. If he cannot make security scans neither can
anybody else. So defacto his job is finished.
For any exception you place into the rules to allow him to scan you must
think VERY carefully what it's effects will be. You might accidentally
open up the internal network to him leading to a false positive detection
from his security scan.
You might sit down with him and work out a plan for what should be done
so he can do his job and you can have the "recent" rule still protecting
your network. Collaboration and education may be your best friend here.
He is, after all, really an ally even when taking on the mantle of an
adversary for security auditing. Besides, you might get the delight of
seeing the lights go on in another person's head when he grasps just what
it is you did which is keeping him, and all others who look like malicious
access attempts, out of your system. Lead him gently to the knowledge and
the results can be more than worth your time and effort.
{^_^}
3:15 a.m.
Am 04.05.2012 03:10, schrieb jdow:
On 2012/05/03 10:57, Reindl Harald wrote:
>
> Am 03.05.2012 19:46, schrieb Paul W. Frields:
>> On Thu, May 03, 2012 at 04:21:20PM +0200, Reindl Harald wrote:
>>> iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m
recent --set
>>> iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW -m
recent --update --seconds 1
>>> --hitcount 75 -j REJECT --reject-with tcp-reset
>>
>> Even when you use comma-separated addresses (allowed when not using
>> the '!' operator), iptables actually creates separate rules in
>> response to the command. I believe that's what you need to do in this
>> situation
>
> in theory yes
> but practically the reject of this rule would be triggered
>
> a secuity auditor from a customer is whining the he no longer
> can make security-scans
Ah, wait a minute. If he cannot make security scans neither can
anybody else. So defacto his job is finished.
this is the naive view :-)
the reality in business-to-business relationships is that
such scans are done with standard software like Nessus
and if this stops the job is not done - if you are
speaking about a big customer with a policy which requires
sec-audits yu can not say "it is done"
a real attacker does not need nessus
he plays around with params manually to find sql-injections
and so on without triggering any rate-control and maybe
even bypassing mod_security or whatever application fierwall
For any exception you place into the rules to allow him to scan you
must
think VERY carefully what it's effects will be. You might accidentally
open up the internal network to him leading to a false positive detection
from his security scan.
i know this, that is the reason why i like to exclude him only
from the rate-control as also done with mod_security
4:37 a.m.
On 2012/05/04 01:15, Reindl Harald wrote:
Am 04.05.2012 03:10, schrieb jdow:
> On 2012/05/03 10:57, Reindl Harald wrote:
>>
>> Am 03.05.2012 19:46, schrieb Paul W. Frields:
>>> On Thu, May 03, 2012 at 04:21:20PM +0200, Reindl Harald wrote:
>>>> iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW
-m recent --set
>>>> iptables -I INPUT -p tcp -i eth0 ! -s $LOCAL_NETWORK -m state --state NEW
-m recent --update --seconds 1
>>>> --hitcount 75 -j REJECT --reject-with tcp-reset
>>>
>>> Even when you use comma-separated addresses (allowed when not using
>>> the '!' operator), iptables actually creates separate rules in
>>> response to the command. I believe that's what you need to do in this
>>> situation
>>
>> in theory yes
>> but practically the reject of this rule would be triggered
>>
>> a secuity auditor from a customer is whining the he no longer
>> can make security-scans
>
> Ah, wait a minute. If he cannot make security scans neither can
> anybody else. So defacto his job is finished.
this is the naive view :-)
the reality in business-to-business relationships is that
such scans are done with standard software like Nessus
and if this stops the job is not done - if you are
speaking about a big customer with a policy which requires
sec-audits yu can not say "it is done"
a real attacker does not need nessus
he plays around with params manually to find sql-injections
and so on without triggering any rate-control and maybe
even bypassing mod_security or whatever application fierwall
> For any exception you place into the rules to allow him to scan you must
> think VERY carefully what it's effects will be. You might accidentally
> open up the internal network to him leading to a false positive detection
> from his security scan.
i know this, that is the reason why i like to exclude him only
from the rate-control as also done with mod_security
So he's not REALLY testing your security profile, is he? He should be
nibbling around the edges, too.
But, then, I note your setting with --recent is not nearly as stringent as
mine. Any given address gets one connection per minute to ssh. That VASTLY
slows down dictionary attacks. Yours is a significant slow down; but, not
so much that somebody could not, as you put it, nibble around the edges to
get in. You have slowed down such attacks, though. That is good.
It would be handy if there was an iptables rule that allowed skipping the
next rule in order if the special rule hit. Alas, I am unaware of such a
trick potential.
{^_^}
4:57 a.m.
Am 04.05.2012 11:37, schrieb jdow:
But, then, I note your setting with --recent is not nearly as
stringent as
mine. Any given address gets one connection per minute to ssh. That VASTLY
slows down dictionary attacks. Yours is a significant slow down; but, not
so much that somebody could not, as you put it, nibble around the edges to
get in. You have slowed down such attacks, though. That is good.
It would be handy if there was an iptables rule that allowed skipping the
next rule in order if the special rule hit. Alas, I am unaware of such a
trick potential.
my sshd has a sepearte rule
the intention of this rule is not to block
it is a rate-control against DOS attacks
since we had "Anonymous" with a distributed DOS attack last
week i can say it works damned good - after replacing a
burned down router :-)
clearly you can not stand the whole DDOS from some thousand
source IPs but it gives you enough time to filter them for
a DROP rule - without this ratecontrol you could not
operate on the machine
before the DDOS it was limited to 100 connections/ip/second
which results in "ab -c 50 -n 50000 http://host-on-machine/"
raise CPU load up to 100% for a short time, go down to 50%
and changing between this both states (sorry baout bad english)
with 75 instead of 100 evebn a "ab -c 4 -n 1000" is completly
broken from outside the own network because "apache benchmark"
thinks the host is dead after 83 connections and stops due too
many errors - well, i guess exactly that is the problem for
Nessus/OpenVAS and such software from outside now
they triggered it all time before with portscans but only
not notice
5:31 p.m.
On 2012/05/04 02:57, Reindl Harald wrote:
Am 04.05.2012 11:37, schrieb jdow:
> But, then, I note your setting with --recent is not nearly as stringent as
> mine. Any given address gets one connection per minute to ssh. That VASTLY
> slows down dictionary attacks. Yours is a significant slow down; but, not
> so much that somebody could not, as you put it, nibble around the edges to
> get in. You have slowed down such attacks, though. That is good.
>
> It would be handy if there was an iptables rule that allowed skipping the
> next rule in order if the special rule hit. Alas, I am unaware of such a
> trick potential.
my sshd has a sepearte rule
the intention of this rule is not to block
it is a rate-control against DOS attacks
since we had "Anonymous" with a distributed DOS attack last
week i can say it works damned good - after replacing a
burned down router :-)
clearly you can not stand the whole DDOS from some thousand
source IPs but it gives you enough time to filter them for
a DROP rule - without this ratecontrol you could not
operate on the machine
before the DDOS it was limited to 100 connections/ip/second
which results in "ab -c 50 -n 50000 http://host-on-machine/"
raise CPU load up to 100% for a short time, go down to 50%
and changing between this both states (sorry baout bad english)
with 75 instead of 100 evebn a "ab -c 4 -n 1000" is completly
broken from outside the own network because "apache benchmark"
thinks the host is dead after 83 connections and stops due too
many errors - well, i guess exactly that is the problem for
Nessus/OpenVAS and such software from outside now
they triggered it all time before with portscans but only
not notice
What happens with something like this (PDL sorta kinda)?
while( 1 )
{
"ab -c 4 -n 50"
Sleep( 2 )
}
I don't know nessus. I am guessing that "-n 1000" part means 1000 trials
and it's running as fast as it can go. The idea is to test up to your
DDOS limit, wait 2 seconds, repeat. Can the test be hacked to keep your
system at its limit but not over its limit?
{^_^}
5:42 p.m.
Am 05.05.2012 00:31, schrieb jdow:
> with 75 instead of 100 evebn a "ab -c 4 -n 1000" is
completly
> broken from outside the own network because "apache benchmark"
> thinks the host is dead after 83 connections and stops due too
> many errors - well, i guess exactly that is the problem for
> Nessus/OpenVAS and such software from outside now
>
> they triggered it all time before with portscans but only
> not notice
What happens with something like this (PDL sorta kinda)?
while( 1 )
{
"ab -c 4 -n 50"
Sleep( 2 )
}
I don't know nessus. I am guessing that "-n 1000" part means 1000 trials
and it's running as fast as it can go. The idea is to test up to your
DDOS limit, wait 2 seconds, repeat. Can the test be hacked to keep your
system at its limit but not over its limit?
no idea, evenif it would not help becasue a company
only doing certified secsancs will never change them
especially if your customer is their customer....
but i found a solution!
with "--remove" you can remove the given IP from the iptables-list
before the REJECT action is triggered and this way add as much
networks / addresses you need
$IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent --set
$IPTABLES -I INPUT -p tcp -i eth0 -s $SECURITY_SCAN -m state --state NEW -m recent
--remove
$IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent --update
--seconds 1 --hitcount 75
-j REJECT --reject-with tcp-reset
$IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent --update
--seconds 1 --hitcount 75
-m limit --limit 60/h -j LOG --log-prefix "Rate-Control: "
7:31 p.m.
On 2012/05/04 15:42, Reindl Harald wrote:
Am 05.05.2012 00:31, schrieb jdow:
>> with 75 instead of 100 evebn a "ab -c 4 -n 1000" is completly
>> broken from outside the own network because "apache benchmark"
>> thinks the host is dead after 83 connections and stops due too
>> many errors - well, i guess exactly that is the problem for
>> Nessus/OpenVAS and such software from outside now
>>
>> they triggered it all time before with portscans but only
>> not notice
>
> What happens with something like this (PDL sorta kinda)?
>
> while( 1 )
> {
> "ab -c 4 -n 50"
> Sleep( 2 )
> }
>
> I don't know nessus. I am guessing that "-n 1000" part means 1000
trials
> and it's running as fast as it can go. The idea is to test up to your
> DDOS limit, wait 2 seconds, repeat. Can the test be hacked to keep your
> system at its limit but not over its limit?
no idea, evenif it would not help becasue a company
only doing certified secsancs will never change them
especially if your customer is their customer....
but i found a solution!
with "--remove" you can remove the given IP from the iptables-list
before the REJECT action is triggered and this way add as much
networks / addresses you need
$IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent --set
$IPTABLES -I INPUT -p tcp -i eth0 -s $SECURITY_SCAN -m state --state NEW -m recent
--remove
$IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent --update
--seconds 1 --hitcount 75
-j REJECT --reject-with tcp-reset
$IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent --update
--seconds 1 --hitcount 75
-m limit --limit 60/h -j LOG --log-prefix "Rate-Control: "
As long as that does not break other iptables based protections it's a
good enough solution. I presume you did audit the iptables setup for that
possibility.
Good luck with it.
(As an aside the scan company should learn to adapt as more and more
customers learn this trick and deploy it.)
{^_^}
1:42 a.m.
Am 05.05.2012 02:31, schrieb jdow:
>> I don't know nessus. I am guessing that "-n
1000" part means 1000 trials
>> and it's running as fast as it can go. The idea is to test up to your
>> DDOS limit, wait 2 seconds, repeat. Can the test be hacked to keep your
>> system at its limit but not over its limit?
>
> no idea, even if it would not help becasue a company
> only doing certified secsancs will never change them
> especially if your customer is their customer....
>
> but i found a solution!
>
> with "--remove" you can remove the given IP from the iptables-list
> before the REJECT action is triggered and this way add as much
> networks / addresses you need
>
>
> $IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent
--set
> $IPTABLES -I INPUT -p tcp -i eth0 -s $SECURITY_SCAN -m state --state NEW -m recent
--remove
> $IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent
--update --seconds 1 --hitcount 75
> -j REJECT --reject-with tcp-reset
> $IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent
--update --seconds 1 --hitcount 75
> -m limit --limit 60/h -j LOG --log-prefix "Rate-Control: "
As long as that does not break other iptables based protections it's a
good enough solution. I presume you did audit the iptables setup for that
possibility.
yes it is fine because it only affects "recent" rules with
"name: DEFAULT" and they are all on top of other rules
there are two others rejecting every connection for two seconds
after trying any port before interesting ones or typically
not public services to make portscans as hard as possible
/sbin/iptables --list --numeric --verbose | grep -v LOG | grep recent
75 4800 REJECT tcp -- eth0 * !LAN 0.0.0.0/0 state NEW
recent: UPDATE seconds:
1 hit_count: 75 name: DEFAULT side: source reject-with tcp-reset
64455 3866K tcp -- eth0 * SECSAN 0.0.0.0/0 state NEW
recent: REMOVE name:
DEFAULT side: source
157K 9125K tcp -- eth0 * !LAN 0.0.0.0/0 state NEW
recent: SET name:
DEFAULT side: source
0 0 REJECT tcp -- !lo * !LAN 0.0.0.0/0 recent:
CHECK seconds: 2 name:
portscan1 side: source reject-with tcp-reset
0 0 tcp -- !lo * !LAN 0.0.0.0/0 recent:
REMOVE name: portscan1
side: source
0 0 REJECT tcp -- !lo * !LAN 0.0.0.0/0 multiport
dports
19,24,52,79,109,142,442,464,548,586,631,992,994,3305 tcp recent: SET name: portscan1 side:
source reject-with tcp-reset
578 28816 REJECT tcp -- !lo * !LAN 0.0.0.0/0 recent:
CHECK seconds: 2 name:
portscan2 side: source reject-with tcp-reset
25 1280 tcp -- !lo * !LAN 0.0.0.0/0 recent:
REMOVE name: portscan2
side: source
338 16864 REJECT tcp -- !lo * !LAN 0.0.0.0/0 multiport
dports
23,137,138,139,445,3389,5900 tcp recent: SET name: portscan2 side: source reject-with
tcp-reset
(As an aside the scan company should learn to adapt as more and more
customers learn this trick and deploy it.)
i fear we are the first ones with such iptables rules and
good mod_security implementation (also had to whitelist)
they saw :-)
6:01 a.m.
On 2012/05/04 23:42, Reindl Harald wrote:
Am 05.05.2012 02:31, schrieb jdow:
>>> I don't know nessus. I am guessing that "-n 1000" part means
1000 trials
>>> and it's running as fast as it can go. The idea is to test up to your
>>> DDOS limit, wait 2 seconds, repeat. Can the test be hacked to keep your
>>> system at its limit but not over its limit?
>>
>> no idea, even if it would not help becasue a company
>> only doing certified secsancs will never change them
>> especially if your customer is their customer....
>>
>> but i found a solution!
>>
>> with "--remove" you can remove the given IP from the iptables-list
>> before the REJECT action is triggered and this way add as much
>> networks / addresses you need
>>
>>
>> $IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent
--set
>> $IPTABLES -I INPUT -p tcp -i eth0 -s $SECURITY_SCAN -m state --state NEW -m
recent --remove
>> $IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent
--update --seconds 1 --hitcount 75
>> -j REJECT --reject-with tcp-reset
>> $IPTABLES -I INPUT -p tcp -i eth0 ! -s $LAN_RANGE -m state --state NEW -m recent
--update --seconds 1 --hitcount 75
>> -m limit --limit 60/h -j LOG --log-prefix "Rate-Control: "
>
> As long as that does not break other iptables based protections it's a
> good enough solution. I presume you did audit the iptables setup for that
> possibility.
yes it is fine because it only affects "recent" rules with
"name: DEFAULT" and they are all on top of other rules
there are two others rejecting every connection for two seconds
after trying any port before interesting ones or typically
not public services to make portscans as hard as possible
/sbin/iptables --list --numeric --verbose | grep -v LOG | grep recent
75 4800 REJECT tcp -- eth0 * !LAN 0.0.0.0/0 state
NEW recent: UPDATE seconds:
1 hit_count: 75 name: DEFAULT side: source reject-with tcp-reset
64455 3866K tcp -- eth0 * SECSAN 0.0.0.0/0 state
NEW recent: REMOVE name:
DEFAULT side: source
157K 9125K tcp -- eth0 * !LAN 0.0.0.0/0 state
NEW recent: SET name:
DEFAULT side: source
0 0 REJECT tcp -- !lo * !LAN 0.0.0.0/0 recent:
CHECK seconds: 2 name:
portscan1 side: source reject-with tcp-reset
0 0 tcp -- !lo * !LAN 0.0.0.0/0 recent:
REMOVE name: portscan1
side: source
0 0 REJECT tcp -- !lo * !LAN 0.0.0.0/0
multiport dports
19,24,52,79,109,142,442,464,548,586,631,992,994,3305 tcp recent: SET name: portscan1
side: source reject-with tcp-reset
578 28816 REJECT tcp -- !lo * !LAN 0.0.0.0/0 recent:
CHECK seconds: 2 name:
portscan2 side: source reject-with tcp-reset
25 1280 tcp -- !lo * !LAN 0.0.0.0/0 recent:
REMOVE name: portscan2
side: source
338 16864 REJECT tcp -- !lo * !LAN 0.0.0.0/0
multiport dports
23,137,138,139,445,3389,5900 tcp recent: SET name: portscan2 side: source reject-with
tcp-reset
> (As an aside the scan company should learn to adapt as more and more
> customers learn this trick and deploy it.)
i fear we are the first ones with such iptables rules and
good mod_security implementation (also had to whitelist)
they saw :-)
It feels REALLY good when you can stay on top of the other professionals,
right? Hopefully that also is keeping you ahead of the malicious creeps.
Congratulations.
{^_^}
4371
days inactive
4373
days old
11 comments
3 participants
participants (3)
-
jdow -
Paul W. Frields -
Reindl Harald