Mailing-List: contact users-help@tomcat.apache.org; run by ezmlm
Precedence: bulk
Reply-To: "Tomcat Users List" <users@tomcat.apache.org>
Subject: Re: Apache/Tomcat vulnerability
To: users@tomcat.apache.org
References: <CAPi8w0MYFr5zs8DSXHjUW3+Z592zTBLiHAyVVZPbMKwfUu4ccw@mail.gmail.com>
 <CA+7hzVPa_p2xD=zerxGYkJVmo58T23Df_gdkckJ52e78xJSTpQ@mail.gmail.com>
 <CAPi8w0NPoBC8z4+ctU38PyKzXFdS8C5rTXX3XUmrTCGwU_D+zA@mail.gmail.com>
 <583AC79E.3090804@ice-sa.com>
 <CAPi8w0P-YJv56CduPYNPfHJ__LvxLLUaZT0HyJ0NUeKyeHPLiA@mail.gmail.com>
 <583AD893.2060804@ice-sa.com>
 <CAPi8w0OF3TXc2O45FCZanYedvVSO_HRK5nNwrxWACPrPHs+_hw@mail.gmail.com>
From: =?UTF-8?Q?Andr=c3=a9_Warnier_=28tomcat=29?= <aw@ice-sa.com>
Message-ID: <583B182A.7030508@ice-sa.com>
Date: Sun, 27 Nov 2016 18:30:18 +0100
User-Agent: Mozilla/5.0 (Windows NT 5.1; rv:38.0) Gecko/20100101
 Thunderbird/38.5.0
MIME-Version: 1.0
In-Reply-To: <CAPi8w0OF3TXc2O45FCZanYedvVSO_HRK5nNwrxWACPrPHs+_hw@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
archived-at: Sun, 27 Nov 2016 17:30:40 -0000

On 27.11.2016 14:26, Jaaz Portal wrote:
> hi,
> everything i know so far is just this single log line that appeared in
> apache error.log
>
> [Fri Nov 25 13:08:00.647835 2016] [mpm_event:error] [pid 13385:tid
> 1397934896385
> 92] AH00484: server reached MaxRequestWorkers setting, consider raising the
> MaxR
> equestWorkers setting
>
> there was nothing else, just this strange line

This is not a "strange" line. It is telling you something.
One problem is that you seem convinced in advance, without serious proof, that there is a 
"bug" or a vulnerability in httpd or tomcat.
Read the explanation of the httpd parameter, here :
http://httpd.apache.org/docs/2.4/mod/mpm_common.html#maxrequestworkers
and try to understand it.

If Apache httpd is writing that line to the log, it means that it has reached the maximum 
number of requests that it is allowed to serve at the same time, considering your 
configuration. That could be normal - if your server is supposed to be heavily loaded - or 
it could be the sign of a "flood" attack. But you seem to be convinced in advance that 
there is no flood. You also do not really answer the questions which you are asked, which 
makes helping you difficult.

About the "bug" or "vulnerability" : a webserver is supposed to serve HTTP requests from 
clients.  That is what you expect of it. It has no choice but to accept any client 
connection and request, up to the maximum it can handle considering its configuration and 
the capacity of the system on which it runs. That is not a bug, it is a feature.

It does not know in advance that these requests are bad. To find out that they are bad, it 
needs to accept the request and read it. For that, there needs to be a process or a thread 
available to do that.
If it accepts the request, and if the URL of the request is in principle valid, it will 
try to process that request.  If there is an application on your server responding to that 
request, then if it takes a long time or freezes the thread which processes the request, 
it may be a bug in your application.

(Note : there /may/ be an unknown vulnerability in Apache httpd or tomcat; but is not 
necessarily so. And there are millions of Apache webservers on the Internet; so if they 
had a bug like that, you would not be the only one to report it.  So be humble and 
realistic; try your best to find out what really happens, before you cry wolf).

The normal Apache httpd access log, will log a request only when it is finished.  If the 
request never finishes, it will not get logged.
That may be why you do not see these requests in the log.
But have a look at this Apache httpd module : 
http://httpd.apache.org/docs/2.4/mod/mod_log_forensic.html

Note : that is also why I was telling you to enable the mod_jk log, and to examine it.
Because mod_jk will also log information before the request produces a response.


and server hanged.

Again, /what/ is "hanged" ? Apache httpd, or tomcat ?

Also in
> access logs there are no clues that it was under any heavy load.
>
> after around hour after discovering that our server hanged-out we have
> restarted tomcat server
> and it worked again

Yes, because that will close all connections between Apache httpd and tomcat, and abort 
all requests that are in the process of being processed by tomcat. So mod_jk will get an 
error from tomcat, and will report an error to httpd, and httpd will communicate that 
error to the clients, and close their connection.
It still does not tell you what the problem was.
The only thing that it suggests, is that the "bad" requests actually make it all the way 
to tomcat.

>
> i think that setting connection-timeout and limiting the number of clients
> by mod_bd i will
> have effect that next time somebody will try this exploit it will block his
> access to the site
> for minute or two, pretty good holistic solution i would say
>
> still, it seems that there is serious vulnerability somewhere in apache,
> mod_jk or tomcat
> i would like to help find it out but need some hints which debug options
> enable to catch the bad guys
> when they will try next time
>
> best regards,
> artur
>
> 2016-11-27 13:58 GMT+01:00 André Warnier (tomcat) <aw@ice-sa.com>:
>
>> On 27.11.2016 13:23, Jaaz Portal wrote:
>>
>>> hi Andre,
>>> thank you very much this was very educative but in my case it is little
>>> bit
>>> different.
>>> The server is no flooded, there is maybe dozen of very sophisticated
>>> connections that somehow hangs apache workers threads
>>>
>>
>> Can you be a bit more specific ?
>> When you say "apache workers threads", do you mean threads in Apache
>> httpd, or threads in Apache Tomcat ? (both are Apache webservers, so it is
>> difficult to tell what you are talking about, unless you are very precise).
>>
>> Let me give you some additional explanations, and maybe you can figure out
>> exactly where it "hangs".
>>
>>  From the Apache httpd front-end point of view, mod_jk (the connector to
>> Apache Tomcat) is basically one among other "response generators".  Apache
>> httpd does not "know" that behind mod_jk, there is one or more Tomcat
>> servers.  Apache httpd receives the original client request, and depending
>> on the URL of the request, it will pass it to mod_jk or to another response
>> generator, to generate the response to the request.
>> That mod_jk in the background is using a Tomcat server to actually
>> generate the response, is none of the business of Apache httpd, and it does
>> not care. All it cares about, is to actually receive the response from
>> mod_jk, and pass it back to the client.
>>
>> If httpd passes a request to mod_jk, it is because you have specified in
>> the Apache configuration, the type of URL that it should pass to mod_jk.
>> That happens via your "JkMount (URL pattern)" directives in Apache httpd.
>>
>> Of course Apache httpd will not pass a request to mod_jk, before it has
>> received at least the URL of the request, and analysed it to determine *if*
>> it should pass it to mod_jk (*).
>>
>> If the mod_jk logging is enabled, you can see in it, exactly *which*
>> requests are passed to mod_jk and to Tomcat.
>> Do you know *which* requests, from which clients, cause this "thread
>> hanging" symptom ?
>> Once you would know this, maybe you can design a strategy to block
>> specifically these requests.
>>
>> and the effect is permanent. Quickly the pool is exhausted
>>>
>>
>> which pool exactly ?
>>
>> and the only
>>
>>> solution that works is to restart tomcat.
>>>
>>> I think it is a bug similar to this one from mod_proxy:
>>> https://tools.cisco.com/security/center/viewAlert.x?alertId=34971
>>>
>>>
>> Maybe, maybe not. As long as we do not know what the requests are, that
>> block things, we do not know this.
>>
>>
>> I think also that your solution with setting connectionTimeout will solve
>>> the problem, at least partially. THANK YOU.
>>>
>>
>> Same thing, we do not know this yet.  It was only giving this explanation,
>> to help you think about where the problem may be.
>>
>>
>>> I would like to help you further investigate this issue, as our server
>>> comes under such attack once or twice in a week.
>>>
>>>
>> Other than giving you hints, there is not much I or anyone else can do to
>> help. You are the one with control of your servers and logfiles, so you
>> have to investigate and provide more precise information.
>>
>>
>> (*) actually, to be precise, Apache httpd passes *all* requests to mod_jk,
>> to ask it "if it wants that request". mod_jk then accepts or declines,
>> depending on the JkMount instructions. If mod_jk declines, then Apache
>> httpd will ask the next response generator if it wants this request, etc...
>>
>>
>>
>>
>>
>> best regards,
>>> artur
>>>
>>> 2016-11-27 12:46 GMT+01:00 André Warnier (tomcat) <aw@ice-sa.com>:
>>>
>>> Hi.
>>>>
>>>> Have a look that the indicated parameters in the two pages below.
>>>> You may be the target of such a variant of DDoS attack : many clients
>>>> open
>>>> a TCP connection to your server (front-end), but then never sends a HTTP
>>>> request on that connection.  In the meantime, the server accepts the TCP
>>>> connection, and passes it on to a "child" process or thread for
>>>> processing.  The child then waits for the HTTP request line to arrive on
>>>> the connection (during a certain time), but it never arrives. After a
>>>> while, this triggers a timeout (see below), but the standard value of
>>>> that
>>>> timeout may be such that in the meantime, a lot of other connections have
>>>> been established by other such nefarious clients, so a lot of resources
>>>> of
>>>> the webserver are tied up, waiting for something that will never come.
>>>> Since there is never any real request sent on the connection, you would
>>>> (probably) not see this in the logs either.
>>>>
>>>> The above is the basic mechanism of such an attack.  There may be
>>>> variations, such as the client not "not sending" a request line, but
>>>> sending it extremely slowly, thus achieving perhaps similar kinds of
>>>> effects.
>>>>
>>>> As someone pointed out, it is quite difficult to do something about this
>>>> at the level of the webserver itself, because by the time you would do
>>>> something about it, the resources have already been consumed and your
>>>> server is probably already overloaded.
>>>> There are specialised front-end devices and software available, to detect
>>>> and protect against this kind of attack.
>>>>
>>>> You may want to have a look at the following parameters, but make sure to
>>>> read the caveats (side-effects, interlocking timeouts etc.), otherwise
>>>> you
>>>> may do more harm than good.
>>>>
>>>> Another thing : the settings below are for Apache Tomcat, which in your
>>>> case is the back-end. It would of course be much better to detect and
>>>> eliminate this at the front-end, or even before.  I had a look at the
>>>> Apache httpd documentation, and could not find a corresponding parameter.
>>>> But I am sure that it must exist. You may want to post this same question
>>>> on the Apache httpd user's list for a better response.
>>>>
>>>> Tomcat configuration settings :
>>>>
>>>> AJP Connector : (http://tomcat.apache.org/tomc
>>>> at-8.5-doc/config/ajp.html#
>>>> Standard_Implementations)
>>>>
>>>> connectionTimeout
>>>>
>>>> The number of milliseconds this Connector will wait, after accepting a
>>>> connection, for the request URI line to be presented. The default value
>>>> for
>>>> AJP protocol connectors is -1 (i.e. infinite).
>>>>
>>>> (You could for example try to set this to 3000 (milliseconds) or even
>>>> lower. That should be more than enough for any legitimate client so send
>>>> the HTTP request line.  Note however that by doing this at the Tomcat
>>>> level, you will probably move the problem to the Apache httpd/mod_jk
>>>> level.  But at least it might confirm that this is the problem that you
>>>> are
>>>> seeing.  The mod_jk logfile at the httpd level may give you some hints
>>>> there.)
>>>>
>>>>
>>>> HTTP Connector : (http://tomcat.apache.org/tomc
>>>> at-8.5-doc/config/http.html#Standard_Implementation)
>>>>
>>>> connectionTimeout
>>>>
>>>> The number of milliseconds this Connector will wait, after accepting a
>>>> connection, for the request URI line to be presented. Use a value of -1
>>>> to
>>>> indicate no (i.e. infinite) timeout. The default value is 60000 (i.e. 60
>>>> seconds) but note that the standard server.xml that ships with Tomcat
>>>> sets
>>>> this to 20000 (i.e. 20 seconds). Unless disableUploadTimeout is set to
>>>> false, this timeout will also be used when reading the request body (if
>>>> any).
>>>>
>>>>
>>>>
>>>> On 26.11.2016 09:57, Jaaz Portal wrote:
>>>>
>>>> hi,
>>>>> sorry, its mod_jk no jk2, my typo. All at latest versions. We tried with
>>>>> mod proxy too.
>>>>>
>>>>> There is no flood of the server. Nobody is flooding us, they use some
>>>>> specific connections after which pool of apache workers is exhausted and
>>>>> blocked
>>>>> and we need to restart tomcat server.
>>>>> It is some kind of exploit but do not know how to log it to obtain
>>>>> details.
>>>>>
>>>>> i had put a limit on connections per client with hope that this will
>>>>> help
>>>>> but once again, it is not a flood.
>>>>> They open several connections that are not dropped by apache when they
>>>>> disconnect. This way whole pool is quickly exhausted and the server
>>>>> broken.
>>>>>
>>>>> i would like to help you to figure details of this attack but this is
>>>>> production server so it is impossible to much debugging options
>>>>>
>>>>> best,
>>>>> artur
>>>>>
>>>>> 2016-11-25 23:44 GMT+01:00 Niranjan Babu Bommu <
>>>>> niranjan.bommu@gmail.com
>>>>>
>>>>>> :
>>>>>>
>>>>>
>>>>> you can find who is flooding site in apache access.log and block them in
>>>>>
>>>>>> firewall.
>>>>>>
>>>>>> ex to find the IP:
>>>>>>
>>>>>> cat /var/log/apache2/access.log |cut -d' ' -f1 |sort |uniq -c|sort -gr
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Fri, Nov 25, 2016 at 8:42 AM, Jaaz Portal <jaazportal@gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>> hi,
>>>>>>
>>>>>>> we are from some weeks struggling with some Polish hackers that are
>>>>>>> bringing our server down. After updating apache to latest version
>>>>>>>
>>>>>>> (2.4.23)
>>>>>>
>>>>>> and tomcat (8.0.38) available for debian systems we still cannot secure
>>>>>>>
>>>>>>> our
>>>>>>
>>>>>> server.
>>>>>>>
>>>>>>> Today it has stopped to respond again and we needed to restart tomcat
>>>>>>> process to get it back alive.
>>>>>>>
>>>>>>> There is no too much clues in the logs. The apache error.log gives
>>>>>>> just
>>>>>>> this line:
>>>>>>>
>>>>>>> [Fri Nov 25 13:08:00.647835 2016] [mpm_event:error] [pid 13385:tid
>>>>>>> 1397934896385
>>>>>>> 92] AH00484: server reached MaxRequestWorkers setting, consider
>>>>>>> raising
>>>>>>>
>>>>>>> the
>>>>>>
>>>>>> MaxR
>>>>>>> equestWorkers setting
>>>>>>>
>>>>>>> seems that somehow tomcat, mod-jk2 or even apache is vulnerable to
>>>>>>> some
>>>>>>>
>>>>>>> new
>>>>>>
>>>>>> exploit, as we certainly does not have such traffic that would block
>>>>>>> our
>>>>>>> server otherwise
>>>>>>>
>>>>>>> for now we have increased MaxRequestWorkers and we have limited number
>>>>>>> of
>>>>>>> connections from one client to 5 by mod_bw and limited number of
>>>>>>> simultaneous connections from one ip by iptables but does not know if
>>>>>>>
>>>>>>> this
>>>>>>
>>>>>> will help
>>>>>>>
>>>>>>> best regards,
>>>>>>> artur
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>> --
>>>>>> *Thanks*
>>>>>> *Niranjan*
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>> ---------------------------------------------------------------------
>>>> To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
>>>> For additional commands, e-mail: users-help@tomcat.apache.org
>>>>
>>>>
>>>>
>>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
>> For additional commands, e-mail: users-help@tomcat.apache.org
>>
>>
>


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org