> Which classes are you worried about as a vulnerability?
You certainly know that we have no serialization vulnerabilities OOTB. Most of the external classes which were still a problem for us one year ago
have been fixed.
I documented that at
https://cwiki.apache.org/confluence/display/OFBIZ/The+infamous+Java+serialization+vulnerability
But still 6 months later:
http://www.pcworld.com/article/3026678/paypal-is-the-latest-victim-of-java-deserialization-bugs-in-web-apps.html
So it's only about warning users about deserialization attacks. They are maybe not totally aware of the issue. How to warn them efficiently? It seems
to me that my proposition does that.
> Perhaps give an example within this context of exactly what happens and how
notsoserial prevents it.
The article about Paypal above states "Security researchers warned at the time that thousands of Java-based Web applications, *including custom-made
enterprise ones*, are likely vulnerable to this attack"
"[snip] it's likely that this type of vulnerability exists in other libraries as
well, waiting to be discovered."
This is not FUD (PayPal!) and it's only 6 months old
About RMI, JMX is using it http://engineering.pivotal.io/post/java-deserialization-jmx/ and if you want to dynamically monitor things, you will use
JMX, won't you?
About JNDI
http://zerothoughts.tumblr.com/post/137831000514/spring-framework-deserialization-rce
OK it's not about OFBiz, but those 2 are less than 6 months old. Who can think
that the problems is definitely resolved? It's not, it can't be.
So no, I have no examples to give you. But one thing I can say, is with notsoserial running in secure mode in the background you will be notified if
any unexpected deserialization attempt happens. https://github.com/kantega/notsoserial#rejecting-deserialization-entirely
I hope I have somehow answered your question. Can you answer mine?
Also, I wonder why both you and Jacopo reject my proposed solution, any
technical or other reasons?
We decided to use notsoserial by default, I don't see why we should change now. I'd have a reassured mind if we committed my proposed change. I see no
reasons to not do it.
Jacques
Le 08/09/2016 à 15:16, Taher Alkhateeb a écrit :
In order to answer your question we have to qualify it. What is your
definition within OFBiz of a deserialization attack. Perhaps give an
example within this context of exactly what happens and how notsoserial
prevents it. Which classes are you worried about as a vulnerability?
On Thursday, 8 September 2016, Jacques Le Roux <[email protected]>
wrote:
It's no only about RMI, deserialization driven attacks can be done on
vulnerable classes.
I think by exposing notsoserial (ie by creating the deserialize-trace.txt
and is-deserialized.txt files in tools\security\notsoserial) we offer a
good chance to users to we warned about the possible risks of
deserialization driven attacks.
It's else difficult for inexperienced users in the security domain to
understand they are at risk and how to cope with it. This is actually the
purpose of the tools\security\notsoserial folder and its content.
I ask you the same question than I asked to Jacopo: "How do you expect to
warn users about deserialization driven attacks?"
Jacques
Le 08/09/2016 à 12:12, Taher Alkhateeb a écrit :
I have a question. How many "naive" users will default to using RMI with
OFBiz?
I think it was agreed long ago that RMI is a bad idea, and things like
DCOM, CORBA and others died for a good reason. If someone is introducing
RMI, they should handle the serialization vulnerabilities according to
their own implementation, it's a choice. I really don't see a big need for
this to be integrated into OFBiz OOTB. Even if you expose notsoserial,
people still need to understand how it works and what to put into the
configured text files. In fact, the knowledge needed to execute
notsoserial
correctly is rather high because you need to know which Java files are
serializable and then which one of those are exposed in your
implementation.
So I recommend making this just an optional feature with good
documentation
in the Wiki where people can use it if they choose to adopt RMI, and we
get
this Jar out of the framework.
On Thu, Sep 8, 2016 at 1:04 PM, Jacques Le Roux <
[email protected]> wrote:
Jacopo,
I agree about releasing more often. Then we need to prepare things
better.
Just few words, I have no ideas how to yet.
You suggest "to implement the best solution". As I said below I see only
2. I don't expect Kantega will take the burden of pushing their jar to
jcenter, even less on each modification, even if I now expect very few,
if
any. They even did not care to answer me!
I don't like much the idea of maintaining a fork, who would do it? I
prefer to automate it, and Gradle with JitPackIt can. My last proposition
(answer to Taher) resolves the notsoserial problem.
If we remove the jar and all the rest, I fear the notsoserial effort will
be definitely thrown away, exposing our "naive" users at the risk of
using
RMI or a vulnerable external classes.
BTW, when you say "We could always bundle it in another release soon" do
you expect to freeze and release R16 very soon?
Jacques
Le 08/09/2016 à 08:47, Jacopo Cappellato a écrit :
I would feel more comfortable in releasing without it, and then work on
the
proposed solutions with more time in order to implement the best
solution.
We could always bundle it in another release soon: in fact with all the
activity in the project, we should consider releasing more often.
Jacopo
On Thu, Sep 8, 2016 at 8:23 AM, Jacques Le Roux <
[email protected]> wrote:
OK wait! I think we skipped a step: access the jar from an accessible
repo.
I see 2 solutions:
1. Push ourselves the notsoserial jar in jcenter => updating a fork now
and then, though notsoserial will not need much changes now
2. Use https://jitpack.io/
I prefer 2, though I have still to
* resolve the notsoserial.jar path in my hasty proposition below
* use|master-SNAPSHOT| instead of last hash (to not break on
changes)
which may need to refresh dependencies (did not investigate yet:
https://jitpack.io/docs/#snapshots)
Index: build.gradle
===================================================================
--- build.gradle (revision 1759557)
+++ build.gradle (working copy)
@@ -32,7 +32,7 @@
// java settings
def jvmArguments = ['-Xms128M', '-Xmx1024M',
- "-javaagent:${rootDir}/tools/security/notsoserial/notsoseria
l-1.0-SNAPSHOT.jar",
+ "-javaagent:C:/Users/Jacques/.gradle/caches/modules-2/files-
2.1/com.github.kantega/notsoserial/f2baaaa/3646e12f5ce4713f9
ad2aa027e3fec81097fcd93/notsoserial-f2baaaa.jar",
"-Dnotsoserial.whitelist=${rootDir}/tools/security/notsoseri
al/empty.txt",
"-Dnotsoserial.dryrun=${rootDir}/tools/security/notsoserial/
is-deserialized.txt",
"-Dnotsoserial.trace=${rootDir}/tools/security/notsoserial/
deserialize-trace.txt"]
@@ -52,6 +52,7 @@
allprojects {
repositories{
jcenter()
+ maven { url "https://jitpack.io"; }
}
}
@@ -119,6 +120,7 @@
compile 'org.zapodot:jackson-databind-java-optional:2.4.2'
compile 'oro:oro:2.0.8'
compile 'wsdl4j:wsdl4j:1.6.2'
+ compile 'com.github.kantega:notsoserial:f2baaaa'
// general framework runtime libs
runtime 'de.odysseus.juel:juel-spi:2.2.7'
I think you get the idea, it works w/o any other modifications, what to
you think?
Jacques
Le 07/09/2016 à 22:37, Jacques Le Roux a écrit :
OK, since we have no issues OOTB that can be done.
But IMO documenting the whole thing in our nososerial readme.txt is
not
enough. We need to make that more prominent. Not sure how yet...
Jacques
Le 07/09/2016 à 20:09, Taher Alkhateeb a écrit :
Scratch that, actually only the -D arguments are ignored, we must
remove
the -javaagent argument because it's not a classpath argument and
would
crash the VM
But for consistency's sake, let's remove them all for now. So simply
we
apply:
Index: build.gradle
===================================================================
--- build.gradle (revision 1759596)
+++ build.gradle (working copy)
@@ -31,11 +31,7 @@
ext.os = System.getProperty('os.name').toLowerCase()
// java settings
-def jvmArguments = ['-Xms128M', '-Xmx1024M',
-
"-javaagent:${rootDir}/tools/security/notsoserial/notsoseria
l-1.0-SNAPSHOT.jar",
-
"-Dnotsoserial.whitelist=${rootDir}/tools/security/notsoseri
al/empty.txt",
-
"-Dnotsoserial.dryrun=${rootDir}/tools/security/notsoserial/
is-deserialized.txt",
-
"-Dnotsoserial.trace=${rootDir}/tools/security/notsoserial/
deserialize-trace.txt"]
+def jvmArguments = ['-Xms128M', '-Xmx1024M']
ext.ofbizMainClass = 'org.apache.ofbiz.base.start.Start'
javadoc.failOnError = false
sourceCompatibility = '1.8'
On Wed, Sep 7, 2016 at 9:04 PM, Jacques Le Roux <
[email protected]> wrote:
OK Cool, if the JVM arguments are simply ignored, then I will proceed
with
an addition in the readme and remove the jar, simple
Jacques
Le 07/09/2016 à 17:16, Jacopo Cappellato a écrit :
Thank you Jacques and Taher.
So it seems we can move on and temporarily remove the jar.
Jacopo
On Wed, Sep 7, 2016 at 5:11 PM, Taher Alkhateeb <
[email protected]>
wrote:
Hi Jacques,
First of all the ofbizSecure task is gone instead everything calls
the
correct jvm arguments by default to fetch notsoserial.
The work to remove notsoserial is almost nothing. You just to
remove a
few
jvm args and that's it. Even if you don't remove the jvm args
nothing
happens because it will just ignore it as missing from the
classpath.
Taher Alkhateeb
On Sep 7, 2016 5:48 PM, "Jacques Le Roux" <
[email protected]>
wrote:
Huho, I was too fast on this. Currently the Gradle "ofbizSecure"
tasks
depends on the notsoserial-1.0-SNAPSHOT.jar
So this would need more work and w/o answers from them I suspect
they
will
not publish the jar.
Now it's a serious security but not OOTB. So I see 2
possibilities.
1. Ask the ASF for a derogation (after all it's a Java issue not
an
OFBiz
one)
2. Do what I said before AND change the Gradle "ofbizSecure"
tasks
Opinions?
Jacques
Le 07/09/2016 à 14:01, Jacques Le Roux a écrit :
Yes I see no problems with that. I just need to add directions
for
users
before. I'll then remove the jars... very soon...
Jacques
Le 07/09/2016 à 13:09, Jacopo Cappellato a écrit :
Jacques, any news from notsoserial?
If not, I think we can proceed by (temporarily) removing the
jars
until
they will publish the jar.
Regards,
Jacopo
On Sat, Aug 20, 2016 at 11:12 AM, Jacques Le Roux <
[email protected]> wrote:
Yes that's what I proposed also, I will try that before the
worse
solution
as Taher called them, would you help?
Jacques
Le 20/08/2016 à 08:32, Pierre Smits a écrit :
Hi Jacques,
Why not try to convince the people behind notsoserial to have
them
push
the
library to maven central and/or jpublish? In stead of this
community
doing
the work?
Best regards,
Pierre Smits
ORRTIZ.COM <http://www.orrtiz.com>
OFBiz based solutions & services
OFBiz Extensions Marketplace
http://oem.ofbizci.net/oci-2/
