Dear Ron, all,
If I understand correctly, according to the JACC specification, the pattern "/" is somehow special and there is no chance of having exact match for a security constraint specified on "/".
If this is the case, shouldn't this be clearly stated as well in the Servlet specification (such that any non JACC container would behave consistently to a JACC one)?
Perhaps I miss something, but I can't personally understand that the pattern "/" can never be matched, by just reading the Servlet specification. From the parts of the document cited by Ron (3.5, 12.2) I just conclude that the "" pattern matches to the context root, but not that "/" instead doesn't.
If I look at other containers, such as Tomcat, my feeling is that I'm not the only one who got confused on this point, as their behaviour is not consistent with Glassfish.
Best regards,
Matteo
________________________________________
From: Ron Monzillo <ron.monzillo_at_oracle.com>
To: Martin Gainty <mgainty_at_hotmail.com>
Cc: users_at_glassfish.java.net, jeff.tancill_at_oracle.com, paul.hendley_at_oracle.com
Subject: Re:
Date: Wed, 08 Aug 2012 16:36:13 -0400
On 8/8/12 9:38 AM, Martin Gainty wrote:
BLU142-W28E5A0D5702103D1216EC1AECD0_at_phx.gbl" type="cite">
http://download.oracle.com/otn-pub/jcp/jacc-1.0-fr-spec-oth-JSpec/jacc-1_0-fr-spec.pdf?AuthParam=1344431271_c27abd82b7b45ea09fbf22ba55ab8381
. If the pattern is a path prefix pattern, it must be qualified by
every path-prefix pattern in the deployment descriptor matched by (the pattern being qualified) *and*
different from the pattern being qualified.
The pattern must also be qualified by every exact pattern appearing in the deployment descriptor that is matched by the pattern being qualified.
. If the pattern is an extension pattern, it must be qualified by
every path-prefix pattern appearing in the deployment descriptor *and*
every exact pattern in the deployment descriptor that is matched by the pattern being qualified.
. If the pattern is the default pattern, "/", it must be qualified by
every other pattern except the default pattern appearing in the deployment descriptor.
. If the pattern is an exact pattern, its qualified form must not contain any qualifying
patterns.
icantly
it seems the spec lacks specific examples for each pattern use case the statistical probability of mis-interpretation for each pattern and mis-implementation is increased
Martin,
In a JACC compatible Servlet container such as Glassfish, when the servlet path as derived from the request equals "/"
an empty-string servlet path is used in its stead in security-constraint matching to ensure that the url-pattern "/"
is never an exact match for any request.
IOW, this was done to ensure that in this corner case, as in all other cases, the default pattern "\' will have the lowest priority.
In the JACC spec this is spelled out in several places, most notable of which may be:
4.1.1 Permission Names for Transport and Pre-Dispatch Decisions
The name of the permission checked in a transport or pre-dispatch decision must
be the unqualified request URI minus the context path. All colon characters
occurring within the name must be represented using escaped encoding1. For the
special case where this transformation of the request URI yields the URLPattern
"/", the empty string URLPattern, "", must be used as the permission name.
For the special case where the empty string must be substituted for the "/"
pattern in the permission evaluation, all target related processing (including
servlet mapping, filter mapping, and form based login processing) must be
performed using the original pattern, "/".
fwiw, there is similar language in the Servlet spec, although its purpose may never have been revealed, and I personally
find it a bit difficult to understand.
As defined in the Servlet Spec Section 3.5 Request Path Elements
Servlet Path: The path section that directly corresponds to the mapping which
activated this request. This path starts with a '/' character except in the case
where the request is matched with the '/*' or "" pattern, in which case it is an
empty string.
and of a similar nature 12.2 ...
"The empty string ("") is a special URL pattern that exactly maps to the
application's context root, i.e., requests of the form
http://host:port/<context-
root>/. In this case the path info is '/' and the servlet path and context path is
empty string ("")."
To your point about examples, I agree that it would be helpful if this aspect of security-constraint
matching was made more visible in developer documentation.
For example, if you want to define a security constraint on the root of your application, you would use the empty-string pattern to do so.
Ron
BLU142-W28E5A0D5702103D1216EC1AECD0_at_phx.gbl" type="cite">
Martin Gainty
______________________________________________
Verzicht und Vertraulichkeitanmerkung/Note de déni et de confidentialité
Diese Nachricht ist vertraulich. Sollten Sie nicht der vorgesehene Empfaenger sein, so bitten wir hoeflich um eine Mitteilung. Jede unbefugte Weiterleitung oder Fertigung einer Kopie ist unzulaessig. Diese Nachricht dient lediglich dem Austausch von Informationen und entfaltet keine rechtliche Bindungswirkung. Aufgrund der leichten Manipulierbarkeit von E-Mails koennen wir keine Haftung fuer den Inhalt uebernehmen.
Ce message est confidentiel et peut être privilégié. Si vous n'êtes pas le destinataire prévu, nous te demandons avec bonté que pour satisfaire informez l'expéditeur. N'importe quelle diffusion non autorisée ou la copie de ceci est interdite. Ce message sert à l'information seulement et n'aura pas n'importe quel effet légalement obligatoire. Étant donné que les email peuvent facilement être sujets à la manipulation, nous ne pouvons accepter aucune responsabilité pour le contenu fourni.
________________________________________
Subject: Re:
From: v.b.kumar.jayanti_at_oracle.com
Date: Wed, 8 Aug 2012 11:49:49 +0530
CC: ron.monzillo_at_oracle.com; jeff.tancill_at_oracle.com; paul.hendley_at_oracle.com
To: users_at_glassfish.java.net
On Aug 8, 2012, at 11:21 AM, KumarJayanti wrote:
On Aug 7, 2012, at 7:41 PM, CASALINO, Matteo Maria wrote:
Hello Everyone,
I believe I found a problem in the Glassfish interpretation of some combinations of security constraints in the deployment descriptor of web applications.
In particular, the problem seems to occur whenever one or more security constraints apply to the context root of a web application (/) and other constraints apply instead to every path under the context root (/*). One such example is given by the following configuration:
<!-- Security Constraint SC1 -->
<security-constraint>
<web-resource-collection>
<web-resource-name/>
<url-pattern>/*</url-pattern>
</web-resource-collection>
</security-constraint>
<!-- Security Constraint SC2 -->
<security-constraint>
<web-resource-collection>
<web-resource-name/>
<url-pattern>/</url-pattern>
</web-resource-collection>
<auth-constraint/>
</security-constraint>
According to Servlet spefication [1, 13.8.3] a HTTP request directed to the context root (such as "GET /") shall be denied in this case, since SC2 has the "best matching" URL pattern. In contrast, Glassfish allows any (even unauthenticated) requests to the context root. Notice that Tomcat behaves instead as prescribed by the Servlet specification and denies all the requests directed to the context root.
Does anyone know if there is any explanation to this behaviour?
It seems like an interesting corner case that i never experimented before, but here are some relevant points from the JSR 115 Spec and Servlet Spec.
-------------------------------------------------------
Servlet URL-Pattern Matching Rules (JSR 115)
-----------------------------------------------
this pattern is the special default pattern,"/",which matches all other patterns.
Qualified URL Pattern Names (JSR 115)
------------------------------------
If the pattern is the default pattern,"/",it must be qualified by every other pattern except the default pattern appearing in the deployment descriptor.
Any pattern, qualified by a pattern that matches it, is overridden and made irrelevant (in the translation) by the qualifying pattern. Specifically, all extension patterns and the default pattern are made irrelevant by the presence of the path prefix pattern "/*" in a deployment descriptor.
Combining Constraints (servlet spec)
----------------------------------
The special case of an authorization constraint that names no roles shall combine with any other constraints to override their affects and cause access to be precluded.
---------------------------------------------
Please note the statement above about the presence of path prefix pattern "/*" making the default pattern "/" irrelevant. So the granted.policy generated in this case for glassfish is :
grant {
permission javax.security.jacc.WebUserDataPermission "/*";
permission javax.security.jacc.WebResourcePermission "/*";
};
And it allows access.
On the other hand in SC1 if you remove the path prefix pattern and change it to something else say "/xyz" then you can see the deny behavior due to SC2 kicking in :
Here is the granted.policy
grant {
permission javax.security.jacc.WebUserDataPermission "/xyz";
permission javax.security.jacc.WebResourcePermission "/xyz";
};
and exclude.policy generated in this case :
grant {
permission javax.security.jacc.WebUserDataPermission "/:/xyz";
permission javax.security.jacc.WebResourcePermission "/:/xyz";
};
As you can see the final result is to deny access to /xyz
Correction : i meant deny access to "/" here.
. So i am tending to believe that GlassFish is correct in your case. I am CCing Ron who is the author of these specs for his comments.
regards,
kumar
[1]
http://jcp.org/aboutJava/communityprocess/final/jsr315/index.html
Matteo Casalino
Research Associate
Security & Trust
SAP Research, SAP Labs France SAS
805, avenue du Dr. Maurice Donat
06254 Mougins Cedex
T +334 92 28 - 63 42
E matteo.maria.casalino_at_sap.com
W
http://www.sap.com/research