Hi all,
in my setup, SATOSA may listen on multiple interfaces/ports/vhosts, and
not all are accessible to all users. Therefore when sending the
authentication response, the IdPs must redirect the users to the
'correct' AssertionConsumerServiceURL. The problem is that the SAML2
backend always selects the first ACS address in the request
(src/satosa/backends/saml2.py:289).
I'd like to select the ACS URL based on the host name of the request
(context["http_headers"]["HTTP_HOST"] specifically). What do you think
about it? Would you consider such a pull request?
I'm still not entirely sure what to do if there's no match. I guess
Shibboleth SP used to specify the ACS URL in the AuthnRequest using
information from HTTP_HOST(?), since I remember seeing error messages on
IdPs when no AssertionConsumerServiceURLs in the metadata matched the
request. Even if I remember right, this might not be the best approach,
because I could think that it'd be more user friendly if SATOSA could
signal the error instead of the IdP, but this might be use case
dependent.
Thanks,
Kristof
Hi,
What is exactly the relationship between the attribute name format and
the mapping within attributemaps?
I couldn't fully understand that part of the code, but empirically it
seems that only the 'last' mapping file is considered, so it's not
possible to have multiple files for the same attrname-format, only one
mapping per name format is allowed. If this is correct, then adfs_v1x.py
and adfs_v20.py being separate files is pretty misleading.
Also, the satosa tree contains a subset (I didn't verify, whether it is
a true subset or not) of the pysaml's default attributemap dir, what is
the purpose of that directory?
Thank you,
Kristof
Hi!
In my calendar there is an IdPy meeting today.
Heather said at the last meeting the she would not be able to make the one today, her being at a meeting in Amsterdam.
Unfortunately I will also be unable to make the meeting today.
— Roland
Attendees:
Roland, Heather, Johan, Scott, Matthew
Regrets:
Ivan, Giuseppe
OIDC Libraries
• session management code: session was defined as one users from one client having one authentication. Based on that and user consent we create a grant. eduTEAMS wanted client authentication with no user involved, so not an OIDC flow but an OAuth flow. They did this by using the client info twice, so Roland has changed session management to grant management, conducting sessions so there can be more clients, no users. Now needs to update the documentation; that's in progress (it's a mess).
• Getting closer to the final OIDC Federation specification. Authors are now meeting weekly to clean up all remaining issues. Roland is now working on implementing the latest version.
Single Logout
• Hannah still working on this. (see notes from last call)
Documentation
• Are poetry and markdown going to be the standard tools/formats for all idpy?
• note that we may be able to use pandoc to convert documentation where necessary
• Matthew is working on the function documentation. It's not complete, but will submit a PR with what he has
Thanks! Heather
Hi all,
As far as I understand, the 'sub' is the standard claim to release a
user identifier value when you use OpenID Connect frontend. But even
though the subject_id is properly set in the internal request, I can not
get the frontend to not change this value when it creates the id_token.
Is there a way to do this?
Or how do others release the user's identifier using OIDC? I can do it
using 'email' or a custom claim, but none of this seems to be The Right
Way.
However, I'm inclined to think that using 'sub' for the user identifier
is overly complex, because I want to be able to populate the subject_id
from a SAML2 NameID (on the backend side), which I can only do with a
response microservice, but I'm not confident that changing the
'subject_id' in a microservice is a safe and supported operation.
Thanks for any insight,
Kristof
Hi!
Having refactored the Session/Grant management I though I would just add some documentation
Famous last words…
We really, really need to rework the documentation.
This isn’t the first time that someone has said this.
Will probably not be the lats time either….
But let’s start small.
Could we at least agree on a layout, a set of chapters and in what order.
Maybe we could begin with deciding on who we are writing for.
Surely we will have readers that comes to us with different backgrounds/needs :
- An architect, who wants a birds eyes view over what the system can or can’t do
- A service provider, who wants a set of steps that will get a server running.
- An apps implementer who wants to know how to interface with the client API and use Google/Github/.. as the OP.
- A developer that wants to add another endpoint/service to the system.
There are probably more.
If we could agree on who we write for we could perhaps start there.
Maybe we should have a set of entry points depending on who you (the reader) are.
Something like the list above.
If I where connected to/employed by a big organisation I could imaging finding persons that fits the
profiles above and ask them what they want to know (the depth of knowledge they want) but I ain’t so
that’s not going to happen.
— Roland
Were it left to me to decide whether we should have a government without newspapers, or newspapers without a government, I should not hesitate a moment to prefer the latter. -Thomas Jefferson, third US president, architect, and author (1743-1826)
(And now for today’s notes)
Attendees:
Johan, Heather, Ivan, Matthew
Regrets:
Scott, Roland
Agenda:
0 - Agenda bash
1 - Architecture
We talked about starting to look forward to the SSI stuff. This needs prep work to gather the specs, setting out what projects we want to build, what projects we want to get from others, and assemble the base layer so we can do things like evolve Satosa. This will start to get into shape after the NORDUnet conference in mid-September.
2 - Documentation
(From Roland)
Could we at least agree on a layout, a set of chapters and in what order.
Maybe we could begin with deciding on who we are writing for.
Surely we will have readers that comes to us with different backgrounds/needs :
- An architect, who wants a birds eyes view over what the system can or can’t do
- A service provider, who wants a set of steps that will get a server running.
- An apps implementer who wants to know how to interface with the client API and use Google/Github/.. as the OP.
- A developer that wants to add another endpoint/service to the system.
Matthew suggests we start with implementers, people who are trying to deploy apps that need federated login help. The examples in the repo aren't sufficient to get there. We need to make it easier for them to come in and user our tools. There's no obvious way to make an IdP object. It's not clear how to use the classes and methods correctly. Example: authentication context in an IdP response. Not obvious that he needed to provide it in the first place, nor how to structure it (until he looked the source code itself; it wasn't in the in-line help). What would be most immediately helpful would be adding more detail to method doc-string. If we could do this, more people could increment the immediate needs of documentation.
Ivan suggests our tools are low level enough that the current audience is people who understand them, and that would be implementers. But if we build packages that integrate with known framework, then that should shortcut a lot of documentation. They would use the tools through the layer of the framework. The adapter is trimming the API space to a more constrained set of default behavior. The rest of the things that can be done are not there; to use those, you have to have a more in-depth understanding of the libraries, and then we need the more low-level, developer-oriented documentation.
Johan agrees with Matthew that the smaller efforts will get immediate wins. Using the new OIDC code is documented like the old code, which means it's wrong for the new libraries. If we can focus on the functions, it'll be the easiest way to get something done now.
Ivan - agree, we can start here then move into restructuring thing. One limit of focusing on method doc-strings: We can tell what the function is doing, but it misses out on the context of why the function is doing it. We need both types of documentation.
Let's focus on requesting doc strings updates and creating documentation for the following questions:
• how to create a custom SAML response,
• how to validate a signature,
• how to create a SAML request,
• how to get custom metadata.
Matthew will work on documentation for the functions he's been working on most recently (stuff he touched for the test IdP), create a small PR, and we can see how that works.
Ivan: we could work on adding typing information. That will be complicated in pySAML2; would need to be added in the code not the doc string. This is something Ivan is working on, but it's slow going.
3 - GitHub review
a. OIDC - https://github.com/IdentityPython (JWTConnect-Python-OidcRP, JWTConnect-Python-CryptoJWT, etc)
No update
b. Satosa - https://github.com/IdentityPython/SATOSA
• SLO - Hannah is working on this; has it working from the perspective from an SP. If an SP requests a logout, that request will go to Satosa and back down to all the other SPs someone is logged into, then back up to the originating IdP. There were some problems getting the logout requests to all the downline SPs that support it, but those seem to be addressed. Now working on the IdP-originated logout and cleaning up the code to submit a PR. Will hopefully have a presentation about this ready next month (after NORDUnet conference).
Architecture ideas for Satosa: we'd talked about switching to a more well-known framework. That's probably going to be FastAPI. Along with that big change, we could also take a few other decisions. Ivan is leaning towards using lxml exclusively (not the built-in python xml parsers). We have long-standing questions about name prefixes. The python built-in parsers rewrite whatever is in the xml document and put in their own prefix. With lxml we can keep our own namespaces and treat them correctly. This helps properly validating signatures. The schema validator we use is lxml-schema, and we also use an underlying library called element-path. They are well made and useful and support lxml, but they can fall back to the python built-in parsers. For other things we use, we were looking to other python projects and how they choose their dependencies. There are two communities providing useful tools:
• https://github.com/encode
• https://github.com/MagicStack
They are using encode/httpx library, which will let us use the async code (we don't need it now, but will with FastAPI).
Ivan is working on a release for Satosa which will include a makefile. There are some other MRs there, but the focus will be on https://github.com/IdentityPython/SATOSA/issues/404 and https://github.com/IdentityPython/SATOSA/pull/405. The way Satosa is built now does not support a path within the base domain you set; the MRs associated with these issues should resolve that.
Another MR involves the ORCID IdP. That MR is making some things optional; Ivan is investigating.
c. pySAML2 - https://github.com/IdentityPython/pysaml2
The changes we talked about will impact pySAML2 somewhat, though users won't see much of a difference.
Ivan released a new version (7.2.0) - https://github.com/IdentityPython/pysaml2/releases/tag/v7.2.0 - no major changes. Mostly fixes, additional schemas (e.g., for eIDAS). There is a new option for a timeout for requests (see request module). Using cryptography to log certificates, which will allow us to support chains of certs.
Working on switching to poetry. This mostly works, but still investigating how to include schema files that are not python files. Also not sure how to update changelog notes automatically. Maybe make this a requirement of PRs?
d. Any other project (pyFF, djangosaml2, etc)
4 - AOB
Have added documentation on how to submit new files to different repos on how to submit code, security issues, etc. Also writing rules in the makefiles.
Thanks! Heather
(Not sure how I missed sending these out; sorry about that!)
Attendees:
Ivan, Matthew, Heather
Notes:
0 - Agenda bash
1 - Administrivia
Note that GÉANT has a tender open for idpy developers. No public link, but if anyone knows anyone that might be interested, a job description has been posted to the #random channel on idpy slack.
2 - Documentation
FAQ vs in-line documentation in the code. The FAQ idea of using responses in issues was very disorganized; Heather spent some time on that and found them not very useful. The in-line doc as they stand today aren't sufficient. The functions aren't documented sufficiently to actually tell someone how to use them. But if we do add the how-to to the doc string, they'll be too long. Ivan sees the separation being "what it is" is in the doc string, but "how to use it" should be in a different place.
Even with the idea that the doc strings being just descriptive, they are falling behind and are not complete.
Matthew points out that he's having to read a lot of source code to figure out how to implement the code. This works well enough for him, but it's hard to teach reverse engineering to new developers. There is good example code out there, but there are places where he doesn't even know he needs to go look at the sample code. What would like to see online is everything: how to use it and the reference material (deep dive on function calls).
We have a short configuration how-to (how to build a small SP or IdP) but it's very basic on readthedocs.io. We could create a list of what we want to display there (e.g., how to create a custom SAML response, how to validate a signature, how to create a SAML request, how to get custom metadata). Part of the answers to those are in the Issues and PRs, and some are in the code itself. We'll need to improve the code (doc strings) at the same time. Let's start with a small list (see examples above).
Ivan is also working on a few small documents: release.md, security.md, developers.md, contributing.md. These answer things like how to create a release, how the code should look, what tools we use, what the doc strings should look like, how to write tests, how to report an incident, where the security policy is located.
See https://docs.github.com/en/repositories/releasing-projects-on-github/automa…
For additional documentation, create a new file here: https://github.com/IdentityPython/pysaml2/tree/master/docs
This will create a new page on readthedocs.io and we can crowdsource answering the questions. Need to use the .rst format. Start by just adding the files to the top level; we can reorganize into folders later.
3 - GitHub review
a. OIDC - https://github.com/IdentityPython (JWTConnect-Python-OidcRP, JWTConnect-Python-CryptoJWT, etc)
b. Satosa - https://github.com/IdentityPython/SATOSA
How should we announce the new docker container? The old repo is still around; should it be removed?
Matthew will draft text for the announcement; aim to post on the website and post in an email. Ivan will point to the official images in the repo.
c. pySAML2 - https://github.com/IdentityPython/pysaml2
d. Any other project (pyFF, djangosaml2, etc)
4 - AOB
Thanks! Heather
