Skip to content

http.client._MAXHEADERS = 100 limit no longer sufficient #131724

Closed

Description

@jmacdone

Bug report

Bug description:

This hard-coded sanity check for HTTP response headers is no longer sufficient to fetch a Microsoft 365 page.

_MAXHEADERS = 100

I do have a case open with Microsoft Support (TrackingID#2503130010002871), but it's not getting much traction as it's not causing problems with full web browsers.

Steps to reproduce:

import http.client
con = http.client.HTTPSConnection('outlook.office365.com')
con.request("GET", "/owa/example.edu")  # any domain seems to trigger
r = con.getresponse()

And that throws a HTTPException

>>> con = http.client.HTTPSConnection('outlook.office365.com')
>>> con.request("GET", "/owa/foo.bar")
>>> r = con.getresponse()
Traceback (most recent call last):
  File "<python-input-8>", line 1, in <module>
    r = con.getresponse()
  File "C:\Users\jmacdone\AppData\Local\Programs\Python\Python313-arm64\Lib\http\client.py", line 1428, in getresponse
    response.begin()
    ~~~~~~~~~~~~~~^^
  File "C:\Users\jmacdone\AppData\Local\Programs\Python\Python313-arm64\Lib\http\client.py", line 350, in begin
    self.headers = self.msg = parse_headers(self.fp)
                              ~~~~~~~~~~~~~^^^^^^^^^
  File "C:\Users\jmacdone\AppData\Local\Programs\Python\Python313-arm64\Lib\http\client.py", line 248, in parse_headers
    headers = _read_headers(fp)
  File "C:\Users\jmacdone\AppData\Local\Programs\Python\Python313-arm64\Lib\http\client.py", line 226, in _read_headers
    raise HTTPException("got more than %d headers" % _MAXHEADERS)
http.client.HTTPException: got more than 100 headers
>>>

It seems to be just spilling over with 101 headers. Though, not consistently. Presumably it depends upon which load balancer node is responding.

$ curl --silent -D - 'https://outlook.office365.com/owa/example.edu' | grep -E "^[a-zA-Z-]+: " | wc -l

returns with 96, 99, 101, etc. headers, depending on Microsoft's mood unknown factors.

For background, it's common to use https://outlook.com/example.edu as a domain hint ("smart link") to go directly to a tenant's identity provider and avoid the "Please provide your email address" step. We have a nagios check for that, which broke recently as the number of Set-Cookie: OpenIdConnect.token.[...] variants continues to grow.

CPython versions tested on:

3.13, 3.9, 3.11

Operating systems tested on:

Linux, Windows

Linked PRs

Activity

  1. StanFromIreland commented on Mar 25, 2025

    @StanFromIreland
    Member

    This seems to only restrict HTTPSConnection and HTTPConnection is not affected.

  2. added
    stdlibStandard Library Python modules in the Lib/ directory
    on Mar 28, 2025
  3. jmacdone commented on Mar 28, 2025

    @jmacdone
    Author

    Just to help clarify, picking a new arbitrary limit allows getresponse() to complete without an exception.

    >>> import http.client
    >>> http.client._MAXHEADERS = 120
    >>> con = http.client.HTTPSConnection('outlook.office365.com')
    >>> con.request("GET", "/owa/example.edu")  # any domain seems to trigger
    >>> r = con.getresponse()
    >>> r.read()
    b'<html><head><title>Object moved</title> [...]'

    @StanFromIreland I don't have an explicit test case, but I think this would affect both classes. HTTPSConnection.getresponse() is inherited from HTTPConnection.getresponse()

    class HTTPSConnection(HTTPConnection):

  4. vstinner commented on Mar 28, 2025

    @vstinner
    Member

    Maybe we should add a parameter to customize this limit?

  5. gostak-dd commented on May 20, 2025

    @gostak-dd
    Contributor

    I tried with both 3.11 and 3.15 and did not get an exception

  6. jmacdone commented on May 20, 2025

    @jmacdone
    Author

    I tried with both 3.11 and 3.15 and did not get an exception

    @GGyll Confirmed. It does indeed appear to be just under the limit today

    >>> len(r.headers)
    95

    And, for completeness, Microsoft closed my ticket a few weeks ago: "working as intended". They're happy they're setting 60+ cookies.

    For the time being, my workaround is setting the module's "private" variable as show in #131724 (comment)

    I suppose that could be a patch option. Rename _MAXHEADERS to MAXHEADERS and document it as part of the module's public interface. Or, maybe it's not worth bothering with just to accommodate one page from one vendor?

  7. vstinner commented on May 20, 2025

    @vstinner
    Member

    Maybe we should add a parameter to customize this limit?

    I mean adding a parameter to HTTPConnection and HTTPSConnection constructors. If not specified, the default value would be http.client._MAXHEADERS.

  8. aeurielesn commented on Jul 19, 2025

    @aeurielesn
    Contributor

    Adding a parameter sounds like a good approach. I can have a look at this.

  9. added 4 commits that reference this issue on Jul 19, 2025
  10. added a commit that references this issue on Jul 20, 2025
  11. encukou commented on Jul 22, 2025

    @encukou
    Member

    Thank you for the feature, @aeurielesn!

  12. added a commit that references this issue on Aug 4, 2025
  13. added a commit that references this issue on Aug 19, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    stdlibStandard Library Python modules in the Lib/ directorytype-bugAn unexpected behavior, bug, or error

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions