Opened 4 years ago

Closed 2 years ago

Last modified 8 months ago

#21608 closed Bug (fixed)

Logged out sessions are resurrected by concurrent requests

Reported by: Jonas Borgström Owned by: Tore Lundqvist
Component: contrib.sessions Version: 1.9
Severity: Normal Keywords:
Cc: m17.admin@…, tlt@… Triage Stage: Ready for checkin
Has patch: yes Needs documentation: no
Needs tests: no Patch needs improvement: no
Easy pickings: no UI/UX: no

Description (last modified by jnnt)

  1. User logs in
  2. User loads a slow page in separate tab or as an ajax request, which modifies the session
  3. User logs out before request in step 2 completes. This will delete the session from the db

Expected behavior

User/session stays logged out since the user explicitly logged out and the session row was delete in step 3.

Actual behavior

The previously deleted session is re-inserted into the database when the request from step 2 completes. So the previously logged out user is now logged in again.

Attachments (2)

session_fix.patch (1.1 KB) - added by Jonas Borgström 4 years ago.
Proposed fix (against django 1.4)
21608.diff (2.7 KB) - added by Tore Lundqvist 2 years ago.

Download all attachments as: .zip

Change History (29)

Changed 4 years ago by Jonas Borgström

Attachment: session_fix.patch added

Proposed fix (against django 1.4)

comment:1 Changed 4 years ago by Jonas Borgström


comment:2 Changed 4 years ago by Russell Keith-Magee

Needs tests: set
Patch needs improvement: set
Triage Stage: UnreviewedAccepted

Seems like a reasonable request, and the patch looks like a decent start -- but it needs tests.

comment:3 Changed 4 years ago by nikl@…

Owner: changed from nobody to anonymous
Status: newassigned

comment:4 Changed 4 years ago by nikl@…

Finalized on the train and airport:

Thanks to everybody at DjangoIsland who helped me tackle this - looking forward to your feedback!

  • Nikl

comment:5 Changed 4 years ago by nikl@…

Has patch: set
Needs tests: unset
Patch needs improvement: unset
Triage Stage: AcceptedReady for checkin

comment:6 Changed 4 years ago by Tim Graham

Triage Stage: Ready for checkinAccepted

Please don't mark your own patch as RFC. Someone who reviews the patch should do that.

comment:7 Changed 4 years ago by Sasha Romijn

I'm not entirely getting this. When a user logs out, the session is flushed. Flushing the session clears it and deletes it. The database session store performs this deletion by actually deleting the record from the DB. The cached_db backend deletes it from the DB and the cache. So basically, all records of this session should be deleted. If you would post a new request with the now deleted session ID, Django will reject it, and assign you a new session with a new session ID.

The reporter says that django re-inserts the session when a request arrives with the old session ID, and will re-insert it with the old session data. But I don't see that anywhere in the code. As far as I can see, Django would reject the session ID, as loading would fail as the session object has been deleted, and the user would be assigned a new session. Even if there were a flaw in that logic: once the session data has been deleted, how would any code know how to recreate the session? The request doesn't contain any hint on what user should be logged in.

The only explanation I can come up with is that we're talking about cookie backed sessions, for which this is a documented limitation: you can't guarantee deletion of a cookie backed session or it's data, no matter what we do in Django: its the nature of cookies.

comment:8 Changed 4 years ago by Tim Graham

Patch needs improvement: set

I couldn't reproduce this using steps 1-3 in the description (SQLite). After logging out in a separate tag, the slow page loaded, but subsequent requests redirected to the admin login page. There also seem to be some concerns from Nick's review on the PR.

comment:9 Changed 4 years ago by Jonas Borgström

I think one key detail missing from the initial reproduction steps is that the "slow page" needs to modify the session to make it dirty. Otherwise the session will not be resurrected.

Anyway, I've now create a complete reproduction test case here:

See README.txt for details.

comment:10 Changed 3 years ago by Collin Anderson

would fix this issue?

comment:11 Changed 3 years ago by Sergey Kolosov

Cc: m17.admin@… added

comment:12 Changed 3 years ago by jnnt

Description: modified (diff)

comment:13 Changed 2 years ago by Tore Lundqvist

I have reproduced this bug with the test site ( on both Django 1.8.7 and 1.9 (using SQLite).

These were the steps I used:

  • Open the admin page in a tab and log in. http://localhost:8000/admin/
  • Switch to a new tab and open the slow page. http://localhost:8000/slow/ Act fast after this step to complete the next three steps before the slow page has finished loaded, <10 sec
  • Switch back to tab with the admin page.
  • Click the "Logout" link on the top right corner of the page. -> Now you are on the logout page
  • Reload page -> Now you are on the login page.
  • Wait for slow page to finish loading.
  • Reload the tab with the login page.
  • Logged in again without entering credentials!

This is a security issue, not critical though, as someone might think that they have logged out but is actually still logged in.
If you logout and leave a public computer while a page is still loading in another tab there is a risk that the next person using that computer can get access to your account.
It would be nice to have this fixed in 1.8 and up.

Last edited 2 years ago by Tore Lundqvist (previous) (diff)

comment:14 Changed 2 years ago by Tore Lundqvist

Version: 1.41.8

comment:15 Changed 2 years ago by Tore Lundqvist

Cc: tlt@… added
Owner: changed from anonymous to Tore Lundqvist
Patch needs improvement: unset
Version: 1.81.9

Changed 2 years ago by Tore Lundqvist

Attachment: 21608.diff added

comment:16 Changed 2 years ago by Tore Lundqvist

Resolution: fixed
Status: assignedclosed

comment:17 Changed 2 years ago by Tore Lundqvist

Run the unit test without the fix to verify the bug. This fix should also be back ported to 1.8 as it is a security fix.

comment:18 Changed 2 years ago by Tore Lundqvist

Resolution: fixed
Status: closednew

comment:19 Changed 2 years ago by Tim Graham

Are you able to submit the patch as a pull request?

comment:20 Changed 2 years ago by Tore Lundqvist

Absolutely, which branch should I create a PR against?

comment:21 Changed 2 years ago by Tim Graham

Master please.

comment:22 Changed 2 years ago by Tore Lundqvist

Hi again, pull request is up.

comment:23 Changed 2 years ago by Esbjörn Ekberg

Triage Stage: AcceptedReady for checkin

comment:24 Changed 2 years ago by Esbjörn Ekberg

I've gone through the patch review checklist and everything is looking good.

comment:25 Changed 2 years ago by Tim Graham <timograham@…>

Resolution: fixed
Status: newclosed

In 3389c5e:

Fixed #21608 -- Prevented logged out sessions being resurrected by concurrent requests.

Thanks Simon Charette for the review.

comment:26 Changed 2 years ago by Tim Graham <timograham@…>

In 5faf745:

Refs #21608 -- Fixed incorrect cache key in cache session backend's save().

The bug was introduced commit 3389c5ea229884a1943873fe7e7ffc2800cefc22.

comment:27 Changed 8 months ago by Dan Tao

I commented here, but just to raise visibility: I'm concerned that the change to fix this bug resulted in a logical error (or at least unintuitive behavior) in Namely: now must_create=False implies must_update=True, which I would argue is wrong. must_create=False should probably mean that either creating or updating is acceptable.

Has this already been discussed elsewhere, and perhaps there's something I'm missing?

Note: See TracTickets for help on using tickets.
Back to Top