1366578 - cookies are not transmitted when handling custom protocol

Status: RESOLVED WORKSFORME

(Core :: Networking: Cookies, defect)

Description

[athorcis]

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:53.0) Gecko/20100101 Firefox/53.0
Build ID: 20170504105526

Steps to reproduce:

- register a custom protocol for a website (i registered magnet protocol)
- click a magnet link


Actual results:

- i was on the right page but without cookies


Expected results:

- the cookies should be sent (aren't they or is this some kind of CSRF protection)

Comment 1

[athorcis]

- I failed to reproduce the bug on "Firefox Developper Edition"
- I also failed to reproduce the bug on a clean install of Firefox 53

So i tried to run firefox without addons but it didn't worked.
Eventually i just refreshed Firefox and then it worked

Sorry to have reported a false bug

Comment 2

[athorcis]

So after searching a while i found the difference the two cases when it's working and when it's not: e10s.
It happens when e10s is disabled, and i known e10s is supposed to be released soon, so will this bug be fixed ?

Comment 3

[Amy Chung [:Amy]]

Hi,
Could you offer me the http log on this bug?
https://developer.mozilla.org/en-US/docs/Mozilla/Debugging/HTTP_logging
Thanks!

Comment 4

[athorcis]

Hi,
Here's the http log: https://paste.ee/p/7F7JY
Normally, only the first HTTP request in the log file is relevant

Comment 5

[Amy Chung [:Amy]]

Hi,
I have confirmed your log, and I found the set-cookie information in response header.
Would you enable the cookie log to let me confirm this cookie which can store in data base?
Sorry forgot to mention you this function.
Please add the flag as below:
export MOZ_LOG=timestamp,rotate:200,nsHttp:5,nsSocketTransport:5,nsStreamPump:5,nsHostResolver:5,cookie:4
Thanks!

Comment 6

[athorcis]

The new log file said: "rejected because third party cookies are blocked for this site"
but then why it's working with e10s enabled
https://paste.ee/p/H3ZzX

Comment 7

[Amy Chung [:Amy]]

Hi,
Would you offer me the log when you enabled the e10s?
Thanks!

Comment 8

[athorcis]

Here's the log with e10s enabled: https://paste.ee/p/Yok1M

Comment 9

[Amy Chung [:Amy]]

Hi,
Would you build the firfox yourself? I would like to give you a patch for testing your case.
https://developer.mozilla.org/en-US/docs/Mozilla/Developer_guide/Build_Instructions
Of would you test your case on the latest version(nightly 56)?
Thanks!

Comment 10

[athorcis]

Hi,
I can build Firefox myself, send me the patch and i'll give you the test results before monday.

Comment 11

[Amy Chung [:Amy]]

Attachment: bug-1366578.patch

Hi,
I have attached the patch, look forward your rely.
Thanks!

Comment 12

[athorcis]

Hi,
I have troubles building Firefox, when i try to build it, the build script crashes when trying to build Stylo.
I'm on Windows should I try to reproduce the bug and build Firefox on a linux based machine ?

Comment 13

[Amy Chung [:Amy]]

Hi,
The latest version has to install llvm for Stylo.
Would you offer me your error message when you built Firefox on windows?
And if you could confirm this bug on linux first is really helpful to me.
Thanks!

Comment 14

[Amy Chung [:Amy]]

Attachment: bug-1366578.patch

Hi,
I have updated new patch which added some debug message.
If you got the same result when you had imported the previous patch, please update the latest patch and attach the log to me.
Thanks!

Comment 15

[athorcis]

Hi,
I succeeded to reproduce the bug on Ubuntu 17.04 but,
My last tests show that the bug does probably not come from Firefox but from an add-on (Self-Destructing Cookies).
But once the add-on is installed the bug is present until i switch back to e10s (the add-on is not compatible with e10s).
That's why i thought the bug was coming from Firefox.

PS: When i was building Firefox i did install llvm before the build (is there a recommended version of llvm because seems to happen because of it

1:59.88 error: failed to run custom build command for `style v0.0.1 (file:///D:/dev/mozilla-source/mozilla-central/servo/components/style)`
 1:59.88 process didn't exit successfully: `d:/dev/mozilla-source/mozilla-central/obj-x86_64-pc-mingw32/toolkit/library\release\build\style-520b2f1fc5513a52\build-script-build` (exit code: 3221226505)
 ...
 2:02.29 --- stderr
 2:02.29 dependent __builtin_offsetof
 2:02.30 UNREACHABLE executed at C:\src\llvm_package_4.0.1\llvm\tools\clang\lib\AST\ExprConstant.cpp:8585!
But this problem should probably have its own bug report

Comment 16

[athorcis]

Even if i said it does not probably come from Firefox and it does not happen before the addon is installed, i can't make up my mind. What do you think ?

Comment 17

[Amy Chung [:Amy]]

Hi,
Would you rebase the firefox version which doesn't have to install llvm?
I think you could try changeset 370427 	8d6a41df1630.
Thanks!

Comment 18

[athorcis]

Hi,

So i put a lot of printf_stderr in netwerk/cookies/nsCookieService and also in dom/base/ThirdPartyUtil and
I found the difference between when e10s is enabled and when e10s is disabled.

When e10s is enabled the external content policy type of the load info is TYPE_DOCUMENT.
When e10s is disabled the external content policy type of the load info is TYPE_OTHER.

So there is probably a problem in the computation of the external content policy type.
Do you known where external content policy type is assigned?

PS: i don't need to rebase since i managed to build a Linux version of Firefox

Comment 19

[Amy Chung [:Amy]]

Hi,
If you used the web api registerContentHandler(), would you offer me the simple test for debugging?
Thanks!

Comment 20

[athorcis]

Do you want me to explain the steps to reproduce to reproduce the bug or an account to access seedbox.athorcis.ovh ?

Comment 21

[Amy Chung [:Amy]]

I would like to you offer me a simple test webpage to access seedbox.athorcis.ovh and I will reproduce the same behavior for debugging.
Thanks!

Comment 22

[athorcis]

You can debug with this page https://seedbox.athorcis.ovh/amy.php

Comment 23

[athorcis]

The cookies are missing only e10s is disabled and https://athorcis.ovh has the following cookie exception "Allow first party only"

Comment 24

[Amy Chung [:Amy]]

Hi,
I could get the cookie string on non-e10s mode, and the result as below:
Cookies

/var/www/athorcis.ovh/seedbox/web/amy.php:21:
array (size=1)
  'PHPSESSID' => string 'b8920tqjpe0q7o99ih71kfala2' (length=26)

And I also could get string from dev tool by using java script.

I tried to add the break point on HttpBaseChannel's constructor to find which function to assign  "TYPE_OTHER", but this testing page can't through the break point.

Would you try to get the cookie from dev tool by using java script?

Thanks!

Comment 25

[Amy Chung [:Amy]]

Hi,
I closed this bug first, if you have any question on this bug, please feel free to reopen again.
Thanks!