561042 - cannot load page send too large cookie

Status: RESOLVED FIXED

Description

[arno renevier]

Hi,
when a page tries to set a big cookie, page does not load. Gecko does not even display about:blank, it just stays on current page. If i copy this url in firefox urlbar and open a new tab with Alt-Enter, all I get is a blank tab with grey background.

http://localhost/tmp/cookie.php
source:
http://localhost/tmp/cookie.txt

Comment 1

[[e-mail]]

Can you attach the files here?

Comment 2

[arno renevier]

Created attachment 440805 [details]
testcase

Comment 3

[[e-mail]]

Sending an HTTP header that large will bust the limit on either the webserver, or the client (Firefox), and result in the connection being dropped. Hence the pageload fail.

This isn't a bug on our side; if anything, you should file a bug with the PHP folks to throw an error if you try to set a cookie larger than 4KB, and maybe if the total HTTP header size exceeds the limit. (I'm not sure exactly what that limit is; it varies between browsers and probably servers too.)

Comment 4

[arno renevier]

(In reply to comment #3)
> Sending an HTTP header that large will bust the limit on either the webserver,
> or the client (Firefox), 

It's probably Firefox: other browsers handle this page well.

> and result in the connection being dropped. Hence the
> pageload fail.

Ok. But may be it's better to load aubout:neterror when firefox drop connection instead of letting a grey background.

> This isn't a bug on our side; if anything, you should file a bug with the PHP
> folks to throw an error if you try to set a cookie larger than 4KB, and maybe
> if the total HTTP header size exceeds the limit. (I'm not sure exactly what
> that limit is; it varies between browsers and probably servers too.)

Even if php modifies his behaviour. I can still send those headers with python, C, whatever. That won't prevent users from having this

Comment 5

[arno renevier]

> http://localhost/tmp/cookie.php
> source:
> http://localhost/tmp/cookie.txt

Oopps, sorry for that. 
Real links were:

http://renevier.net/bugs/bigcookie.txt
http://renevier.net/bugs/bigcookie.php

Comment 6

[arno renevier]

(In reply to comment #3)
> Sending an HTTP header that large will bust the limit on either the webserver,
> or the client (Firefox), 

It's probably Firefox: other browsers handle this page well.

> and result in the connection being dropped. Hence the
> pageload fail.

Ok. But may be it's better to load about:neterror when firefox drop connection instead of doing nothing.

> This isn't a bug on our side; if anything, you should file a bug with the PHP
> folks to throw an error if you try to set a cookie larger than 4KB, and maybe
> if the total HTTP header size exceeds the limit. (I'm not sure exactly what
> that limit is; it varies between browsers and probably servers too.)

Even if php modifies his behaviour. I can still send those headers with python, C, whatever.

Comment 7

[arno renevier]

Created attachment 440977 [details]
grey background when opening in a new tab

Comment 8

[[e-mail]]

Other browsers don't die, you say?

Worth another look, then.

Comment 9

[Boris Zbarsky [:bz] (vacation Aug 14-27)]

nsHttpTransaction::ParseLineSegment has:

721     if (mLineBuf.Length() + len > MAX_LINEBUF_LENGTH) {
722         LOG(("excessively long header received, canceling transaction [trans=%x]", this));
723         return NS_ERROR_ABORT;
724     }

MAX_LINEBUF_LENGTH is (1024 * 10).

That code was added in bug 152697.

I suppose we could do an about:neterror here by using a status code nsDocShell::EndLoad recognizes or adding NS_ERROR_ABORT there, or using a new code just for this case.  We could also increase (or remove?) the limit here.

Comment 10

[[e-mail]]

Sounds like a sensible thing to do, especially since a single cookie can be 4K. :)

Comment 11

[Boris Zbarsky [:bz] (vacation Aug 14-27)]

Which one does?

I just checked and chrome has no obvious header length limit, btw.

Comment 12

[[e-mail]]

Either making it much bigger or removing it. Having such a small limit to prevent DOS isn't such a strong argument nowadays, I think.

Ideally we'd also have a neterror page, but that can be a separate bug...

Comment 13

[Boris Zbarsky [:bz] (vacation Aug 14-27)]

My gut feel is to nix the check.  Jason, thoughts?

Comment 14

[Jason Duell [:jduell] (needinfo me)]

Sure, my gut feel is fine with nixing the check.  So long as we also make sure nothing breaks as a result :)

Comment 15

[arno renevier]

Created attachment 546303 [details] [diff] [review]
patch proposal

patch proposal: remove http header length check.

Comment 16

[Boris Zbarsky [:bz] (vacation Aug 14-27)]

Comment on attachment 546303 [details] [diff] [review]
patch proposal

r=me

Comment 17

[Marco Bonardo [::mak]]

http://hg.mozilla.org/mozilla-central/rev/9c8b0c5a5591