693925 - Redirect loop on one page (maybe more, but not all pages) on sap.com with browsers that implement new anti-chosen-plaintext-attack mitigations

Status: RESOLVED WORKSFORME

(Tech Evangelism Graveyard :: English US, defect, P1)

Description

[Francesco Lodolo [:flod]]

When trying to visit the URL https://www.sap.com/campaign/2010_curr_sap_crystal_reports_2008/index.epx?kNtBzmUK9zU, I get the error "Firefox has detected that the server is redirecting the request for this address in a way that will never complete."

Build ID: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:10.0a1) Gecko/20111011 Firefox/10.0a1

I can visit that page without problems with Firefox 7.0, Safari or Chrome on the same Mac. Tried also with a fresh new profile without changes.

Comment 1

[Francesco Lodolo [:flod]]

Some more details.

Works with: 
Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:10.0a1) Gecko/20111007 Firefox/10.0a1

Doesn't work with: 
Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:10.0a1) Gecko/20111008 Firefox/10.0a1

Comment 2

[Matthias Versen [:Matti]]

confirming with Mozilla/5.0 (Windows NT 6.1; rv:10.0a1) Gecko/20111011 Firefox/10.0a1 SeaMonkey/2.7a1

Comment 3

[Boris Zbarsky [:bzbarsky]]

Looks like the range is http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=c3a50afc2243&tochange=6c780dcb4b99 but nothing jumps out at me there...

Comment 4

[Boris Zbarsky [:bzbarsky]]

Ah, maybe the NSS upgrade?

Comment 5

[Boris Zbarsky [:bzbarsky]]

Yep, hg bisect confirms that.

Requesting tracking for this regression.

Comment 6

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

The same problem occurs with Google Chrome Canary: Error 310 (net::ERR_TOO_MANY_REDIRECTS): There were too many redirects.

I will investigate it more thoroughly Friday. This is probably related to the 1/(n-1) record splitting implemented in bug 665814.

Comment 7

[Adam Langley]

Yes, it's 1/n-1 record splitting. This is a WontFix from Chrome; the site will have to be fixed.

(Brian: the current list of known broken things are Brocade SSL terminators (at least some of them) and lighttpd < 1.4.27)

Comment 8

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

See bug 69820 comment 6 for an explanation.

Comment 9

[Wan-Teh Chang]

Just to clarify: the broken code on the server side is most likely
not the SSL library but rather some code *above* SSL.  So
"broken SSL implementation" is a little misleading.

All SSL libraries should be able to handle an application_data
record containing one byte of data.  Most likely, some higher-layer
code does not handle a partial read from SSL correctly.

Comment 10

[Francesco Lodolo [:flod]]

No more a bug, working fine now.