nginx/1.3.2 - 4xx responses are returned without HTTP headers, bug?

[Sjon Hortensius]

On 07/09/2012 02:46 PM, Igor Sysoev wrote:
> On Mon, Jul 09, 2012 at 01:01:21PM +0200, Sjon Hortensius wrote:
>> On 07/09/2012 11:52 AM, Maxim Dounin wrote:
>>> Hello!
>>>
>>> On Mon, Jul 09, 2012 at 10:41:19AM +0200, Sjon Hortensius wrote:
>>>
>>>> Hi,
>>>>
>>>> I think I stumbled upon a bug in nginx but I would like to get
>>>> someone else's opinion on this.
>>>>
>>>> It seems all 4xx errors are not prefixed by headers in HTTP
>>>> responses. I have a simple test-script in bash which you can run to
>>>> verify this behavior:
>>>>
>>>> function req { echo -n "GET /bugtest?id="; for i in `seq 15000
>>>> 17100` ; do echo -n "%7C$i" ; done ; echo -en "HTTP/1.1\nUser-Agent:
>>>> Mozilla/5.0\nConnection: Close\nHost: wiki.nginx.org\r\n\r\n" ; } ;
>>>> req | nc wiki.nginx.org 80
>>>>
>>>> I can also reproduce this for other 4xx responses. My interpretation
>>>> of the RFC is that this is unwanted, see
>>>> http://tools.ietf.org/html/rfc2616#section-6.1
>>>
>>> Headers aren't sent if nginx assumes the request is in HTTP/0.9
>>> format.  The above script makes nginx to think it's HTTP/0.9 as
>>> uri used dosn't fit into header buffer and nginx isn't able to
>>> see "HTTP/1.1" before it returns 414.
>>>
>>> Maxim Dounin
>>
>> Thanks for your reply! I have been looking for RFC documents describing
>> the maximum request-length in HTTP/0.9, but I cannot find any. I found
>> http://www.w3.org/Protocols/HTTP/Request.html (which links to
>> http://www.w3.org/Addressing/URL/5_BNF.html to describe URLs) and
>> http://www.ietf.org/rfc/rfc1738.txt, both of which describe no maximum
>> length.
>>
>> If nginx assumes HTTP/0.9 for the request, it doesn't seem very logical
>> to return an error which cannot exist within that very protocol?
>
> Well, it does not return error. It just returns a page with text:
> "414 Request-URI Too Large". What can you suggest to return for
> a request larger than configured buffer ?

When I said it returns an error I obviously meant it returns a HTTP 
error. Since HTTP/0.9 doesn't define a maximum length for a URL, it 
seems strange to return an error about this very length in this specific 
protocol.

I think the only problem I have with this behavior is nginx falling back 
to HTTP/0.9; a protocol which isn't used on any network I know. I would 
prefer falling back to HTTP/1.0 which more clients will support, 
although I don't know what impact this has.

--

Regards,

Sjon Hortensius

| React B.V.
| http://www.react.nl
| Julianastraat 30
| 2012 ES, Haarlem
| 023  553 90 00