Hi, I'm getting a whole bunch of these in my private web server logs:

[Wed Jul 18 11:22:26 2012] [error] [client 193.64.20.129] Invalid
method in request ------------------------------c5a4fa5b03a9
[Wed Jul 18 11:24:37 2012] [error] [client 193.64.20.129] Invalid
method in request ------------------------------5faa291f1fd6
[Wed Jul 18 11:26:48 2012] [error] [client 193.64.20.129] Invalid
method in request ------------------------------a921c3ae8ed4
[Wed Jul 18 11:27:54 2012] [error] [client 193.64.20.129] Invalid
method in request ------------------------------e947bc9524fa

It looks like server expects a new request, POST /xxx HTTP xxx, but
instead gets a multipart post separator.

All (most) the clients are pycurl/libcurl/openssl with http keepalives
and ssl session caching and they send multipart POST requests, only
very small percentage of these requests triggers this.

All the broken requests come over 3G connection, I suspect
slow/unreliable network connection triggers the bug.

Since all traffic is wrapped in openssl, the bug must be either in
libcurl, its usage or apache.

I suspect some timeout gets hit in 100-continue/timeout code.

Client is licurl-7.24.0 c-ares-1.7.5 openssl 1.0.1 python-pycurl
7.19.0-6 arch arm zlib is compiled in, but I think is not used.

Cursory search shows there were bugs like this in apache 1.3 that's
almost 10 years ago, but nothing recent.

Did anyone see anything like this before?

Where do I start digging?

Should the socket be closed if client sent POST headers and failed to
submit post data per spec?

d.
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2012-07-19