Summary
DoS protection support for spiffy
Metadata
- Id: 9087f3ffb3d4eb90abe13b97b8fe739f2987f265
- Trac id: 279
- Type: enhancement
- Reporter: mario
- Owner: sjamaan
- Cc:
- Status: closed
- Component: extensions
- Estimated difficulty:
- Resolution: wontfix
- Priority: not urgent at all
- Milestone:
- Version: 4.5.x
- Changetime: 2011-09-23 21:11:31 UTC
- Created: 2010-07-09 21:28:53 UTC
- Keywords: spiffy, dos
Description
Some tips Brian Mastenbrook gave on #scheme. I'm pasting them here in case we want a DoS protection module for spiffy someday. Here are the relevant parts:
<chandler> mario-goulart: I think a sufficient approach would be to
(a) limit the number of active connections, (b) kill
connections according to a least-recently-transmitted
policy, and (c) set a timeout for connections in the
receiving headers phase, and another timeout for sending
data.
<mario-goulart> chandler: wouldn't (c) mess up with things like comet
and long live connection intentionally requested to
avoid the multiple requests overhead?
<chandler> If the server is blocking before responding to a request,
don't time out.
<chandler> But the client shouldn't be allowed to connect or start
sending a request and then block indefinitely before
finishing sending headers.
<chandler> mario-goulart: Furthermore, if the server has data to send
to the client but hasn't actually been able to send it for
several seconds, the connection should be terminated as
well.
Changes and comments
[2010-08-02 14:45:02 UTC] felix removed milestone 4.6.0
[2011-02-19 19:08:29 UTC] sjamaan changed status from new to assigned
[2011-02-19 19:08:29 UTC] sjamaan set owner to sjamaan
[2011-07-13 20:32:21 UTC] sjamaan wrote:
Are there good ways to test something like this? How do you know what is sane behaviour? If someone initiates a DDoS attack, this shouldn't start semi-randomly killing existing connections of real users.
[2011-09-23 21:11:31 UTC] sjamaan changed status from assigned to closed
[2011-09-23 21:11:31 UTC] sjamaan set resolution to wontfix
[2011-09-23 21:11:31 UTC] sjamaan wrote:
Closing this since I have absolutely no clue how to fix this. If you have a concrete test or patch, please reopen.