svn commit: r266806 - head/sys/netgraph

[Gleb Smirnoff]

On Wed, May 28, 2014 at 11:41:23PM +0800, Julian Elischer wrote:
J> On 5/28/14, 10:56 PM, Gleb Smirnoff wrote:
J> > On Wed, May 28, 2014 at 10:12:32PM +0800, Julian Elischer wrote:
J> > J> On 5/28/14, 9:15 PM, Gleb Smirnoff wrote:
J> > J> > Author: glebius
J> > J> > Date: Wed May 28 13:15:14 2014
J> > J> > New Revision: 266806
J> > J> > URL: http://svnweb.freebsd.org/changeset/base/266806
J> > J> >
J> > J> > Log:
J> > J> >    Use M_WAITOK for the NGM_PIPE_SET_CFG control message. We expect it to
J> > J> >    arrive from userland only.
J> > J> >
J> > J> >    Submitted by:	Dmitry Luhtionov <dmitryluhtionov gmail.com>
J> > J> what's to stop another node from generating it and sending it on?
J> > J> generally a message may come from anywhere.
J> > J> Just becasue YOU don't have module that
J> > J> sends messages to ng_pipe, doesn't mean there never will be..
J> > J> also there are cases when the locking may force a message to be
J> > J> delivered asynchronously.
J> >
J> > I know that. After resolving many issues with netgraph, I feel that
J> > our policy should be towards putting some invariants on what events
J> > SHOULD come from userland only and which events SHOULD be serviced
J> > without memory failures.
J> >
J> > Current paradigma that messages are fully symmetrical and can come
J> > from anywhere are quite a curious thought experiment. I liked that
J> > for a long time. But in practice if we want to build a robust software
J> > we should make more strict rules of using it.
J> >
J> > You could disagree, but if you try to fix this particular one liner
J> > in the paradigma of "messages come from anywhere", then you will end
J> > up with smth like 20 lines of code to this particular module. Next
J> > comes the need to fix any software or script that sends NGM_PIPE_SET_CFG,
J> > it now should be taught of dealing with ENOMEM. So, instead of one
J> > liner you will bury yourself under tons of work.
J> >
J> >
J> Then I suggest that we increment the protocol, and add support to messages
J> to say whether they come from user space.
J> and some providers only accept such messages.
J> 
J> I have lost track of the code so I don't know if my comment about
J> messages getting queued instead of delivered is still true.
J> However if it is, then a message could be delivered by a kernel agent 
J> even
J> if it is initiated by a userspace program.

I don't think we need to increment protocol. Does it change? Right now
we have a de facto standard, that some messages are expected from userland
only. It is not enforced, it is just a habit. We should just keep this
direction.

-- 
Totus tuus, Glebius.