Re: fs, net: deadlock between bind/splice on af_unix

Al Viro [off-list ref] writes:

On Thu, Dec 08, 2016 at 10:32:00PM -0800, Cong Wang wrote:

quoted

quoted

Why do we do autobind there, anyway, and why is it conditional on
SOCK_PASSCRED?  Note that e.g. for SOCK_STREAM we can bloody well get
to sending stuff without autobind ever done - just use socketpair()
to create that sucker and we won't be going through the connect()
at all.

In the case Dmitry reported, unix_dgram_sendmsg() calls unix_autobind(),
not SOCK_STREAM.

Yes, I've noticed.  What I'm asking is what in there needs autobind triggered
on sendmsg and why doesn't the same need affect the SOCK_STREAM case?

With respect to the conditionality on SOCK_PASSCRED those are the linux
semantics.  Semantically that is the way the code has behaved since
2.1.15 when support for passing credentials was added to the code.
So I presume someone thought it was a good idea to have a name for
a socket that is sending credentials to another socket.  It certainly
seems reasonable at first glance.

With socketpair the only path that doesn't enforce this with
SOCK_STREAM and SOCK_PASSCRED that is either an oversight or a don't
care because we already know who is at the other end.

I can imagine two possible fixes:
1) Declare that splice is non-sense in the presence of SOCK_PASSCRED.
2) Someone adds a preparation operation that can be called on
   af_unix sockets that will ensure the autobind happens before
   any problematic locks are taken.

Eric