[PATCH] c/r: Add AF_UNIX support (v3)

Oren Laadan orenl at cs.columbia.edu
Mon Jul 6 12:06:40 PDT 2009



Dan Smith wrote:
> OL> Because in the kernel: "->peer != NULL" if-and-only-if "h->raddr
> OL> is meaningful" so I wonder whether you should detect such input
> OL> and complain, in case a malicious user does that ?
> 
> Ah, sorry, I misunderstood.  Yep, agreed.  I was thinking I was
> covered on that by the use of the regular functions but my makeaddr
> function would let it slip by, so I've fixed that.
> 
> OL> Well, it should not be more than allowed by ->rcvbuf, or some
> OL> other system limits (which I'm sure exists). At least apply the
> OL> restrictions put by a regular read() syscall.
> 
> Relying on the stored rcvbuf doesn't help prevent a DoS, right?

Yes, of course, I meant a sane ->rcvbuf :)

> 
> OL> One exception is when a socket's rcvbuf fills with large amount of
> OL> data, and then the user calls setsockopt() to reduce the buffer to
> OL> a lower size... in which case relying on the restored value for
> OL> -> rcvbuf limit is clearly wrong.
> 
> Hmm, I don't know why that's the case.  Even still, I don't see why we
> should rely on the stored version of rcvbuf at all, aside from maybe
> just checking that the socket buffers are smaller than the claimed
> rcvbuf size of correctness' sake.  However, checking against an
> existing sysctl or other limit (if there is one, still need to look)
> is obvious a good idea.

Yes.

In fact, if you take a look at sock_setsockopt(), it already has the
logic to impose (or not to) a limit on sk_rcvbuf.

Maybe do something like:
	...
	sock_setsockopt(sock, ..., saved_rcvbuf_length);
	sock_read_buffers(..., saved_recvbuf_length)
	sock_setsockopt(sock, ..., saved_rcvbuf_val);
	...

So first you set the limit to the actual size you are going to restore,
if that succeeds, you restore the buffer (guaranteed to have enough
room), then restore the saved sk_rcvbuf value.  Permission tests are
on the house.

(You'll need to slightly refactor sock_setsockopt() for that).

> 
> OL> The connect() syscall will auto-bind only if the socket isn't
> OL> already bound. And in the case of ad_unix, even that happens only
> OL> when SOCK_PASSCRED bit is set in sock->flags.
> 
> Well, I'm not really sure why that makes sense (not arguing that it's
> true).  Perhaps a comment about this potential issue will suffice for
> now?

Comment about what ?  I was just describing the behavior ...

> 
> OL> Even an established socket may have a local address that needs to
> OL> be saved, to make getsockname() consistent across c/r.
> 
> ...right, but that's already restored properly, no?  My test case
> shows that established STREAM sockets have the same output of
> getsockname() before and after restart.

Indeed this test passes. But it is insufficient. Looking at v4 of
the patch, for established sockets, for the auto-bind case of
connect(), you are mostly correct. But what about --

1) 	s = socket(.., SOCK_DGRAM,...);
	bind(s, any_addr);
--> that is, s isn't connected to/from another socket.

2) 	s = socket(.., SOCK_DGRAM,...);
	bind(s, any_addr);
	connect(s, other_addr, ...);
--> now s is connected, but after restart you can't connect another
socket to it because the address wasn't bind() properly.

3)	s = socket(..., SOCK_STREAM,...);
	bind(s, any_addr);
	connect(s, other_addr, ...);
--> here, too, s is not properly bind(), so if after restart someone
tries to:
	r = socket(..., ..., ...);
	bind(r, any_addr);
the bind() will succeed after restart, but will have failed on the
original system where the checkpoint was taken.

IOW: when you restore the address of a socket, you need to emulate
the operation of bind(), not merely memcpy the address into place.
That's the reason why I originally had suggested to imitate the
usual sequence of connection sockets to restore them.

(And if the address was a pathname, but already unlinked, then
also unlink after the bind, FWIW).

Oren.

> 
> OL> s1 = socket(); bind(s1, addr); unlink(addr);
> OL> s2 = socket(); bind(s1, addr);
> OL> 			^^^^^^^^
> OL>                    (s2, addr)    <-- should have been	
> 
> OL> Now with the fixed example, the problem is if s2 is restored first,
> OL> the restore of s1 will fail, or succeed and make s1 unreachable.
> 
> I think you meant "restore of s1 will fail, or succeed and make *s2*
> unreachable".
> 
> OL> However, we know that s1 is no longer reachable through connect()
> OL> (or sendto() for dgram) - so during restart we want to manually
> OL> set the local address, without the side-effect of creating an
> OL> inode in the filesystem, for those sockets that are 1) bound with
> OL> pathname address, and 2) the corresponding socket inode isn't
> OL> unlinked.
> 
> Okay, done.
> 
> OL> I meant "cwd at the time of the bind()". To find it, you can
> OL> "subtract" the sock-address from the total pathname of the inode
> OL> (assuming, of course, that the address is relative).
> 
> Right, right.  I was thinking that the socket kept an inode which
> would make it hard to determine the full path, but it's a dentry.
> I'll work that in.
> 
> Thanks!
> 


More information about the Containers mailing list