Re: problems with latest smbfs changes on 2.4.34 and security backports

Hi Santiago !

On Mon, Jan 22, 2007 at 09:54:00AM +0100, Santiago Garcia Mantinan wrote:
Hi again!

I tried to replicate the problem at home during the weekend with my laptop,
but I couldn't get it to show links with previous kernels, so I guess I had
something different on my samba server or similar, I'm at the real machines
now so I have done the real tests and they look promising. I'm getting
completely different results than those of Grant, which seems really weird.

I applied just this patch:

--- kernel-source-2.4.27.orig/fs/smbfs/proc.c 2007-01-19 17:53:57.247695476 -0700
+++ kernel-source-2.4.27/fs/smbfs/proc.c 2007-01-19 17:49:07.480161733 -0700
@@ -1997,7 +1997,7 @@
fattr->f_mode = (server->mnt->dir_mode & (S_IRWXU | S_IRWXG | S_IRWXO)) | S_IFDIR;
else if ( (server->mnt->flags & SMB_MOUNT_FMODE) &&
!(S_ISDIR(fattr->f_mode)) )
- fattr->f_mode = (server->mnt->file_mode & (S_IRWXU | S_IRWXG | S_IRWXO)) | S_IFREG;
+ fattr->f_mode = (server->mnt->file_mode & (S_IRWXU | S_IRWXG | S_IRWXO)) | (fattr->f_mode & S_IFMT);

}

To an unpatched 2.4.34, the client is an IBM NetworkStation 1000 (a PowerPC
based thin client), and the server is a normal amd64 based PC running
2.6.19.1, both running Debian, the client runs Sarge and the Server Etch.
I'm descriving this to see if differences on the architectures could be
causing the differences on behaviour between my tests and Grant's.

client running 2.4.34 with above patch, server is running 2.6.19.2 to
eliminate it from the problem space (hopefully ;) :
grant@sempro:/home/other$ uname -r
2.4.34b
grant@sempro:/home/other$ ls -l
total 9
drwxr-xr-x 1 grant wheel 4096 2007-01-21 11:44 dir/
drwxr-xr-x 1 grant wheel 4096 2007-01-21 11:44 dirlink/
-rwxr-xr-x 1 grant wheel 15 2007-01-21 11:43 file*
-rwxr-xr-x 1 grant wheel 15 2007-01-21 11:43 filelink*

It seems to me that there is a difference, because filelink now appears the
same size as file. It's just as if we had hard links instead of symlinks.

Here is what I did, I mounted the remote filesystem on /mnt on my client,
the share on the server has a normal Debian Sarge PowerPC filesystem on it.

$ pwd
/mnt/usr
$ ls -l
total 0
drwxr-xr-x 1 root root 0 Feb 15 2005 X11R6
drwxr-xr-x 1 root root 0 Jan 16 2007 bin
drwxr-xr-x 1 root root 0 Jan 16 2007 doc
drwxr-xr-x 1 root root 0 Feb 10 2005 games
drwxr-xr-x 1 root root 0 Jan 16 2007 include
lrwxr-xr-x 1 root root 10 Jan 16 2007 info -> share/info
drwxr-xr-x 1 root root 0 Jan 16 2007 lib
drwxr-xr-x 1 root root 0 Feb 10 2005 local
drwxr-xr-x 1 root root 0 Jan 16 2007 sbin
drwxr-xr-x 1 root root 0 Jan 5 2006 share
drwxr-xr-x 1 root root 0 Dec 15 2004 src
$ ls -l info/
total 249856
-rwxr-xr-x 1 root root 150109 Jul 16 2004 coreutils.info.gz
-rwxr-xr-x 1 root root 1299 Jan 16 2007 dir
-rwxr-xr-x 1 root root 1299 Jan 16 2007 dir.old
-rwxr-xr-x 1 root root 28019 Mar 20 2005 find.info.gz
-rwxr-xr-x 1 root root 26136 Nov 22 2004 grep.info.gz
-rwxr-xr-x 1 root root 12914 Sep 16 2006 gzip.info.gz
-rwxr-xr-x 1 root root 12316 Sep 18 2005 ipc.info.gz
-rwxr-xr-x 1 root root 21432 Jan 23 2005 rl5userman.info.gz
-rwxr-xr-x 1 root root 26647 Dec 1 2004 sed.info.gz
-rwxr-xr-x 1 root root 123382 Dec 1 2006 tar.info.gz
-rwxr-xr-x 1 root root 54876 May 23 2005 wget.info.gz
$ cd ../bin
$ ls -l sh
lrwxr-xr-x 1 root root 4 Jan 16 2007 sh -> bash
$ dd if=sh bs=1 count=6
ELF6+0 records in
6+0 records out
6 bytes transferred in 0.001432 seconds (4190 bytes/sec)

As you can see I now can see the symbolic links perfectly and they work as
expected.

In fact, this patch is working so well that it poses a security risk, as now
the devices on my /mnt/dev directory are not only seen as devices (like they
were seen on 2.4.33) but they also work (which didn't happen on 2.4.33).

Why do you consider this a security problem ? Is any user able to create a
device entry with enough permissions ? As a general rule of thumb, networked
file systems should be mounted with the "nodev" option.

So... for me now the remote filesystem works as if it was a local
filesystem, without any difference of behaviour, not even on special files
like devices or whatever.

As I said before... this behaviour of having the remote device files work...
seems a security problem and I don't think is desirable, other than that it
seems to work well on my PowerPC, I'll try to run the tests on a normal x86
client and report back.

Thanks very much for your tests.

Grant, just to be sure, are you really certain that you tried the fixed kernel ?
It is possible that you booted a wrong kernel during one of your tests. I'm
intrigued by the fact that it changed nothing for you and that it fixed the
problem for Santiago.

Best regards,
Willy

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Relevant Pages

  • NFS problems with through 2.5.x to 2.6.0-test9
    ... When the server is running the ... kernel, as a client the 2.6 series seem to work perfectly, excluding ... Interesting problem arose when I attempted switch the server's kernel to ... with and without nfsv4 support compiled in (was considering testing it at ...
    (Linux-Kernel)
  • "CIFS VFS: server not responding" with some client/server combinations
    ... As it might be kernel related and others ... A Samba server running Debian Etch with latest updates. ... I don't know what's different with Etch on the client, ...
    (Linux-Kernel)
  • Re: Using COM OutProc server on WinCE
    ... nk.exe (kernel process) and you can't popup UI from nk.exe in CE 6.0. ... Microsoft Corporation ... to the client so I guess in that scenario services.exe won't be helpful. ... Server (you'd need SYSGEN_DCOM, see ...
    (microsoft.public.windowsce.app.development)
  • Re: NFS EINVAL on open(... | O_TRUNC) on 2.6.23.9
    ... Thanks to Chuck's help i finally decided to proceed to a git bisect and found the bad patch. ... What isn't quite clear to me is whether this commit causes your user- space server to start failing suddenly, or it causes the client to start sending the special non-standard time stamps in the SETATTR request. ... it would be helpful if you could run this test with a constant kernel version on one side while varying it on the other. ...
    (Linux-Kernel)
  • Re: What doesnt lend itself to OO?
    ... >> proxy and instructs the server to constuct the real object. ... rather than client code. ... If 'clock' is instantiated in the server, ... > for the server interface at the OOA level. ...
    (comp.object)