[linux-cifs-client] Improving auto.smb for user mounts

[Joe Krahn]

simo wrote:
> On Tue, 2008-01-29 at 11:29 -0500, Joe Krahn wrote:
> 
>> My understanding is that Windows shares are designed for user-level
>> network connections, rather than system level auto-mounting, which is
>> why a new user requires a new session.
> 
> And this is the only sane option for network mounts unless you want to
> trust the client machine fully.
User-level access is usually better, but a system-level mount can be
useful in some cases, if the client can be fully trusted.
> 
>> But, POSIX is designed around
>> system-level network mounts, which is why Samba created smbsh in order
>> to emulate a user-level virtual network file system.
> 
> Not sure why you mention smbsh, anyway system-level mounts are not at
> odds with user authenticated connections.
Are you talking about a "file browser", or mapping user authenticated
connections to actual filesystem mount points?

> 
>> Don't you need to implement either a user-level virtual file system, or
>> user-level access for files in the same system-level network mount
>> point? Or, is the plan for each user to have their own "/smb" automount
>> base? All of these seem a bit 'fancy' to me. If it is already planned,
>> is there documentation somewhere discussing the design?
> 
> The kernel always knows which user is accessing a mount point so
> "roughly" all you need is to switch to the right user authenticated
> session depending on which user is making the specific request to the
> filesystem.
> This means a new authentication for each user walking a cifs mount
> point, but at least it does not mean a new connection as cifs can
> multiplex multiple sessions.
> 
> Simo.
This makes sense, but how do you present a different connection to
different users on the same mount point? I mention smbsh because it was
created to emulate this outside of the kernel. You are talking about
changes in the way the kernel interacts with the filesystem drivers, right?

Joe