Re: [PATCH v2 3/3] nbd: fix race between nbd_alloc_config() and module removal

[Wouter Verhelst <w@uter.be>]

On Mon, Sep 13, 2021 at 12:32:37PM +0800, Hou Tao wrote:
> Hi Christoph,
> 
> On 9/9/2021 2:40 PM, Christoph Hellwig wrote:
> > On Tue, Sep 07, 2021 at 11:04:16AM +0800, Hou Tao wrote:
> >> Let me explain first. The reason it works is due to genl_lock_all() in netlink code.
> > Btw, please properly format your mail.  These overly long lines are really
> > hard to read.
> Thanks for reminding.
> >> If the module removal happens before calling try_module_get(), nbd_cleanup() will
> >>
> >> call genl_unregister_family() first, and then genl_lock_all(). genl_lock_all() will
> >>
> >> prevent ops in nbd_connect_genl_ops() from being called, because the calling
> >>
> >> of nbd ops happens in genl_rcv() which needs to acquire cb_lock first.
> > Good.
> >
> >> I have checked multiple genl_ops, it seems that the reason why these genl_ops
> >>
> >> don't need try_module_get() is that these ops don't create new object through
> >>
> >> genl_ops and just control it. However genl_family_rcv_msg_dumpit() will try to
> >>
> >> call try_module_get(), but according to the history (6dc878a8ca39 "netlink: add reference of module in netlink_dump_start"),
> >>
> >> it is because inet_diag_handler_cmd() will call __netlink_dump_start().
> > And now taking a step back:  Why do we even need this extra module
> > reference?  For the case where nbd_alloc_config is called from nbd_open
> > we obviously don't need it.  In the case where it is called from
> > nbd_genl_connect that prevents unloading nbd when there is a configured
> > but not actually nbd device.  Which isn't reallyed need and counter to
> > how other drivers work.
> Yes, the purpose of module ref-counting in nbd_alloc_config() is to force
> the user to disconnect the nbd device manually before module removal.
> And loop device works in the same way. If a file is attached to a loop device,
> an extra module reference will be taken in loop_configure() and the removal
> of loop module will fail. The only difference is that loop driver takes the
> extra ref-count by ioctl, and nbd does it through netlink.

Haven't checked the actual patch, but just wanted to point out:

nbd should do it through netlink *and* ioctl -- the older way to
configure nbd was through ioctl, which we should still support for
backcompat reasons.

(if that's already the case, then ignore what I said :-)

-- 
     w@uter.{be,co.za}
wouter@{grep.be,fosdem.org,debian.org}