linux-kernel - Re: [PATCH linux-next] PCI: Fix the order in unregister path

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 30 Aug 2021 09:55:54 -0500
From:   Rob Herring <robh@...nel.org>
To:     Yajun Deng <yajun.deng@...ux.dev>
Cc:     Bjorn Helgaas <bhelgaas@...gle.com>, Arnd Bergmann <arnd@...db.de>,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        PCI <linux-pci@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH linux-next] PCI: Fix the order in unregister path

On Thu, Aug 26, 2021 at 9:39 PM <yajun.deng@...ux.dev> wrote:
>
> August 26, 2021 8:01 PM, "Rob Herring" <robh@...nel.org> wrote:
>
> > On Wed, Aug 25, 2021 at 10:57 PM <yajun.deng@...ux.dev> wrote:
> >
> >> August 25, 2021 9:55 PM, "Rob Herring" <robh@...nel.org> wrote:
> >>
> >> On Wed, Aug 25, 2021 at 3:34 AM Yajun Deng <yajun.deng@...ux.dev> wrote:
> >>
> >> device_del() should be called first and then called put_device() in
> >> unregister path, becase if that the final reference count, the device
> >> will be cleaned up via device_release() above. So use device_unregister()
> >> instead.
> >>
> >> Fixes: 9885440b16b8 (PCI: Fix pci_host_bridge struct device release/free handling)
> >> Signed-off-by: Yajun Deng <yajun.deng@...ux.dev>
> >> ---
> >> drivers/pci/probe.c | 4 +---
> >> 1 file changed, 1 insertion(+), 3 deletions(-)
> >>
> >> NAK.
> >>
> >> The current code is correct. Go read the comments for device_add/device_del.
> >>
> >> But the device_unregister() is only contains device_del() and put_device(). It just put
> >> device_del() before put_device().
> >
> > And that is the wrong order as we want to undo what the code above
> > did. The put_device here is for the get_device we did. The put_device
> > in device_unregister is for the get_device that device_register did
> > (on success only).
> >
> > Logically, it is wrong too to call unregister if register failed. That
> > would be like doing this:

You are right that the register and unregister are different devices.
However, your change is still wrong. The device_register is actually
irrelevant.

> >
> > p = malloc(1);
> > if (!p)
> > free(p);
> >
> This is the raw code:
>         err = device_register(&bus->dev);
>         if (err)
>                 goto unregister;
> unregister:
>         put_device(&bridge->dev);
>         device_del(&bridge->dev);

The pertinent parts are this:

err = device_add(&bridge->dev);  // which calls get_device() itself,
so there's the first ref
if (err) {
    put_device(&bridge->dev);
    goto free;
}
bus->bridge = get_device(&bridge->dev);  // This is the 2nd ref which
the PCI core holds
...
unregister:
    put_device(&bridge->dev);  // This is the put for the get_device
just above here.
    device_del(&bridge->dev);  // Then this does the 2nd put.

The get_device and put_device are paired, and the device_add and
device_del are paired.

As I said earlier, go read the kerneldoc for device_add. For your
convenience, here's the important part:

device_add:
 * Rule of thumb is: if device_add() succeeds, you should call
 * device_del() when you want to get rid of it. If device_add() has
 * *not* succeeded, use *only* put_device() to drop the reference
 * count.

device_del:
 * NOTE: this should be called manually _iff_ device_add() was
 * also called manually.


Rob