Re: [PATCH RESEND] PCI: s390: Fix use-after-free of PCI bus resources with s390 per-function hotplug

[Bjorn Helgaas]

[+cc Lukas]

On Wed, Feb 22, 2023 at 05:54:55PM +0100, Niklas Schnelle wrote:
> On Mon, 2023-02-20 at 13:53 +0100, Niklas Schnelle wrote:
> > On Fri, 2023-02-17 at 17:15 -0600, Bjorn Helgaas wrote:
> > > On Tue, Feb 14, 2023 at 10:49:10AM +0100, Niklas Schnelle wrote:
> ---8<---

> > >   - What about zpci_bus_scan_device()?  Why does it call both
> > >     pci_bus_add_device() and pci_bus_add_devices()?  The latter will
> > >     just call the former, so it looks redundant.  And the latter is
> > >     locked but not the former?
> > 
> > Hmm. great find. This seems to have been weird and redundant since I
> > first used that pattern in 3047766bc6ec ("s390/pci: fix enabling a
> > reserved PCI function"). I think maybe then the reason for this was
> > that prior to 960ac3626487 ("s390/pci: allow zPCI zbus without a
> > function zero") when the newly enabled is devfn == 0 there could be
> > functions from the same bus which would not have been added yet. I'm
> > not sure though. That was definitely the idea behind the
> > zpci_bus_scan_bus() in zpci_scan_configured_devices() that is also
> > redundant now as we can now scan each function as it appears.
> 
> I'm working on cleaning this up but I'm a little confused by what
> exactly needs to be under the pci_rescan_remove lock. For example the
> pci_bus_add_device(virtfn) at the end of pci_iov_add_virtfn() doesn't
> seem to be under the lock while most calls to pci_bus_add_devices()
> are, most prominently the one in acpi_pci_root_add() which I assume is
> what is used on most x86 systems. Any hints?
> 
> Also I think my original thought here might have been a premature worry
> about PCI-to-PCI bridges thinking that adding the new device could lead
> to more devices appearing. Of course actually thinking about it a bit
> more there are quite a few other things that won't work without further
> changes if we wanted to add bridges e.g. we would need to create
> zpci_dev structs for these somewhere.

Hmm.  Good question.  Off the top of my head, I can't explain the
difference between pci_rescan_remove_lock and pci_bus_sem, so I'm
confused, too.  I added Lukas in case he has a ready explanation.

Bjorn