]> git.kernelconcepts.de Git - karo-tx-linux.git/commit
PCI: Check parent kobject in pci_destroy_dev()
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tue, 14 Jan 2014 19:04:51 +0000 (12:04 -0700)
committerBjorn Helgaas <bhelgaas@google.com>
Wed, 15 Jan 2014 17:34:13 +0000 (10:34 -0700)
commit8a4c5c329de716996eea03d93753ccbb5406072b
tree4a9c74354b7f7bbb39ae0426246d4b9baff3cc7f
parenta83919e0940f6eb8f77ab1d602a063f8a6703117
PCI: Check parent kobject in pci_destroy_dev()

If pci_stop_and_remove_bus_device() is run concurrently for a device and
its parent bridge via remove_callback(), both code paths attempt to acquire
pci_rescan_remove_lock.  If the child device removal acquires it first,
there will be no problems.  However, if the parent bridge removal acquires
it first, it will eventually execute pci_destroy_dev() for the child
device, but that device object will not be freed yet due to the reference
held by the concurrent child removal.  Consequently, both
pci_stop_bus_device() and pci_remove_bus_device() will be executed for that
device unnecessarily and pci_destroy_dev() will see a corrupted list head
in that object.  Moreover, an excess put_device() will be executed for that
device in that case which may lead to a use-after-free in the final
kobject_put() done by sysfs_schedule_callback_work().

To avoid that problem, make pci_destroy_dev() check if the device's parent
kobject is NULL, which only happens after device_del() has already run for
it.  Make pci_destroy_dev() return immediately whithout doing anything in
that case.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
drivers/pci/remove.c