hw/vfio/pci: Fix double free of migration_blocker
commit1335d64323be87ee14c766c59fabfb7e9acd7af7
authorMichal Privoznik <mprivozn@redhat.com>
Mon, 18 Nov 2019 17:41:47 +0000 (18 10:41 -0700)
committerAlex Williamson <alex.williamson@redhat.com>
Mon, 18 Nov 2019 17:41:47 +0000 (18 10:41 -0700)
tree1ad9d7c7b33eba26f5eb0cfaf4aae2481bde56f3
parent1bd0f1c9c149c2fb738f381099cec7ad0ee224a9
hw/vfio/pci: Fix double free of migration_blocker

When user tries to hotplug a VFIO device, but the operation fails
somewhere in the middle (in my testing it failed because of
RLIMIT_MEMLOCK forbidding more memory allocation), then a double
free occurs. In vfio_realize() the vdev->migration_blocker is
allocated, then something goes wrong which causes control to jump
onto 'error' label where the error is freed. But the pointer is
left pointing to invalid memory. Later, when
vfio_instance_finalize() is called, the memory is freed again.

In my testing the second hunk was sufficient to fix the bug, but
I figured the first hunk doesn't hurt either.

==169952== Invalid read of size 8
==169952==    at 0xA47DCD: error_free (error.c:266)
==169952==    by 0x4E0A18: vfio_instance_finalize (pci.c:3040)
==169952==    by 0x8DF74C: object_deinit (object.c:606)
==169952==    by 0x8DF7BE: object_finalize (object.c:620)
==169952==    by 0x8E0757: object_unref (object.c:1074)
==169952==    by 0x45079C: memory_region_unref (memory.c:1779)
==169952==    by 0x45376B: do_address_space_destroy (memory.c:2793)
==169952==    by 0xA5C600: call_rcu_thread (rcu.c:283)
==169952==    by 0xA427CB: qemu_thread_start (qemu-thread-posix.c:519)
==169952==    by 0x80A8457: start_thread (in /lib64/libpthread-2.29.so)
==169952==    by 0x81C96EE: clone (in /lib64/libc-2.29.so)
==169952==  Address 0x143137e0 is 0 bytes inside a block of size 48 free'd
==169952==    at 0x4A342BB: free (vg_replace_malloc.c:530)
==169952==    by 0xA47E05: error_free (error.c:270)
==169952==    by 0x4E0945: vfio_realize (pci.c:3025)
==169952==    by 0x76A4FF: pci_qdev_realize (pci.c:2099)
==169952==    by 0x689B9A: device_set_realized (qdev.c:876)
==169952==    by 0x8E2C80: property_set_bool (object.c:2080)
==169952==    by 0x8E0EF6: object_property_set (object.c:1272)
==169952==    by 0x8E3FC8: object_property_set_qobject (qom-qobject.c:26)
==169952==    by 0x8E11DB: object_property_set_bool (object.c:1338)
==169952==    by 0x5E7BDD: qdev_device_add (qdev-monitor.c:673)
==169952==    by 0x5E81E5: qmp_device_add (qdev-monitor.c:798)
==169952==    by 0x9E18A8: do_qmp_dispatch (qmp-dispatch.c:132)
==169952==  Block was alloc'd at
==169952==    at 0x4A35476: calloc (vg_replace_malloc.c:752)
==169952==    by 0x51B1158: g_malloc0 (in /usr/lib64/libglib-2.0.so.0.6000.6)
==169952==    by 0xA47357: error_setv (error.c:61)
==169952==    by 0xA475D9: error_setg_internal (error.c:97)
==169952==    by 0x4DF8C2: vfio_realize (pci.c:2737)
==169952==    by 0x76A4FF: pci_qdev_realize (pci.c:2099)
==169952==    by 0x689B9A: device_set_realized (qdev.c:876)
==169952==    by 0x8E2C80: property_set_bool (object.c:2080)
==169952==    by 0x8E0EF6: object_property_set (object.c:1272)
==169952==    by 0x8E3FC8: object_property_set_qobject (qom-qobject.c:26)
==169952==    by 0x8E11DB: object_property_set_bool (object.c:1338)
==169952==    by 0x5E7BDD: qdev_device_add (qdev-monitor.c:673)

Fixes: f045a0104c8c ("vfio: unplug failover primary device before migration")
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
hw/vfio/pci.c