On 10/9/20 10:36 AM, Michael Ellerman wrote:
Vasant Hegde <hegdevas...@linux.vnet.ibm.com> writes:
diff --git a/arch/powerpc/platforms/powernv/opal-dump.c
b/arch/powerpc/platforms/powernv/opal-dump.c
index 543c816fa99e..7e6eeedec32b 100644
--- a/arch/powerpc/platforms/powernv/opal-dump.c
+++ b/arch/powerpc/platforms/powernv/opal-dump.c
@@ -346,21 +345,39 @@ static struct dump_obj *create_dump_obj(uint32_t id,
size_t size,
rc = kobject_add(&dump->kobj, NULL, "0x%x-0x%x", type, id);
if (rc) {
kobject_put(&dump->kobj);
- return NULL;
+ return;
}
+ /*
+ * As soon as the sysfs file for this dump is created/activated there is
+ * a chance the opal_errd daemon (or any userspace) might read and
+ * acknowledge the dump before kobject_uevent() is called. If that
+ * happens then there is a potential race between
+ * dump_ack_store->kobject_put() and kobject_uevent() which leads to a
+ * use-after-free of a kernfs object resulting in a kernel crash.
+ *
+ * To avoid that, we need to take a reference on behalf of the bin file,
+ * so that our reference remains valid while we call kobject_uevent().
+ * We then drop our reference before exiting the function, leaving the
+ * bin file to drop the last reference (if it hasn't already).
+ */
+
+ /* Take a reference for the bin file */
+ kobject_get(&dump->kobj);
rc = sysfs_create_bin_file(&dump->kobj, &dump->dump_attr);
if (rc) {
kobject_put(&dump->kobj);
- return NULL;
+ /* Drop reference count taken for bin file */
+ kobject_put(&dump->kobj);
+ return;
}
pr_info("%s: New platform dump. ID = 0x%x Size %u\n",
__func__, dump->id, dump->size);
kobject_uevent(&dump->kobj, KOBJ_ADD);
-
- return dump;
+ /* Drop reference count taken for bin file */
+ kobject_put(&dump->kobj);
}
I think this would be better if it was reworked along the lines of:
aea948bb80b4 ("powerpc/powernv/elog: Fix race while processing OPAL error log
event.")
Sure. Will fix it in v2.
-Vasant
