Re: [PATCH 2/4] test: firmware_class: add asynchronous request trigger

Hi Kees,

On Wed, Dec 09, 2015 at 01:09:02PM -0800, Kees Cook wrote:

On Tue, Dec 8, 2015 at 6:38 PM, Brian Norris
[off-list ref] wrote:

quoted

diff --git a/lib/test_firmware.c b/lib/test_firmware.c
index 841191061816..ba0a12d0301d 100644
--- a/lib/test_firmware.c
+++ b/lib/test_firmware.c

@@ -12,6 +12,7 @@
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/printk.h>
+#include <linux/completion.h>
 #include <linux/firmware.h>
 #include <linux/device.h>
 #include <linux/fs.h>

@@ -81,6 +82,57 @@ out:
 }
 static DEVICE_ATTR_WO(trigger_request);

+static DECLARE_COMPLETION(async_fw_done);
+
+static void trigger_async_request_cb(const struct firmware *fw, void *context)
+{
+       test_firmware = fw;
+       complete(&async_fw_done);
+}
+
+static ssize_t trigger_async_request_store(struct device *dev,
+                                          struct device_attribute *attr,
+                                          const char *buf, size_t count)
+{
+       int rc;
+       char *name;
+
+       name = kzalloc(count + 1, GFP_KERNEL);
+       if (!name)
+               return -ENOSPC;
+       memcpy(name, buf, count);

It strikes me that this (and the existing code) should use kstrndup
instead, since the request_firmware* interfaces will ignore \0 bytes
in the name:

    name = kstrndup(buf, count, GFP_KERNEL);
    if (!name)
        return -ENOSPC;

Thought of that at some point, then for some reason I didn't do it.
Probably laziness...

Will do in a v2, along with the more important fix below.

quoted

+
+       pr_info("loading '%s'\n", name);
+
+       mutex_lock(&test_fw_mutex);
+       release_firmware(test_firmware);
+       test_firmware = NULL;
+       rc = request_firmware_nowait(THIS_MODULE, 1, name, dev, GFP_KERNEL,
+                                    NULL, trigger_async_request_cb);
+       /* Free 'name' ASAP, to test for race conditions */
+       kfree(name);
+       if (rc) {
+               pr_info("async load of '%s' failed: %d\n", name, rc);

Well, that's a little TOO soon. :) The pr_info uses it still.

Haha, yeah... nice catch.

I was also thinking, since use-after-free isn't necessarily immediately
obvious (this worked fine in my testing), that maybe we could poison the
buffer before kfree()'ing? Like:

	name = ...;
	len = strlen(name);

	...

	rc = request_firmware_nowait(...);
	if (rc) {
		pr_info("...");
		kfree(name);
		goto out;
	}
	/*
	 * Clear out the name, to test for race conditions with the
	 * async request
	 */
	memset(name, 0, len);
	kfree(name);

quoted

+               goto out;
+       }
+
+       wait_for_completion(&async_fw_done);
+
+       if (test_firmware) {
+               pr_info("loaded: %zu\n", test_firmware->size);
+               rc = count;
+       } else {
+               pr_err("failed to async load firmware\n");
+               rc = -ENODEV;
+       }
+
+out:
+       mutex_unlock(&test_fw_mutex);
+
+       return rc;
+}
+static DEVICE_ATTR_WO(trigger_async_request);
+
 static int __init test_firmware_init(void)
 {
        int rc;

...

Brian