On Tue, 20 Jul 2021 14:13:13 +0200
Hans Verkuil [off-list ref] wrote:

quoted

On 06/07/2021 16:50, Pavel Skripkin wrote:

quoted

Syzbot reported corrupted list in em28xx driver. The problem was in
non-reinitialized lists on disconnect. Since all 2 lists are global
variables and driver can be connected and disconnected many times we
should call INIT_LIST_HEAD() in .disconnect method to prevent
corrupted list entries.

Fixes: 1a23f81b7dc3 ("V4L/DVB (9979): em28xx: move usb probe code
to a proper place") Reported-by:
syzbot+a6969ef522a36d3344c9@syzkaller.appspotmail.com
Signed-off-by: Pavel Skripkin <redacted> ---
 drivers/media/usb/em28xx/em28xx-cards.c | 2 ++
 drivers/media/usb/em28xx/em28xx-core.c  | 6 ++++++
 drivers/media/usb/em28xx/em28xx.h       | 1 +
 3 files changed, 9 insertions(+)

diff --git a/drivers/media/usb/em28xx/em28xx-cards.c

b/drivers/media/usb/em28xx/em28xx-cards.c index
ba9292e2a587..8b1ff79c37a0 100644 ---
a/drivers/media/usb/em28xx/em28xx-cards.c +++
b/drivers/media/usb/em28xx/em28xx-cards.c @@ -4148,6 +4148,8 @@
static void em28xx_usb_disconnect(struct usb_interface *intf)
dev->dev_next = NULL; }
 	kref_put(&dev->ref, em28xx_free_device);
+
+	em28xx_reset_lists();
 }
 
 static int em28xx_usb_suspend(struct usb_interface *intf,

diff --git a/drivers/media/usb/em28xx/em28xx-core.c

b/drivers/media/usb/em28xx/em28xx-core.c index
584fa400cd7d..03970ed00dba 100644 ---
a/drivers/media/usb/em28xx/em28xx-core.c +++
b/drivers/media/usb/em28xx/em28xx-core.c @@ -1131,6 +1131,12 @@
void em28xx_init_extension(struct em28xx *dev)
mutex_unlock(&em28xx_devlist_mutex); }
 
+void em28xx_reset_lists(void)
+{
+	INIT_LIST_HEAD(&em28xx_devlist);
+	INIT_LIST_HEAD(&em28xx_extension_devlist);

This needs a mutex_lock(&em28xx_devlist_mutex);

But actually, I don't think this is right: if there are multiple
em28xx devices, then I think if you disconnect one, then the other is
- with this code - also removed from the list.

Can you give a link to the actual syzbot bug? I'm not at all sure you
are fixing the right thing here.

Hi, Hans!

I guess, I missed my coffee that morning, sorry :) This patch looks
complety wrong, of course.

I've took a close look at this bug one more time, and I found the true
root case of this bug.

If em28xx dev has dev_next pointer we need to close dev->next extension
on disconnect to avoid UAF and corrupted list bug. So, something like
this should work:

diff --git a/drivers/media/usb/em28xx/em28xx-cards.c b/drivers/media/usb/em28xx/em28xx-cards.c
index c1e0dccb7408..d56b040e1bd7 100644
--- a/drivers/media/usb/em28xx/em28xx-cards.c
+++ b/drivers/media/usb/em28xx/em28xx-cards.c

@@ -4139,8 +4139,10 @@ static void em28xx_usb_disconnect(struct usb_interface *intf)
 
 	em28xx_close_extension(dev);
 
-	if (dev->dev_next)
+	if (dev->dev_next) {
 		em28xx_release_resources(dev->dev_next);
+		em28xx_close_extension(dev->dev_next);
+	}
 	em28xx_release_resources(dev);
 
 	if (dev->dev_next) {

How to You feel about it? I am going to send this patch for syzbot
testing