Re: set dev->rfkill to NULL in device cleanup routine

On 01/09/2021 09:39, LinMa wrote:
> In nfc_unregister_device() function, the dev->rfkill is forgotten to set to NULL after the rfkill_destroy(). This may lead to possible cocurrency UAF in other functions like nfc_dev_up().

Commit msg should be wrapper at 75 char.
https://elixir.bootlin.com/linux/v5.13/source/Documentation/process/submitting-patches.rst#L124


Use also scripts/get_maintainers.pl to get list of people and lists
you need to CC. You skipped Networking maintainers and two mailing lists.

> 
> The FREE chain is like
> 

Please trim multiple blank lines and organize the commit msg to be readable.
No need to paste existing code into the commit msg.

> 
> void nfc_unregister_device(struct nfc_dev *dev)
> {
>   int rc;
>   pr_debug("dev_name=%s\n", dev_name(&dev->dev));
>   if (dev->rfkill) {
>     rfkill_unregister(dev->rfkill);
>     rfkill_destroy(dev->rfkill);
>   // ......
> }
> 
> 
> 
> The USE chain is like
> 
> 
> static int nfc_genl_dev_up(struct sk_buff *skb, struct genl_info *info)
> {
>   struct nfc_dev *dev;
>   int rc;
>   u32 idx;
>   if (!info->attrs[NFC_ATTR_DEVICE_INDEX])
>     return -EINVAL;
>   idx = nla_get_u32(info->attrs[NFC_ATTR_DEVICE_INDEX]);
>   dev = nfc_get_device(idx);
>   if (!dev)
>     return -ENODEV;
>   rc = nfc_dev_up(dev);
> 
>   // ......
> }
> 
> 
> int nfc_dev_up(struct nfc_dev *dev)
> {
>   int rc = 0;
>   pr_debug("dev_name=%s\n", dev_name(&dev->dev));
>   device_lock(&dev->dev);
>   if (dev->rfkill && rfkill_blocked(dev->rfkill)) { // dev->rfkill is not NULL here
>     rc = -ERFKILL;
>     goto error;
>   }
>   // ......
> }
> 
> 
> The FREE chain and USE chain can be like below (as there is no locking protection).

Something is missing.

> 
> 
> Therefore, the below patch can be added.

Use imperative form:
https://elixir.bootlin.com/linux/v5.13/source/Documentation/process/submitting-patches.rst#L89

> 
> 
> Signed-off-by: Lin Ma <[email protected]>
> ---
>  net/nfc/core.c | 1 +
>  1 file changed, 1 insertion(+)
> diff --git a/net/nfc/core.c b/net/nfc/core.c
> index 573c80c6ff7a..d0b3224e65d7 100644
> --- a/net/nfc/core.c
> +++ b/net/nfc/core.c
> @@ -1157,6 +1157,7 @@ void nfc_unregister_device(struct nfc_dev *dev)
>   if (dev->rfkill) {
>   rfkill_unregister(dev->rfkill);
>   rfkill_destroy(dev->rfkill);
> + dev->rfkill = NULL;

This is not a valid patch. Does not match the code.
For example, use git format-patch and git send-email.

About the topic:
Your code does not prevent a race condition, since you say there is no
locking. Even if you move dev->rfkill=NULL before rfkill_unregister(),
still nfc_dev_up() could happen between.

The questions are:
1. Whether nfc_unregister_device() can happen after nfc_get_device()?
2. Whether netlink nfc_genl_dev_up() can happen after nfc_unregister_device()
started.



>   }
>   if (dev->ops->check_presence) {
> --
> 2.32.0
> _______________________________________________
> Linux-nfc mailing list -- linux-nfc(a)lists.01.org
> To unsubscribe send an email to linux-nfc-leave(a)lists.01.org
> %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
> 


Best regards,
Krzysztof