[PATCH] Bluetooth: RFCOMM: Use skb_trim to trim checksum

[PATCH] Bluetooth: RFCOMM: Use skb_trim to trim checksum

[Soenke Huster]

Use the skb helper instead of direct manipulation. This fixes the
following page fault, when connecting my Android phone:

    BUG: unable to handle page fault for address: ffffed1021de29ff
    #PF: supervisor read access in kernel mode
    #PF: error_code(0x0000) - not-present page
    RIP: 0010:rfcomm_run+0x831/0x4040 (net/bluetooth/rfcomm/core.c:1751)

Signed-off-by: Soenke Huster <soenke.huster@eknoes.de>
---
 net/bluetooth/rfcomm/core.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/net/bluetooth/rfcomm/core.c b/net/bluetooth/rfcomm/core.c
index 7324764384b6..7360e905d045 100644
--- a/net/bluetooth/rfcomm/core.c
+++ b/net/bluetooth/rfcomm/core.c
@@ -1747,8 +1747,8 @@ static struct rfcomm_session *rfcomm_recv_frame(struct rfcomm_session *s,
 	type = __get_type(hdr->ctrl);
 
 	/* Trim FCS */
-	skb->len--; skb->tail--;
-	fcs = *(u8 *)skb_tail_pointer(skb);
+	skb_trim(skb, skb->len - 1);
+	fcs = *(skb->data + skb->len);
 
 	if (__check_fcs(skb->data, type, fcs)) {
 		BT_ERR("bad checksum in packet");
-- 
2.36.1

Re: [PATCH] Bluetooth: RFCOMM: Use skb_trim to trim checksum

[Eric Dumazet]

On Wed, Jun 8, 2022 at 6:51 AM Soenke Huster <soenke.huster@eknoes.de> wrote:
>
> Use the skb helper instead of direct manipulation. This fixes the
> following page fault, when connecting my Android phone:
>
>     BUG: unable to handle page fault for address: ffffed1021de29ff
>     #PF: supervisor read access in kernel mode
>     #PF: error_code(0x0000) - not-present page
>     RIP: 0010:rfcomm_run+0x831/0x4040 (net/bluetooth/rfcomm/core.c:1751)
>
> Signed-off-by: Soenke Huster <soenke.huster@eknoes.de>
> ---
>  net/bluetooth/rfcomm/core.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/net/bluetooth/rfcomm/core.c b/net/bluetooth/rfcomm/core.c
> index 7324764384b6..7360e905d045 100644
> --- a/net/bluetooth/rfcomm/core.c
> +++ b/net/bluetooth/rfcomm/core.c
> @@ -1747,8 +1747,8 @@ static struct rfcomm_session *rfcomm_recv_frame(struct rfcomm_session *s,
>         type = __get_type(hdr->ctrl);
>
>         /* Trim FCS */
> -       skb->len--; skb->tail--;
> -       fcs = *(u8 *)skb_tail_pointer(skb);
> +       skb_trim(skb, skb->len - 1);
> +       fcs = *(skb->data + skb->len);
>

Hmmm... I do not see any difference before/after in term of memory
dereference to get fcs.

I think you should give more details on how exactly the bug triggers.

Re: [PATCH] Bluetooth: RFCOMM: Use skb_trim to trim checksum

[Sönke Huster]

Hi Eric,

On 08.06.22 17:33, Eric Dumazet wrote:
> On Wed, Jun 8, 2022 at 6:51 AM Soenke Huster <soenke.huster@eknoes.de> wrote:
>>
>> Use the skb helper instead of direct manipulation. This fixes the
>> following page fault, when connecting my Android phone:
>>
>>     BUG: unable to handle page fault for address: ffffed1021de29ff
>>     #PF: supervisor read access in kernel mode
>>     #PF: error_code(0x0000) - not-present page
>>     RIP: 0010:rfcomm_run+0x831/0x4040 (net/bluetooth/rfcomm/core.c:1751)
>>
>> Signed-off-by: Soenke Huster <soenke.huster@eknoes.de>
>> ---
>>  net/bluetooth/rfcomm/core.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/net/bluetooth/rfcomm/core.c b/net/bluetooth/rfcomm/core.c
>> index 7324764384b6..7360e905d045 100644
>> --- a/net/bluetooth/rfcomm/core.c
>> +++ b/net/bluetooth/rfcomm/core.c
>> @@ -1747,8 +1747,8 @@ static struct rfcomm_session *rfcomm_recv_frame(struct rfcomm_session *s,
>>         type = __get_type(hdr->ctrl);
>>
>>         /* Trim FCS */
>> -       skb->len--; skb->tail--;
>> -       fcs = *(u8 *)skb_tail_pointer(skb);
>> +       skb_trim(skb, skb->len - 1);
>> +       fcs = *(skb->data + skb->len);
>>
> 
> Hmmm... I do not see any difference before/after in term of memory
> dereference to get fcs.
> 
> I think you should give more details on how exactly the bug triggers.

Sorry, yesterday I was not able to track down why exactly it crashes,
but by now I think I figured it out.

The crash happens when using Bluetooth in a virtual machine.
On connecting my Android phone to the physical controller which I use 
inside the virtual machine via the VirtIO driver, after some seconds
the crash occurs.

Before the trimming step, I examined the skb in gdb and saw, that 
skb->tail is zero. Thus, skb->tail--; modifies the unsigned integer to -1
resp. MAX_UINT. In skb_tail_pointer, skb->head + skb->tail is calculated
which results in the page fault.

By using skb_trim, skb->tail is set to the accurate value and thus the
issue is fixed.

I am not an expert in the Linux kernel area, do you think there is an
underlying issue anywhere else? When using my Android phone on my host
computer, I do not have that problem - it might be in some 
(e.g. virtio_bt?) driver? On the other hand, with the patch my problem
is solved and the phone is usable in the virtual machine!