RE: [PATCH] net: tls: fix a memory leak bug

[Vakul Garg]

> -----Original Message-----
> From: Jakub Kicinski <jakub.kicinski@xxxxxxxxxxxxx>
> Sent: Thursday, April 25, 2019 6:39 AM
> To: Wenwen Wang <wang6495@xxxxxxx>
> Cc: Boris Pismenny <borisp@xxxxxxxxxxxx>; Aviad Yehezkel
> <aviadye@xxxxxxxxxxxx>; Dave Watson <davejwatson@xxxxxx>; John
> Fastabend <john.fastabend@xxxxxxxxx>; Daniel Borkmann
> <daniel@xxxxxxxxxxxxx>; David S. Miller <davem@xxxxxxxxxxxxx>; open
> list:NETWORKING [TLS] <netdev@xxxxxxxxxxxxxxx>; open list <linux-
> kernel@xxxxxxxxxxxxxxx>; Vakul Garg <vakul.garg@xxxxxxx>
> Subject: Re: [PATCH] net: tls: fix a memory leak bug
> 
> On Wed, 24 Apr 2019 15:18:07 -0500, Wenwen Wang wrote:
> > In decrypt_internal(), a memory block 'mem' is allocated through
> > kmalloc() to hold aead_req, sgin[], sgout[], aad, and iv. This memory
> > block should be freed after it is used, before this function is
> > returned. However, if the return value of the function invocation of
> > tls_do_decryption() is -EINPROGRESS, this memory block is actually not
> > freed, which is a memory leak bug.
> >
> > To fix this issue, free the allocated block before the error code
> > -EINPROGRESS is returned.
> >
> > Signed-off-by: Wenwen Wang <wang6495@xxxxxxx>
> 
> Did you find this by code inspection or is this provable at runtime (kmemleak
> or such)?
> 
> > diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c index
> > b50ced8..22445bb 100644
> > --- a/net/tls/tls_sw.c
> > +++ b/net/tls/tls_sw.c
> > @@ -1445,8 +1445,10 @@ static int decrypt_internal(struct sock *sk,
> struct sk_buff *skb,
> >  	/* Prepare and submit AEAD request */
> >  	err = tls_do_decryption(sk, skb, sgin, sgout, iv,
> >  				data_len, aead_req, async);
> > -	if (err == -EINPROGRESS)
> > +	if (err == -EINPROGRESS) {
> > +		kfree(mem);
> >  		return err;
> 
> Mm... don't think so.
> 
> -EINPROGRESS is special, it means something is working on the request
> asynchronously here.
> 
> I think this memory is freed in tls_decrypt_done():
> 
> 	kfree(aead_req);
> 
> Note that aead_req is the first member of the allocated buffer.
> 
> CCing Vakul

The patch is wrong. 
Valid reasons have been given against the patch.

> 
> > +	}
> >
> >  	/* Release the pages in case iov was mapped to pages */
> >  	for (; pages > 0; pages--)