netdev - Re: [Patch bpf-next v3 2/4] skmsg: implement slab allocator cache for sk

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <aDipm6P+RWGD8j4M@pop-os.localdomain>
Date: Thu, 29 May 2025 11:38:19 -0700
From: Cong Wang <xiyou.wangcong@...il.com>
To: Zijian Zhang <zijianzhang@...edance.com>
Cc: John Fastabend <john.fastabend@...il.com>, netdev@...r.kernel.org,
	bpf@...r.kernel.org, zhoufeng.zf@...edance.com,
	jakub@...udflare.com, Cong Wang <cong.wang@...edance.com>
Subject: Re: [Patch bpf-next v3 2/4] skmsg: implement slab allocator cache
 for sk_msg

On Wed, May 28, 2025 at 05:49:22PM -0700, Zijian Zhang wrote:
> On 5/28/25 5:04 PM, John Fastabend wrote:
> > On 2025-05-19 13:36:26, Cong Wang wrote:
> > > From: Zijian Zhang <zijianzhang@...edance.com>
> > > 
> > > Optimizing redirect ingress performance requires frequent allocation and
> > > deallocation of sk_msg structures. Introduce a dedicated kmem_cache for
> > > sk_msg to reduce memory allocation overhead and improve performance.
> > > 
> > > Reviewed-by: Cong Wang <cong.wang@...edance.com>
> > > Signed-off-by: Zijian Zhang <zijianzhang@...edance.com>
> > > ---
> > >   include/linux/skmsg.h | 21 ++++++++++++---------
> > >   net/core/skmsg.c      | 28 +++++++++++++++++++++-------
> > >   net/ipv4/tcp_bpf.c    |  5 ++---
> > >   3 files changed, 35 insertions(+), 19 deletions(-)
> > > 
> > > diff --git a/include/linux/skmsg.h b/include/linux/skmsg.h
> > > index d6f0a8cd73c4..bf28ce9b5fdb 100644
> > > --- a/include/linux/skmsg.h
> > > +++ b/include/linux/skmsg.h
> > > @@ -121,6 +121,7 @@ struct sk_psock {
> > >   	struct rcu_work			rwork;
> > >   };
> > > +struct sk_msg *sk_msg_alloc(gfp_t gfp);
> > >   int sk_msg_expand(struct sock *sk, struct sk_msg *msg, int len,
> > >   		  int elem_first_coalesce);
> > >   int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
> > > @@ -143,6 +144,8 @@ int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
> > >   		   int len, int flags);
> > >   bool sk_msg_is_readable(struct sock *sk);
> > > +extern struct kmem_cache *sk_msg_cachep;
> > > +
> > >   static inline void sk_msg_check_to_free(struct sk_msg *msg, u32 i, u32 bytes)
> > >   {
> > >   	WARN_ON(i == msg->sg.end && bytes);
> > > @@ -319,6 +322,13 @@ static inline void sock_drop(struct sock *sk, struct sk_buff *skb)
> > >   	kfree_skb(skb);
> > >   }
> > > +static inline void kfree_sk_msg(struct sk_msg *msg)
> > > +{
> > > +	if (msg->skb)
> > > +		consume_skb(msg->skb);
> > > +	kmem_cache_free(sk_msg_cachep, msg);
> > > +}
> > > +
> > >   static inline bool sk_psock_queue_msg(struct sk_psock *psock,
> > >   				      struct sk_msg *msg)
> > >   {
> > > @@ -330,7 +340,7 @@ static inline bool sk_psock_queue_msg(struct sk_psock *psock,
> > >   		ret = true;
> > >   	} else {
> > >   		sk_msg_free(psock->sk, msg);
> > > -		kfree(msg);
> > > +		kfree_sk_msg(msg);
> > 
> > Isn't this a potential use after free on msg->skb? The sk_msg_free() a
> > line above will consume_skb() if it exists and its not nil set so we would
> > consume_skb() again?
> > 
> 
> Thanks to sk_msg_free, after consuming the skb, it invokes sk_msg_init
> to make msg->skb NULL to prevent further double free.
> 
> To avoid the confusion, we can replace kfree_sk_msg here with
> kmem_cache_free.
> 

Right, the re-initialization in sk_msg_free() is indeed confusing, maybe
it is time to clean up its logic? For example, separate sk_msg_init()
out from sk_msg_free().

I can add a separate patch for this in next update, if people prefer.

Thanks!