[bpf-next,v1,2/5] af_unix: add unix_stream_proto for sockmap

unix_stream_proto is similar to unix_dgram_proto.
Also implement stream related sockmap functions.

Add dgram key words to those dgram specific functions.

Signed-off-by: Jiang Wang <jiang.wang@bytedance.com>
Reviewed-by: Cong Wang <cong.wang@bytedance.com>.
---
 include/net/af_unix.h |  8 +++-
 net/core/sock_map.c   |  8 +++-
 net/unix/af_unix.c    | 77 ++++++++++++++++++++++++++++++-----
 net/unix/unix_bpf.c   | 93 +++++++++++++++++++++++++++++++++----------
 4 files changed, 151 insertions(+), 35 deletions(-)

Jiang Wang wrote:
> unix_stream_proto is similar to unix_dgram_proto.
> Also implement stream related sockmap functions.
> 
> Add dgram key words to those dgram specific functions.
> 
> Signed-off-by: Jiang Wang <jiang.wang@bytedance.com>
> Reviewed-by: Cong Wang <cong.wang@bytedance.com>.
> ---

Overall LGTM a few small question/comments below.

>  include/net/af_unix.h |  8 +++-
>  net/core/sock_map.c   |  8 +++-
>  net/unix/af_unix.c    | 77 ++++++++++++++++++++++++++++++-----
>  net/unix/unix_bpf.c   | 93 +++++++++++++++++++++++++++++++++----------
>  4 files changed, 151 insertions(+), 35 deletions(-)
> 
> diff --git a/include/net/af_unix.h b/include/net/af_unix.h
> index 435a2c3d5..5d04fbf8a 100644
> --- a/include/net/af_unix.h
> +++ b/include/net/af_unix.h
> @@ -84,6 +84,8 @@ long unix_outq_len(struct sock *sk);
>  
>  int __unix_dgram_recvmsg(struct sock *sk, struct msghdr *msg, size_t size,
>  			 int flags);
> +int __unix_stream_recvmsg(struct sock *sk, struct msghdr *msg, size_t size,
> +			  int flags);
>  #ifdef CONFIG_SYSCTL
>  int unix_sysctl_register(struct net *net);
>  void unix_sysctl_unregister(struct net *net);
> @@ -93,9 +95,11 @@ static inline void unix_sysctl_unregister(struct net *net) {}
>  #endif
>  
>  #ifdef CONFIG_BPF_SYSCALL
> -extern struct proto unix_proto;
> +extern struct proto unix_dgram_proto;
> +extern struct proto unix_stream_proto;
>  
> -int unix_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
> +int unix_dgram_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
> +int unix_stream_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
>  void __init unix_bpf_build_proto(void);
>  #else
>  static inline void __init unix_bpf_build_proto(void)
> diff --git a/net/core/sock_map.c b/net/core/sock_map.c
> index ae5fa4338..42f50ea7a 100644
> --- a/net/core/sock_map.c
> +++ b/net/core/sock_map.c
> @@ -517,9 +517,15 @@ static bool sk_is_tcp(const struct sock *sk)
>  	       sk->sk_protocol == IPPROTO_TCP;
>  }
>  
> +static bool sk_is_unix_stream(const struct sock *sk)
> +{
> +	return sk->sk_type == SOCK_STREAM &&
> +	       sk->sk_protocol == PF_UNIX;
> +}
> +
>  static bool sock_map_redirect_allowed(const struct sock *sk)
>  {
> -	if (sk_is_tcp(sk))
> +	if (sk_is_tcp(sk) || sk_is_unix_stream(sk))
>  		return sk->sk_state != TCP_LISTEN;
>  	else
>  		return sk->sk_state == TCP_ESTABLISHED;
> diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
> index 32eeb4a6a..c68d13f61 100644
> --- a/net/unix/af_unix.c
> +++ b/net/unix/af_unix.c
> @@ -791,17 +791,35 @@ static void unix_close(struct sock *sk, long timeout)
>  	 */
>  }
>  
> -struct proto unix_proto = {
> -	.name			= "UNIX",
> +static void unix_unhash(struct sock *sk)
> +{
> +	/* Nothing to do here, unix socket does not need a ->unhash().
> +	 * This is merely for sockmap.
> +	 */
> +}

Do we really need an unhash hook for unix_stream? I'm doing some testing
now to pull it out of TCP side as well. It seems to be an artifact of old
code that is no longer necessary. On TCP side at least just using close()
looks to be enough now.

> +
> +struct proto unix_dgram_proto = {
> +	.name			= "UNIX-DGRAM",
> +	.owner			= THIS_MODULE,
> +	.obj_size		= sizeof(struct unix_sock),
> +	.close			= unix_close,
> +#ifdef CONFIG_BPF_SYSCALL
> +	.psock_update_sk_prot	= unix_dgram_bpf_update_proto,
> +#endif
> +};
> +
> +struct proto unix_stream_proto = {
> +	.name			= "UNIX-STREAM",
>  	.owner			= THIS_MODULE,
>  	.obj_size		= sizeof(struct unix_sock),
>  	.close			= unix_close,
> +	.unhash			= unix_unhash,
>  #ifdef CONFIG_BPF_SYSCALL
> -	.psock_update_sk_prot	= unix_bpf_update_proto,
> +	.psock_update_sk_prot	= unix_stream_bpf_update_proto,
>  #endif
>  };
>  
> -static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
> +static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, int type)
>  {
>  	struct sock *sk = NULL;
>  	struct unix_sock *u;
> @@ -810,7 +828,17 @@ static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
>  	if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
>  		goto out;
>  
> -	sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
> +	if (type != 0) {
> +		if (type == SOCK_STREAM)
> +			sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_stream_proto, kern);
> +		else /*for seqpacket */
> +			sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_dgram_proto, kern);
> +	} else {
> +		if (sock->type == SOCK_STREAM)
> +			sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_stream_proto, kern);
> +		else
> +			sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_dgram_proto, kern);
> +	}
>  	if (!sk)
>  		goto out;
>  
> @@ -872,7 +900,7 @@ static int unix_create(struct net *net, struct socket *sock, int protocol,
>  		return -ESOCKTNOSUPPORT;
>  	}
>  
> -	return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
> +	return unix_create1(net, sock, kern, 0) ? 0 : -ENOMEM;
>  }
>  
>  static int unix_release(struct socket *sock)
> @@ -1286,7 +1314,7 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
>  	err = -ENOMEM;
>  
>  	/* create new sock for complete connection */
> -	newsk = unix_create1(sock_net(sk), NULL, 0);
> +	newsk = unix_create1(sock_net(sk), NULL, 0, sock->type);
>  	if (newsk == NULL)
>  		goto out;
>  
> @@ -2214,7 +2242,7 @@ static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg, size_t si
>  	struct sock *sk = sock->sk;
>  
>  #ifdef CONFIG_BPF_SYSCALL
> -	if (sk->sk_prot != &unix_proto)
> +	if (sk->sk_prot != &unix_dgram_proto)
>  		return sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
>  					    flags & ~MSG_DONTWAIT, NULL);
>  #endif
> @@ -2533,6 +2561,21 @@ static int unix_stream_read_actor(struct sk_buff *skb,
>  	return ret ?: chunk;
>  }
>  
> +int __unix_stream_recvmsg(struct sock *sk, struct msghdr *msg,
> +			  size_t size, int flags)
> +{
> +	struct socket *sock = sk->sk_socket;
> +	struct unix_stream_read_state state = {
> +		.recv_actor = unix_stream_read_actor,
> +		.socket = sock,
> +		.msg = msg,
> +		.size = size,
> +		.flags = flags
> +	};
> +
> +	return unix_stream_read_generic(&state, true);
> +}
> +
>  static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
>  			       size_t size, int flags)
>  {
> @@ -2544,6 +2587,13 @@ static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
>  		.flags = flags
>  	};
>  
> +	struct sock *sk = sock->sk;
> +
> +#ifdef CONFIG_BPF_SYSCALL
> +	if (sk->sk_prot != &unix_stream_proto)
> +		return sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
> +					    flags & ~MSG_DONTWAIT, NULL);
> +#endif
>  	return unix_stream_read_generic(&state, true);
>  }
>  
> @@ -2993,7 +3043,13 @@ static int __init af_unix_init(void)
>  
>  	BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
>  
> -	rc = proto_register(&unix_proto, 1);
> +	rc = proto_register(&unix_dgram_proto, 1);

Can you add a note in the commit message on why they proto_register is
needed. I think it might be helpful later.

> +	if (rc != 0) {
> +		pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
> +		goto out;
> +	}
> +
> +	rc = proto_register(&unix_stream_proto, 1);
>  	if (rc != 0) {
>  		pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
>  		goto out;
> @@ -3009,7 +3065,8 @@ static int __init af_unix_init(void)
>  static void __exit af_unix_exit(void)
>  {
>  	sock_unregister(PF_UNIX);
> -	proto_unregister(&unix_proto);
> +	proto_unregister(&unix_dgram_proto);
> +	proto_unregister(&unix_stream_proto);
>  	unregister_pernet_subsys(&unix_net_ops);
>  }
>  
> diff --git a/net/unix/unix_bpf.c b/net/unix/unix_bpf.c
> index db0cda29f..9067210d3 100644
> --- a/net/unix/unix_bpf.c
> +++ b/net/unix/unix_bpf.c
> @@ -38,9 +38,18 @@ static int unix_msg_wait_data(struct sock *sk, struct sk_psock *psock,
>  	return ret;
>  }
>  
> -static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
> -				  size_t len, int nonblock, int flags,
> -				  int *addr_len)
> +static int __unix_recvmsg(struct sock *sk, struct msghdr *msg,
> +			   size_t len, int flags)
> +{
> +	if (sk->sk_type == SOCK_DGRAM)
> +		return __unix_dgram_recvmsg(sk, msg, len, flags);
> +	else
> +		return __unix_stream_recvmsg(sk, msg, len, flags);
> +}
> +
> +static int unix_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
> +			    size_t len, int nonblock, int flags,
> +			    int *addr_len)
>  {
>  	struct unix_sock *u = unix_sk(sk);
>  	struct sk_psock *psock;
> @@ -48,12 +57,12 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
>  
>  	psock = sk_psock_get(sk);
>  	if (unlikely(!psock))
> -		return __unix_dgram_recvmsg(sk, msg, len, flags);
> +		return __unix_recvmsg(sk, msg, len, flags);
>  
>  	mutex_lock(&u->iolock);
>  	if (!skb_queue_empty(&sk->sk_receive_queue) &&
>  	    sk_psock_queue_empty(psock)) {
> -		ret = __unix_dgram_recvmsg(sk, msg, len, flags);
> +		ret = __unix_recvmsg(sk, msg, len, flags);

Will need rebase after Cong's fix for iolock goes in.

>  		goto out;
>  	}
>  
> @@ -68,7 +77,7 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
>  		if (data) {
>  			if (!sk_psock_queue_empty(psock))
>  				goto msg_bytes_ready;
> -			ret = __unix_dgram_recvmsg(sk, msg, len, flags);
> +			ret = __unix_recvmsg(sk, msg, len, flags);
>  			goto out;
>  		}
>  		copied = -EAGAIN;
> @@ -80,30 +89,55 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
>  	return ret;
>  }
>  
> -static struct proto *unix_prot_saved __read_mostly;
> -static DEFINE_SPINLOCK(unix_prot_lock);
> -static struct proto unix_bpf_prot;
> +static struct proto *unix_dgram_prot_saved __read_mostly;
> +static DEFINE_SPINLOCK(unix_dgram_prot_lock);
> +static struct proto unix_dgram_bpf_prot;
> +
> +static struct proto *unix_stream_prot_saved __read_mostly;
> +static DEFINE_SPINLOCK(unix_stream_prot_lock);
> +static struct proto unix_stream_bpf_prot;
> +
> +static void unix_dgram_bpf_rebuild_protos(struct proto *prot, const struct proto *base)
> +{
> +	*prot        = *base;
> +	prot->close  = sock_map_close;
> +	prot->recvmsg = unix_bpf_recvmsg;
> +}
>  
> -static void unix_bpf_rebuild_protos(struct proto *prot, const struct proto *base)
> +static void unix_stream_bpf_rebuild_protos(struct proto *prot,
> +					   const struct proto *base)
>  {
>  	*prot        = *base;
>  	prot->close  = sock_map_close;
> -	prot->recvmsg = unix_dgram_bpf_recvmsg;
> +	prot->recvmsg = unix_bpf_recvmsg;
> +	prot->unhash  = sock_map_unhash;

Still unsure whats different between stream and dgram that means we now
need the unhash hook.

>  }
>  
> -static void unix_bpf_check_needs_rebuild(struct proto *ops)
> +static void unix_dgram_bpf_check_needs_rebuild(struct proto *ops)
>  {
> -	if (unlikely(ops != smp_load_acquire(&unix_prot_saved))) {
> -		spin_lock_bh(&unix_prot_lock);
> -		if (likely(ops != unix_prot_saved)) {
> -			unix_bpf_rebuild_protos(&unix_bpf_prot, ops);
> -			smp_store_release(&unix_prot_saved, ops);
> +	if (unlikely(ops != smp_load_acquire(&unix_dgram_prot_saved))) {
> +		spin_lock_bh(&unix_dgram_prot_lock);
> +		if (likely(ops != unix_dgram_prot_saved)) {
> +			unix_dgram_bpf_rebuild_protos(&unix_dgram_bpf_prot, ops);
> +			smp_store_release(&unix_dgram_prot_saved, ops);
>  		}
> -		spin_unlock_bh(&unix_prot_lock);
> +		spin_unlock_bh(&unix_dgram_prot_lock);
>  	}
>  }
>  
> -int unix_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
> +static void unix_stream_bpf_check_needs_rebuild(struct proto *ops)
> +{
> +	if (unlikely(ops != smp_load_acquire(&unix_stream_prot_saved))) {
> +		spin_lock_bh(&unix_stream_prot_lock);
> +		if (likely(ops != unix_stream_prot_saved)) {
> +			unix_stream_bpf_rebuild_protos(&unix_stream_bpf_prot, ops);
> +			smp_store_release(&unix_stream_prot_saved, ops);
> +		}
> +		spin_unlock_bh(&unix_stream_prot_lock);
> +	}
> +}
> +
> +int unix_dgram_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
>  {
>  	if (restore) {
>  		sk->sk_write_space = psock->saved_write_space;
> @@ -111,12 +145,27 @@ int unix_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
>  		return 0;
>  	}
>  
> -	unix_bpf_check_needs_rebuild(psock->sk_proto);
> -	WRITE_ONCE(sk->sk_prot, &unix_bpf_prot);
> +	unix_dgram_bpf_check_needs_rebuild(psock->sk_proto);
> +	WRITE_ONCE(sk->sk_prot, &unix_dgram_bpf_prot);
> +	return 0;
> +}
> +
> +int unix_stream_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
> +{
> +	if (restore) {
> +		sk->sk_write_space = psock->saved_write_space;
> +		WRITE_ONCE(sk->sk_prot, psock->sk_proto);
> +		return 0;
> +	}
> +
> +	unix_stream_bpf_check_needs_rebuild(psock->sk_proto);
> +	WRITE_ONCE(sk->sk_prot, &unix_stream_bpf_prot);
>  	return 0;
>  }
>  
>  void __init unix_bpf_build_proto(void)
>  {
> -	unix_bpf_rebuild_protos(&unix_bpf_prot, &unix_proto);
> +	unix_dgram_bpf_rebuild_protos(&unix_dgram_bpf_prot, &unix_dgram_proto);
> +	unix_stream_bpf_rebuild_protos(&unix_stream_bpf_prot, &unix_stream_proto);
> +
>  }
> -- 
> 2.20.1
>

On Tue, Jul 27, 2021 at 9:37 AM John Fastabend <john.fastabend@gmail.com> wrote:
> Do we really need an unhash hook for unix_stream? I'm doing some testing

> now to pull it out of TCP side as well. It seems to be an artifact of old

> code that is no longer necessary. On TCP side at least just using close()

> looks to be enough now.


How do you handle the disconnection from remote without ->unhash()?

For all stream sockets, we still only allow established sockets to stay
in sockmap, which means we have to remove it if it is disconnected
or closed.

But it seems Jiang forgot to call ->unhash() when disconnecting.

Thanks.

Cong Wang wrote:
> On Tue, Jul 27, 2021 at 9:37 AM John Fastabend <john.fastabend@gmail.com> wrote:

> > Do we really need an unhash hook for unix_stream? I'm doing some testing

> > now to pull it out of TCP side as well. It seems to be an artifact of old

> > code that is no longer necessary. On TCP side at least just using close()

> > looks to be enough now.

> 

> How do you handle the disconnection from remote without ->unhash()?


Would close() not work for stream/dgram sockets?

> 

> For all stream sockets, we still only allow established sockets to stay

> in sockmap, which means we have to remove it if it is disconnected

> or closed.


+1.

> 

> But it seems Jiang forgot to call ->unhash() when disconnecting.


Aha so we need to add it in af_unix code I guess. Anyways looking forward
to v2.

Thanks.

On Wed, Jul 28, 2021 at 11:44 AM John Fastabend
<john.fastabend@gmail.com> wrote:
>

> Cong Wang wrote:

> > On Tue, Jul 27, 2021 at 9:37 AM John Fastabend <john.fastabend@gmail.com> wrote:

> > > Do we really need an unhash hook for unix_stream? I'm doing some testing

> > > now to pull it out of TCP side as well. It seems to be an artifact of old

> > > code that is no longer necessary. On TCP side at least just using close()

> > > looks to be enough now.

> >

> > How do you handle the disconnection from remote without ->unhash()?

>

> Would close() not work for stream/dgram sockets?


close() is called when the local closes the sockets, but when the remote
closes or disconnects it, unhash() is called. This is why TCP calls unhash()
to remove the socket from established socket hash table. unhash() itself
might not make much sense for AF_UNIX as it probably does not need a
hash table to track established ones, however, the idea is the same, that
is, we have to handle remote disconnections here.

Thanks.

Cong Wang wrote:
> On Wed, Jul 28, 2021 at 11:44 AM John Fastabend

> <john.fastabend@gmail.com> wrote:

> >

> > Cong Wang wrote:

> > > On Tue, Jul 27, 2021 at 9:37 AM John Fastabend <john.fastabend@gmail.com> wrote:

> > > > Do we really need an unhash hook for unix_stream? I'm doing some testing

> > > > now to pull it out of TCP side as well. It seems to be an artifact of old

> > > > code that is no longer necessary. On TCP side at least just using close()

> > > > looks to be enough now.

> > >

> > > How do you handle the disconnection from remote without ->unhash()?

> >

> > Would close() not work for stream/dgram sockets?

> 

> close() is called when the local closes the sockets, but when the remote

> closes or disconnects it, unhash() is called. This is why TCP calls unhash()

> to remove the socket from established socket hash table. unhash() itself

> might not make much sense for AF_UNIX as it probably does not need a

> hash table to track established ones, however, the idea is the same, that

> is, we have to handle remote disconnections here.


Following up on this series. Leaving a socket in the sockmap until close()
happens is not paticularly problematic, but does consume space in the map
so unhash() is slightly better I guess. Thanks.

> 

> Thanks.