From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from out-179.mta1.migadu.com (out-179.mta1.migadu.com [95.215.58.179]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 7E40F30B53A for ; Thu, 5 Mar 2026 11:04:15 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=95.215.58.179 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1772708658; cv=none; b=un023vA5f7MwmM7BPkEKuCU1MBGHD7mCQaefeBXDkitEN2DJxuV1Q5tiLnbLWkXtu2YzQYEA3mciOV8M/lGjhWI/ldgJYkKHZ1BfoGDR9txFResLJf7y8hKy9TyC28zJcPSYprcBS1oEn+U5g2XPt0hg0mjN1ynwn4ltntExob8= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1772708658; c=relaxed/simple; bh=cfgzteg9YNjoqMF83yogwjIurVatxMzEUqaq3jY34ms=; h=MIME-Version:Date:Content-Type:From:Message-ID:Subject:To:Cc: In-Reply-To:References; b=JyfRKQZVO9uhd6R7k9rNwWeiHxCjql3/5jXFzjuW/RYQadvn+h05ciYLD+MN3nBRX3I2ZQ1plh37rlwxJ6cIe1SkQ6/At0Ua+kCFxYuRto9yX5KSzj3X8HV+L/PlNrUBVw+bXd3GAzxZw0XID6odmxgubvY13ZvOvzkF7mKiasE= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.dev; spf=pass smtp.mailfrom=linux.dev; dkim=pass (1024-bit key) header.d=linux.dev header.i=@linux.dev header.b=pH9RNeMy; arc=none smtp.client-ip=95.215.58.179 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.dev Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=linux.dev Authentication-Results: smtp.subspace.kernel.org; dkim=pass (1024-bit key) header.d=linux.dev header.i=@linux.dev header.b="pH9RNeMy" Precedence: bulk X-Mailing-List: netdev@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1772708653; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=GV9z8/gbc0Lbehx4Jxn10pNeFtNhIHnkxBumfZAougU=; b=pH9RNeMydC1tHCknzI9JuETaSXrVWp5m4gCHLQmYCdlR3ZTvBmGrzJoBbqE8c3/joucn1u PB7ltj3lQt5x7bgzu7YO3CrmAdK99PWWAiAyUBPRrNz826+0TSHbe44+8zuWF27qrpX9sA taHe9lg0HUjS1wn7127hI4ntiAv8qoE= Date: Thu, 05 Mar 2026 11:04:10 +0000 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. From: "Jiayuan Chen" Message-ID: <2006d401a04274b735dc15b86e58b497fd524540@linux.dev> TLS-Required: No Subject: Re: [PATCH v4 bpf/net 6/6] sockmap: Fix broken memory accounting for UDP. To: "Kuniyuki Iwashima" Cc: "John Fastabend" , "Jakub Sitnicki" , "Willem de Bruijn" , "Kuniyuki Iwashima" , bpf@vger.kernel.org, netdev@vger.kernel.org, syzbot+5b3b7e51dda1be027b7a@syzkaller.appspotmail.com In-Reply-To: <3e278eea8830a7b68054ae0b5a99e1bc7ae5fbf3@linux.dev> References: <20260221233234.3814768-1-kuniyu@google.com> <20260221233234.3814768-7-kuniyu@google.com> <389162b2e117d8a5675b02286c7bafcc342086b6@linux.dev> <3e278eea8830a7b68054ae0b5a99e1bc7ae5fbf3@linux.dev> X-Migadu-Flow: FLOW_OUT March 5, 2026 at 18:45, "Jiayuan Chen" wrote: >=20 >=20March 5, 2026 at 17:27, "Kuniyuki Iwashima" wrote: >=20 >=20>=20 >=20> On Thu, Mar 5, 2026 at 12:30 AM Jiayuan Chen wrote: > >=20=20 >=20>=20=20 >=20> March 5, 2026 at 15:48, "Kuniyuki Iwashima" wrote: > >=20=20 >=20> On Wed, Mar 4, 2026 at 10:37 PM Jiayuan Chen wrote: > >=20=20 >=20> > > > > On Sat, Feb 21, 2026 at 11:30:53PM +0800, Kuniyuki Iwashima wrote: > > > [...] > > > > > > > > > Fixes: d7f571188ecf ("udp: Implement ->read_sock() for sockmap") > > > Reported-by: syzbot+5b3b7e51dda1be027b7a@syzkaller.appspotmail.com > > > Closes: https://lore.kernel.org/netdev/698f84c6.a70a0220.2c38d7.00= cb.GAE@google.com/ > > > Signed-off-by: Kuniyuki Iwashima > > > --- > > > v4: Fix deadlock when requeued > > > v2: Fix build failure when CONFIG_INET=3Dn > > > --- > > > include/net/udp.h | 9 +++++++++ > > > net/core/skmsg.c | 29 +++++++++++++++++++++++++---- > > > net/ipv4/udp.c | 9 +++++++++ > > > 3 files changed, 43 insertions(+), 4 deletions(-) > > > > > > diff --git a/include/net/udp.h b/include/net/udp.h > > > index 700dbedcb15f..ae38a4da9388 100644 > > > --- a/include/net/udp.h > > > +++ b/include/net/udp.h > > > @@ -455,6 +455,15 @@ struct sock *__udp6_lib_lookup(const struct n= et *net, > > > struct sk_buff *skb); > > > struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, > > > __be16 sport, __be16 dport); > > > + > > > +#ifdef CONFIG_INET > > > +void udp_sock_rfree(struct sk_buff *skb); > > > +#else > > > +static inline void udp_sock_rfree(struct sk_buff *skb) > > > +{ > > > +} > > > +#endif > > > + > > > int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); > > > > > > /* UDP uses skb->dev_scratch to cache as much information as possi= ble and avoid > > > diff --git a/net/core/skmsg.c b/net/core/skmsg.c > > > index 96f43e0dbb17..dd9134a45663 100644 > > > --- a/net/core/skmsg.c > > > +++ b/net/core/skmsg.c > > > @@ -7,6 +7,7 @@ > > > > > > #include > > > #include > > > +#include > > > #include > > > #include > > > > > > @@ -576,6 +577,7 @@ static int sk_psock_skb_ingress(struct sk_psoc= k *psock, struct sk_buff *skb, > > > u32 off, u32 len, bool take_ref) > > > { > > > struct sock *sk =3D psock->sk; > > > + bool is_udp =3D sk_is_udp(sk); > > > struct sk_msg *msg; > > > int err =3D -EAGAIN; > > > > > > @@ -583,13 +585,20 @@ static int sk_psock_skb_ingress(struct sk_ps= ock *psock, struct sk_buff *skb, > > > if (!msg) > > > goto out; > > > > > > - if (skb->sk !=3D sk && take_ref) { > > > + if (is_udp) { > > > + if (unlikely(skb->destructor =3D=3D udp_sock_rfree)) > > > + goto enqueue; > > > + > > > + spin_lock_bh(&sk->sk_receive_queue.lock); > > > + } > > > + > > > + if (is_udp || (skb->sk !=3D sk && take_ref)) { > > > if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) > > > - goto free; > > > + goto unlock; > > > > > > if (!sk_rmem_schedule(sk, skb, skb->truesize)) > > > - goto free; > > > - } else { > > > + goto unlock; > > > + } else if (skb->sk =3D=3D sk || !take_ref) { > > > /* This is used in tcp_bpf_recvmsg_parser() to determine whether t= he > > > * data originates from the socket's own protocol stack. No need to > > > * refcount sk because msg's lifetime is bound to sk via the ingres= s_msg. > > > @@ -604,11 +613,23 @@ static int sk_psock_skb_ingress(struct sk_ps= ock *psock, struct sk_buff *skb, > > > * into user buffers. > > > */ > > > skb_set_owner_r(skb, sk); > > > + > > > + if (is_udp) { > > > + spin_unlock_bh(&sk->sk_receive_queue.lock); > > > + > > > + skb->destructor =3D udp_sock_rfree; > > > + } > > > + > > > +enqueue: > > > err =3D sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg= , take_ref); > > > if (err < 0) > > > goto free; > > > out: > > > return err; > > > + > > > +unlock: > > > + if (is_udp) > > > + spin_unlock_bh(&sk->sk_receive_queue.lock); > > > free: > > > kfree(msg); > > > goto out; > > > diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c > > > index 422c96fea249..831d26748a90 100644 > > > --- a/net/ipv4/udp.c > > > +++ b/net/ipv4/udp.c > > > @@ -2039,6 +2039,15 @@ struct sk_buff *__skb_recv_udp(struct sock = *sk, unsigned int flags, > > > } > > > EXPORT_SYMBOL(__skb_recv_udp); > > > > > > +void udp_sock_rfree(struct sk_buff *skb) > > > +{ > > > + struct sock *sk =3D skb->sk; > > > + > > > + spin_lock_bh(&sk->sk_receive_queue.lock); > > > + sock_rfree(skb); > > > + spin_unlock_bh(&sk->sk_receive_queue.lock); > > > +} > > > + > > > int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) > > > { > > > struct sk_buff *skb; > > > -- > > > 2.53.0.371.g1d285c8824-goog > > > > > > There are too many protocol-specific checks in the generic skmsg c= ode > > > here. This should be abstracted out. > > > > > > Also TCP also has a similar issue with sk_forward_alloc concurrenc= y > > > between the backlog charge and recvmsg uncharge paths so the abstr= action > > > is necessary. > > > > > > I've put together a simple diff based on your patch for reference. > > > I haven't tested it thoroughly, but it at least handles TCP and UD= P > > > separately through a callback, keeping the generic code clean. > > > > > I can follow up on TCP, but TCP needs another fix, which > > cannot be factored out. > >=20=20 >=20> Some wrong assumptions in your patch: > >=20=20 >=20> 1) tcp_psock_ingress_charge() uses psock->ingress_lock, > > but it does not protect any tcp_sock fields, especially > > sk->sk_forwad_alloc. > >=20=20 >=20> 2) TCP memory accounting happens under bh_lock_sock() > > _or_ lock_sock(). sendmsg()/recvmsg() works under > > lock_sock() and TCP fast path puts skb to backlog if > > lock_sock() is held by userspace. This means even > > a simple bh_lock_sock() in tcp_psock_ingress_charge() > > race with memory accounting happening under lock_sock(). > >=20=20 >=20> Since sk_psock_skb_ingress() is called from both GFP_KERNEL > > and GFP_ATOMIC context, the simplest fix would be to put > > lock_sock() in sk_psock_backlog() for TCP, which is ugly though. > >=20=20 >=20> OK, thanks. > >=20=20 >=20> Now my only concern is keeping UDP-specific logic > > out of skmsg.c. > >=20=20 >=20> We already have TCP-only msg->sk logic there that > > you added. It just does not look like TCP code. > >=20 >=20The msg->sk field marks ingress-self vs ingress-other, which is > protocol-agnostic metadata. TCP happens to consume it for seq > tracking, but the assignment itself has no TCP-specific check > (no is_tcp_sk() or similar). >=20 >=20>=20 >=20> If really needed, we can do such cleanup in bpf-next > > for TCP and UDP altogether (probably after another > > TCP fix ?). > >=20=20 >=20>=20=20 >=20> We could use a function pointer so that UDP > > implements its own ingress charge in udp_bpf.c, while other > > protocols just use a default no-op or even NULL. > >=20=20 >=20> If TCP requires locking outside of the scope, there's > > no point adding one more extra layer/complexity just > > for UDP. > >=20=20 >=20> Even if we go that way, most likely we cannot get rid > > of protocol-specific code (UDP/TCP-only hook) and > > end up with the indirect call helpers, which will be > > expensive than simple if logic. > >=20 >=20An alternative would be to add such sk_psock_udp_skb_ingress() in > udp_bpf.c, without adding any indirection layer. >=20 >=20This way we don't need to design a new abstraction before the > approach is settled. Just a UDP-specific ingress helper, which > keeps things clean enough for now, and avoid scattering is_udp > checks all over sk_psock_skb_ingress, just need one if (is_udp). >=20 >=20Thanks. > After thinking a while, since the UDP-specific behaviour only appears in the redirect-to-self path, how about implementing sk_psock_udp_skb_ingress_self() in udp_bpf.c and calling it from sk_psock_skb_ingress_self()? That gives us a single, contained check point rather than scattering is_udp all over skmsg.c, which seems more aligned with the "simple" you described before.