Netdev List
 help / color / mirror / Atom feed
From: Dominik Maier <dmnk@google.com>
To: Eric Dumazet <edumazet@kernel.org>,
	Neal Cardwell <ncardwell@google.com>,
	 Kuniyuki Iwashima <kuniyu@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>,
	Joanne Koong <joannelkoong@gmail.com>,
	netdev@vger.kernel.org,  dmnk+artist@google.com,
	Dominik Maier <dmnk@google.com>
Subject: [PATCH net] inet: release newly acquired port on inet_csk_listen_start() hash error
Date: Thu, 19 Nov 2026 20:21:33 +0100	[thread overview]
Message-ID: <20261119192133.2632936-1-dmnk@google.com> (raw)

When inet_csk_listen_start() is called on a socket that has not been
explicitly bound, sk->sk_prot->get_port() allocates an ephemeral port
and links the socket into bhash and bhash2 without setting
SOCK_BINDPORT_LOCK. If sk->sk_prot->hash() then fails (for example,
when a prior failed connect() left icsk->unhashed_state set to
TCP_SYN_SENT so inet_hash() returns -EINVAL), inet_csk_listen_start()
resets sk_state to TCP_CLOSE and returns the error without releasing
the newly acquired port.

Because inet->inet_num remains non-zero while SOCK_BINDPORT_LOCK is
unset, a subsequent Fast Open connect on a dual-stack AF_INET6 socket
can update sk_v6_rcv_saddr via tcp_v4_connect() without rehashing the
bhash2 entry. Disconnecting with AF_UNSPEC then invokes
__inet_put_port(), which locks the new address's bhash2 bucket while
unlinking and freeing tb2 from the original bhash2 bucket without
holding its spinlock.

Record whether icsk->icsk_bind_hash was already populated before
get_port(). On error, after transitioning back to TCP_CLOSE and before
purging the Fast Open RST queue, clear inet->inet_sport and call
inet_put_port(sk) if get_port() newly bound the socket. Sockets that
were explicitly bound before listen() retain their user-bound port.

This issue was discovered with an AI-assisted code review tool (ARTIST)
and reproduced in QEMU/KVM on an unmodified v7.4-rc3 kernel without
kernel modifications or error injection; a standalone reproducer is
available upon maintainers' request. An earlier revision of this fix was
verified against the reproducer in QEMU/KVM; this version, rebased onto
commit 342e62fd34c2 ("tcp: purge Fast Open RST queue when listen() fails"),
was checked with checkpatch.pl --strict.

Fixes: 8cc3aef0cb19 ("tcp: Do not allow buggy transitions between ehash and lhash2.")
Fixes: 28044fc1d495 ("net: Add a bhash2 table hashed by port and address")
Cc: stable@vger.kernel.org
Signed-off-by: Dominik Maier <dmnk@google.com>
---
 net/ipv4/inet_connection_sock.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/net/ipv4/inet_connection_sock.c b/net/ipv4/inet_connection_sock.c
index 9a3b05e..2f367a1 100644
--- a/net/ipv4/inet_connection_sock.c
+++ b/net/ipv4/inet_connection_sock.c
@@ -1369,6 +1369,7 @@ int inet_csk_listen_start(struct sock *sk)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
 	struct inet_sock *inet = inet_sk(sk);
+	bool bound;
 	int err;
 
 	err = inet_ulp_can_listen(sk);
@@ -1385,6 +1386,7 @@ int inet_csk_listen_start(struct sock *sk)
 	 * It is OK, because this socket enters to hash table only
 	 * after validation is complete.
 	 */
+	bound = !!icsk->icsk_bind_hash;
 	inet_sk_state_store(sk, TCP_LISTEN);
 	err = sk->sk_prot->get_port(sk, inet->inet_num);
 	if (!err) {
@@ -1398,6 +1400,10 @@ int inet_csk_listen_start(struct sock *sk)
 	}
 
 	inet_sk_set_state(sk, TCP_CLOSE);
+	if (!bound && icsk->icsk_bind_hash) {
+		inet->inet_sport = 0;
+		inet_put_port(sk);
+	}
 
 	/* A Fast Open child accepted before a previous shutdown() of this
 	 * listener might have queued its request in rskq_rst_head while
-- 
2.56.1.809.ga0b61035b5-goog


                 reply	other threads:[~2026-11-19 19:21 UTC|newest]

Thread overview: [no followups] expand[flat|nested]  mbox.gz  Atom feed

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20261119192133.2632936-1-dmnk@google.com \
    --to=dmnk@google.com \
    --cc=dmnk+artist@google.com \
    --cc=edumazet@kernel.org \
    --cc=joannelkoong@gmail.com \
    --cc=kuba@kernel.org \
    --cc=kuniyu@google.com \
    --cc=ncardwell@google.com \
    --cc=netdev@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox