]> git.ipfire.org Git - thirdparty/linux.git/commit
net/handshake: Take a long-lived file reference at submit
authorChuck Lever <chuck.lever@oracle.com>
Mon, 25 May 2026 16:51:18 +0000 (12:51 -0400)
committerPaolo Abeni <pabeni@redhat.com>
Thu, 28 May 2026 11:35:31 +0000 (13:35 +0200)
commit09dba37eee70d0596e26645015f1aa95a9848e9d
treeb54ea39d12dcbbbf2e25231af0df732842373ffc
parent6b22d433aa13f68e3cd9534ca9a5f4277bfa01c2
net/handshake: Take a long-lived file reference at submit

handshake_nl_accept_doit() needs the file pointer backing
req->hr_sk->sk_socket to survive the window between
handshake_req_next() and the subsequent FD_PREPARE() and get_file().
The submit-side sock_hold() does not provide that.  sk_refcnt keeps
struct sock alive, but struct socket is owned by sock->file: when
the consumer fputs the last file reference, sock_release() tears
the socket down regardless of any sock_hold.

Add an hr_file pointer to struct handshake_req and acquire an
explicit reference on sock->file during handshake_req_submit().
handshake_complete() and handshake_req_cancel() release the
reference on the completion-bit-winning path.

The submit error path must also release the file reference, but
after rhashtable insertion a concurrent handshake_req_cancel() can
discover the request and race the error path.  Gate the error-path
cleanup -- sk_destruct restoration, fput, and request destruction
-- with test_and_set_bit(HANDSHAKE_F_REQ_COMPLETED), the same
serialization handshake_complete() and handshake_req_cancel()
already use.  When cancel has already claimed ownership, the submit
error path returns without touching the request; socket teardown
handles final destruction.

The accept-side dereferences are not yet retargeted; that change
comes in the next patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://patch.msgid.link/20260525-handshake-file-pin-v3-4-66c616906ead@oracle.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
net/handshake/handshake.h
net/handshake/netlink.c
net/handshake/request.c