]> git.baikalelectronics.ru Git - kernel.git/commitdiff
nvme-tcp: can't set sk_user_data without write_lock
authorMaurizio Lombardi <mlombard@redhat.com>
Fri, 2 Jul 2021 08:11:21 +0000 (10:11 +0200)
committerChristoph Hellwig <hch@lst.de>
Mon, 5 Jul 2021 08:14:44 +0000 (10:14 +0200)
The sk_user_data pointer is supposed to be modified only while
holding the write_lock "sk_callback_lock", otherwise
we could race with other threads and crash the kernel.

we can't take the write_lock in nvmet_tcp_state_change()
because it would cause a deadlock, but the release_work queue
will set the pointer to NULL later so we can simply remove
the assignment.

Fixes: b5332a9f3f3d ("nvmet-tcp: fix incorrect locking in state_change sk callback")
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
drivers/nvme/target/tcp.c

index d8aceef83284680fbfeae752154eeea37b6f2960..07ee347ea3f3cf6d7cd5142e740dfa07ac9e5293 100644 (file)
@@ -1497,7 +1497,6 @@ static void nvmet_tcp_state_change(struct sock *sk)
        case TCP_CLOSE_WAIT:
        case TCP_CLOSE:
                /* FALLTHRU */
-               sk->sk_user_data = NULL;
                nvmet_tcp_schedule_release_queue(queue);
                break;
        default: