[stgt] [PATCH] delayed closing iscsi conn. when tasks are still in scsi

[FUJITA Tomonori]

On Tue, 30 Nov 2010 17:06:04 +0200
Alexander Nezhinsky <alexandern at Voltaire.COM> wrote:

> When iscsi conn gets closed, some tasks may be still processed in scsi layer,
> and can't be released immediatelly, thus their release is delayed.
> But the conn release should be delayed too. To do this we can increment conn

Why do we need this?

iscsi_alloc_task calls conn_get() so as long as a task lives, its conn
never be freed?


> refcount per each task in scsi. Thus the last conn_put() in conn_close() will
> not lead to the conn destruction. Then we decrement the refcount when the
> tasks complete and ensure that the conn is freed when the last such task
> returns from scsi.
> 
> Signed-off-by: Alexander Nezhinsky <alexandern at voltaire.com>
> ---
>  usr/iscsi/conn.c   |   11 ++++++++++-
>  usr/iscsi/iscsid.c |    8 +++++++-
>  2 files changed, 17 insertions(+), 2 deletions(-)
> 
> diff --git a/usr/iscsi/conn.c b/usr/iscsi/conn.c
> index a3af6a4..fa4096c 100644
> --- a/usr/iscsi/conn.c
> +++ b/usr/iscsi/conn.c
> @@ -173,8 +173,17 @@ void conn_close(struct iscsi_connection *conn)
>  		 * This task is in SCSI. We need to wait for I/O
>  		 * completion.
>  		 */
> -		if (task_in_scsi(task))
> +		if (task_in_scsi(task)) {
> +			/* We increment connection's refcount to allow
> +			 * the task to complete in scsi layer.
> +			 * Then we'll decrement	it back and dispose of
> +			 * the connection when no such tasks remain.
> +			 */
> +			dprintf("task %" PRIx64 " in_scsi, release delayed\n",
> +				task->tag);
> +			conn_get(conn);
>  			continue;
> +		}
>  		iscsi_free_task(task);
>  	}
>  done:
> diff --git a/usr/iscsi/iscsid.c b/usr/iscsi/iscsid.c
> index 3a79d93..050047e 100644
> --- a/usr/iscsi/iscsid.c
> +++ b/usr/iscsi/iscsid.c
> @@ -1184,9 +1184,15 @@ static int iscsi_scsi_cmd_done(uint64_t nid, int result, struct scsi_cmd *scmd)
>  	 * We could delay the closing of the conn in some cases and send
>  	 * the response with a little extra code or we can check if this
>  	 * task got reassinged to another connection.
> +	 * When the connection was closing, its refcount got incremented
> +	 * for each in_scsi task whose release was delayed.
> +	 * Thus we need to decrement it back to enable proper
> +	 * connection and session release.
>  	 */
> -	if (task->conn->state == STATE_CLOSE) {
> +	if (task->conn->closed) {
> +		dprintf("task %" PRIx64 " from scsi, releasing\n", task->tag);
>  		iscsi_free_cmd_task(task);
> +		conn_put(task->conn);
>  		return 0;
>  	}
>  
> --
> 1.7.3.2
> --
> To unsubscribe from this list: send the line "unsubscribe stgt" in
> the body of a message to majordomo at vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe stgt" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html