On Mon, Feb 25, 2019 at 09:41:30AM -0800, Lee Duncan wrote:
> From: Lee Duncan <lduncan@xxxxxxxx>
>
> If there is an error queueing an iscsi command in
> iscsi_queuecommand(), for example if the transport fails
> to take the command in sessuin->tt->xmit_task(), then
> the error path can call iscsi_complete_task() without
> first aquiring the back_lock as required. This can
> lead to things like ITT pool can get corrupt, resulting
> in duplicate ITTs being sent out.
>
> The solution is to hold the back_lock around
> iscsi_complete_task() calls, and to add a little commenting
> to help others understand when back_lock must be held.
>
> Signed-off-by: Lee Duncan <lduncan@xxxxxxxx>
> ---
Lee,
Quick question, can you confirm that you tested this with lockdep?
It seems right to me, it's just that we've been hit with lockdep
problems dealing with these locks before.
- Chris
> drivers/scsi/libiscsi.c | 22 ++++++++++++++++++++--
> 1 file changed, 20 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c
> index b8d325ce8754..ef7871e8c6bd 100644
> --- a/drivers/scsi/libiscsi.c
> +++ b/drivers/scsi/libiscsi.c
> @@ -838,7 +838,7 @@ EXPORT_SYMBOL_GPL(iscsi_conn_send_pdu);
> * @datalen: len of buffer
> *
> * iscsi_cmd_rsp sets up the scsi_cmnd fields based on the PDU and
> - * then completes the command and task.
> + * then completes the command and task. called under back_lock
> **/
> static void iscsi_scsi_cmd_rsp(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
> struct iscsi_task *task, char *data,
> @@ -941,6 +941,9 @@ static void iscsi_scsi_cmd_rsp(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
> * @conn: iscsi connection
> * @hdr: iscsi pdu
> * @task: scsi command task
> + *
> + * iscsi_data_in_rsp sets up the scsi_cmnd fields based on the data received
> + * then completes the command and task. called under back_lock
> **/
> static void
> iscsi_data_in_rsp(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
> @@ -1025,6 +1028,16 @@ static int iscsi_send_nopout(struct iscsi_conn *conn, struct iscsi_nopin *rhdr)
> return 0;
> }
>
> +/**
> + * iscsi_nop_out_rsp - SCSI NOP Response processing
> + * @task: scsi command task
> + * @nop: the nop structure
> + * @data: where to put the data
> + * @datalen: length of data
> + *
> + * iscsi_nop_out_rsp handles nop response from use or
> + * from user space. called under back_lock
> + **/
> static int iscsi_nop_out_rsp(struct iscsi_task *task,
> struct iscsi_nopin *nop, char *data, int datalen)
> {
> @@ -1791,7 +1804,9 @@ int iscsi_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc)
> return 0;
>
> prepd_reject:
> + spin_lock_bh(&session->back_lock);
> iscsi_complete_task(task, ISCSI_TASK_REQUEUE_SCSIQ);
> + spin_unlock_bh(&session->back_lock);
> reject:
> spin_unlock_bh(&session->frwd_lock);
> ISCSI_DBG_SESSION(session, "cmd 0x%x rejected (%d)\n",
> @@ -1799,7 +1814,9 @@ int iscsi_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc)
> return SCSI_MLQUEUE_TARGET_BUSY;
>
> prepd_fault:
> + spin_lock_bh(&session->back_lock);
> iscsi_complete_task(task, ISCSI_TASK_REQUEUE_SCSIQ);
> + spin_unlock_bh(&session->back_lock);
> fault:
> spin_unlock_bh(&session->frwd_lock);
> ISCSI_DBG_SESSION(session, "iscsi: cmd 0x%x is not queued (%d)\n",
> @@ -3121,8 +3138,9 @@ fail_mgmt_tasks(struct iscsi_session *session, struct iscsi_conn *conn)
> state = ISCSI_TASK_ABRT_SESS_RECOV;
> if (task->state == ISCSI_TASK_PENDING)
> state = ISCSI_TASK_COMPLETED;
> + spin_lock_bh(&session->back_lock);
> iscsi_complete_task(task, state);
> -
> + spin_unlock_bh(&session->back_lock);
> }
> }
>
> --
> 2.16.4
>
