[PATCH] drbd: Fix potential deadlock on &session->back_lock and &session->frwd_lock

From: Chengfeng Ye
Date: Tue Sep 26 2023 - 07:00:34 EST


Timer interrupt could introduce inverse lock ordering between
&session->frwd_lock and &session->back_lock if qedi_tmf_resp_work()
is interrupted by timer.

<Deadlock #1>

[CPU0] | [CPU1]
qedi_tmf_resp_work() |
--> spin_lock(&session->back_lock) |
| iscsi_eh_cmd_timed_out()
| --> spin_lock_bh(&session->frwd_lock)
|
| --> spin_lock(&session->back_lock)
<timer interrupt> |
--> iscsi_tmf_timedout() |
--> spin_lock(&session->frwd_lock) |

<Deadlock #2>

[CPU0] | [CPU1]
qedi_tmf_resp_work() |
--> __iscsi_complete_pdu() |
--> spin_lock(&conn->session->frwd_lock) |
| qedi_process_async_mesg()
| --> spin_lock_bh(&session->back_lock)
|
| --> __iscsi_complete_pdu()
| --> spin_lock(&conn->session->frwd_lock)
<timer interrupt> |
--> iscsi_check_transport_timeouts() |
--> iscsi_send_nopout() |
--> iscsi_alloc_mgmt_task() |
--> iscsi_put_task() |
--> spin_lock_bh(&session->back_lock) |

This flaw was found by an experimental static analysis tool I am
developing for irq-related deadlock.

To prevent the potential deadlock, the patch use spin_lock_bh()
on &session->back_lock inside qedi_tmf_resp_work() to prevent the
timer interrupts.

Signed-off-by: Chengfeng Ye <dg573847474@xxxxxxxxx>
---
drivers/scsi/qedi/qedi_fw.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/scsi/qedi/qedi_fw.c b/drivers/scsi/qedi/qedi_fw.c
index 6901738324da..60a168ca178c 100644
--- a/drivers/scsi/qedi/qedi_fw.c
+++ b/drivers/scsi/qedi/qedi_fw.c
@@ -162,9 +162,9 @@ static void qedi_tmf_resp_work(struct work_struct *work)
if (rval)
goto exit_tmf_resp;

- spin_lock(&session->back_lock);
+ spin_lock_bh(&session->back_lock);
__iscsi_complete_pdu(conn, (struct iscsi_hdr *)resp_hdr_ptr, NULL, 0);
- spin_unlock(&session->back_lock);
+ spin_unlock_bh(&session->back_lock);

exit_tmf_resp:
kfree(resp_hdr_ptr);
--
2.17.1