The lpfc_sli4_nvmet_xri_aborted routine takes out the abts_buf_list_lock
and traverses the buffer contexts to match the xri. Upon match, it then
takes the context lock before potentially removing the context from the
associated buffer list. This violates the lock hierarchy used elsewhere
in the driver of locking context, then the abts_buf_list_lock - thus a
possible deadlock.
Resolve by: after matching, release the abts_buf_list_lock, then take
the context lock, and if to be deleted from the list, retake the
abts_buf_list_lock, maintaining lock hierarchy. This matches same list
lock hierarchy as elsewhere in the driver
Reported-by: Jia-Ju Bai <baijiaju1990@xxxxxxxxx>
Signed-off-by: James Smart <jsmart2021@xxxxxxxxx>
---
drivers/scsi/lpfc/lpfc_nvmet.c | 11 +++++++----
1 file changed, 7 insertions(+), 4 deletions(-)
diff --git a/drivers/scsi/lpfc/lpfc_nvmet.c b/drivers/scsi/lpfc/lpfc_nvmet.c
index f2d9a3580887..6e3dd0b9bcfa 100644
--- a/drivers/scsi/lpfc/lpfc_nvmet.c
+++ b/drivers/scsi/lpfc/lpfc_nvmet.c
@@ -1797,19 +1797,22 @@ lpfc_sli4_nvmet_xri_aborted(struct lpfc_hba *phba,
if (ctxp->ctxbuf->sglq->sli4_xritag != xri)
continue;
- spin_lock(&ctxp->ctxlock);
+ spin_unlock_irqrestore(&phba->sli4_hba.abts_nvmet_buf_list_lock,
+ iflag);
+
+ spin_lock_irqsave(&ctxp->ctxlock, iflag);
/* Check if we already received a free context call
* and we have completed processing an abort situation.
*/
if (ctxp->flag & LPFC_NVME_CTX_RLS &&
!(ctxp->flag & LPFC_NVME_ABORT_OP)) {
+ spin_lock(&phba->sli4_hba.abts_nvmet_buf_list_lock);
list_del_init(&ctxp->list);
+ spin_unlock(&phba->sli4_hba.abts_nvmet_buf_list_lock);
released = true;
}
ctxp->flag &= ~LPFC_NVME_XBUSY;
- spin_unlock(&ctxp->ctxlock);
- spin_unlock_irqrestore(&phba->sli4_hba.abts_nvmet_buf_list_lock,
- iflag);
+ spin_unlock_irqrestore(&ctxp->ctxlock, iflag);
rrq_empty = list_empty(&phba->active_rrq_list);
ndlp = lpfc_findnode_did(phba->pport, ctxp->sid);
--
2.26.2
