From: Justin Tee justin.tee@broadcom.com
[ Upstream commit 4281f44ea8bfedd25938a0031bebba1473ece9ad ]
Current dev_loss_tmo handling checks whether there has been a previous call to unregister with SCSI transport. If so, the NDLP kref count is decremented a second time in dev_loss_tmo as the final kref release. However, this can sometimes result in a reference count underflow if there is also a race to unregister with NVMe transport as well. Add a check for NVMe transport registration before decrementing the final kref. If NVMe transport is still registered, then the NVMe transport unregistration is designated as the final kref decrement.
Signed-off-by: Justin Tee justin.tee@broadcom.com Link: https://lore.kernel.org/r/20241031223219.152342-8-justintee8345@gmail.com Signed-off-by: Martin K. Petersen martin.petersen@oracle.com Signed-off-by: Sasha Levin sashal@kernel.org --- drivers/scsi/lpfc/lpfc_hbadisc.c | 36 +++++++++++++++++++++----------- 1 file changed, 24 insertions(+), 12 deletions(-)
diff --git a/drivers/scsi/lpfc/lpfc_hbadisc.c b/drivers/scsi/lpfc/lpfc_hbadisc.c index 9241075f72fa4..6e8d8a96c54fb 100644 --- a/drivers/scsi/lpfc/lpfc_hbadisc.c +++ b/drivers/scsi/lpfc/lpfc_hbadisc.c @@ -155,6 +155,7 @@ lpfc_dev_loss_tmo_callbk(struct fc_rport *rport) struct lpfc_hba *phba; struct lpfc_work_evt *evtp; unsigned long iflags; + bool nvme_reg = false;
ndlp = ((struct lpfc_rport_data *)rport->dd_data)->pnode; if (!ndlp) @@ -177,38 +178,49 @@ lpfc_dev_loss_tmo_callbk(struct fc_rport *rport) /* Don't schedule a worker thread event if the vport is going down. */ if (test_bit(FC_UNLOADING, &vport->load_flag) || !test_bit(HBA_SETUP, &phba->hba_flag)) { + spin_lock_irqsave(&ndlp->lock, iflags); ndlp->rport = NULL;
+ if (ndlp->fc4_xpt_flags & NVME_XPT_REGD) + nvme_reg = true; + /* The scsi_transport is done with the rport so lpfc cannot - * call to unregister. Remove the scsi transport reference - * and clean up the SCSI transport node details. + * call to unregister. */ - if (ndlp->fc4_xpt_flags & (NLP_XPT_REGD | SCSI_XPT_REGD)) { + if (ndlp->fc4_xpt_flags & SCSI_XPT_REGD) { ndlp->fc4_xpt_flags &= ~SCSI_XPT_REGD;
- /* NVME transport-registered rports need the - * NLP_XPT_REGD flag to complete an unregister. + /* If NLP_XPT_REGD was cleared in lpfc_nlp_unreg_node, + * unregister calls were made to the scsi and nvme + * transports and refcnt was already decremented. Clear + * the NLP_XPT_REGD flag only if the NVME Rport is + * confirmed unregistered. */ - if (!(ndlp->fc4_xpt_flags & NVME_XPT_REGD)) + if (!nvme_reg && ndlp->fc4_xpt_flags & NLP_XPT_REGD) { ndlp->fc4_xpt_flags &= ~NLP_XPT_REGD; + spin_unlock_irqrestore(&ndlp->lock, iflags); + lpfc_nlp_put(ndlp); /* may free ndlp */ + } else { + spin_unlock_irqrestore(&ndlp->lock, iflags); + } + } else { spin_unlock_irqrestore(&ndlp->lock, iflags); - lpfc_nlp_put(ndlp); - spin_lock_irqsave(&ndlp->lock, iflags); }
+ spin_lock_irqsave(&ndlp->lock, iflags); + /* Only 1 thread can drop the initial node reference. If * another thread has set NLP_DROPPED, this thread is done. */ - if (!(ndlp->fc4_xpt_flags & NVME_XPT_REGD) && - !(ndlp->nlp_flag & NLP_DROPPED)) { - ndlp->nlp_flag |= NLP_DROPPED; + if (nvme_reg || (ndlp->nlp_flag & NLP_DROPPED)) { spin_unlock_irqrestore(&ndlp->lock, iflags); - lpfc_nlp_put(ndlp); return; }
+ ndlp->nlp_flag |= NLP_DROPPED; spin_unlock_irqrestore(&ndlp->lock, iflags); + lpfc_nlp_put(ndlp); return; }