| Message ID | 20220920082503.224189-1-bmt@zurich.ibm.com (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
| Series | RDMA/siw: Fix QP destroy to wait for all references dropped. | expand |
On Tue, 20 Sep 2022 10:25:03 +0200, Bernard Metzler wrote: > Delay QP destroy completion until all siw references to QP are > dropped. The calling RDMA core will free QP structure after > successful return from siw_qp_destroy() call, so siw must not > hold any remaining reference to the QP upon return. > A use-after-free was encountered in xfstest generic/460, while > testing NFSoRDMA. Here, after a TCP connection drop by peer, > the triggered siw_cm_work_handler got delayed until after > QP destroy call, referencing a QP which has already freed. > > [...] Applied, thanks! [1/1] RDMA/siw: Fix QP destroy to wait for all references dropped. https://git.kernel.org/rdma/rdma/c/a3c278807a459e Best regards,
diff --git a/drivers/infiniband/sw/siw/siw.h b/drivers/infiniband/sw/siw/siw.h index df03d84c6868..2f3a9cda3850 100644 --- a/drivers/infiniband/sw/siw/siw.h +++ b/drivers/infiniband/sw/siw/siw.h @@ -418,6 +418,7 @@ struct siw_qp { struct ib_qp base_qp; struct siw_device *sdev; struct kref ref; + struct completion qp_free; struct list_head devq; int tx_cpu; struct siw_qp_attrs attrs; diff --git a/drivers/infiniband/sw/siw/siw_qp.c b/drivers/infiniband/sw/siw/siw_qp.c index 7e01f2438afc..e6f634971228 100644 --- a/drivers/infiniband/sw/siw/siw_qp.c +++ b/drivers/infiniband/sw/siw/siw_qp.c @@ -1342,6 +1342,6 @@ void siw_free_qp(struct kref *ref) vfree(qp->orq); siw_put_tx_cpu(qp->tx_cpu); - + complete(&qp->qp_free); atomic_dec(&sdev->num_qp); } diff --git a/drivers/infiniband/sw/siw/siw_verbs.c b/drivers/infiniband/sw/siw/siw_verbs.c index 8dedae7ae79e..3e814cfb298c 100644 --- a/drivers/infiniband/sw/siw/siw_verbs.c +++ b/drivers/infiniband/sw/siw/siw_verbs.c @@ -480,6 +480,8 @@ int siw_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *attrs, list_add_tail(&qp->devq, &sdev->qp_list); spin_unlock_irqrestore(&sdev->lock, flags); + init_completion(&qp->qp_free); + return 0; err_out_xa: @@ -624,6 +626,7 @@ int siw_destroy_qp(struct ib_qp *base_qp, struct ib_udata *udata) qp->scq = qp->rcq = NULL; siw_qp_put(qp); + wait_for_completion(&qp->qp_free); return 0; }
Delay QP destroy completion until all siw references to QP are dropped. The calling RDMA core will free QP structure after successful return from siw_qp_destroy() call, so siw must not hold any remaining reference to the QP upon return. A use-after-free was encountered in xfstest generic/460, while testing NFSoRDMA. Here, after a TCP connection drop by peer, the triggered siw_cm_work_handler got delayed until after QP destroy call, referencing a QP which has already freed. Fixes: 303ae1cdfdf7 ("rdma/siw: application interface") Reported-by: Olga Kornievskaia <kolga@netapp.com> Signed-off-by: Bernard Metzler <bmt@zurich.ibm.com> --- drivers/infiniband/sw/siw/siw.h | 1 + drivers/infiniband/sw/siw/siw_qp.c | 2 +- drivers/infiniband/sw/siw/siw_verbs.c | 3 +++ 3 files changed, 5 insertions(+), 1 deletion(-)