rxe: IB_WR_REG_MR does not capture MR's iova field
authorChuck Lever <chuck.lever@oracle.com>
Sun, 25 Nov 2018 22:13:08 +0000 (17:13 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 26 Jan 2019 08:37:02 +0000 (09:37 +0100)
[ Upstream commit b024dd0eba6e6d568f69d63c5e3153aba94c23e3 ]

FRWR memory registration is done with a series of calls and WRs.
1. ULP invokes ib_dma_map_sg()
2. ULP invokes ib_map_mr_sg()
3. ULP posts an IB_WR_REG_MR on the Send queue

Step 2 generates an iova. It is permissible for ULPs to change this
iova (with certain restrictions) between steps 2 and 3.

rxe_map_mr_sg captures the MR's iova but later when rxe processes the
REG_MR WR, it ignores the MR's iova field. If a ULP alters the MR's iova
after step 2 but before step 3, rxe never captures that change.

When the remote sends an RDMA Read targeting that MR, rxe looks up the
R_key, but the altered iova does not match the iova stored in the MR,
causing the RDMA Read request to fail.

Reported-by: Anna Schumaker <schumaker.anna@gmail.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/infiniband/sw/rxe/rxe_req.c

index de853bcc2384d92ede39f6f8d30c5c3749445d6d..08ae4f3a6a3795cf052cb59a7388ef0f24e3ad73 100644 (file)
@@ -640,6 +640,7 @@ next_wqe:
                        rmr->access = wqe->wr.wr.reg.access;
                        rmr->lkey = wqe->wr.wr.reg.key;
                        rmr->rkey = wqe->wr.wr.reg.key;
+                       rmr->iova = wqe->wr.wr.reg.mr->iova;
                        wqe->state = wqe_state_done;
                        wqe->status = IB_WC_SUCCESS;
                } else {