The optimization done in "xprtrdma: Simplify rpcrdma_mr_pop" was a bit too optimistic. MRs left over after a reconnect still need to be recycled, not added back to the free list, since they could be in flight or actually fully registered. Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx> --- net/sunrpc/xprtrdma/frwr_ops.c | 35 +++++++++++++++++++++++++++-------- net/sunrpc/xprtrdma/rpc_rdma.c | 2 +- net/sunrpc/xprtrdma/xprt_rdma.h | 1 + 3 files changed, 29 insertions(+), 9 deletions(-) diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c index 368cdf3..30065a2 100644 --- a/net/sunrpc/xprtrdma/frwr_ops.c +++ b/net/sunrpc/xprtrdma/frwr_ops.c @@ -88,15 +88,8 @@ void frwr_release_mr(struct rpcrdma_mr *mr) kfree(mr); } -/* MRs are dynamically allocated, so simply clean up and release the MR. - * A replacement MR will subsequently be allocated on demand. - */ -static void -frwr_mr_recycle_worker(struct work_struct *work) +static void frwr_mr_recycle(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr *mr) { - struct rpcrdma_mr *mr = container_of(work, struct rpcrdma_mr, mr_recycle); - struct rpcrdma_xprt *r_xprt = mr->mr_xprt; - trace_xprtrdma_mr_recycle(mr); if (mr->mr_dir != DMA_NONE) { @@ -114,6 +107,32 @@ void frwr_release_mr(struct rpcrdma_mr *mr) frwr_release_mr(mr); } +/* MRs are dynamically allocated, so simply clean up and release the MR. + * A replacement MR will subsequently be allocated on demand. + */ +static void +frwr_mr_recycle_worker(struct work_struct *work) +{ + struct rpcrdma_mr *mr = container_of(work, struct rpcrdma_mr, + mr_recycle); + + frwr_mr_recycle(mr->mr_xprt, mr); +} + +/* frwr_recycle - Discard MRs + * @req: request to reset + * + * Used after a reconnect. These MRs could be in flight, we can't + * tell. Safe thing to do is release them. + */ +void frwr_recycle(struct rpcrdma_req *req) +{ + struct rpcrdma_mr *mr; + + while ((mr = rpcrdma_mr_pop(&req->rl_registered))) + frwr_mr_recycle(mr->mr_xprt, mr); +} + /* frwr_reset - Place MRs back on the free list * @req: request to reset * diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c index 67e1684..19dd29a 100644 --- a/net/sunrpc/xprtrdma/rpc_rdma.c +++ b/net/sunrpc/xprtrdma/rpc_rdma.c @@ -867,7 +867,7 @@ static bool rpcrdma_prepare_msg_sges(struct rpcrdma_xprt *r_xprt, * chunks. Very likely the connection has been replaced, * so these registrations are invalid and unusable. */ - frwr_reset(req); + frwr_recycle(req); /* This implementation supports the following combinations * of chunk lists in one RPC-over-RDMA Call message: diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h index bd1befa..65e6b0e 100644 --- a/net/sunrpc/xprtrdma/xprt_rdma.h +++ b/net/sunrpc/xprtrdma/xprt_rdma.h @@ -542,6 +542,7 @@ static inline bool rpcrdma_regbuf_dma_map(struct rpcrdma_xprt *r_xprt, /* Memory registration calls xprtrdma/frwr_ops.c */ bool frwr_is_supported(struct ib_device *device); +void frwr_recycle(struct rpcrdma_req *req); void frwr_reset(struct rpcrdma_req *req); int frwr_open(struct rpcrdma_ia *ia, struct rpcrdma_ep *ep); int frwr_init_mr(struct rpcrdma_ia *ia, struct rpcrdma_mr *mr);