On 2018/5/11 18:46, Sowmini Varadhan wrote:
On (05/11/18 15:48), Yanjun Zhu wrote:diff --git a/net/rds/ib_rdma.c b/net/rds/ib_rdma.c index e678699..2228b50 100644 --- a/net/rds/ib_rdma.c +++ b/net/rds/ib_rdma.c @@ -539,11 +539,17 @@ void rds_ib_flush_mrs(void) void *rds_ib_get_mr(struct scatterlist *sg, unsigned long nents, struct rds_sock *rs, u32 *key_ret) { - struct rds_ib_device *rds_ibdev; + struct rds_ib_device *rds_ibdev = NULL; struct rds_ib_mr *ibmr = NULL; - struct rds_ib_connection *ic = rs->rs_conn->c_transport_data; + struct rds_ib_connection *ic = NULL; int ret; + if (rs->rs_bound_addr == 0) { + ret = -EPERM; + goto out; + } + + ic = rs->rs_conn->c_transport_data; rds_ibdev = rds_ib_get_device(rs->rs_bound_addr); if (!rds_ibdev) { ret = -ENODEV; I made this raw patch. If you can reproduce this bug, please make tests with it.I dont think this solves the problem, I think it just changes the timing under which it can still happen. what if the rds_remove_bound() in rds_bind() happens after the check for if (rs->rs_bound_addr == 0) added above by the patch I believe you need some type of synchronization (either through mutex, or some atomic flag in the rs or similar) to make sure rds_bind() and rds_ib_get_mr() are mutually exclusive.
Sure. I agree with you. Maybe mutex is a good choice. Zhu Yanjun
--Sowmini
