| Message ID | 20230129234441.116310-14-michael.christie@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | target: TMF and recovery fixes | expand |
po 30. 1. 2023 v 0:45 odesÃlatel Mike Christie <michael.christie@oracle.com> napsal: > > We can end up freeing a command too early during conn cleanup by doing: > > 1. Send iserts N 1MB writes. > 2. Pull cable or down the port isert is using. > 3. isert_wait_conn's call to isert_put_unsol_pending_cmds races > with a running isert_send_done -> isert_completion_put -> isert_put_cmd > where isert_put_unsol_pending_cmds sees the cmd on the conn_cmd_list > and it does a isert_put_cmd which will free the cmd. Then isert_send_done > is run and will access the freed command while doing it's normal command > completion: > isert_completion_put -> isert_put_cmd -> transport_generic_free_cmd > > This patch has us increment write_data_done, so > isert_put_unsol_pending_cmds can correctly detect commands which will we > will not be calling isert_send_done for. > > Fixes: 38a2d0d429f1 ("IB/isert: convert to the generic RDMA READ/WRITE > API") > [code written and suggested by] > Suggested-by: Sagi Grimberg <sagi@grimberg.me> > > Signed-off-by: Mike Christie <michael.christie@oracle.com> > --- > drivers/infiniband/ulp/isert/ib_isert.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c > index 516fa37494e1..a44da60352f6 100644 > --- a/drivers/infiniband/ulp/isert/ib_isert.c > +++ b/drivers/infiniband/ulp/isert/ib_isert.c > @@ -1636,7 +1636,7 @@ isert_rdma_read_done(struct ib_cq *cq, struct ib_wc *wc) > if (isert_prot_cmd(isert_conn, se_cmd)) > ret = isert_check_pi_status(se_cmd, isert_cmd->rw.reg->mr); > isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn); > - cmd->write_data_done = 0; > + cmd->write_data_done = se_cmd->data_length; /* done fetching data */ > > isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd); > spin_lock_bh(&cmd->istate_lock); > -- > 2.25.1 > Reviewed-by: Maurizio Lombardi <mlombard@redhat.com>
diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index 516fa37494e1..a44da60352f6 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -1636,7 +1636,7 @@ isert_rdma_read_done(struct ib_cq *cq, struct ib_wc *wc) if (isert_prot_cmd(isert_conn, se_cmd)) ret = isert_check_pi_status(se_cmd, isert_cmd->rw.reg->mr); isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn); - cmd->write_data_done = 0; + cmd->write_data_done = se_cmd->data_length; /* done fetching data */ isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd); spin_lock_bh(&cmd->istate_lock);
We can end up freeing a command too early during conn cleanup by doing: 1. Send iserts N 1MB writes. 2. Pull cable or down the port isert is using. 3. isert_wait_conn's call to isert_put_unsol_pending_cmds races with a running isert_send_done -> isert_completion_put -> isert_put_cmd where isert_put_unsol_pending_cmds sees the cmd on the conn_cmd_list and it does a isert_put_cmd which will free the cmd. Then isert_send_done is run and will access the freed command while doing it's normal command completion: isert_completion_put -> isert_put_cmd -> transport_generic_free_cmd This patch has us increment write_data_done, so isert_put_unsol_pending_cmds can correctly detect commands which will we will not be calling isert_send_done for. Fixes: 38a2d0d429f1 ("IB/isert: convert to the generic RDMA READ/WRITE API") [code written and suggested by] Suggested-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Mike Christie <michael.christie@oracle.com> --- drivers/infiniband/ulp/isert/ib_isert.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)