L2CAP COC: Preventing a race condition about mynewt-nimble HOT 4 CLOSED

Fixed

from mynewt-nimble.

I looked a bit further and I think a client workaround seems unlikely, since this pretty much happens inside NimBLE.

In ble_l2cap_coc:492 it says

        conn = ble_hs_conn_find_assert(chan->conn_handle);
        rc = ble_l2cap_tx(conn, chan, txom);

        if (rc) {
            /* txom is consumed by l2cap */
            txom = NULL;
            goto failed;
        } else {
            tx->credits--;
            tx->data_offset += len - sdu_size_offset;
        }

ble_hs_conn_find_assert is supposed to trigger an assertion in debug code, when the conn_handle is nowhere to be found in the active connections. For production code, this (of course) doesn't happen -- but an assertion might be too much of a weapon here anyways. I suppose something along the lines of:

        conn = ble_hs_conn_find_assert(chan->conn_handle);
        if (!conn) { /* connection might have been dropped in the meantime */
            goto failed;
        }
        rc = ble_l2cap_tx(conn, chan, txom);

        if (rc) {
            /* txom is consumed by l2cap */
            txom = NULL;
            goto failed;
        } else {
            tx->credits--;
            tx->data_offset += len - sdu_size_offset;
        }

This change helps in my tests. The upper layers then properly detect that send fails, but since it's during shutdown, they're silently ignored.

from mynewt-nimble.

Hi, could you please try applying this patch? #1766 And please let me know if this works

from mynewt-nimble.

@KKopyscinski This patch works very well. Instead of a crash, I get the expected -ENOTCONN now.

E (9486) NimBLEL2CAPChannel: ble_l2cap_send failed: 7

Thanks!

from mynewt-nimble.