This patchset fixes the race between smc_release triggered by close(2) and cdc_handle triggered by underlaying RDMA device. The race is caused because the smc_connection may been released before the pending tx CDC messages got its CQEs. In order to fix this, I add a counter to track how many pending WRs we have posted through the smc_connection, and only release the smc_connection after there is no pending WRs on the connection. The first patch prevents posting WR on a QP that is not in RTS state. This patch is needed because if we post WR on a QP that is not in RTS state, ib_post_send() may success but no CQE will return, and that will confuse the counter tracking the pending WRs. The second patch add a counter to track how many WRs were posted through the smc_connection, and don't reset the QP on link destroying to prevent leak of the counter. Dust Li (2): net/smc: don't send CDC/LLC message if link not ready net/smc: fix kernel panic caused by race of smc_sock net/smc/smc.h | 5 +++++ net/smc/smc_cdc.c | 52 +++++++++++++++++++++------------------------- net/smc/smc_cdc.h | 2 +- net/smc/smc_core.c | 27 ++++++++++++++++++------ net/smc/smc_core.h | 6 ++++++ net/smc/smc_ib.c | 4 ++-- net/smc/smc_ib.h | 1 + net/smc/smc_llc.c | 2 +- net/smc/smc_wr.c | 45 +++++---------------------------------- net/smc/smc_wr.h | 5 ++--- 10 files changed, 68 insertions(+), 81 deletions(-) -- 2.19.1.3.ge56e4f7
