From: Kaike Wan kaike.wan@intel.com
commit a70ed0f2e6262e723ae8d70accb984ba309eacc2 upstream.
Each user context is allocated a certain number of RcvArray (TID) entries and these entries are managed through TID groups. These groups are put into one of three lists in each user context: tid_group_list, tid_used_list, and tid_full_list, depending on the number of used TID entries within each group. When TID packets are expected, one or more TID groups will be allocated. After the packets are received, the TID groups will be freed. Since multiple user threads may access the TID groups simultaneously, a mutex exp_mutex is used to synchronize the access. However, when the user file is closed, it tries to release all TID groups without acquiring the mutex first, which risks a race condition with another thread that may be releasing its TID groups, leading to data corruption.
This patch addresses the issue by acquiring the mutex first before releasing the TID groups when the file is closed.
Fixes: 3abb33ac6521 ("staging/hfi1: Add TID cache receive init and free funcs") Link: https://lore.kernel.org/r/20200210131026.87408.86853.stgit@awfm-01.aw.intel.... Reviewed-by: Mike Marciniszyn mike.marciniszyn@intel.com Signed-off-by: Kaike Wan kaike.wan@intel.com Signed-off-by: Dennis Dalessandro dennis.dalessandro@intel.com Signed-off-by: Jason Gunthorpe jgg@mellanox.com Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
--- drivers/infiniband/hw/hfi1/user_exp_rcv.c | 2 ++ 1 file changed, 2 insertions(+)
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c +++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c @@ -165,10 +165,12 @@ void hfi1_user_exp_rcv_free(struct hfi1_ if (fd->handler) { hfi1_mmu_rb_unregister(fd->handler); } else { + mutex_lock(&uctxt->exp_mutex); if (!EXP_TID_SET_EMPTY(uctxt->tid_full_list)) unlock_exp_tids(uctxt, &uctxt->tid_full_list, fd); if (!EXP_TID_SET_EMPTY(uctxt->tid_used_list)) unlock_exp_tids(uctxt, &uctxt->tid_used_list, fd); + mutex_unlock(&uctxt->exp_mutex); }
kfree(fd->invalid_tids);