When blk_rq_timed_out_timer run, only requests that are not completed are selected for processing with LLD callback function.
In the case of fc_bsg_job_timeout, if BLK_EH_NOT_HANDLED returned, the block layer will be signaled to do nothing for the request, but the request still remains uncompleted since it is already marked with REQ_ATOM_COMPLETE by the timer function and blk_complete_request() will check it. Thus mm leakage is triggered. It is fixed by returning BLK_EH_HANDLED, instead of BLK_EH_NOT_HANDLED, with warning also recorded. Thanks all for comments when preparing this work. Signed-off-by: Hillf Danton <[email protected]> --- --- a/drivers/scsi/scsi_transport_fc.c 2010-11-01 19:54:12.000000000 +0800 +++ b/drivers/scsi/scsi_transport_fc.c 2010-11-24 20:55:12.000000000 +0800 @@ -3626,9 +3626,9 @@ fc_bsg_job_timeout(struct request *req) /* the blk_end_sync_io() doesn't check the error */ if (done) - return BLK_EH_NOT_HANDLED; - else - return BLK_EH_HANDLED; + printk(KERN_WARNING "%s: fixing mm leak...done\n", __func__); + + return BLK_EH_HANDLED; } static int _______________________________________________ devel mailing list [email protected] http://www.open-fcoe.org/mailman/listinfo/devel
