Bart, >> While the WARN_ON here looks helpful shouldn't we instead ensure that >> sd_setup_rw6_cmnd never gets called with a 0 argument instead of bailing >> out like this? > > Before I posted this patch I searched for code that submits read or > write requests with length 0 but I haven't found any. do_iter_read() > and do_iter_write() in fs/read_write.c do not submit any block layer > requests if tot_len == 0. Are you perhaps aware of kernel code that > can submit zero-length read or write requests? At some level I would have liked the sanity check in the top rw prep command. However, given that a zero transfer length is mostly a hypothetical scenario I tend to concur with sticking the check in rw6 outside of the hot path. -- Martin K. Petersen Oracle Linux Engineering
