> So if you ran ps at exactly the right time (while one of the > procs on the machine was opening a #-ed file name and > was between strlen and sprint), you'd win the race. > You might improve your chances if the machine were > close to out of memory and the smalloc had to wait for > more memory. > > Oops. Changing those lines to use up->genbuf instead > of p->genbuf should eliminate the race.
many thanks. i've been chasing this bug for some time. often the backtrace has been much more confusing. in fact, ps is run fairly often on this machine because i am sanguine about dns' stability. if ever there were a canidate for publish subscribe, it would be dns. - erik
