On 29/11/19 11:13, Igor Mammedov wrote: >> Agreed, mkstemp+shm_open seems better. Perhaps this could be done in >> hostmem-memfd.c though, basically as a fallback option? In principle >> one could even use getmntent to search for a hugetlbfs mount. > So far fall backs proved to be a pain to deal with, as end users can't > be sure what machine they are getting eventually. > I'd prefer if we fail cleanly if asked config isn't possible and > let user fix vm configuration instead. >
As far as I know memfd vs. mktemp+shm_open+shm_unlink is pretty much the same thing. memfd provide additional features such as sealing, but unless someone explicitly checks for memfd features, the two should look the same. Paolo