On Thu, Mar 09, 2017 at 07:43:12AM -0500, Jeff Layton wrote: > > Maybe we need a systemwide (or fs-level) tunable that makes ENOSPC a > transient error? Just have it hang until we get enough space when that > tunable is enabled? Or maybe we need a new kernel-internal errno (ala ERESTARSYS) which means it's a "soft ENOSPC"? It would get translated to ENOSPC if it gets propagated to userspace, but that way for devices like dm-thin or other storage array with thin volumes, it could send back a soft ENOSPC, while for file systems where "ENOSPC means ENOSPC", we can treat those as a hard ENOSPC. - Ted
