On Fri 13-10-17 08:52:28, Dan Williams wrote: > On Fri, Oct 13, 2017 at 12:22 AM, Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote: > > On Wed, Oct 11, 2017 at 03:11:21PM -0700, Dan Williams wrote: > >> > + /* > >> > + * We don't support synchronous mappings for non-DAX files. At least > >> > + * until someone comes with a sensible use case. > >> > + */ > >> > + if (!IS_DAX(file_inode(file)) && (map_flags & MAP_SYNC)) > >> > + return -EOPNOTSUPP; > >> > >> Perhaps EPERM instead to differentiate the unsupported flags case? > >> There's precedent for mmap returning EPERM for reasons other than > >> incompatible PROT flags. > > > > Why would we want to voluntarily use arcane errors for a entirely > > new interface under our control? > > > > -EOPNOTSUPP is nice and explicit about what is going on. > > We have this general and perennial problem of parsing why the kernel > is throwing an error. It saves a debug step if the error code is > unique, and in this case would indicate that the filesystem supports > MAP_SYNC but the administrator failed to arrange for the device to be > in DAX mode. So I understand this wish however I think the error codes should also reflect the nature of the problem. And EPERM has something to do with access permissions, not whether file supports DAX access or not. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
