> > > > FWIW, with FUSE_PASSTHROUGH, I plan that a shared mmap of an inode > > in "passthrough mode" (i.e. has an open FOPEN_PASSTHROUGH file) will > > be allowed (maps the backing file) regardless of fc->direct_io_allow_mmap. > > FOPEN_PARALLEL_DIRECT_WRITES will also be allowed on an inode in > > "passthrough mode", because an inode in "passthrough mode" cannot have > > any pending page cache writes. > > > > This makes me realize that I will also need to handle passthrough of > > ->direct_IO() on an FOPEN_PASSTHROUGH file. > > I really need to take a few hours to look at your patches. > Only if you want to glimpse at WIP. There is still time before the direct_io/passthrough + mmap logic is worked out. Thanks, Amir.
