On Wed, Apr 15, 2020 at 9:03 AM Theodore Y. Ts'o <tytso@xxxxxxx> wrote: > > On Mon, Apr 13, 2020 at 09:00:25PM -0700, ira.weiny@xxxxxxxxx wrote: > > From: Ira Weiny <ira.weiny@xxxxxxxxx> > > > > Encryption and DAX are incompatible. Changing the DAX mode due to a > > change in Encryption mode is wrong without a corresponding > > address_space_operations update. > > > > Make the 2 options mutually exclusive by returning an error if DAX was > > set first. > > > > Signed-off-by: Ira Weiny <ira.weiny@xxxxxxxxx> > > The encryption flag is inherited from the containing directory, and > directories can't have the DAX flag set, so anything we do in > ext4_set_context() will be safety belt / sanity checking in nature. > > But we *do* need to figure out what we do with mount -o dax=always > when the file system might have encrypted files. My previous comments > about the verity flag and dax flag applies here. > > Also note that encrypted files are read/write so we must never allow > the combination of ENCRPYT_FL and DAX_FL. So that may be something > where we should teach __ext4_iget() to check for this, and declare the > file system as corrupted if it sees this combination. (For VERITY_FL > && DAX_FL that is a combo that we might want to support in the future, > so that's probably a case where arguably, we should just ignore the > DAX_FL for now.) We also have a pending consideration for what MKTME (inline memory encryption with programmable hardware keys) means for file-encryption + dax. Certainly kernel based software encryption is incompatible with dax, but one of the hallway track discussions I wanted to have at LSF is whether fscrypt is the right interface for managing inline memory encryption. For now, disallowing ENCRPYT_FL + DAX_FL seems ok if ENCRPYT_FL always means software encryption, but this is something to circle back to once we get MKTME implemented for volume encryption and start to look at finer grained (per-directory key) encryption.