Hi Jan, On Tue, May 23, 2017 at 10:24:10AM +0200, Jan Kara wrote: > On Mon 22-05-17 17:53:16, Eric Biggers wrote: > > From: Eric Biggers <ebiggers@xxxxxxxxxx> > > > > Currently we don't allow direct I/O on encrypted regular files, so in > > such cases we return 0 early in ext4_direct_IO(). There was also an > > additional BUG_ON() check in ext4_direct_IO_write(), but it can never be > > hit because of the earlier check for the exact same condition in > > ext4_direct_IO(). There was also no matching check on the read path, > > which made the write path specific check seem very ad-hoc. > > > > Just remove the unnecessary BUG_ON(). > > > > Signed-off-by: Eric Biggers <ebiggers@xxxxxxxxxx> > > Yeah, the check is rather before the BUG_ON so I guess that there's no big > point in the BUG_ON. When looking at this code I have one question though: > > So when you mount the filesystem with 'dioread_nolock', do overwriting > direct write to the file, and just after we do inode_unlock() in > ext4_direct_IO_write() someone calls EXT4_IOC_SET_ENCRYPTION_POLICY ioctl > on the file, the BUG_ON could actually trigger. So I think you need > to wait for outstanding direct IO for the file when setting encryption > policy. Likely in ext4_set_context() or maybe in the generic fscrypt code > (you need to wait after acquiring inode_lock), I'm not sure how other > filesystems using fscrypt handle this and whether it would make more sense > in the generic code or in ext4 specific one. > That's not possible because the ioctl can only set an encryption policy on a directory, and specifically an empty one. Other files can only acquire an encryption policy through inheritance. There have been thoughts about implementing "in-place" encryption but it's not something we currently support. Eric
