Hi! So I'm going through the twisty labyrinth of ext4_map_blocks() and related functions to fix DAX related issues. Since I don't want to just bolt changes DAX needs somewhere on the side of that monster (it has already enough cruft bolted at various places) I'm looking into factoring some parts out. One thing I've noticed is the handling of delayed && unwritten extents. Currently we have quite some code to handle this case and it seems a bit disproportionate given how rare it is - AFAIU this can happen only when someone fallocates blocks under delalloc blocks. Also I'm somewhat uneasy that the whole unwritten extent gets marked as dealloc if any block in it was delalloc. That seems to have a potential for delalloc accounting issues (especially in bigalloc case), SEEK_DATA will report such unwritten extents as delayed (only a potential performance issue when someone then reads whole unwritten extent only to find out there's a bit of data in it), FIEMAP won't report delalloc flag for such extent at all... I see two solutions here: 1) We can fixup these issues (probably would need some massaging of extent cache code, fixup in FIEMAP). 2) We could just forbid such state - we could do that by fsyncing() range before it will be fallocated and exclude page faults for the file while fallocate is running (using new i_mmap_sem ext4 will have similarly to xfs). I'm in favor of solution 2) since IMHO fallocate under delalloc blocks mostly signals an application problem and so performance doesn't matter. But other's may have different opinion... Anyone? Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
