Re: vm-pageout-throttling.patch: hanging in throttle_vm_writeout/blk_congestion_wait

[Andrew Morton <akpm@osdl.org>]

Hans Reiser <reiser@namesys.com> wrote:
>
> >>>loop-AES-2.2d)...  I think (!) it caused this deadlock.
>  >>>      
>  >>>
>  >>That's not at all surprising.  See the swap_extent work Andrew did
>  >>for 2.5 (in mm/swapfile.c), by which swap to a swapfile now avoids
>  >>the filesystem altogether (except while swapon prepares the map of
>  >>disk blocks).  By swapping to a loop device over a file, you're
>  >>sneaking past his work, and putting the filesystem back under swap.
>  >>    
>  >>
> 
>  Does Andrew's approach prevent putting swap on a compressed file (useful 
>  for reiser4 once the compression plugin is stable, not reiserfs)? (And 
>  no, I don't have any idea what the performance effect of that would be 
>  before it is tried and benchmarked....)

swapfiles bypass the filesystem completely, so if you're implementing
compression at the writepage() level, swap will cheerfully ignore all that
and will launch submit_bio() direct against your blockdev anyway.

encrypted swap should be done via dm-crypt.  compressed swap would I guess
require dm-compress.