OK, most dragons seem to be dead or at least severely wounded.  I hope
I will post a patch against test8 today or at the latest tomorrow. 
For anyone who is interested, there is enough here to verify whether
this tailmerging strategy is effective at saving Ext2 filesystem space
or not (seems pretty effective to me) and what the performance impact
is (I haven't checked that).  It's guaranteed broken for concurrent
access because I haven't even begun to look at the issues yet.

It's now possible to fully merge a filesystem, use it for a while
causing some files to become unmerged, then fully unmerge it and pass
a fsck.  At least, it's possible with the test cases I've tried so
far.  My guess: there are still two bugs lurking in this code, then
the dragons are planning to attack again when I get into concurrency.
:-)

Done:
  - mount options: tails=merge/unmerge/on/off
  - full merge on mount, only if not already merged
  - full unmerge on mount
  - read/write in-place in merged files
  - write across/past merged tail with unmerge
  - truncate always unmerges (could stay merged on truncate within
tail)
  - delete inode

Not Done:
  - diffs against test7 and forward port to test8
  - package everything up with docs and test cases
  - mmap not finished - just does an unmerge now (bad for r/o)
  - full merge on mount can't handled aready-merged fs
  - Concurrency issues not addressed at all
  - need to ensure global locking in inode order
  - on last unmerge, could reuse tail instead of freeing
  - test across architectures
  - Ext2 compatibility and feature bits
  - modify e2fsck

Notes:
  - Patch exports one internal function (create_empty_buffers)
  - All other patches are against fs/ext2/* and include/linux/
  - Implements fstype "mrg2", a temporary name.
  - This code for developers only
  - Tracing code is still embedded
  - Includes sample test cases

--
Daniel
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]

Reply via email to