Subject: Removing tmpfs' experimental status, take 2
To: None <>
From: Julio M. Merino Vidal <>
List: tech-kern
Date: 11/09/2006 16:47:13

It has been almost three weeks since I proposed removing tmpfs's
experimental status.  At that time several concerns were raised and
all of them have been addressed.  These were:

* NFS-exportability.  tmpfs now works over NFS; at the very least
  its test suite can be executed successfully over a remote tmpfs.
* Data corruption: Two issues were raised.  Once was a real
  problem in tmpfs where it would not deal with mtimes properly
  during renames and the other was due to misconfiguration on
  the tester's machine.  In other words, there was no "data
  corruption" as such.
* vnd over files in tmpfs: This now works.
* Linux binaries cannot readdir tmpfs properly: This has been
  addressed and now applications such as (the
  one against which the bug was filled) can see the contents of
  tmpfs volumes.
* kqueue notifications have been audited and they now match the
  behavior of MFS (FFS).

Aside the raised issues:

* A tmpfs LKM has been added.
* A bug in tmpfs has been fixed that caused a crash (in DEBUG
  kernels only) when accessing the current directory after it was
* Maybe something else...

There still remains the problem of "too much memory usage" but fixing
that has a lot of chances of breaking stability.  Plus there is no
time to do that safely before 4.0 ships (e.g. I don't know where to
start yet, and it seems certainly non-trivial).

So... what do you think about removing the experimental status now and
enabling tmpfs by default in all GENERIC kernels?


Julio M. Merino Vidal <>
The Julipedia -