Re: [overlayfs/port] overlayfs: v13 port attempt to kernel 3.5

Re: [overlayfs/port] overlayfs: v13 port attempt to kernel 3.5

[Sedat Dilek]

On Fri, Aug 17, 2012 at 3:05 AM, Andrew Watts <akwatts@ymail.com> wrote:
>> Might be easier to collect/extract commits listed in [1]?
>>
>> If you are on Ubuntu: Check mainline kernels in [2] (quantal) and [3] (precise).
>>
>> - Sedat -
>>
>
> That indeed was much easier and I have it working great on 3.5.2. Many
> thanks for the link to the git repo.
>
> It would be nice to see overlayfs ack'd into mainline at some point. Is
> there any chance of this?
>

[ TO Linus and CC LKML and Daniel (maintainer of debian-live project) ]
( I could or should CC even more folks... )

Hi,

There exists a lot of threads on LKML and here on linux-fsdevel ML
about the topic "Union FileSystems".
In some of these discussions you will read about pros and cons about
"OverlayFS".
More precisely: Endless or even fruitless(?) discussions as after so
many years there is no upstream solution in sight.

Unfortunately, some kernel developers still didn't realize that
***Linux live-systems*** and ***Embedded Linux*** are IMHO two
examples for great "success stories" of Linux!

To run both successfully, you need this fundamental filesystem components:

(1) a "Compressed Read-Only File System" (like SquashFS)
(2) a "Union FileSystem" (like AUFS, OverlayFS, "Union Mounts", etc.)

Someday we got all happy with SquashFS accepted in upstream - this was
the first success!

The other part to run a Linux live-system is a "Union FileSystem" -
this part is missing (speaking of upstream).

Since years AUFS seems to be the choice #1 in a lot of distros to
workaround the problem.
NOTE: AUFS was rejected from upstream (to say not accepted).

I talked with so many distro kernel-teams and especially with Daniel
Baumann (maintainer of debian-live project):
The responsibles say if OverlayFS is upstreamed we will take it else
NOPE followed by an "OkThxBye" from impolite people!
[Q] So, why the hell do you include AUFS (not upstreamed) but not
OverlayFS (also not upstreamed) :-)?

As a sidenote, Linux-v3.6-rc1 got some first bits of "Union Mounts"
which was chosen as the ***preferred*** (longterm!) upstream solution.
But this special topic seems to me to get a ***never-ending story***.
Maintainers changed - new developers carried (or not?) - patchsets
were sent to MLs but got never really reviewed or finished.
YES, ***finished***!
I remember the "nVidia, F*** you!" video, not because of the famous(?)
F-word, but the part where Linus talks about "visions" VS. "I like to
see/get stuff finished".

One thing is a bit strange to me: Where does the development of "Union
Mounts" happen?
Normally, someone would assume - as this is gonna be a part of the
Linux kernel - this happens in the public.
I don't know of a repository - do you know?

Personally, I would like to see OverlayFS as a or the "small" solution
besides $union-filesystem-preferred-by-upstream.
OverlayFS is pretty small - a very good choice for embedded Linux systems.

After looking into the pile #1 of VFS pull-request for Linux-v3.6-rc1,
I noticed this...

     "* preparatory bits from unionmount series (from dhowells)."

So, I talked with some responsibles on #ubuntu-kernel and this was
appreciated (and lead to some adaptation of the OverlayFS code and
resulted in v14).
On the one hand it is nice to see that something happens in upstream,
on the other no one knows when "Union Mounts" will get a "working"
solution (to say "finished").

Just to name two projects:
Ubuntu (Linux distribution) and OpenWrt (Linux embedded system) use
OverlayFS in their live-system and build-system environments.

But... I am a person which is *not* in charge or responsible for the
development in the (near) future - nor in the position to decide.

I did some early testing (against linux-next) and contributed a few lines.
After that I am regularly testing OverlayFS since v11 till present.

Anyway, I would be really happy to see any working ***upstream solution***!

It might be worth to escalate this topic to Linus directly (Oops, I did?).

I hope people rest calm and do NOT throw me S*** and F*** words - be
always a Gentleman :-)!

My 0.02EUR.

- Sedat -

> ~ Andy

Re: [overlayfs/port] overlayfs: v13 port attempt to kernel 3.5

[J. R. Okajima]

Sedat Dilek:
> The other part to run a Linux live-system is a "Union FileSystem" -
> this part is missing (speaking of upstream).
>
> Since years AUFS seems to be the choice #1 in a lot of distros to
> workaround the problem.
> NOTE: AUFS was rejected from upstream (to say not accepted).

Exactly.
The reason was that linux rejects all union-type filesystems but
UnionMount (which is union-type mount).
Later, the development of UnionMount seems to be almost stopped. The
essential design of overlayfs is based upon UnionMount, and I have
pointed out several issues such as
- for users, the inode number may change silently. eg. copy-up.
- hardlinks may break by copy-up.
- read(2) may get an obsoleted filedata (fstat(2) for metadata too).
- fcntl(F_SETLK) may be broken by copy-up.
- unnecessary copy-up may happen, for example mmap(MAP_PRIVATE) after
  open(O_RDWR).
- Later I noticed one more thing. /proc/PID/{fd/,exe} may not work
  correctly for overlayfs ...
- etc...
They are called "unPOSIXy behavior", and unforunately many of them are
NOT seem to be addressed in recent patches either.

Also I have posted
	If the development of UnionMount is really stopped, then I'd ask people
	to consider merging aufs as well as overlayfs.
but I am not sure whether LKML people are still waiting for UnionMount
and rejecting aufs.


J. R. Okajima

Re: [overlayfs/port] overlayfs: v13 port attempt to kernel 3.5

[Sedat Dilek]

On Mon, Aug 20, 2012 at 12:38 PM, J. R. Okajima <hooanon05@yahoo.co.jp> wrote:
>
> Sedat Dilek:
>> The other part to run a Linux live-system is a "Union FileSystem" -
>> this part is missing (speaking of upstream).
>>
>> Since years AUFS seems to be the choice #1 in a lot of distros to
>> workaround the problem.
>> NOTE: AUFS was rejected from upstream (to say not accepted).
>
> Exactly.
> The reason was that linux rejects all union-type filesystems but
> UnionMount (which is union-type mount).
> Later, the development of UnionMount seems to be almost stopped. The
> essential design of overlayfs is based upon UnionMount, and I have
> pointed out several issues such as
> - for users, the inode number may change silently. eg. copy-up.
> - hardlinks may break by copy-up.
> - read(2) may get an obsoleted filedata (fstat(2) for metadata too).
> - fcntl(F_SETLK) may be broken by copy-up.
> - unnecessary copy-up may happen, for example mmap(MAP_PRIVATE) after
>   open(O_RDWR).
> - Later I noticed one more thing. /proc/PID/{fd/,exe} may not work
>   correctly for overlayfs ...
> - etc...
> They are called "unPOSIXy behavior", and unforunately many of them are
> NOT seem to be addressed in recent patches either.
>
> Also I have posted
>         If the development of UnionMount is really stopped, then I'd ask people
>         to consider merging aufs as well as overlayfs.
> but I am not sure whether LKML people are still waiting for UnionMount
> and rejecting aufs.
>

Okajima san /o\,

thanks for the short summary of history, quick overview of AUFS
features and OverlayFS design.
I can't say much about AUFS, I just simply used it for so long.

But it looks like you agree with me that something should happen in
case of Union Filesystem's.
Anyway, as said I would like to see an upstream solution in the near future.
If this will be AUFS, I am OK with that decision.

Regards,
- Sedat -

> J. R. Okajima