LazyFS for 2.6.16

Hello.

The current alpha version of LazyFS on Kernel 2.6.16 hangs like below.
This is reproducable issue, but I can not figure out the condition 
this happens.

Anybody gives me any hint?
How to debug this kind of issue? Any help?

                  --- Okajima, Jun. Tokyo, Japan.

-----------
[4294682.718000] Unable to handle kernel paging request at virtual address 447e8b30

[4294682.718000]  printing eip:

[4294682.718000] c01e452a

[4294682.718000] *pde = 00000000

[4294682.718000] Oops: 0000 [#1]

[4294682.718000] Modules linked in: lazyfs0d1d26 squashfs loop mptspi mptscsih mptbase

[4294682.718000] CPU:    0

[4294682.718000] EIP:    0060:[<c01e452a>]    Not tainted VLI

[4294682.718000] EFLAGS: 00010097   (2.6.16 #3) 

Re: LazyFS for 2.6.16

I fount much more important issue ---.
I dont know why but in some system, LazyFS does not run with tmpfs.
I mean, if you mount tmpfs on /var/cache/zero-inst, it does not run.
The key issue is, even you use (for example) 2.6.8 with the genie
LazyFS from sourceforge, it does not run.

Somebody give me help?

BTW, it fails here.
In short, it got host_dentry with lookup_hash(),
but the struct has no d_inode. Why???

static struct dentry *try_get_host_dentry(struct dentry *dentry)
{
....
if (!host_dentry->d_inode)
	goto fetch;
}

                --- Okajima, Jun. Tokyo, Japan.

>
>Hello.
>
>The current alpha version of LazyFS on Kernel 2.6.16 hangs like below.
>This is reproducable issue, but I can not figure out the condition 
>this happens.
>
>Anybody gives me any hint?

Re: LazyFS for 2.6.16

In other words, in what condition this happens?
In what condition "negative or the wrong type" inode is returned?

                   --- Okajima.

fetch:
	/* host_dentry exists, but is negative or the wrong type */
	dput(host_dentry);
	dec(R_HOST_DENTRY);
	return NULL;
}

------------------------
I fount much more important issue ---.
I dont know why but in some system, LazyFS does not run with tmpfs.
I mean, if you mount tmpfs on /var/cache/zero-inst, it does not run.
The key issue is, even you use (for example) 2.6.8 with the genie
LazyFS from sourceforge, it does not run.

Somebody give me help?

BTW, it fails here.
In short, it got host_dentry with lookup_hash(),
but the struct has no d_inode. Why???

static struct dentry *try_get_host_dentry(struct dentry *dentry)
{
....
if (!host_dentry->d_inode)

Re: LazyFS for 2.6.16

On Sun, 04 Jun 2006 18:16:35 +0900, Jun OKAJIMA wrote:

> I fount much more important issue ---. I dont know why but in some system,
> LazyFS does not run with tmpfs. I mean, if you mount tmpfs on
> /var/cache/zero-inst, it does not run.

Did you mount tmpfs before or after mounting lazyfs? Lazyfs attaches to
the root dentry when you mount it, so if you later mount something else on
top of the cache it will ignore it and continue trying to use the
filesystem under the mount (but the user-space helper will be saving files
into the new tmpfs filesystem).

--

-- 
Dr Thomas Leonard		http://rox.sourceforge.net
GPG: 9242 9807 C985 3C07 44A6  8B9A AE07 8280 59A5 3CC1

Re: LazyFS for 2.6.16

THANX THANX THANX!!!!!!!!!!!!!!!

I will try this now.

Anyway, thank you very much.

I have spent all my weekend for this problem. Sigh.
It is 7:45 PM Sunday in Tokyo, and hope very much 
if I can solve this issue within this Sunday....

            --- Okajima.

>
>> I fount much more important issue ---. I dont know why but in some system,
>> LazyFS does not run with tmpfs. I mean, if you mount tmpfs on
>> /var/cache/zero-inst, it does not run.
>
>Did you mount tmpfs before or after mounting lazyfs? Lazyfs attaches to
>the root dentry when you mount it, so if you later mount something else on
>top of the cache it will ignore it and continue trying to use the
>filesystem under the mount (but the user-space helper will be saving files
>into the new tmpfs filesystem).
>
>
>-- 
>Dr Thomas Leonard		http://rox.sourceforge.net
>GPG: 9242 9807 C985 3C07 44A6  8B9A AE07 8280 59A5 3CC1
>
>

Compiling GTK 2.4 applications

I've made a little script that sets up a GTK-2.4 build environment
using Zero Install. You can get it like this:

$ 0alias gtk-2.4 http://0install.net/2006/interfaces/GTK-2.4.xml

Running the 'gtk-2.4' command will give you a sub-shell with
LIBRARY_PATH and PKG_CONFIG_PATH set to the locations of the GTK, ATK,
Pango and Glib headers (all downloaded by Zero Install, of course).

For example, say you have 2.8.17 installed:

$ pkg-config --modversion gtk+-2.0
2.8.17

Then running 'gtk-2.4' will give:

$ gtk-2.4
Giving you a new shell... compile your program in this environment.
gtk2.4 $ pkg-config --modversion gtk+-2.0
2.4.14

You can now build your program as normal:

gtk2.4 $ ./configure && make
gtk2.4 $ exit

Exiting GTK 2.4 environment
$

The advantages of this are:

Flash demos

Thomas Formella has created a Flash video showing how to use 0launch:

  http://0install.net/flash.html

I've created a similar (smaller) version in English and put it on the same
page.

Anyone else got suggestions for making the system easier to understand?

--

-- 
Dr Thomas Leonard		http://rox.sourceforge.net
GPG: 9242 9807 C985 3C07 44A6  8B9A AE07 8280 59A5 3CC1

P2P support (slashdot effect)

Hi.

I'd be interested in continuing the "Slashdot effect" discussion.  In
particular, I've implemented a BitTorrent client, and integrated it with
FUSE to provide a download-as-needed capability that seems to work ok
(see https://sourceforge.net/projects/btslave/).

A good peer-to-peer filesystem backing zero-install seems to me like
potentially a very good thing.  In particular, it could:

-- Provide the massive bandwidth needed during slashdot effect events
-- Provide consistent fast downloads even when a source site is slow or
off-line, or even if it no longer exists
-- Help users find packages using distributed hash tables (DHTs)
-- Greatly reduce the bandwidth cost of hosting popular packages

BitTorrent is a simple innovative system for file distribution.
However, BitTorrent is not very good at supporting a dynamically
changing world of small distributed packages.  It's really best suited
for it's most famous use: downloading HUGE files, like RedHat
distributions or feature-length movies.

For this application, I think a new P2P system would need to be
developed.  It happens to be an area I've played around with a bit, and
I enjoy dabbling with P2P systems.  If you think there's potential in
the idea, I'd be interested in implementing it.

Bill

Re: P2P support (slashdot effect)

On Mon, 12 Jun 2006 22:00:10 -0400, Bill Cox wrote:

> Hi.
> 
> I'd be interested in continuing the "Slashdot effect" discussion.  In
> particular, I've implemented a BitTorrent client, and integrated it with
> FUSE to provide a download-as-needed capability that seems to work ok
> (see https://sourceforge.net/projects/btslave/).

One of the problems with implementing it as a filesystem is that the
interface with the user becomes very complicated. E.g., a full bittorrent
client lets you see all the peers and change your settings (port ranges,
upload and download rates, etc). Having a central daemon process do this
makes this kind of interface impossible, because you can't trust different
users to all control the shared daemon (just viewing it is tricky).

However, in the newer injector model the user downloads the software using
their own client (which can be as complex as you like), and then submits
the completed download to the sharing system (which just checks the
digest). I think this would work better for you than the old filesystem
implementation.

> BitTorrent is a simple innovative system for file distribution.
> However, BitTorrent is not very good at supporting a dynamically
> changing world of small distributed packages.  It's really best suited
> for it's most famous use: downloading HUGE files, like RedHat
> distributions or feature-length movies.
> 
> For this application, I think a new P2P system would need to be
> developed.  It happens to be an area I've played around with a bit, and

Re: P2P support (slashdot effect)

Am Dienstag, 13. Juni 2006 04:00 schrieb Bill Cox:
<snip>
> 
> For this application, I think a new P2P system would need to be
> developed.  It happens to be an area I've played around with a bit, and
> I enjoy dabbling with P2P systems.  If you think there's potential in
> the idea, I'd be interested in implementing it.
> 
May I can lead your eye to GNU's p2p network: http://gnunet.org/

Especially this GNUnet feature sounds very interesting:

<quote>
Peers in GNUnet monitor each others behavior with respect to resource usage; peers that contribute to the
network are rewarded with better service.'
</quote>

Regards,
Thomas

Am Dienstag, 13. Juni 2006 04:00 schrieb Bill Cox:
<snip>
> 
> For this application, I think a new P2P system would need to be
> developed.  It happens to be an area I've played around with a bit, and
> I enjoy dabbling with P2P systems.  If you think there's potential in
> the idea, I'd be interested in implementing it.
> 
May I can lead your eye to GNU's p2p network: http://gnunet.org/