[DISCUSSION] Journal playback

[DISCUSSION] Journal playback

[Vladimir Serbinenko]

I've seen that journal playback is one of TODO items. I think it's bad
idea because:

1) It'll increase the size of FS modules. And fs modules are often embed
in core image. So their size is critical
2) It'll decrease safety because now GRUB doesn't write to FS and then
it'll be so the GRUB's bugs could damage FS.
3) It'll be impossible to use resume/resume2 because the fs environment
is not more the same as at the hibernation.

What do you think about this?
                                                                        
                                          Vladimir

Re: [DISCUSSION] Journal playback

[Vincent Pelletier]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Vladimir Serbinenko wrote:
> 1) It'll increase the size of FS modules. And fs modules are often embed
> in core image. So their size is critical

True (on x86).

> 2) It'll decrease safety because now GRUB doesn't write to FS and then
> it'll be so the GRUB's bugs could damage FS.

I don't think the journal replay would actualy be written to hte FS. As
I understood the changelog item, it would only be used to read the files
grub uses with the advantage of journalized FS. Maybe a special module
could be written to allow actually applying the changes to the FS, to
recover - say - a broken kernel image.

For 3) I don't know enought about hibernation.

Vincent Pelletier
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)

iD8DBQFC90M4FEQoKRQyjtURAtBdAKCgofsak4i3kxQOmq8owK+zYgM29wCfSRLi
111CbxnPlfGLUKJJpGH/Ww8=
=ZsuG
-----END PGP SIGNATURE-----

	

	
		
___________________________________________________________________________ 
Appel audio GRATUIT partout dans le monde avec le nouveau Yahoo! Messenger 
Tйlйchargez cette version sur http://fr.messenger.yahoo.com

Re: [DISCUSSION] Journal playback

[Marco Gerards]

Vladimir Serbinenko <phcoder@gmail.com> writes:

> I've seen that journal playback is one of TODO items. I think it's bad
> idea because:
>
> 1) It'll increase the size of FS modules. And fs modules are often embed
> in core image. So their size is critical

The size of a filesystem module is now about 4KB.  We do not know yet
how much the size will increase when journaling is possible.

> 2) It'll decrease safety because now GRUB doesn't write to FS and then
> it'll be so the GRUB's bugs could damage FS.
> 3) It'll be impossible to use resume/resume2 because the fs environment
> is not more the same as at the hibernation.

The data should not be written to the filesystem.  Instead the
journal should be played back for the data that is read.  I am not yet
sure if it is possible, easy and fast enough.  But I do think it is
worthwhile to consider journaling support.

--
Marco

Re: [DISCUSSION] Journal playback

[Vladimir Serbinenko]

Marco Gerards wrote:

>Vladimir Serbinenko <phcoder@gmail.com> writes:
>
>  
>
>>I've seen that journal playback is one of TODO items. I think it's bad
>>idea because:
>>
>>1) It'll increase the size of FS modules. And fs modules are often embed
>>in core image. So their size is critical
>>    
>>
>
>The size of a filesystem module is now about 4KB.  We do not know yet
>how much the size will increase when journaling is possible.
>
>  
>
>>2) It'll decrease safety because now GRUB doesn't write to FS and then
>>it'll be so the GRUB's bugs could damage FS.
>>3) It'll be impossible to use resume/resume2 because the fs environment
>>is not more the same as at the hibernation.
>>    
>>
>
>The data should not be written to the filesystem.  Instead the
>journal should be played back for the data that is read.  I am not yet
>sure if it is possible, easy and fast enough.  But I do think it is
>worthwhile to consider journaling support.
>
>  
>
It'll give for sure considerable amount of work but the question if
it'll give real advantages?
Journaling was made to keep *FS* from the corruption. *Not* the data. So
even if data needed
by GRUB is corrupted (that is not very probable because kernels, initrds
and GRUB data)
there is a high probability that journaling will simply remove this
corrupted data. On the other
hand it can perhaps create strange bugs and make maintnaining harde
                                                                     
Vladimir

Re: [DISCUSSION] Journal playback

[Marco Gerards]

Vladimir Serbinenko <phcoder@gmail.com> writes:

> It'll give for sure considerable amount of work but the question if
> it'll give real advantages?  Journaling was made to keep *FS* from
> the corruption. *Not* the data. So even if data needed by GRUB is
> corrupted (that is not very probable because kernels, initrds and
> GRUB data) there is a high probability that journaling will simply
> remove this corrupted data. On the other hand it can perhaps create
> strange bugs and make maintnaining harde

Right, I know what journaling is used for.  But this makes it possible
to boot from a corrupted filesystem that can not be used without
journaling support.

About how hard it is, we will see about it when we will work on it.
It is not an important feature; at least not to me.  But it is
something we have to keep in mind because, unlike lilo, GRUB does
depend on the consistency of the metadata.

--
Marco

Re: [DISCUSSION] Journal playback

[Yoshinori K. Okuji]

On Monday 08 August 2005 14:27, Vladimir Serbinenko wrote:
> It'll give for sure considerable amount of work but the question if
> it'll give real advantages?

Yes, it does. Especially when the implementation of a journaling is loose, the 
superblock of a journaling filesystem often breaks, when you cut the power 
supply while your OS is replaying a journal. In this case, GRUB cannot read 
anything from the filesystem, because the metadata of the root directory is 
corrupted. Due to this, ext3 is more fragile than ext2, in my experience.

Okuji

Re: [DISCUSSION] Journal playback

[Yoshinori K. Okuji]

On Monday 08 August 2005 13:46, Marco Gerards wrote:
> The size of a filesystem module is now about 4KB.  We do not know yet
> how much the size will increase when journaling is possible.

According to the experience in GRUB Legacy, it will increase the size by 1KB 
or so. Since pre-loaded modules are compressed, it won't be be that much.

> The data should not be written to the filesystem.  Instead the
> journal should be played back for the data that is read.  I am not yet
> sure if it is possible, easy and fast enough.  But I do think it is
> worthwhile to consider journaling support.

It should be feasible, since Jochen succeeded to implement journaling support 
for ReiserFS in GRUB Legacy. In GRUB 2, it should be much easier, because we 
can use more memory dynamically.

Okuji