Quota patches

Quota patches

[Jan Kara]

  Hello,

  In following mails I'll send (because patches are big, I'll post them just
diretly to Linus - others see ftp below) quota patches for 2.5.15 (patches
apply well on 2.5.16 too). Currently they implement:
  * new quota format (allows 32 uids, accounting in bytes -> mainly for
    Reiserfs)
  * needed infrastructure for XFS quota
  * quota statistics in /proc (we can drop Q_GETSTATS call; it's a lot
    easier to change in future)
  * implements correct syncing of quota
  * introduces interface which allows usage of both quota formats in kernel
  * implemented filesystem callback function on certain quota operations
    (needed for journaled quota, Ext3)
  * implements ioctl() for reporting occupied space in bytes (not just blocks)

  The patches can be downloaded at:
ftp://atrey.karlin.mff.cuni.cz/pub/local/jack/quota/v2.5/2.5.15/

  Old quota tools should work with patches if you configure old quota interface
in '.config'. There are also quota utilities capable of communicating with new
generic interface. You can download them at:

http://www.sf.net/projects/linuxquota/

or you can checkout version from SourceForge CVS.

  Any comments & bugreports welcome.

								Honza

Below is a bit more detailed list of changes:
  The changes are split into 13 parts (should I create one empty patch? ;-)):
quota-2.5.15-1-newlocks - this patch changes counting of references on dquot structures
  so filesystem can be sure we never call it during DQUOT_ALLOC/DQUOT_FREE/..
  calls.
quota-2.5.15-2-formats - this patch removes most format dependent code from dquot.c
  and quota.h and puts calls of callback functions instead
quota-2.5.15-3-register - this patch implements registering/unregistering of quota
  formats
quota-2.5.15-4-getstats - this patch removes Q_GETSTATS call and creates /proc/fs/quota
  entry instead
quota-2.5.15-5-bytes - implements accounting of used space in bytes on quota side
quota-2.5.15-6-bytes - implements accounting of used space in bytes on VFS side -
  - neccessary functions are added to fs.h
quota-2.5.15-7-quotactl - implementation of generic quotactl interface (probably the
  biggest patch). Interface is moved from dquot.c to quota.c file. Pointers
  to quota operations in superblock are now not filled on quota_on() but
  on mount so filesystem can override them (for example ext3 would like to
  check on quota_on() that quotafile lies on proper device and turn on
  data-journaling on it - at least when we'll have journaled quota :)).
quota-2.5.15-8-format1 - implementation of old quota format (mainly moved stuff from
  dquot.c + some interface functions)
quota-2.5.15-9-format2 - implementation of new quota format (mainly just copied from
  patches used in -ac kernels)
quota-2.5.15-10-inttype - replace silly usage of 'short' by 'int'
quota-2.5.15-11-sync - implements correct syncing of quota - also quota info stored
  in superblock is stored (here 2.4.18 and 2.5.6 patches significantly differ
  - in 2.5.6 it's a bit simplier to do it)
quota-2.5.15-12-compat - implements backward compatible quotactl() interface. It's
  configurable whether it should be used at all and whether is should behave
  as interface in Linus's (the oldest interface) or Alan's (old interface for
  new quota format) kernel.
quota-2.5.15-13-ioctl - implements ioctl for getting file size in bytes. I placed
  this patch as the last one because I consider it ugly to create ioctl() for
  such thing (changing stat() would be cleaner but this change isn't probably
  important enough for it to be worth yet another stat() change). So if it
  is decided the patch won't be included there won't be problems...

-- 
Jan Kara <jack@suse.cz>
SuSE CR Labs

Re: Quota patches

[Christoph Hellwig]

On Mon, May 20, 2002 at 03:55:31PM +0200, Jan Kara wrote:
> quota-2.5.15-3-register - this patch implements registering/unregistering of quota
>   formats

Please don't use the big kernel lock for a newly added list.
Also using <linux/lists.h> would clean up the list handling.

> quota-2.5.15-4-getstats - this patch removes Q_GETSTATS call and creates /proc/fs/quota
>   entry instead

Yuck, even more /proc abuse.  Please convert it to the seq_file interface
at least. Using individual sysctls per value would be much better.

> quota-2.5.15-7-quotactl - implementation of generic quotactl interface (probably the
>   biggest patch). Interface is moved from dquot.c to quota.c file. Pointers
>   to quota operations in superblock are now not filled on quota_on() but
>   on mount so filesystem can override them (for example ext3 would like to
>   check on quota_on() that quotafile lies on proper device and turn on
>   data-journaling on it - at least when we'll have journaled quota :)).

The vfs_get*/vfs_set* names sound too generic, could you please rename them
to vfs_get_quota*/vfs_set_quota*?

Also I think any quota supporting filesystem should set the quota operations
explicitly to make the intention clearer.

> quota-2.5.15-12-compat - implements backward compatible quotactl() interface. It's
>   configurable whether it should be used at all and whether is should behave
>   as interface in Linus's (the oldest interface) or Alan's (old interface for
>   new quota format) kernel.

I don't think we want to keep old userspace interface in 2.5, it just
bloats the kernel and requiring quota tools for a development kernel that
are already required by all vendor kernels sounds sane to me.

Else your patches look very good to me, I look forward to finally see
properly working quota support in a mainline kernel.

	Christoph

Re: Quota patches

[Jan Kara]

> On Mon, May 20, 2002 at 03:55:31PM +0200, Jan Kara wrote:
> > quota-2.5.15-3-register - this patch implements registering/unregistering of quota
> >   formats
> 
> Please don't use the big kernel lock for a newly added list.
  We should get rid of big kernel lock in whole quota subsystem (which I want to do
as soon as Linus accepts the patches). So I agree with you I wanted to do some
extra lock it later.

> Also using <linux/lists.h> would clean up the list handling.
  I intentionally didn't use <linux/lists.h> as the use of the list
is so trivial and structure so small that it's IMHO overkill.

> > quota-2.5.15-4-getstats - this patch removes Q_GETSTATS call and creates /proc/fs/quota
> >   entry instead
> 
> Yuck, even more /proc abuse.  Please convert it to the seq_file interface
> at least. Using individual sysctls per value would be much better.
  I'll have a look at it...

> > quota-2.5.15-7-quotactl - implementation of generic quotactl interface (probably the
> >   biggest patch). Interface is moved from dquot.c to quota.c file. Pointers
> >   to quota operations in superblock are now not filled on quota_on() but
> >   on mount so filesystem can override them (for example ext3 would like to
> >   check on quota_on() that quotafile lies on proper device and turn on
> >   data-journaling on it - at least when we'll have journaled quota :)).
> 
> The vfs_get*/vfs_set* names sound too generic, could you please rename them
> to vfs_get_quota*/vfs_set_quota*?
  Good point... I'll change it.

> Also I think any quota supporting filesystem should set the quota operations
> explicitly to make the intention clearer.
  Hmm.. I don't know if it's cleaner but I start to like idea that this way quota_on()
and other operations will fail on filesystem not supporting quotas (currently everything
is silent, just quota is not counted...).

> > quota-2.5.15-12-compat - implements backward compatible quotactl() interface. It's
> >   configurable whether it should be used at all and whether is should behave
> >   as interface in Linus's (the oldest interface) or Alan's (old interface for
> >   new quota format) kernel.
> 
> I don't think we want to keep old userspace interface in 2.5, it just
> bloats the kernel and requiring quota tools for a development kernel that
> are already required by all vendor kernels sounds sane to me.
  Actually I included the patch mainly because I have it created for 2.4 where it's
reasonable to have it (I also have a backport for 2.4 because it will
take a lot of time before 2.6 will exist and be stable enough) and so I wanted to have
the patch also tested in 2.5... But I agree that it's bloating the kernel and
so if it won't be included I won't mind too much.

> Else your patches look very good to me, I look forward to finally see
> properly working quota support in a mainline kernel.

-- 
Jan Kara <jack@suse.cz>
SuSE CR Labs

Re: Quota patches

[OGAWA Hirofumi]

Hi,

Jan Kara <jack@suse.cz> writes:

>   Hello,
> 
>   In following mails I'll send (because patches are big, I'll post them just
> diretly to Linus - others see ftp below) quota patches for 2.5.15 (patches
> apply well on 2.5.16 too). Currently they implement:
>   * new quota format (allows 32 uids, accounting in bytes -> mainly for
>     Reiserfs)
>   * needed infrastructure for XFS quota
>   * quota statistics in /proc (we can drop Q_GETSTATS call; it's a lot
>     easier to change in future)
>   * implements correct syncing of quota
>   * introduces interface which allows usage of both quota formats in kernel
>   * implemented filesystem callback function on certain quota operations
>     (needed for journaled quota, Ext3)
>   * implements ioctl() for reporting occupied space in bytes (not just blocks)
> 
>   The patches can be downloaded at:
> ftp://atrey.karlin.mff.cuni.cz/pub/local/jack/quota/v2.5/2.5.15/
> 
>   Old quota tools should work with patches if you configure old quota interface
> in '.config'. There are also quota utilities capable of communicating with new
> generic interface. You can download them at:
> 
> http://www.sf.net/projects/linuxquota/
> 
> or you can checkout version from SourceForge CVS.
> 
>   Any comments & bugreports welcome.

What do you think of the following patches for kernel without
quota support? We already have weak symbol for sys_quotactl(). 

--- linux-bk/fs/Makefile        Wed May 22 01:17:48 2002
+++ linux-2.5.17/fs/Makefile    Thu May 23 03:23:30 2002
@@ -15,7 +15,7 @@
                namei.o fcntl.o ioctl.o readdir.o select.o fifo.o locks.o \
                dcache.o inode.o attr.o bad_inode.o file.o iobuf.o dnotify.o \
                filesystems.o namespace.o seq_file.o xattr.o libfs.o \
-               fs-writeback.o quota.o
+               fs-writeback.o

 ifneq ($(CONFIG_NFSD),n)
 ifneq ($(CONFIG_NFSD),)
@@ -81,7 +81,7 @@

 obj-$(CONFIG_BINFMT_ELF)       += binfmt_elf.o

-obj-$(CONFIG_QUOTA) += dquot.o
+obj-$(CONFIG_QUOTA) += dquot.o quota.o
 obj-$(CONFIG_QFMT_V1) += quota_v1.o
 obj-$(CONFIG_QFMT_V2) += quota_v2.o

Regards.
-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Re: Quota patches

[Jan Kara]

  Hello,
  
> Jan Kara <jack@suse.cz> writes:
> 
> >   Hello,
> > 
> >   In following mails I'll send (because patches are big, I'll post them just
> > diretly to Linus - others see ftp below) quota patches for 2.5.15 (patches
> > apply well on 2.5.16 too). Currently they implement:
> >   * new quota format (allows 32 uids, accounting in bytes -> mainly for
> >     Reiserfs)
> >   * needed infrastructure for XFS quota
> >   * quota statistics in /proc (we can drop Q_GETSTATS call; it's a lot
> >     easier to change in future)
> >   * implements correct syncing of quota
> >   * introduces interface which allows usage of both quota formats in kernel
> >   * implemented filesystem callback function on certain quota operations
> >     (needed for journaled quota, Ext3)
> >   * implements ioctl() for reporting occupied space in bytes (not just blocks)
> > 
> >   The patches can be downloaded at:
> > ftp://atrey.karlin.mff.cuni.cz/pub/local/jack/quota/v2.5/2.5.15/
> > 
> >   Old quota tools should work with patches if you configure old quota interface
> > in '.config'. There are also quota utilities capable of communicating with new
> > generic interface. You can download them at:
> > 
> > http://www.sf.net/projects/linuxquota/
> > 
> > or you can checkout version from SourceForge CVS.
> > 
> >   Any comments & bugreports welcome.
> 
> What do you think of the following patches for kernel without
> quota support? We already have weak symbol for sys_quotactl(). 
  I know this but this file is also needed by XFS people so it isn't
so easy.  But as XFS stuff is not currently in kernel the change is
probably for now (I'll check whether there's not some hidden reference
I forgot about and submit patch to Linus).

								Honza

Re: Quota patches

[Nathan Scott]

hi there,

On Thu, May 23, 2002 at 04:46:28AM +0900, OGAWA Hirofumi wrote:
> What do you think of the following patches for kernel without
> quota support? We already have weak symbol for sys_quotactl(). 
> 
> --- linux-bk/fs/Makefile        Wed May 22 01:17:48 2002
> +++ linux-2.5.17/fs/Makefile    Thu May 23 03:23:30 2002
> @@ -15,7 +15,7 @@
>                 namei.o fcntl.o ioctl.o readdir.o select.o fifo.o locks.o \
>                 dcache.o inode.o attr.o bad_inode.o file.o iobuf.o dnotify.o \
>                 filesystems.o namespace.o seq_file.o xattr.o libfs.o \
> -               fs-writeback.o quota.o
> +               fs-writeback.o
> 
>  ifneq ($(CONFIG_NFSD),n)
>  ifneq ($(CONFIG_NFSD),)
> @@ -81,7 +81,7 @@
> 
>  obj-$(CONFIG_BINFMT_ELF)       += binfmt_elf.o
> 
> -obj-$(CONFIG_QUOTA) += dquot.o
> +obj-$(CONFIG_QUOTA) += dquot.o quota.o
>  obj-$(CONFIG_QFMT_V1) += quota_v1.o
>  obj-$(CONFIG_QFMT_V2) += quota_v2.o
> 

On Wed, May 22, 2002 at 09:57:00PM +0200, Jan Kara wrote:
>   I know this but this file is also needed by XFS people so it isn't
> so easy.  But as XFS stuff is not currently in kernel the change is


A change to the patch like this would make this work for us too:

replacing:
-obj-$(CONFIG_QUOTA) += dquot.o
+obj-$(CONFIG_QUOTA) += dquot.o quota.o

with just:
+obj-$(CONFIG_QUOTACTL) += quota.o


Needs to be fleshed out a little bit, of course - both CONFIG_QUOTA
and CONFIG_XFS_QUOTA (from the XFS patches) would need fs/Config.in
changes to add the new $CONFIG_QUOTACTL dependency, but that should
just about do the trick I think.

cheers.

-- 
Nathan

Re: Quota patches

[Nathan Scott]

On Thu, May 23, 2002 at 09:30:10AM +1000, Nathan Scott wrote:
> ... but that should just about do the trick I think.

How does the patch below look Jan?

-- 
Nathan


diff -Naur pristine-linux-2.5.17/fs/Config.in quota-linux-2.5.17/fs/Config.in
--- pristine-linux-2.5.17/fs/Config.in	Tue May 21 15:07:31 2002
+++ quota-linux-2.5.17/fs/Config.in	Thu May 23 10:33:32 2002
@@ -8,10 +8,13 @@
 dep_tristate '  Old quota format support' CONFIG_QFMT_V1 $CONFIG_QUOTA
 dep_tristate '  VFS v0 quota format support' CONFIG_QFMT_V2 $CONFIG_QUOTA
 dep_mbool '  Compatible quota interfaces' CONFIG_QIFACE_COMPAT $CONFIG_QUOTA
-if [ "$CONFIG_QUOTA" = "y" -a "$CONFIG_QIFACE_COMPAT" = "y" ]; then
-   choice '    Compatible quota interfaces' \
-	"Original	CONFIG_QIFACE_V1 \
-	 VFSv0		CONFIG_QIFACE_V2" Original
+if [ "$CONFIG_QUOTA" = "y" ]; then
+   define_bool CONFIG_QUOTACTL y
+   if [ "$CONFIG_QIFACE_COMPAT" = "y" ]; then
+       choice '    Compatible quota interfaces' \
+		"Original	CONFIG_QIFACE_V1 \
+		 VFSv0		CONFIG_QIFACE_V2" Original
+   fi
 fi
 tristate 'Kernel automounter support' CONFIG_AUTOFS_FS
 tristate 'Kernel automounter version 4 support (also supports v3)' CONFIG_AUTOFS4_FS
diff -Naur pristine-linux-2.5.17/fs/Makefile quota-linux-2.5.17/fs/Makefile
--- pristine-linux-2.5.17/fs/Makefile	Tue May 21 15:07:30 2002
+++ quota-linux-2.5.17/fs/Makefile	Thu May 23 10:43:00 2002
@@ -15,7 +15,7 @@
 		namei.o fcntl.o ioctl.o readdir.o select.o fifo.o locks.o \
 		dcache.o inode.o attr.o bad_inode.o file.o iobuf.o dnotify.o \
 		filesystems.o namespace.o seq_file.o xattr.o libfs.o \
-		fs-writeback.o quota.o
+		fs-writeback.o
 
 ifneq ($(CONFIG_NFSD),n)
 ifneq ($(CONFIG_NFSD),)
@@ -81,9 +81,10 @@
 
 obj-$(CONFIG_BINFMT_ELF)	+= binfmt_elf.o
 
-obj-$(CONFIG_QUOTA) += dquot.o
-obj-$(CONFIG_QFMT_V1) += quota_v1.o
-obj-$(CONFIG_QFMT_V2) += quota_v2.o
+obj-$(CONFIG_QUOTA)		+= dquot.o
+obj-$(CONFIG_QFMT_V1)		+= quota_v1.o
+obj-$(CONFIG_QFMT_V2)		+= quota_v2.o
+obj-$(CONFIG_QUOTACTL)		+= quota.o
 
 # persistent filesystems
 obj-y += $(join $(subdir-y),$(subdir-y:%=/%.o))

Re: Quota patches

[Alan Cox]

> On Thu, May 23, 2002 at 09:30:10AM +1000, Nathan Scott wrote:
> > ... but that should just about do the trick I think.
> 
> How does the patch below look Jan?

Doesn't let me select both ?

> +if [ "$CONFIG_QUOTA" = "y" ]; then
> +   define_bool CONFIG_QUOTACTL y
> +   if [ "$CONFIG_QIFACE_COMPAT" = "y" ]; then
> +       choice '    Compatible quota interfaces' \
> +		"Original	CONFIG_QIFACE_V1 \
> +		 VFSv0		CONFIG_QIFACE_V2" Original
> +   fi
>  fi

Re: Quota patches

[Nathan Scott]

hi Alan,

On Thu, May 23, 2002 at 02:56:43AM +0100, Alan Cox wrote:
> > On Thu, May 23, 2002 at 09:30:10AM +1000, Nathan Scott wrote:
> > > ... but that should just about do the trick I think.
> > 
> > How does the patch below look Jan?
> 
> Doesn't let me select both ?
> 
> > +if [ "$CONFIG_QUOTA" = "y" ]; then
> > +   define_bool CONFIG_QUOTACTL y
> > +   if [ "$CONFIG_QIFACE_COMPAT" = "y" ]; then
> > +       choice '    Compatible quota interfaces' \
> > +		"Original	CONFIG_QIFACE_V1 \
> > +		 VFSv0		CONFIG_QIFACE_V2" Original
> > +   fi
> >  fi
> 

I think that was Jan's intention (the patch I sent doesn't change
that aspect of things)...

-if [ "$CONFIG_QUOTA" = "y" -a "$CONFIG_QIFACE_COMPAT" = "y" ]; then
-   choice '    Compatible quota interfaces' \
-       "Original       CONFIG_QIFACE_V1 \
-        VFSv0          CONFIG_QIFACE_V2" Original


So all I've really added is the line:
+   define_bool CONFIG_QUOTACTL y


wrt having both, iirc there was overlap between the original ABI
and Jan's new 32bit one - I suspect thats why both together is
disallowed, but Jan will be able to give the definitive answers
there.

cheers.

-- 
Nathan

Re: Quota patches

[Jan Kara]

> > On Thu, May 23, 2002 at 09:30:10AM +1000, Nathan Scott wrote:
> > > ... but that should just about do the trick I think.
> > 
> > How does the patch below look Jan?
> 
> Doesn't let me select both ?
  Yes. Only one of compatible interfaces is allowed. The reason is
mainly due to QUOTAON. Both V1 and V2 interfaces used Q_QUOTAON
to turn quotas on (looking back I admit it was stupid but it happened).
So now we would have to recognize which quota file was actually given
to us and turn on proper quota format - and I dislike such magic in
kernel.. especially when it's not needed. When user has really old
quota tools (<=2.00) he will turn on V1 interface. If he has newer tools
(<3.05) he has to decide depending on format he wants to use... 3.05
and newer don't use this interface. Actually the easiest is always to
upgrade quota tools :) and this compatible interface mess is there
mainly due to possibility of backport into 2.4 kernels...
  
> > +if [ "$CONFIG_QUOTA" = "y" ]; then
> > +   define_bool CONFIG_QUOTACTL y
> > +   if [ "$CONFIG_QIFACE_COMPAT" = "y" ]; then
> > +       choice '    Compatible quota interfaces' \
> > +		"Original	CONFIG_QIFACE_V1 \
> > +		 VFSv0		CONFIG_QIFACE_V2" Original
> > +   fi
> >  fi

								Honza
-- 
Jan Kara <jack@suse.cz>
SuSE CR Labs

Re: Quota patches

[Linus Torvalds]

On Thu, 23 May 2002, Jan Kara wrote:
> > Doesn't let me select both ?
>   Yes. Only one of compatible interfaces is allowed. The reason is
> mainly due to QUOTAON. Both V1 and V2 interfaces used Q_QUOTAON
> to turn quotas on (looking back I admit it was stupid but it happened).
> So now we would have to recognize which quota file was actually given
> to us and turn on proper quota format - and I dislike such magic in
> kernel.. especially when it's not needed. When user has really old
> quota tools (<=2.00) he will turn on V1 interface. If he has newer tools
> (<3.05) he has to decide depending on format he wants to use...

This makes me pretty certain we just do not want to have the backwards-
compatibility layer in 2.5.x

Are there _any_ reasons to use the old stuff, if the fix is just to
upgrade to a newer quota tool?

		Linus

Re: Quota patches

[Nathan Scott]

hi Linus,

On Thu, May 23, 2002 at 10:03:50AM -0700, Linus Torvalds wrote:
> 
> On Thu, 23 May 2002, Jan Kara wrote:
> > ... . If he has newer tools
> > (<3.05) he has to decide depending on format he wants to use...
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^

> This makes me pretty certain we just do not want to have the backwards-
> compatibility layer in 2.5.x
> 
> Are there _any_ reasons to use the old stuff, if the fix is just to
> upgrade to a newer quota tool?

Moving to newer interfaces implies use of the new ondisk format
for the quota files (exclusively) - I'd imagine that's the main
reason behind providing a choice.  Whether or not that reason is
sufficently compelling though... dunno.  If one wanted to be able
to switch between booting either 2.4 (unpatched) and 2.5+, and
also maintain quota information on filestystems, then the choice
would be useful in that situation.

cheers.

-- 
Nathan

Re: Quota patches

[Linus Torvalds]

On Fri, 24 May 2002, Nathan Scott wrote:
>
> Moving to newer interfaces implies use of the new ondisk format
> for the quota files (exclusively)

Sure. But I'd assume that just running quotacheck should initialize that,
so..

		Linus

Re: Quota patches

[Jan Kara]

  Hi all,
  
> On Thu, May 23, 2002 at 10:03:50AM -0700, Linus Torvalds wrote:
> > 
> > On Thu, 23 May 2002, Jan Kara wrote:
> > > ... . If he has newer tools
> > > (<3.05) he has to decide depending on format he wants to use...
>                       ^^^^^^^^^^^^^^^^^^^^^^^^^^
> 
> > This makes me pretty certain we just do not want to have the backwards-
> > compatibility layer in 2.5.x
> > 
> > Are there _any_ reasons to use the old stuff, if the fix is just to
> > upgrade to a newer quota tool?
> 
> Moving to newer interfaces implies use of the new ondisk format
> for the quota files (exclusively) - I'd imagine that's the main
> reason behind providing a choice.  Whether or not that reason is
> sufficently compelling though... dunno.  If one wanted to be able
> to switch between booting either 2.4 (unpatched) and 2.5+, and
> also maintain quota information on filestystems, then the choice
> would be useful in that situation.
  Latest quota interface is able to handle both formats together
(structures passed throught Q_GETQUOTA, Q_SETQUOTA,... are independent
of quota format and Q_QUOTAON takes as an argument in 'id' the quota format
number). So if user wants to stay at old format he can...
  So I think Linus is right here that there's no real reason for keeping
compatibility code in 2.5... Linus, I'll send you the patch which kicks
out the compatibility stuff.

								Honza
-- 
Jan Kara <jack@suse.cz>
SuSE CR Labs

Re: Quota patches

[Martin Dalecki]

Uz.ytkownik Jan Kara napisa?:
>   Hi all,
>   
> 
>>On Thu, May 23, 2002 at 10:03:50AM -0700, Linus Torvalds wrote:
>>
>>>On Thu, 23 May 2002, Jan Kara wrote:
>>>
>>>>... . If he has newer tools
>>>>(<3.05) he has to decide depending on format he wants to use...
>>>
>>                      ^^^^^^^^^^^^^^^^^^^^^^^^^^
>>
>>
>>>This makes me pretty certain we just do not want to have the backwards-
>>>compatibility layer in 2.5.x
>>>
>>>Are there _any_ reasons to use the old stuff, if the fix is just to
>>>upgrade to a newer quota tool?
>>
>>Moving to newer interfaces implies use of the new ondisk format
>>for the quota files (exclusively) - I'd imagine that's the main
>>reason behind providing a choice.  Whether or not that reason is
>>sufficently compelling though... dunno.  If one wanted to be able
>>to switch between booting either 2.4 (unpatched) and 2.5+, and
>>also maintain quota information on filestystems, then the choice
>>would be useful in that situation.
> 
>   Latest quota interface is able to handle both formats together
> (structures passed throught Q_GETQUOTA, Q_SETQUOTA,... are independent
> of quota format and Q_QUOTAON takes as an argument in 'id' the quota format
> number). So if user wants to stay at old format he can...
>   So I think Linus is right here that there's no real reason for keeping
> compatibility code in 2.5... Linus, I'll send you the patch which kicks
> out the compatibility stuff.

As a side note:

If we can do it for quota - we could possible remove the
IPC_OLD variant away as well. It's looong overdue by now,
becouse the IPC_OLD was not standard conformant anyway.

I would be really really glad to do it iff ACK-ed.

Re: Quota patches

[Alan Cox]

> If we can do it for quota - we could possible remove the
> IPC_OLD variant away as well. It's looong overdue by now,
> becouse the IPC_OLD was not standard conformant anyway.
> 
> I would be really really glad to do it iff ACK-ed.

More code that takes almost no space, ensures old systems still work and 
old XFree86 still runs on new kernels. Why remove it ?

If you want to design a mathematically elegant and small ultra clean OS go
do it. Linux however has to work in the real world not in the happy clueless
world of pure mathematical elegance

Re: Quota patches

[Martin Dalecki]

Uz.ytkownik Alan Cox napisa?:
>>If we can do it for quota - we could possible remove the
>>IPC_OLD variant away as well. It's looong overdue by now,
>>becouse the IPC_OLD was not standard conformant anyway.
>>
>>I would be really really glad to do it iff ACK-ed.
> 
> 
> More code that takes almost no space, ensures old systems still work and 
> old XFree86 still runs on new kernels. Why remove it ?

It is an illusion to think that you can actually run *that old*
a.out binaries on a modern kernel I think.

BTW. (almost no space) * (many times) == huge number.

> If you want to design a mathematically elegant and small ultra clean OS go
> do it. Linux however has to work in the real world not in the happy clueless
> world of pure mathematical elegance

Re: Quota patches

[Christoph Hellwig]

On Fri, May 24, 2002 at 04:31:39PM +0200, Martin Dalecki wrote:
> It is an illusion to think that you can actually run *that old*
> a.out binaries on a modern kernel I think.

Of course you can.  Even the latest OpenLinux release (shipping 2.4.13-ac)
uses a libc4/a.out based installer fo space reasons.  Not to forget the
old quake1 binary from some redhat 4.x CD I run from time to time :)

Re: Quota patches

[Martin Dalecki]

Uz.ytkownik Christoph Hellwig napisa?:
> On Fri, May 24, 2002 at 04:31:39PM +0200, Martin Dalecki wrote:
> 
>>It is an illusion to think that you can actually run *that old*
>>a.out binaries on a modern kernel I think.
> 
> 
> Of course you can.  Even the latest OpenLinux release (shipping 2.4.13-ac)
> uses a libc4/a.out based installer fo space reasons.  Not to forget the
> old quake1 binary from some redhat 4.x CD I run from time to time :)

OK thanks for the *substantial* answer. That was the reason I was asking about.
Somehow this is of course surprising me of course.

Re: Quota patches

[Alan Cox]

> > Of course you can.  Even the latest OpenLinux release (shipping 2.4.13-ac)
> > uses a libc4/a.out based installer fo space reasons.  Not to forget the
> > old quake1 binary from some redhat 4.x CD I run from time to time :)
> 
> OK thanks for the *substantial* answer. That was the reason I was asking about.
> Somehow this is of course surprising me of course.

So why didn't you -test- the theory before suggesting it. It btw goes beyond
Libc4. Currently we have almost 100% compatibility back to libc 2.2.2. The
dated libc before that doesn't work because we dropped some very very
early obscure versions of a few syscalls.

Is it too much to ask that you go and look through the syscall tables of
old and new kernels ? 

Alan

Re: Quota patches

[Christoph Hellwig]

On Fri, May 24, 2002 at 05:12:05PM +0100, Alan Cox wrote:
> > > Of course you can.  Even the latest OpenLinux release (shipping 2.4.13-ac)
> > > uses a libc4/a.out based installer fo space reasons.  Not to forget the
> > > old quake1 binary from some redhat 4.x CD I run from time to time :)
> > 
> > OK thanks for the *substantial* answer. That was the reason I was asking about.
> > Somehow this is of course surprising me of course.
> 
> So why didn't you -test- the theory before suggesting it. It btw goes beyond
> Libc4. Currently we have almost 100% compatibility back to libc 2.2.2. The
> dated libc before that doesn't work because we dropped some very very
> early obscure versions of a few syscalls.
> 
> Is it too much to ask that you go and look through the syscall tables of
> old and new kernels ? 

For 2.5 I have some plans to make obsolete syscalls depend on CONFIG_COMPAT_*,
this allows to compile big and bloated kernel for compatiblity and smaller
kernels without that (e.g. for embedded devices).  And in fact we have quite
a loft of cruft that can go away for setups only having very modern userspace..

Re: Quota patches

[Martin Dalecki]

Uz.ytkownik Christoph Hellwig napisa?:
> On Fri, May 24, 2002 at 05:12:05PM +0100, Alan Cox wrote:
> 
>>>>Of course you can.  Even the latest OpenLinux release (shipping 2.4.13-ac)
>>>>uses a libc4/a.out based installer fo space reasons.  Not to forget the
>>>>old quake1 binary from some redhat 4.x CD I run from time to time :)
>>>
>>>OK thanks for the *substantial* answer. That was the reason I was asking about.
>>>Somehow this is of course surprising me of course.
>>
>>So why didn't you -test- the theory before suggesting it. It btw goes beyond
>>Libc4. Currently we have almost 100% compatibility back to libc 2.2.2. The
>>dated libc before that doesn't work because we dropped some very very
>>early obscure versions of a few syscalls.
>>
>>Is it too much to ask that you go and look through the syscall tables of
>>old and new kernels ? 
> 
> 
> For 2.5 I have some plans to make obsolete syscalls depend on CONFIG_COMPAT_*,
> this allows to compile big and bloated kernel for compatiblity and smaller
> kernels without that (e.g. for embedded devices).  And in fact we have quite
> a loft of cruft that can go away for setups only having very modern userspace..

I second this!

Re: Quota patches

[Alan Cox]

> For 2.5 I have some plans to make obsolete syscalls depend on CONFIG_COMPAT_*,
> this allows to compile big and bloated kernel for compatiblity and smaller
> kernels without that (e.g. for embedded devices).  And in fact we have quite
> a loft of cruft that can go away for setups only having very modern userspace..

I'm sure we do. For embedded you also want config options to remove the 
block layer and so forth. I'd been thinking about a set of options buried
in a config menu item like "Fine tune configuration for small/embedded devices" CONFIG_SMALL 

Re: Quota patches

[Alan Cox]

> It is an illusion to think that you can actually run *that old*
> a.out binaries on a modern kernel I think.
> 
> BTW. (almost no space) * (many times) == huge number.

Uninline copy*user and a few other things then. Fix the size of struct
page. Add a CONFIG_SMALL where users can pick to have very small hash tables
on older systems with little RAM.  Add the two extra common slab sizes 
people identified

This kind of stuff tends to save you hundreds of K yet nobody is doing that
before even worrying about some trivial 150 bytes.

Re: Quota patches

[David Woodhouse]

alan@lxorguk.ukuu.org.uk said:
>  Uninline copy*user and a few other things then. Fix the size of
> struct page. Add a CONFIG_SMALL where users can pick to have very
> small hash tables on older systems with little RAM.  Add the two extra
> common slab sizes  people identified

Don't forget CONFIG_BLK_DEV

--
dwmw2

Re: Quota patches

[Thunder from the hill]

Hi,

On Fri, 24 May 2002, Alan Cox wrote:
>       Add a CONFIG_SMALL where users can pick to have very small hash
> tables on older systems with little RAM.

On Fri, 24 May 2002, Alan Cox wrote:
> I'd been thinking about a set of options buried in a config menu item
> like "Fine tune configuration for small/embedded devices" CONFIG_SMALL

I'd prefer CONFIG_TIGHT_MEMORY. You never know what things in the world 
might be called small.

Regards,
Thunder
-- 
Was it a black who passed along in the sand?
Was it a white who left his footprints?
Was it an african? An indian?
Sand says, 'twas human.

Re: Quota patches

[Alan Cox]

> This makes me pretty certain we just do not want to have the backwards-
> compatibility layer in 2.5.x
> 
> Are there _any_ reasons to use the old stuff, if the fix is just to
> upgrade to a newer quota tool?

Most people use 2.4 with quota tools and 32bit uid quota already, so its
not much of a breakage at all. The 2.4 quota base code is unusable in the
real world so the problem got settled by the vendor trees

Re: Quota patches

[Jan Kara]

> On Thu, May 23, 2002 at 09:30:10AM +1000, Nathan Scott wrote:
> > ... but that should just about do the trick I think.
> 
> How does the patch below look Jan?
> 
  The patch is OK with me. Can you apply it Linus?

								Thanks
  									Honza

diff -Naur pristine-linux-2.5.17/fs/Config.in quota-linux-2.5.17/fs/Config.in
--- pristine-linux-2.5.17/fs/Config.in	Tue May 21 15:07:31 2002
+++ quota-linux-2.5.17/fs/Config.in	Thu May 23 10:33:32 2002
@@ -8,10 +8,13 @@
 dep_tristate '  Old quota format support' CONFIG_QFMT_V1 $CONFIG_QUOTA
 dep_tristate '  VFS v0 quota format support' CONFIG_QFMT_V2 $CONFIG_QUOTA
 dep_mbool '  Compatible quota interfaces' CONFIG_QIFACE_COMPAT $CONFIG_QUOTA
-if [ "$CONFIG_QUOTA" = "y" -a "$CONFIG_QIFACE_COMPAT" = "y" ]; then
-   choice '    Compatible quota interfaces' \
-	"Original	CONFIG_QIFACE_V1 \
-	 VFSv0		CONFIG_QIFACE_V2" Original
+if [ "$CONFIG_QUOTA" = "y" ]; then
+   define_bool CONFIG_QUOTACTL y
+   if [ "$CONFIG_QIFACE_COMPAT" = "y" ]; then
+       choice '    Compatible quota interfaces' \
+		"Original	CONFIG_QIFACE_V1 \
+		 VFSv0		CONFIG_QIFACE_V2" Original
+   fi
 fi
 tristate 'Kernel automounter support' CONFIG_AUTOFS_FS
 tristate 'Kernel automounter version 4 support (also supports v3)' CONFIG_AUTOFS4_FS
diff -Naur pristine-linux-2.5.17/fs/Makefile quota-linux-2.5.17/fs/Makefile
--- pristine-linux-2.5.17/fs/Makefile	Tue May 21 15:07:30 2002
+++ quota-linux-2.5.17/fs/Makefile	Thu May 23 10:43:00 2002
@@ -15,7 +15,7 @@
 		namei.o fcntl.o ioctl.o readdir.o select.o fifo.o locks.o \
 		dcache.o inode.o attr.o bad_inode.o file.o iobuf.o dnotify.o \
 		filesystems.o namespace.o seq_file.o xattr.o libfs.o \
-		fs-writeback.o quota.o
+		fs-writeback.o
 
 ifneq ($(CONFIG_NFSD),n)
 ifneq ($(CONFIG_NFSD),)
@@ -81,9 +81,10 @@
 
 obj-$(CONFIG_BINFMT_ELF)	+= binfmt_elf.o
 
-obj-$(CONFIG_QUOTA) += dquot.o
-obj-$(CONFIG_QFMT_V1) += quota_v1.o
-obj-$(CONFIG_QFMT_V2) += quota_v2.o
+obj-$(CONFIG_QUOTA)		+= dquot.o
+obj-$(CONFIG_QFMT_V1)		+= quota_v1.o
+obj-$(CONFIG_QFMT_V2)		+= quota_v2.o
+obj-$(CONFIG_QUOTACTL)		+= quota.o
 
 # persistent filesystems
 obj-y += $(join $(subdir-y),$(subdir-y:%=/%.o))