mapfs-user — Discussion and help for users of MapFS

[mapfs-user] Mapfs module insertion problem

[Kiran P. - N. Chennai. <ki...@hc...>]

Hi,

=20

I have compiled the mapfs with linux-2.6.12 kernel and able to get =
mapfs.ko
module.

But when I try to insert the module I am getting some kallsyms lookup
message.

Kindly suggest me what more modules do I need to include.

Kallsyms feature is enabled.

=20

[root@initiator1 mapfs-1.0-0.1035]# insmod mapfs.ko

insmod: error inserting 'mapfs.ko': -1 Invalid parameters

=20

[root@initiator1 mapfs-1.0-0.1035]# dmesg

Address for "kallsyms_lookup_name" must be specified.

=20

Thanks and Regards,

Kiran.

=20

=20


Disclaimer:

This message and any attachment(s) contained here are information that =
is confidential, proprietary to HCL Technologies and its customers, =
privileged or otherwise protected by law. The information is solely =
intended for the individual or the entity it is addressed to. If you are =
not the intended recipient of this message, you are not authorized to =
read, forward, print, retain, copy or disseminate this message or any =
part of it. If you have received this e-mail in error, please notify the =
sender immediately by return e-mail and delete it from your computer.

[mapfs-user] Unionfs

[Sanghyeon S. <sa...@gm...>]

How does MapFS compare to Unionfs? Two look very similar to me.

Seo Sanghyeon

Re: [mapfs-user] how to make a read-only fs writable?

[Albrecht G. <alb...@un...>]

On Thu, Nov 03, 2005 at 04:29:58AM -0800, Nathaniel Stahl wrote:
> Albrecht Gebhardt wrote:
>=20
> >On Thu, Nov 03, 2005 at 02:58:04AM -0800, Nathaniel Stahl wrote:
> >=20
> ...
>=20
> >ok, I'll try it.=20
> >
> >Does it mean that the rw-tree will not automatically see files which
> >appear after the mapfs mount in the ro-tree? (this can not occur for a
> >cdrom but e.g. is possible for a readonly nfs mount)
> >
> >Then I have to re-run the populate-view script?
> >
> ...
>=20
> rw-tree should see new files that are created in the View.  It will also=
=20
> notice if a file is removed through the View directly and not through=20
> MapFS...  but you might get an error the first time you try to access=20
> the newly removed file.
>=20
> Don't recreate all the View files... you'll invalidate the current MapFS=
=20
> files because the View inodes will change.  Just create files that are=20
> new and not in the View already.  If a file is currently open, MapFS=20
> will properly hold the reference open until it closes... but dcache=20
> entries may remain until the kernel decides to shrink the dcache or you=
=20
> try a lookup on a name that was unlinked and recreated.
>=20
> warning: If a file in MapFS refers to a file on a writable filesystem,=20
> MapFS will remove it if you remove the file through MapFS (it thinks=20
> you're manipulating a potentially old Overlay).
>=20
> NFS support is still a little shakey.  It's what I'm working on=20
> polishing now.  NFS is kinda the naughtiest filesystem I've found so far=
=20
> in terms of what it does under the covers.  It should mostly work.

Currently I'm using overlay filesystems for readonly cdrom and network
installations (cloop, NFS and mostly AFS).  I use it only for the /etc
subtree, currently I use mulafs after I tried unionfs (with many
kernel Oopses) and minifo_fs (still with some Ooopses) without
success. mulafs is more or less stable (only one unnecessary init
script crahses after a kernel Oops). But now I'll give a try to mapfs
=2E..

Albrecht

PS:
I prepared an ubuntu/debian package for mapfs, available at
http://ubuntu.uni-klu.ac.at/

The module packages are for our local patched version of breezy kernel
2.6.12, but the mapfs-source package can be used with module-assistent
to build binary packages for other kernels (I guess only 2.6 will
work, I modified the Makefile only for this case). Currently I have
also problems with th k7 version, it links the _read_lock,
_read_unlock, _spin_unlock and _spin_lock symbols which are only
exported in my k7-smp and 686-smp kernels. The 686 module works fine.

To rebuild all from scratch grab the source package components
mapfs-source* from=20
http://ubuntu.uni-klu.ac.at/ubuntu.uniklu/dists/hoary/uniklu/source/
and use dpkg-source, debuild and module-assistant.

>=20
> -Nate
>=20

--=20
// Albrecht Gebhardt          Tel.: (++43 463) 2700/3118
// Institut fuer Mathematik   Fax : (++43 463) 2700/3198
// Universitaet Klagenfurt    mailto:alb...@un...
// Universitaetsstr. 65       http://www.math.uni-klu.ac.at/~agebhard
// A-9020 Klagenfurt, Austria
// GPG PK: http://www.math.uni-klu.ac.at/~agebhard/agebhard.asc
// GPG FP: F46F 656E E83C 9323 CE30  FF8F 9DBA D1A3 B55A 78A6

Re: [mapfs-user] how to make a read-only fs writable?

[Nathaniel S. <nr...@le...>]

Albrecht Gebhardt wrote:

>On Thu, Nov 03, 2005 at 02:58:04AM -0800, Nathaniel Stahl wrote:
>  
>
>>Almost right.  You need to populate the View so that it refers to the 
>>files in the read/only filesystem.  For files in the View: directories 
>>represent themselves, data files contain a path to the host file that 
>>backs them (can be newline terminated or not) relative to the lookup 
>>root (root=) mount parameter, other special files (symbolic links, 
>>fifos, device nodes) just represent themselves.
>>
>>
>>    
>>
<snip>

>ok, I'll try it. 
>
>Does it mean that the rw-tree will not automatically see files which
>appear after the mapfs mount in the ro-tree? (this can not occur for a
>cdrom but e.g. is possible for a readonly nfs mount)
>
>Then I have to re-run the populate-view script?
>
>Thanks for the quick answer
>
>Albrecht
>
>
>  
>

rw-tree should see new files that are created in the View.  It will also 
notice if a file is removed through the View directly and not through 
MapFS...  but you might get an error the first time you try to access 
the newly removed file.

Don't recreate all the View files... you'll invalidate the current MapFS 
files because the View inodes will change.  Just create files that are 
new and not in the View already.  If a file is currently open, MapFS 
will properly hold the reference open until it closes... but dcache 
entries may remain until the kernel decides to shrink the dcache or you 
try a lookup on a name that was unlinked and recreated.

warning: If a file in MapFS refers to a file on a writable filesystem, 
MapFS will remove it if you remove the file through MapFS (it thinks 
you're manipulating a potentially old Overlay).

NFS support is still a little shakey.  It's what I'm working on 
polishing now.  NFS is kinda the naughtiest filesystem I've found so far 
in terms of what it does under the covers.  It should mostly work.

-Nate

[mapfs-user] how to make a read-only fs writable? (resent)

[Albrecht G. <alb...@un...>]

the mailing-list archiver did not like my copy-pasted script snippet,
so here a little bit modified again for the archive only:

Hi,

reading the announcement I tought it should be possible to make
a readonly mounted path (eg. part of a mounted cdrom) writable by adding
a COW layer on top of it (as other layer filesystems like unionfs,
minifo_fs, mulafs do ...).

I tried the following

cd /tmp
dd if=3D/dev/zero of=3D/tmp/ro.img bs=3D1M count=3D2
mkfs.ext2 -F /tmp/ro.img
mkdir /tmp/ro-tree
mount -o loop /tmp/ro.img /tmp/ro-tree
echo "test" > /tmp/ro-tree/test.txt=20
mkdir /tmp/ro-tree/dir
cp /tmp/ro-tree/test.txt /tmp/ro-tree/dir
sync
umount /tmp/ro-tree
mount -o loop,ro /tmp/ro.img /tmp/ro-tree

Now I have i readonly filesystem mounted at /tmp/ro-tree with some files
in it:

# ls /tmp/ro-tree
dir  lost+found  test.txt
# touch /tmp/ro-tree/xxx
touch: cannot touch `/tmp/ro-tree/xxx': Read-only file system

Now I want to mount this path with mapfs to /tmp/rw-tree where I would
like to see the files from /tmp/ro-tree an would like to be able to
modify/add/delete files in the /tmp/rw-tree (virtually including the
files from /tmp/ro-tree).

The rw directories:

mkdir /tmp/rw-tree
mkdir -p /tmp/mapfs/view
mkdir -p /tmp/mapfs/overlay

I tried

mount -t mapfs -o \
root=3D/tmp/ro-tree,\
view=3D/tmp/mapfs/view,\
overlay=3D/tmp/mapfs/overlay,\
cow_base=3D/overlay none /tmp/rw-tree

Does not work, the test.txt files from /tmp/ro-path are not visible
in /tmp/rw-tree (Tough I can create and edit files in /tmp/rw-tree
resulting in new files in the view and overlay location)

# ls -l /tmp/rw-tree
total 0


I even tried to move the (rw) view and overlay directories with mount
--bind into the ro-tree:

umount /tmp/rw-tree
umount /tmp/ro-tree
mount -o loop /tmp/ro.img /tmp/ro-tree
mkdir /tmp/ro-tree/mapfs
sync
umount /tmp/ro-tree
mount -o loop,ro /tmp/ro.img /tmp/ro-tree
mount --bind /tmp/mapfs /tmp/ro-tree/mapfs

and now:

mount -t mapfs -o \
root=3D/tmp/ro-tree,\
view=3D/tmp/ro-tree/mapfs/view,\
overlay=3D/tmp/ro-tree/mapfs/overlay,\
cow_base=3D/mapfs/overlay none /tmp/rw-tree

Still /tmp/rw-tree is empty.

Do I miss something?

Thanks

Albrecht


--=20
// Albrecht Gebhardt          Tel.: (++43 463) 2700/3118
// Institut fuer Mathematik   Fax : (++43 463) 2700/3198
// Universitaet Klagenfurt    mailto:alb...@un...
// Universitaetsstr. 65       http://www.math.uni-klu.ac.at/~agebhard
// A-9020 Klagenfurt, Austria
// GPG PK: http://www.math.uni-klu.ac.at/~agebhard/agebhard.asc
// GPG FP: F46F 656E E83C 9323 CE30  FF8F 9DBA D1A3 B55A 78A6

Re: [mapfs-user] how to make a read-only fs writable?

[Albrecht G. <alb...@un...>]

On Thu, Nov 03, 2005 at 02:58:04AM -0800, Nathaniel Stahl wrote:
>=20
> Almost right.  You need to populate the View so that it refers to the=20
> files in the read/only filesystem.  For files in the View: directories=20
> represent themselves, data files contain a path to the host file that=20
> backs them (can be newline terminated or not) relative to the lookup=20
> root (root=3D) mount parameter, other special files (symbolic links,=20
> fifos, device nodes) just represent themselves.
>=20
> export VIEW=3D/path/to/mapfs/view
> export ROBASE=3D/path/to/robase
> export RELROBASE=3D/path/to/robase/from/mapfs/lookup/root
> cd $ROBASE
> find . -print0 | xargs -0 --replace=3DFOO sh -c "
> if [ -h FOO ]; then
>        cp -dp FOO $VIEW/FOO;
> elif [ -d FOO ]; then
>        mkdir $VIEW/FOO;
>        chown --reference=3DFOO $VIEW/FOO;
>        chmod --reference=3DFOO $VIEW/FOO;
> elif [ -f FOO ]; then
>        echo "$RELROBASE/FOO" > $VIEW/FOO;
>        chown --reference=3DFOO $VIEW/FOO;
>        chmod --reference=3DFOO $VIEW/FOO;
> else
>        cp -dp FOO $VIEW/FOO;
> fi"
>=20

ok, I'll try it.=20

Does it mean that the rw-tree will not automatically see files which
appear after the mapfs mount in the ro-tree? (this can not occur for a
cdrom but e.g. is possible for a readonly nfs mount)

Then I have to re-run the populate-view script?

Thanks for the quick answer

Albrecht


--=20
// Albrecht Gebhardt          Tel.: (++43 463) 2700/3118
// Institut fuer Mathematik   Fax : (++43 463) 2700/3198
// Universitaet Klagenfurt    mailto:alb...@un...
// Universitaetsstr. 65       http://www.math.uni-klu.ac.at/~agebhard
// A-9020 Klagenfurt, Austria
// GPG PK: http://www.math.uni-klu.ac.at/~agebhard/agebhard.asc
// GPG FP: F46F 656E E83C 9323 CE30  FF8F 9DBA D1A3 B55A 78A6

Re: [mapfs-user] how to make a read-only fs writable?

[Nathaniel S. <nr...@le...>]

Almost right.  You need to populate the View so that it refers to the 
files in the read/only filesystem.  For files in the View: directories 
represent themselves, data files contain a path to the host file that 
backs them (can be newline terminated or not) relative to the lookup 
root (root=) mount parameter, other special files (symbolic links, 
fifos, device nodes) just represent themselves.

export VIEW=/path/to/mapfs/view
export ROBASE=/path/to/robase
export RELROBASE=/path/to/robase/from/mapfs/lookup/root
cd $ROBASE
find . -print0 | xargs -0 --replace=FOO sh -c "
if [ -h FOO ]; then
        cp -dp FOO $VIEW/FOO;
elif [ -d FOO ]; then
        mkdir $VIEW/FOO;
        chown --reference=FOO $VIEW/FOO;
        chmod --reference=FOO $VIEW/FOO;
elif [ -f FOO ]; then
        echo "$RELROBASE/FOO" > $VIEW/FOO;
        chown --reference=FOO $VIEW/FOO;
        chmod --reference=FOO $VIEW/FOO;
else
        cp -dp FOO $VIEW/FOO;
fi"

Albrecht Gebhardt wrote:

>Hi,
>
>reading the announcement I tought it should be possible to make
>a readonly mounted path (eg. part of a mounted cdrom) writable by adding
>a COW layer on top of it (as other layer filesystems like unionfs,
>minifo_fs, mulafs do ...).
>
>I tried the following
>
>cd /tmp
>dd if=/dev/zero of=/tmp/ro.img bs=1M count=2
>mkfs.ext2 -F /tmp/ro.img
>mkdir /tmp/ro-tree
>mount -o loop /tmp/ro.img /tmp/ro-tree
>cat > /tmp/ro-tree/test.txt <<EOT
>test
>EOT
>mkdir /tmp/ro-tree/dir
>cp /tmp/ro-tree/test.txt /tmp/ro-tree/dir
>sync
>umount /tmp/ro-tree
>mount -o loop,ro /tmp/ro.img /tmp/ro-tree
>
>Now I have i readonly filesystem mounted at /tmp/ro-tree with some files
>in it:
>
># ls /tmp/ro-tree
>dir  lost+found  test.txt
># touch /tmp/ro-tree/xxx
>touch: cannot touch `/tmp/ro-tree/xxx': Read-only file system
>
>Now I want to mount this path with mapfs to /tmp/rw-tree where I would
>like to see the files from /tmp/ro-tree an would like to be able to
>modify/add/delete files in the /tmp/rw-tree (virtually including the
>files from /tmp/ro-tree).
>
>The rw directories:
>
>mkdir /tmp/rw-tree
>mkdir -p /tmp/mapfs/view
>mkdir -p /tmp/mapfs/overlay
>
>I tried
>
>mount -t mapfs -o \
>root=/tmp/ro-tree,\
>view=/tmp/mapfs/view,\
>overlay=/tmp/mapfs/overlay,\
>cow_base=/overlay none /tmp/rw-tree
>
>Does not work, the test.txt files from /tmp/ro-path are not visible
>in /tmp/rw-tree (Tough I can create and edit files in /tmp/rw-tree
>resulting in new files in the view and overlay location)
>
># ls -l /tmp/rw-tree
>total 0
>
>
>  
>

Right... the View is empty, and it defines what appears in the mounted 
MapFS filesystem (in this case /tmp/rw-tree).  The MapFS Overlay 
directory and any other filesystems you are going to want to map files 
from into the MapFS view need to share a common parent so MapFS lookups 
(which originate from the MapFS lookup root) can succeed for all of 
them.  cow_base should be set appropriately if the lookup root isn't /.

So, if the MapFS lookup root was /tmp (root=/tmp), then cow_base should 
be /mapfs/overlay in your above command.

export VIEW=/tmp/mapfs/view
export ROBASE=/tmp/ro-tree
export RELROBASE=/ro-tree
cd $ROBASE

find ... (from above)

mount -t mapfs \
-o root=/tmp,view=/tmp/mapfs/view,\
overlay=/tmp/mapfs/overlay,\
cow_base=/mapfs/overlay \
none /tmp/rw-tree

>I even tried to move the (rw) view and overlay directories with mount
>--bind into the ro-tree:
>
>umount /tmp/rw-tree
>umount /tmp/ro-tree
>mount -o loop /tmp/ro.img /tmp/ro-tree
>mkdir /tmp/ro-tree/mapfs
>sync
>umount /tmp/ro-tree
>mount -o loop,ro /tmp/ro.img /tmp/ro-tree
>mount --bind /tmp/mapfs /tmp/ro-tree/mapfs
>
>and now:
>
>mount -t mapfs -o \
>root=/tmp/ro-tree,\
>view=/tmp/ro-tree/mapfs/view,\
>overlay=/tmp/ro-tree/mapfs/overlay,\
>cow_base=/mapfs/overlay none /tmp/rw-tree
>
>Still /tmp/rw-tree is empty.
>
>Do I miss something?
>
>Thanks
>
>Albrecht
>
>
>  
>

[mapfs-user] how to make a read-only fs writable?

[Albrecht G. <alb...@un...>]

Hi,

reading the announcement I tought it should be possible to make
a readonly mounted path (eg. part of a mounted cdrom) writable by adding
a COW layer on top of it (as other layer filesystems like unionfs,
minifo_fs, mulafs do ...).

I tried the following

cd /tmp
dd if=3D/dev/zero of=3D/tmp/ro.img bs=3D1M count=3D2
mkfs.ext2 -F /tmp/ro.img
mkdir /tmp/ro-tree
mount -o loop /tmp/ro.img /tmp/ro-tree
cat > /tmp/ro-tree/test.txt <<EOT
test
EOT
mkdir /tmp/ro-tree/dir
cp /tmp/ro-tree/test.txt /tmp/ro-tree/dir
sync
umount /tmp/ro-tree
mount -o loop,ro /tmp/ro.img /tmp/ro-tree

Now I have i readonly filesystem mounted at /tmp/ro-tree with some files
in it:

# ls /tmp/ro-tree
dir  lost+found  test.txt
# touch /tmp/ro-tree/xxx
touch: cannot touch `/tmp/ro-tree/xxx': Read-only file system

Now I want to mount this path with mapfs to /tmp/rw-tree where I would
like to see the files from /tmp/ro-tree an would like to be able to
modify/add/delete files in the /tmp/rw-tree (virtually including the
files from /tmp/ro-tree).

The rw directories:

mkdir /tmp/rw-tree
mkdir -p /tmp/mapfs/view
mkdir -p /tmp/mapfs/overlay

I tried

mount -t mapfs -o \
root=3D/tmp/ro-tree,\
view=3D/tmp/mapfs/view,\
overlay=3D/tmp/mapfs/overlay,\
cow_base=3D/overlay none /tmp/rw-tree

Does not work, the test.txt files from /tmp/ro-path are not visible
in /tmp/rw-tree (Tough I can create and edit files in /tmp/rw-tree
resulting in new files in the view and overlay location)

# ls -l /tmp/rw-tree
total 0


I even tried to move the (rw) view and overlay directories with mount
--bind into the ro-tree:

umount /tmp/rw-tree
umount /tmp/ro-tree
mount -o loop /tmp/ro.img /tmp/ro-tree
mkdir /tmp/ro-tree/mapfs
sync
umount /tmp/ro-tree
mount -o loop,ro /tmp/ro.img /tmp/ro-tree
mount --bind /tmp/mapfs /tmp/ro-tree/mapfs

and now:

mount -t mapfs -o \
root=3D/tmp/ro-tree,\
view=3D/tmp/ro-tree/mapfs/view,\
overlay=3D/tmp/ro-tree/mapfs/overlay,\
cow_base=3D/mapfs/overlay none /tmp/rw-tree

Still /tmp/rw-tree is empty.

Do I miss something?

Thanks

Albrecht


--=20
// Albrecht Gebhardt          Tel.: (++43 463) 2700/3118
// Institut fuer Mathematik   Fax : (++43 463) 2700/3198
// Universitaet Klagenfurt    mailto:alb...@un...
// Universitaetsstr. 65       http://www.math.uni-klu.ac.at/~agebhard
// A-9020 Klagenfurt, Austria
// GPG PK: http://www.math.uni-klu.ac.at/~agebhard/agebhard.asc
// GPG FP: F46F 656E E83C 9323 CE30  FF8F 9DBA D1A3 B55A 78A6

[mapfs-user] simplified module loading

[Albrecht G. <alb...@un...>]

Hi,

with this oneliner (make sure it goes really all in one long line)

mapfsargs=$(for ksym in $(strings /lib/modules/$(uname -r)/fs/mapfs.ko |
awk '/from System.map/ {match($0,"^parm=([^:]+):",a); print a[1];}'); do
sym=${ksym#ksym_}; addr=$(awk '/ '$sym'$/ {print
$1}' /boot/System.map-$(uname -r)); echo " $ksym=0x$addr "; done)

is it possible to load mapfs with correct address parameters
automatically:

modprobe mapfs $mapfsargs

best regards

Albrecht