[uml-user] how can i do a _real_ check of the rootfs?

[<skoehler@up...>]

hi,

it seems that the rootfs (a reiserfs) is damanged. I thought that i just 
have to create the file /forcefsck so that /etc/init.d/checkroot will 
run fsck on the filesystem, but the command that is executed is "fsck -C 
-a -f", but that command doesn't check the filesystem if it is a 
reiserfs. fsck.reiserfs skipps all checks as it seems, since the 
filesystem has been marked as cleanly unmounted.

The problem is, that i cannot remount the rootfs read-only after the 
system has been booted completly. So i have the following questions:

- how can i tell gentoo to do a real complete check of the rootfs at 
startup (where the rootfs should be read-only mounted)
- how can i do that via ssh on a running system
- which paramters can i pass when booting gentoo, so that it boots into 
maintainance mode where the rootfs is still mounted read-only?

Thx
   Sven

Re: [uml-user] how can i do a _real_ check of the rootfs?

[<skoehler@up...>]

whoops, that mail was ment for the gentoo user mailinglist.

Re: [uml-user] how can i do a _real_ check of the rootfs?

[Blaisorblade <blaisorblade@ya...>]

On Thursday 13 January 2005 04:28, Sven K=F6hler wrote:
> hi,
>
> it seems that the rootfs (a reiserfs) is damanged. I thought that i just
> have to create the file /forcefsck so that /etc/init.d/checkroot will
> run fsck on the filesystem, but the command that is executed is "fsck -C
> -a -f", but that command doesn't check the filesystem if it is a
> reiserfs. fsck.reiserfs skipps all checks as it seems, since the
> filesystem has been marked as cleanly unmounted.
>
> The problem is, that i cannot remount the rootfs read-only after the
> system has been booted completly. So i have the following questions:
>
> - how can i tell gentoo to do a real complete check of the rootfs at
> startup (where the rootfs should be read-only mounted)
> - how can i do that via ssh on a running system
> - which paramters can i pass when booting gentoo, so that it boots into
> maintainance mode where the rootfs is still mounted read-only?
Adding "1" to the command line makes it boot in single-user mode (I think I=
've=20
tested it), like for any distro (the kernel does not handle it, so it passe=
s=20
it to init, which uses it, since it's a number, as runlevel selection). And=
=20
in single-user mode you *can* remount the root_fs as read-only (in some=20
distros it's kept read-only, IIRC).

And "telinit 1" as root *turns* the system to single user mode (where ssh=20
*does* not run).

Btw, why don't you do the check on the host? Real probleems or will to do i=
t=20
properly?
=2D-=20
Paolo Giarrusso, aka Blaisorblade
Linux registered user n. 292729
http://www.user-mode-linux.org/~blaisorblade

Re: [uml-user] how can i do a _real_ check of the rootfs?

[<skoehler@up...>]

> Btw, why don't you do the check on the host? Real probleems or will to do it 
> properly?

Thx for the answer, but i just klicked the wrong "newsgroup" in my 
mozilla. The mail was ment to go to gentoo-user mailinglist.

I was talking about a real system without UML, and the problem is, that 
even with "mount -f -o remount,ro /" fsck.reiserfs still think that the 
rootfs is mounted writable. I don't know why. mount shows "ro", but that 
might not mean that much.

I would do the fsck.reiserfs remotely via ssh if i could, but obviously 
i'm to stupid, or it just won't work. It seems that i should jump into 
my car tomorrow, and check the fs while i'm sitting in front of the 
computer.

Re: [uml-user] how can i do a _real_ check of the rootfs?

[Blaisorblade <blaisorblade@ya...>]

On Thursday 13 January 2005 04:50, Sven K=F6hler wrote:
> > Btw, why don't you do the check on the host? Real probleems or will to =
do
> > it properly?
>
> Thx for the answer, but i just klicked the wrong "newsgroup" in my
> mozilla. The mail was ment to go to gentoo-user mailinglist.
About /forcefsck, maybe it ignores it if /halt or something like that is=20
absent....
> I was talking about a real system without UML, and the problem is, that
> even with "mount -f -o remount,ro /" fsck.reiserfs still think that the
> rootfs is mounted writable. I don't know why. mount shows "ro", but that
> might not mean that much.
Check inside /proc/mounts, it's reliable (rootfs is always rw, you care to =
the=20
reiserfs mount) - mount cannot update /etc/mtab since it would require=20
writing to the rootfs. Or maybe /etc/mtab still says it's rw...

Seriously - have you checked the "check" field in /etc/fstab to make sure i=
t's=20
"1"? I have played with that flag sometimes....

Also, what you say about fsck.reiserfs is strange... while for ext3, if the=
 fs=20
is clean, no checking at all is done, for reiserfs a light check (replaying=
=20
the journal and walking the tree) is always done on newer versions, (Mandra=
ke=20
9.1 didn't do it, 10.0 does the check, which is annoying).
> I would do the fsck.reiserfs remotely via ssh if i could, but obviously
> i'm to stupid, or it just won't work.
ssh won't like a read-only root fs, I think... at least, normally it record=
s=20
the login, his logs... (Maybe it can be made to behave, I can't exclude it).
> It seems that i should jump into=20
> my car tomorrow, and check the fs while i'm sitting in front of the
> computer.

=2D-=20
Paolo Giarrusso, aka Blaisorblade
Linux registered user n. 292729
http://www.user-mode-linux.org/~blaisorblade

Re: [uml-user] how can i do a _real_ check of the rootfs?

[<skoehler@up...>]

>>I was talking about a real system without UML, and the problem is, that
>>even with "mount -f -o remount,ro /" fsck.reiserfs still think that the
>>rootfs is mounted writable. I don't know why. mount shows "ro", but that
>>might not mean that much.
> 
> Check inside /proc/mounts, it's reliable (rootfs is always rw, you care to the 
> reiserfs mount) - mount cannot update /etc/mtab since it would require 
> writing to the rootfs. Or maybe /etc/mtab still says it's rw...

Problem solved: i stopped all running services, and suddenly a normal 
"mount -o remount,ro /" worked, and fsck.reiserfs is working too now.
I tried the same thing yesterday, and it didn't work - strange.

> ssh won't like a read-only root fs, I think... at least, normally it records 
> the login, his logs... (Maybe it can be made to behave, I can't exclude it).

i still can login. i guess all the information is logged to the console 
now, since the syslog daemon is stopped.

Re: [uml-user] how can i do a _real_ check of the rootfs?

[<skoehler@up...>]

>>>Btw, why don't you do the check on the host? Real probleems or will to do
>>>it properly?
>>
>>Thx for the answer, but i just klicked the wrong "newsgroup" in my
>>mozilla. The mail was ment to go to gentoo-user mailinglist.
> 
> About /forcefsck, maybe it ignores it if /halt or something like that is 
> absent....

No, it isn't ignored. Gentoo prints a warning on the screen, that a 
check has been forced, so the file is still present when Gentoo boots.

>>I was talking about a real system without UML, and the problem is, that
>>even with "mount -f -o remount,ro /" fsck.reiserfs still think that the
>>rootfs is mounted writable. I don't know why. mount shows "ro", but that
>>might not mean that much.
> 
> Check inside /proc/mounts, it's reliable (rootfs is always rw, you care to the 
> reiserfs mount) - mount cannot update /etc/mtab since it would require 
> writing to the rootfs. Or maybe /etc/mtab still says it's rw...

all possible. i will check that.

> Seriously - have you checked the "check" field in /etc/fstab to make sure it's 
> "1"? I have played with that flag sometimes....

yes, it's "1", but the "check" field is ignored by /etc/init.d/checkroot 
anyway.

> Also, what you say about fsck.reiserfs is strange... while for ext3, if the fs 
> is clean, no checking at all is done, for reiserfs a light check (replaying 
> the journal and walking the tree) is always done on newer versions, (Mandrake 
> 9.1 didn't do it, 10.0 does the check, which is annoying).

fsck.reiserfs always performs that light check, but i'd like to force a 
full check on the next boot, and there doesn't seem to a proper way to 
do that. Gentoo's /etc/init.d/checkroot does a "fsck -C -a -f /" if the 
file /forcefsck is present, but fsck.reiserfs still does its light 
check. I guess fsck.ext2/ext3 would do a full check.

>>I would do the fsck.reiserfs remotely via ssh if i could, but obviously
>>i'm to stupid, or it just won't work.
> 
> ssh won't like a read-only root fs, I think... at least, normally it records 
> the login, his logs... (Maybe it can be made to behave, I can't exclude it).

good point, so a check on the next boot would be my choice, but how?