[RFC][Patch] Allow not mounting a root fs

[RFC][Patch] Allow not mounting a root fs

[Bodo Eggert]

This patch adds the option to not mount another root filesystem 
by specifying root=initramfs.

TODO: Documentation
---

BTW: Is it possible to mount a tmpfs on / before extracting the cpio?


While I'm at it:

In init/do_mounts.c, mount_root(void):
ROOT_NFS: Is it desirable to use the floppy as a root device if nfs-root 
failed? If I tell my system to d A, i usually dislike it to do B instead.

Is the boot-from-floppy code here still usable, even though booting from 
floppy is no longer supported?



diff -X dontdiff -pruN 2.6.21.ori/init/do_mounts.c 2.6.21/init/do_mounts.c
--- 2.6.21.ori/init/do_mounts.c	2007-07-06 03:31:57.000000000 +0200
+++ 2.6.21/init/do_mounts.c	2007-07-06 03:27:33.000000000 +0200
@@ -427,6 +427,9 @@ void __init prepare_namespace(void)
 			mount_block_root(root_device_name, root_mountflags);
 			goto out;
 		}
+		if (!strncmp(root_device_name, "initramfs", 3)) {
+			goto out_nomount;
+		}
 		ROOT_DEV = name_to_dev_t(root_device_name);
 		if (strncmp(root_device_name, "/dev/", 5) == 0)
 			root_device_name += 5;
@@ -444,6 +447,7 @@ void __init prepare_namespace(void)
 out:
 	sys_mount(".", "/", NULL, MS_MOVE, NULL);
 	sys_chroot(".");
+out_nomount:
 	security_sb_post_mountroot();
 }
 
-- 
Top 100 things you don't want the sysadmin to say:
65. What do you mean /home was on that disk?  I umounted it!

Re: [RFC][Patch] Allow not mounting a root fs

[H. Peter Anvin]

Bodo Eggert wrote:
> This patch adds the option to not mount another root filesystem 
> by specifying root=initramfs.

Uhm, the kernel doesn't mount anything if you're using an initramfs.

> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?

Not in the stock kernel.  There have been some patches floating around
for that, I think.

	-hpa

Re: [RFC][Patch] Allow not mounting a root fs

[Bodo Eggert]

On Thu, 5 Jul 2007, H. Peter Anvin wrote:

> Bodo Eggert wrote:
> > This patch adds the option to not mount another root filesystem 
> > by specifying root=initramfs.
> 
> Uhm, the kernel doesn't mount anything if you're using an initramfs.

Yes, instead it panics trying to mount the non-existing "root=".
I've put the complete system into initrams, and I just want the kernel to
run /sbin/init.

> > BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
> 
> Not in the stock kernel.  There have been some patches floating around
> for that, I think.

Are they messy, or would it just be a case of calling sys_mount before 
unpacking the cpio?
-- 
The complexity of a weapon is inversely proportional to the IQ of the
weapon's operator.

Re: [RFC][Patch] Allow not mounting a root fs

[Dan Aloni]

On Fri, Jul 06, 2007 at 05:00:40PM +0200, Bodo Eggert wrote:
> On Thu, 5 Jul 2007, H. Peter Anvin wrote:
> 
> > Bodo Eggert wrote:
> > > This patch adds the option to not mount another root filesystem 
> > > by specifying root=initramfs.
> > 
> > Uhm, the kernel doesn't mount anything if you're using an initramfs.
> 
> Yes, instead it panics trying to mount the non-existing "root=".
> I've put the complete system into initrams, and I just want the kernel to
> run /sbin/init.

If I understood you correctly, you're trying to get the kernel 
to execute the "/sbin/init" file of your initramfs. However, "/init" 
is the sole default initramfs entry point. To fix this, you can pass 
rdinit=/sbin/init _or_ modify your initramfs image so that "/init" 
is the entrypoint.

Normally, if your initramfs has a /init script your kernel would 
skip the entire stage where it panics in the perpare_namespace() 
function for trying to mount "root=" by itself (see 
init/main.c:init()).

-- 
Dan Aloni
XIV LTD, http://www.xivstorage.com
da-x (at) monatomic.org, dan (at) xiv.co.il

Re: [RFC][Patch] Allow not mounting a root fs

[Andre Noll]

[-- Attachment #1: Type: text/plain, Size: 343 bytes --]

On 17:00, Bodo Eggert wrote:

> I've put the complete system into initrams, and I just want the kernel to
> run /sbin/init.

Did you try to exec /sbin/init as the last thing in the init script
of your initramfs? That worked for me some time ago.

Andre
-- 
The only person who always got his work done by Friday was Robinson Crusoe

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [RFC][Patch] Allow not mounting a root fs

[H. Peter Anvin]

Andre Noll wrote:
> On 17:00, Bodo Eggert wrote:
> 
>> I've put the complete system into initrams, and I just want the kernel to
>> run /sbin/init.
> 
> Did you try to exec /sbin/init as the last thing in the init script
> of your initramfs? That worked for me some time ago.
> 

Just add a symlink:

/init -> /sbin/init

	-hpa

Re: [RFC][Patch] Allow not mounting a root fs

[Jan Engelhardt]

On Jul 5 2007 19:08, H. Peter Anvin wrote:
>
>> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
>
>Not in the stock kernel.  There have been some patches floating around
>for that, I think.

What would it buy? rootfs is a tmpfs, is not it?


	Jan
-- 

Re: [RFC][Patch] Allow not mounting a root fs

[H. Peter Anvin]

Jan Engelhardt wrote:
> On Jul 5 2007 19:08, H. Peter Anvin wrote:
>>> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
>> Not in the stock kernel.  There have been some patches floating around
>> for that, I think.
> 
> What would it buy? rootfs is a tmpfs, is not it?

No, rootfs is ramfs.

	-hpa

Re: [RFC][Patch] Allow not mounting a root fs

[Jan Engelhardt]

On Jul 7 2007 13:56, H. Peter Anvin wrote:
>Jan Engelhardt wrote:
>> On Jul 5 2007 19:08, H. Peter Anvin wrote:
>>>> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
>>> Not in the stock kernel.  There have been some patches floating around
>>> for that, I think.
>> 
>> What would it buy? rootfs is a tmpfs, is not it?
>
>No, rootfs is ramfs.

Thanks for clarifying. Well, why is it a ramfs, and not tmpfs?
(I think I know the difference, but the question stands.
A system without any swap somewhat equals a ramfs, or?)


	Jan
-- 

Re: [RFC][Patch] Allow not mounting a root fs

[H. Peter Anvin]

Jan Engelhardt wrote:
> On Jul 7 2007 13:56, H. Peter Anvin wrote:
>> Jan Engelhardt wrote:
>>> On Jul 5 2007 19:08, H. Peter Anvin wrote:
>>>>> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
>>>> Not in the stock kernel.  There have been some patches floating around
>>>> for that, I think.
>>> What would it buy? rootfs is a tmpfs, is not it?
>> No, rootfs is ramfs.
> 
> Thanks for clarifying. Well, why is it a ramfs, and not tmpfs?
> (I think I know the difference, but the question stands.
> A system without any swap somewhat equals a ramfs, or?)

It gets initialized very early, and a lot of the setup needed for tmpfs
to work isn't ready yet.

	-hpa

Re: [RFC][Patch] Allow not mounting a root fs

[Rob Landley]

On Saturday 07 July 2007 5:17:31 pm H. Peter Anvin wrote:
> Jan Engelhardt wrote:
> > On Jul 7 2007 13:56, H. Peter Anvin wrote:
> >> Jan Engelhardt wrote:
> >>> On Jul 5 2007 19:08, H. Peter Anvin wrote:
> >>>>> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
> >>>>
> >>>> Not in the stock kernel.  There have been some patches floating around
> >>>> for that, I think.
> >>>
> >>> What would it buy? rootfs is a tmpfs, is not it?
> >>
> >> No, rootfs is ramfs.
> >
> > Thanks for clarifying. Well, why is it a ramfs, and not tmpfs?
> > (I think I know the difference, but the question stands.
> > A system without any swap somewhat equals a ramfs, or?)
>
> It gets initialized very early, and a lot of the setup needed for tmpfs
> to work isn't ready yet.

Out of curiosity, do you know which setup?  I know that it won't actually swap 
anything out yet because no swap is mounted, but I thought that mounting 
tmpfs and doing a swapon afterwards was ok...

> 	-hpa

Rob
-- 
"One of my most productive days was throwing away 1000 lines of code."
  - Ken Thompson.

Re: [RFC][Patch] Allow not mounting a root fs

[H. Peter Anvin]

Rob Landley wrote:
> 
> Out of curiosity, do you know which setup?  I know that it won't actually swap 
> anything out yet because no swap is mounted, but I thought that mounting 
> tmpfs and doing a swapon afterwards was ok...
> 

Nope, I just know that the "obvious" way to do it crashes the kernel
horribly.  It looks based on the patchsets that have flown about
recently as it's mostly a matter of initializing the icache.

	-hpa

Re: [RFC][Patch] Allow not mounting a root fs

[Roland M. Kruggel]

-- 
cu

Roland Kruggel  mailto: rk.liste at bbf7.de
System: Intel, Debian etch, 2.6.16.16, xfce4 KDE 3.5

Test vom script...

Re: [RFC][Patch] Allow not mounting a root fs

[Bodo Eggert]

On Sat, 7 Jul 2007, Jan Engelhardt wrote:
> On Jul 5 2007 19:08, H. Peter Anvin wrote:

> >> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
> >
> >Not in the stock kernel.  There have been some patches floating around
> >for that, I think.
> 
> What would it buy? rootfs is a tmpfs, is not it?

As far as I understand, it is a ramfs aka. non-swappable tmpfs without any 
limit and including a chance of DoSing the system.

Since I'm toying with root-on-initcpio, I'm looking for things that might 
make the task some easier.

Currently I'm thinking about changing the root=initramfs to root=tmpfs and
additionally mounting a tmpfs on / before unpacking the cpios. OTOH, this 
seems to be a little bit messy. OTOH again, in _that_ boot code, nobody
should complain :)

-- 
"Whoever said the pen is mightier than the sword obviously never
encountered automatic weapons."
-Gen. Douglas MacArthur

Re: [RFC][Patch] Allow not mounting a root fs

[Jan Engelhardt]

On Jul 7 2007 23:18, Bodo Eggert wrote:
>> >> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
>> >
>> >Not in the stock kernel.  There have been some patches floating around
>> >for that, I think.
>> 
>> What would it buy? rootfs is a tmpfs, is not it?
>
>As far as I understand, it is a ramfs aka. non-swappable tmpfs without any 
>limit and including a chance of DoSing the system.
>
>Since I'm toying with root-on-initcpio, I'm looking for things that might 
>make the task some easier.
>
>Currently I'm thinking about changing the root=initramfs to root=tmpfs and
>additionally mounting a tmpfs on / before unpacking the cpios. OTOH, this 
>seems to be a little bit messy. OTOH again, in _that_ boot code, nobody
>should complain :)

Come to speak of it, I think you can have it much easier by having the kernel
exporting the cpio image as a virtual file inside rootfs, so that you could
re-extract it inside a tmpfs. In other words:

mount -t tmpfs tmpfs /mnt;
cd /mnt;
cpio -diuv </initramfs.cpio;
#
# switch_root nukes rootfs, pivot_roots and execs some init
klibc_switch_root .; # or something like that

Of course this needs double the memory than directly mounting tmpfs as 
rootfs, but it's an idea too.


	Jan
-- 

Re: [RFC][Patch] Allow not mounting a root fs

[H. Peter Anvin]

Jan Engelhardt wrote:
> 
> Come to speak of it, I think you can have it much easier by having the kernel
> exporting the cpio image as a virtual file inside rootfs, so that you could
> re-extract it inside a tmpfs. In other words:
> 
> mount -t tmpfs tmpfs /mnt;
> cd /mnt;
> cpio -diuv </initramfs.cpio;
> #
> # switch_root nukes rootfs, pivot_roots and execs some init
> klibc_switch_root .; # or something like that
> 
> Of course this needs double the memory than directly mounting tmpfs as 
> rootfs, but it's an idea too.
> 

Well, we're doing exactly this for non-initramfs initrd, however, it
would definitely be cleaner just to overmount rootfs with a tmpfs before
extraction -- it's an absolutely trivial amount of code; the biggest
complexity would be spotting the kernel command line option to invoke it.

Yet another variant, which works on existing kernels, is to have a
nested cpio, where your rootfs consists of a trivial /init which mounts
tmpfs and extracts another tar- or cpioball.

	-hpa

Re: [RFC][Patch] Allow not mounting a root fs

[Willy Tarreau]

On Sat, Jul 07, 2007 at 11:18:59PM +0200, Bodo Eggert wrote:
> On Sat, 7 Jul 2007, Jan Engelhardt wrote:
> > On Jul 5 2007 19:08, H. Peter Anvin wrote:
> 
> > >> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
> > >
> > >Not in the stock kernel.  There have been some patches floating around
> > >for that, I think.
> > 
> > What would it buy? rootfs is a tmpfs, is not it?
> 
> As far as I understand, it is a ramfs aka. non-swappable tmpfs without any 
> limit and including a chance of DoSing the system.
> 
> Since I'm toying with root-on-initcpio, I'm looking for things that might 
> make the task some easier.
> 
> Currently I'm thinking about changing the root=initramfs to root=tmpfs and
> additionally mounting a tmpfs on / before unpacking the cpios. OTOH, this 
> seems to be a little bit messy. OTOH again, in _that_ boot code, nobody
> should complain :)

Well, maybe it would be time to assign limits to ramfs. I have a patch for
this in all my own 2.4 kernels that I use on appliances, network boot and
bootable CDs. It was a bit tricky to get the limits to work well in 2.4
because it was not easy to track which pages were used by whom. But I presume
that rmap in 2.6 should help a lot.

If someone is willing to give this a try, here's the 2.4 patch :

  http://linux.1wt.eu/kernel/patches-2.4.34-wt1/pool/2.4.32-ramfs-limits-5

Regards,
Willy
 

Re: [RFC][Patch] Allow not mounting a root fs

[Al Boldi]

Bodo Eggert wrote:
> On Sat, 7 Jul 2007, Jan Engelhardt wrote:
> > On Jul 5 2007 19:08, H. Peter Anvin wrote:
> > >> BTW: Is it possible to mount a tmpfs on / before extracting the cpio?
> > >
> > >Not in the stock kernel.  There have been some patches floating around
> > >for that, I think.
> > 
> > What would it buy? rootfs is a tmpfs, is not it?
> 
> As far as I understand, it is a ramfs aka. non-swappable tmpfs without any 
> limit and including a chance of DoSing the system.
> 
> Since I'm toying with root-on-initcpio, I'm looking for things that might 
> make the task some easier.
> 
> Currently I'm thinking about changing the root=initramfs to root=tmpfs and
> additionally mounting a tmpfs on / before unpacking the cpios. OTOH, this 
> seems to be a little bit messy. OTOH again, in _that_ boot code, nobody
> should complain :)

Try this "[PATCH] initramfs: Allow rootfs to use tmpfs instead of ramfs".

It's trivial and should be in mainline.


Thanks!

--
Al

Re: [RFC][Patch] Allow not mounting a root fs

[H. Peter Anvin]

Al Boldi wrote:
> 
> Try this "[PATCH] initramfs: Allow rootfs to use tmpfs instead of ramfs".
> 
> It's trivial and should be in mainline.
> 

Simple, yes, trivial, no.

It should be pushed to -mm before mainline to give it some smokeout time.

	-hpa

Re: [RFC][Patch] Allow not mounting a root fs

[Al Boldi]

H. Peter Anvin wrote:
> Al Boldi wrote:
> > Try this "[PATCH] initramfs: Allow rootfs to use tmpfs instead of
> > ramfs".
> >
> > It's trivial and should be in mainline.
>
> Simple, yes, trivial, no.
>
> It should be pushed to -mm before mainline to give it some smokeout time.

Great, maybe somebody can pick it up and adjust it for inclusion into -mm.


Thanks!

--
Al