2.6.0-test3-mm1 and rootflags

2.6.0-test3-mm1 and rootflags

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 756 bytes --]

OK.. I'm stumped..

While testing something, I tried to boot with 'rootflags=noatime', and
found the system wouldn't boot, as ext3, ext2, and reiserfs all failed to
recognize the option.  Looking at the code in fs/ext3/super.c:parse_options()
and init/do_mounts.c:root_data_setup(), it appears to be impossible
to set any of the filesystem-independent flags via rootflags, which explains
the special-case code for the 'ro' and 'rw' flags.  However, there doesn't
seem to be any way to pass nodev, noatime, nodiratime, or any of the other
flags.  (And yes, all 3 of those make sense in my environment - it's a laptop
and I don't need atime, and I use devfs so nodev on the root makes sense too).

Am I missing something?  Or in fact, is this an non-doable?

[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: 2.6.0-test3-mm1 and rootflags

[Andrew Morton]

Valdis.Kletnieks@vt.edu wrote:
>
> OK.. I'm stumped..
> 
> While testing something, I tried to boot with 'rootflags=noatime', and
> found the system wouldn't boot, as ext3, ext2, and reiserfs all failed to
> recognize the option.  Looking at the code in fs/ext3/super.c:parse_options()
> and init/do_mounts.c:root_data_setup(), it appears to be impossible
> to set any of the filesystem-independent flags via rootflags, which explains
> the special-case code for the 'ro' and 'rw' flags.  However, there doesn't
> seem to be any way to pass nodev, noatime, nodiratime, or any of the other
> flags.  (And yes, all 3 of those make sense in my environment - it's a laptop
> and I don't need atime, and I use devfs so nodev on the root makes sense too).
> 

The fs-independent options are parsed in user space by mount(8), and are
passed into the kernel as individual bits in a `flags' argument.

So we'd need a new `rootopts=0x0040' thingy to support this.  But given
that most things can be set after boot with `mount / -o remount,noatime',
it may not be necessary.

Re: 2.6.0-test3-mm1 and rootflags

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1098 bytes --]

On Tue, 12 Aug 2003 14:25:19 PDT, Andrew Morton said:

> The fs-independent options are parsed in user space by mount(8), and are
> passed into the kernel as individual bits in a `flags' argument.

OK.. that's what I *thought* the code was doing.. ;)

> So we'd need a new `rootopts=0x0040' thingy to support this.  But given
> that most things can be set after boot with `mount / -o remount,noatime',
> it may not be necessary.

Can be reset with remount: rdonly, sync, mandlock, atime, diratime.

Can't be reset: nosuid, nodev, noexec, dirsync, and a few others.  Of those,
nosuid and noexec seem pointless for a root file system (though they may make
sense if you have a *really* minimal / and separate /bin /sbin /etc filesystems
- I seem to remember one minimalist config trying to get a r/o / so /etc was a
separate mount...), so at least for my current application, nodev is the only
thing I can't get with remounting....

If there's a consensus that a 'rootopts=' would be a Good Thing, I'll write a
patch.  If not, I'll just fix the initial value of root_mountflags in init/do_mount.c ;)


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: 2.6.0-test3-mm1 and rootflags

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 572 bytes --]

On Tue, 12 Aug 2003 23:41:37 EDT, Valdis.Kletnieks@vt.edu said:

> If there's a consensus that a 'rootopts=' would be a Good Thing, I'll write a
> patch.  If not, I'll just fix the initial value of root_mountflags in init/do_mount.c ;)

OK.. I'm a dork. ;)

Turns out that root_mountflags (at least in *my* config) ends up applying to
the initrd root - so to do what I *wanted* to do I needed to fix the 'mount'
command in /linuxrc:

echo Mounting root filesystem
mount -o noatime,nodev --ro -t ext3 /dev/rootvg/root /sysroot
pivot_root /sysroot /sysroot/initrd

D'oh! :)

[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: 2.6.0-test3-mm1 and rootflags

[Greg Depoire--Ferrer]

Hello.

I configure my kernel to boot without an initrd, with "rw root=PARTUUID=...
rootflags=...". I would like the kernel to mount my root partition with
MS_NOATIME but it is currently not possible without using an initrd that
remounts the root partition with the new flags, as was described in this
thread.

Because the kernel already supports setting filesystem dependent flags, I think
it would make sense for it to also support setting filesystem independent flags.
Also, there is the issue of flags that can't be reset later by initrd (see the
previous messages).

In reply to this message, there is a patch that implements what was suggested,
adding a 'rootopts=' parameter that takes a bitmap value for the mount flags.

It's my first time using a mailing list. I'm not sure if you're supposed to
reply to very old messages like this one or create a new thread instead. Sorry
if I did something wrong.

[PATCH] init: add rootopts=%d for rootfs mount flags

[Greg Depoire--Ferrer]

Add the rootopts command line parameter to specify mount flags for the
root filesystem.

The kernel provides the rootflags parameter to specify filesystem
dependent options when mouting the root filesystem but did not provide
any way to set the filesystem independent flags.

This is not an issue on systems using an initrd because on these
systems, the rootflags parameter is passed to the /sbin/mount binary
which understands filesystem independent flags in the option string,
such as noatime (MS_NOATIME) or nodev (MS_NODEV).

Signed-off-by: Greg Depoire--Ferrer <greg.depoire@gmail.com>
---
 Documentation/admin-guide/kernel-parameters.txt |  2 ++
 init/do_mounts.c                                | 11 +++++++++++
 2 files changed, 13 insertions(+)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 26453f250683..01e1217583f5 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -4851,6 +4851,8 @@
 
 	rootfstype=	[KNL] Set root filesystem type
 
+	rootopts=	[KNL] Set root filesystem mount flags bitmap
+
 	rootwait	[KNL] Wait (indefinitely) for root device to show up.
 			Useful for devices that are detected asynchronously
 			(e.g. USB and MMC devices).
diff --git a/init/do_mounts.c b/init/do_mounts.c
index a78e44ee6adb..ebe51fc7cf86 100644
--- a/init/do_mounts.c
+++ b/init/do_mounts.c
@@ -326,6 +326,16 @@ static int __init root_data_setup(char *str)
 	return 1;
 }
 
+static int __init root_flags_setup(char *str)
+{
+	int flags;
+
+	if (kstrtoint(str, 0, &flags))
+		return 0;
+	root_mountflags |= flags;
+	return 1;
+}
+
 static char * __initdata root_fs_names;
 static int __init fs_names_setup(char *str)
 {
@@ -341,6 +351,7 @@ static int __init root_delay_setup(char *str)
 }
 
 __setup("rootflags=", root_data_setup);
+__setup("rootopts=", root_flags_setup);
 __setup("rootfstype=", fs_names_setup);
 __setup("rootdelay=", root_delay_setup);
 
-- 
2.31.1

Re: 2.6.0-test3-mm1 and rootflags

[Randy Dunlap]

On 6/29/21 10:20 AM, Greg Depoire--Ferrer wrote:
> Hello.
> 
> I configure my kernel to boot without an initrd, with "rw root=PARTUUID=...
> rootflags=...". I would like the kernel to mount my root partition with
> MS_NOATIME but it is currently not possible without using an initrd that
> remounts the root partition with the new flags, as was described in this
> thread.
> 
> Because the kernel already supports setting filesystem dependent flags, I think
> it would make sense for it to also support setting filesystem independent flags.
> Also, there is the issue of flags that can't be reset later by initrd (see the
> previous messages).
> 
> In reply to this message, there is a patch that implements what was suggested,
> adding a 'rootopts=' parameter that takes a bitmap value for the mount flags.
> 
> It's my first time using a mailing list. I'm not sure if you're supposed to
> reply to very old messages like this one or create a new thread instead. Sorry
> if I did something wrong.

Hi,

It would be better to start a new email thread and include a link to
the previous discussion. E.g.:

Reference: or Link: https://lore.kernel.org/lkml/200308121855.h7CIt6St002437@turing-police.cc.vt.edu/

and also Cc: any interested parties.


thanks.
-- 
~Randy

Re: 2.6.0-test3-mm1 and rootflags

["Andrey Borzenkov"]

> If there's a consensus that a 'rootopts=' would be a Good Thing, I'll write a
> patch.  If not, I'll just fix the initial value of root_mountflags > in init/do_mount.c ;)

Mandrake does it in initrd - if mkinitrd detects any non-standard
rootfs option in use it makes it be mounted from within initrd.

Drawback is you can't change them on the fly. Unfortunately parsing
rootfstype/rootflags in initrd is not trivial given size constraints.

-andrey