On Fri, Aug 31, 2001 at 10:54:20PM +0400, Andrej Borsenkow wrote:
>
> The actual question is, why it tries to mknod at all.
Indeed. I guess I was being obtuse about that point. :-)
> Consider lvm.c and
> lvm-fs.c in -12 source tree:
>
> lvm.c:
> lvm_chr_ioctl()
> case VG_CREATE:
> /* create a VGDA, assume VG number is filled in */
> return lvm_do_vg_create(-1, arg);
> lvm_do_vg_create():
> /* create devfs and procfs entries */
> lvm_fs_create_vg(vg_ptr);
>
> and lvm-fs.c:
> lvm_fs_create():
> vg_devfs_handle[vg_ptr->vg_number] =
> devfs_mk_dir(0, vg_ptr->vg_name, NULL);
> ch_devfs_handle[vg_ptr->vg_number] = devfs_register(
> vg_devfs_handle[vg_ptr->vg_number] , "group",
> DEVFS_FL_DEFAULT, LVM_CHAR_MAJOR, vg_ptr->vg_number,
> S_IFCHR | S_IRUSR | S_IWUSR | S_IRGRP,
> &lvm_chr_fops, NULL);
Oh, I see it alright.
> i.e. creating group automatically creates /dev entries.
In the case of devfs, sure defvs_register() is the right thing to do,
but in the case of no devfs, mknod() would be, surely -- which seems to
be left up to the userspace tools rather than the kernel. (See
conclusion below)
> I understand that vgscan tries to make sure all needed specials are
> there.
That's what it would seem.
> But it is not needed in case of devfs, when all device entries
> are created automatically.
Indeed. It seems as though the userspace tools were built un-devfs
aware. Maybe the userspace tools can't be built for both devfs and
non-devfs systems. i.e. the choice between mknod()ing and not is a
compile time attribute and not a run-time one.
> vgscan should check for mounted devfs and refrain from creating /dev
> entries.
Yup.
However, all of this still does not address my empty
/dev/<volume group> directory. It is missing the actual device nodes
(group, and <lv>s).
This latter part might be due to my using the initrd however. I am of
thought that the /dev/<volume group> nodes are created within the
following order of operations:
1. mount initrd
2. mount devfs
3 load lvm module (which creates devfs nodes)
3. vgscan, vchange -a y
4. mount real root (on lvm partition)
5. mount devfs again (on real root)
6. unmount initrd
I think the lvm nodes are created successfully in step 3 but get lost
when devfs is mounted a second time in step 5.
I am at a loss as to how to correct this however as I don't seem to be
able to create nodes on the devfs. I guess it's time to get on over
to the lvm mailing list for some ideas.
Anyone here is welcome to make suggestions of course. Also,
confirmation of what I suspect is happening in steps 1-6 above would
be appreciated as well as it is only a theory so far.
b.
--
Brian J. Murrell
