On Tue, Nov 26, 2024 at 04:35:13PM +0100, Roger Price wrote: > On Tue, 26 Nov 2024, to...@tuxteam.de wrote: > > > On Tue, Nov 26, 2024 at 03:53:28PM +0100, Roger Price wrote: > > > "except via mdadm" : exactly the point I would like to make. mdadm needs > > > to > > > be able to address the individual underlying devices. Only /dev/sdxn > > > style > > > addressing can do this, not duplicate UUIDs, or duplicate LABELs. > > > > If those we are talking about are file system UUIDs/LABELs (and all evidence > > supports that), they *are not* duplicate. They are *the same*. Change "one" > > and *poof* the "other" will change along. Because they are on the file > > system, > > which is sitting astride your MD devices. > > My understanding is that a Linux file system is a hierachical structure > starting with the root directory (/) which organises the directories and > files. The files are stored on various devices which have identities such > as /dev/sdxn, UUID or LABEL. These identities are for the devices, not > parts of the file system.
File system is indeed ambiguous here. In this context it was meant as
the data structure existing in a block device organizing its files.
For example, in my case:
tomas@caliban:~$ lsblk
NAME MAJ:MIN RM SIZE RO TYPE MOUNTPOINTS
sda 8:0 0 1.8T 0 disk
├─sda1 8:1 0 512M 0 part /boot/efi
├─sda2 8:2 0 488M 0 part /boot
└─sda3 8:3 0 1.8T 0 part
└─sda3_crypt 254:0 0 1.8T 0 crypt
├─caliban--vg-root 254:1 0 23.3G 0 lvm /
├─caliban--vg-home 254:2 0 1.6T 0 lvm /home
├─caliban--vg-var 254:3 0 160G 0 lvm /var
└─caliban--vg-tmp 254:4 0 40.3G 0 lvm /tmp
all of /dev/sda1, /dev/sda2, /dev/mapper/caliban--vg-root (which is
actually /dev/dm-1, but let's pretend) and so on have a file system
on them; most of them actually ext4, except /dev/sda2, which is ext2,
and /dev/sda1, which is vfat (as efi, it has to).
Thanks to mount, their file system trees are grafted together into
one single big file system tree (what you rightfully call "the file
system" above).
> The file system and the underlying devices are separate notions, which is
> why I don't understand your phrase "file system UUIDs/LABELs".
Each of those file systems has a slot for a label. Ext has also a slot for
an UUID, which gets generated afresh whenever you mkfs.ext. ISTR that the
FATs don't have that, but I'm too lazy to look it up now.
Here [1] is the ext4 superblock structure: at offsets 0x68 and 0x78 you
can find the places for the UUID and the name, aka label.
Cheers
[1] https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html
--
t
signature.asc
Description: PGP signature
