btrfs-tools has a script for initramfs which load the btrfs module
and does a "btrfs device scan".

I suggest to replace the "btrfs device scan" with a udev rule (the
one provided by the package udev, see Bug#772744) which uses
the udev builtin method.
Because the module may be inserted after that some devices appeared,
after the module insert,we have to be do a "udevadm trigger -s block"
to redoing a device rescan for btrfs.

The goal of this change is to armonize how the btrfs device scan is
done in intramfs to after the initramfs

Below two patch to highlight this change.

BR
G.Baroncelli
--- /usr/share/initramfs-tools/scripts/local-premount/btrfs	2014-11-22 15:01:04.000000000 +0100
+++ usr/share/initramfs-tools/scripts/local-premount/btrfs	2014-12-10 20:31:31.289746904 +0100
@@ -19,5 +19,5 @@
 if [ -x /sbin/btrfs ]
 then
 	modprobe btrfs
-	/sbin/btrfs device scan 2> /dev/null
+	udevadm trigger -s block
 fi

--- /usr/share/initramfs-tools/hooks/btrfs	2014-11-22 15:01:04.000000000 +0100
+++ usr/share/initramfs-tools/hooks/btrfs	2014-12-10 20:34:07.181751787 +0100
@@ -26,4 +26,13 @@
 	then
 	    copy_exec /sbin/fsck.btrfs /sbin
 	fi
+
+	mkdir -p $DESTDIR/lib/udev/rules.d/
+	for rules in 64-btrfs.rules; do
+	  if   [ -e /etc/udev/rules.d/$rules ]; then
+	    cp -p /etc/udev/rules.d/$rules $DESTDIR/lib/udev/rules.d/
+	  elif [ -e /lib/udev/rules.d/$rules ]; then
+	    cp -p /lib/udev/rules.d/$rules $DESTDIR/lib/udev/rules.d/
+	  fi
+	done
 fi

I disagree, udev in initramfs is optional and the premount script is
there precisely because at the moment udev based discovery is not
sufficient. Also it's not btrfs package/hook responsibility to call
udevadm trigger. On a running system this is only done after all local
filesystems have been mounted.

I have a separate bug report requesting to add an extra systemd unit
to call device scan before local filesystems have been mounted.

Also 64-btrfs.rules should not be installed by btrfs package, if udev
package ships it. Looking at the systemd source package, the
64-btrfs.rules udev rule does something different to the btrfs-tools:
btrfs device scan calls BTRFS_IOC_SCAN_DEV whilst systemd's rule calls
BTRFS_IOC_DEVICES_READY. I'm not too sure, but if ramdisk is not using
systemd and not using udev it really needs device scan.

Do you have a bug report against current integration? If not, I will
close this bug report as won't fix.

Regards,

Dimitri.

If udev is optional in initramfs, my request has to be rejected.
discovery is not sufficient ? Or are you referring about the
fact that udev is not mandatory in initramfs.

[...]

BR
G.Baroncelli

udev in initramfs is optional, moreover initramfs is optional on
debian and plenty of people run things without it.

mosty yes, refereing to the fact that udev is not mandatory in the
Debian's initramfs. There is also ongoing bug stating that udev based
discovery is not sufficient in the initramfsless system, i have not
investigated it completely.

I read something about that. The report stated that btrfs didn't honor the
device=/dev/<disk> mount options; In this case would be impossible to
mount a multi-device filesystem.
But I didn't understand the reason....

Goffredo