Merge ~sil2100/cloud-initramfs-tools:growroot-do-not-waitroot into cloud-initramfs-tools:master

[Łukasz Zemczak]

When performing growroot, we don't need to perform a wait-for-root call as we already waited for the whole of udev to settle before that.

When performing growroot, we don't need to perform a wait-for-root call as we already waited for the whole of udev to settle before that.

The reason I'm proposing this change is actually as a workaround to a weird problem I have encountered with growpart and udev (might be kernel related?). Basically the problem there is that after growpart the rootfs block device loses all the ID_* udev properties, and wait-for-root basically loops until timeout looking for ID_FS_TYPE. Yes, this is a real bug that we need to chase down, but I think that's a separate thing.

I don't think there is *any* merit in doing a `wait-for-root` here as it was. In the regular local initrd setup, wait-for-root is (if I understand correctly) used to wait for the root device to be fully available without having to wait for the whole of udev to settle (per the comment in scripts/local's local_device_setup() ). In case of growroot, we already do an `udevadm settle` right before the wait-for-root, so there's not much more we can do actually - the call is pointless. Sure, we fetch the FSTYPE as well, but it's not needed (and supported to be optional), as for all of our initrd we anyway use get_fstype() for that (which uses blkid) - and if no filesystem type is given to mount, mount defaults to using blkid for the FS identification.

So even though this is a workaround to get growroot working in this weird broken usecase, I think either way this call is redundant and doesn't make any sense to be there. Looks like it got copied over just-in-case anyway. Though, maybe I'm missing context and it's needed? If yes, please comment!

[Scott Moser (smoser)]

well, lots of things have changed since this code was written and it probably did take a safe approach even then. Generally speaking, one 'settle' or (wait-for-root) immediately after another is going to happen "real fast", so its not really wasteful.

In theory, we've worked out all the bugs in growpart at cloud-utils version 0.32, and there is no longer any need for *any* settle as growpart should not return until everything has been done. (see 6b98f49ffb2c7, 957ba4cff0, d99b2d7664d5).

that might be something to consider.

I won't approve or deny this, simply because I don't really have the time to think it all the way through. I'll let someone else do that.

Unmerged commits

c02d734... by Łukasz Zemczak on 2021-11-02

When performing growroot, we don't need to perform a wait-for-root call as we already waited for the whole of udev to settle before that.

diff --git a/growroot/scripts/local-bottom/growroot b/growroot/scripts/local-bottom/growroot
index b4efa24..4c27d66 100644
--- a/growroot/scripts/local-bottom/growroot
+++ b/growroot/scripts/local-bottom/growroot
@@ -98,12 +98,12 @@ fi
 # so that the root partition is available when we try and mount it.
 udevadm settle --timeout ${ROOTDELAY:-30}
 
-# this is taken from 'mountroot' function
-#   see /usr/share/initramfs-tools/scripts/local
-FSTYPE=$(wait-for-root "${ROOT}" ${ROOTDELAY:-30})
+# We don't need to do a call of wait-for-root here as we already waited for
+# the whole of udev to settle, nor do we need to know the FSTYPE - mount will
+# use blkid to guess it anyway.
 roflag="-r"
 [ "${readonly}" = "y" ] || roflag="-w"
-mount ${roflag} ${FSTYPE:+-t ${FSTYPE} }${ROOTFLAGS} ${ROOT} ${rootmnt} ||
+mount ${roflag} ${ROOTFLAGS} ${ROOT} ${rootmnt} ||
 	fail "failed to re-mount ${ROOT}. this is bad!"
 
 # write to /etc/grownroot-grown. most likely this wont work (readonly)