bug#27242: Fail to load LUKS encrypted rootfs, attempts to open luks device before it's ready.

bug#27242: Fail to load LUKS encrypted rootfs, attempts to open luks device before it's ready.

[Adam Van Ymeren]

On my machine when booting GuixSD, it fails to find the rootfs from the
initramfs.  It errors out attempting to find the luks device node.
Shortly after it errors is when I see the kernel dmesg output of the
nvme0 device becomining ready.

It looks like we need to wait for the device node to become ready, or
poll/sleep a few times if we fail to locate the device.

Thanks!
-Adam

bug#27242: Fail to load LUKS encrypted rootfs, attempts to open luks device before it's ready.

[Mark H Weaver]

Adam Van Ymeren <adam@vany.ca> writes:

> On my machine when booting GuixSD, it fails to find the rootfs from the
> initramfs.  It errors out attempting to find the luks device node.
> Shortly after it errors is when I see the kernel dmesg output of the
> nvme0 device becomining ready.
>
> It looks like we need to wait for the device node to become ready, or
> poll/sleep a few times if we fail to locate the device.

I ran into the same problem at one point, and have applied the following
patch to my private branch of Guix.  Perhaps it should be applied to
master.

      Mark


From b7c1947d050a944937b983532d940491821fa75f Mon Sep 17 00:00:00 2001
From: Mark H Weaver <mhw@netris.org>
Date: Tue, 28 Mar 2017 05:52:28 -0400
Subject: [PATCH] DRAFT: Cope with delayed appearance of LUKS source.

---
 gnu/system/mapped-devices.scm | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/gnu/system/mapped-devices.scm b/gnu/system/mapped-devices.scm
index 2959802c9..3659ac2d6 100644
--- a/gnu/system/mapped-devices.scm
+++ b/gnu/system/mapped-devices.scm
@@ -101,6 +101,7 @@
     #~(let ((source #$source))
         ;; XXX: 'use-modules' should be at the top level.
         (use-modules (rnrs bytevectors)           ;bytevector?
+                     (srfi srfi-1)
                      ((gnu build file-systems)
                       #:select (find-partition-by-luks-uuid)))
 
@@ -114,7 +115,11 @@
                         ;; udev-populated /dev/disk/by-id directory but udev may
                         ;; be unavailable at the time we run this.
                         (if (bytevector? source)
-                            (or (find-partition-by-luks-uuid source)
+                            (or (let loop ((tries-left 10))
+                                  (and (positive? tries-left)
+                                       (or (find-partition-by-luks-uuid source)
+                                           (begin (sleep 1)
+                                                  (loop (- tries-left 1))))))
                                 (error "LUKS partition not found" source))
                             source)
 
-- 
2.13.0

bug#27242: Fail to load LUKS encrypted rootfs, attempts to open luks device before it's ready.

[Adam Van Ymeren]

Mark H Weaver <mhw@netris.org> writes:
>
> I ran into the same problem at one point, and have applied the following
> patch to my private branch of Guix.  Perhaps it should be applied to
> master.

Thanks, this patch works for me.  Something like this upstream would be
nice :)

This can probably be unified with the resolve function defined in build/file-systems.scm

bug#27242: Fail to load LUKS encrypted rootfs, attempts to open luks device before it's ready.

[Mark H Weaver]

Adam Van Ymeren <adam@vany.ca> writes:

> Mark H Weaver <mhw@netris.org> writes:
>>
>> I ran into the same problem at one point, and have applied the following
>> patch to my private branch of Guix.  Perhaps it should be applied to
>> master.
>
> Thanks, this patch works for me.  Something like this upstream would be
> nice :)

I pushed a similar patch to master as commit
f45878a80d412dd79c95e9274c3ee5dd24e1cec9.

> This can probably be unified with the resolve function defined in
> build/file-systems.scm

Thanks for pointing this out.  I don't currently have the spare time to
think about it, but feel free to propose a patch to rework this in a
nicer way :)

I'm closing this bug for now.

     Thanks!
       Mark