[PATCH 0/4] Allow mounting root on tmpfs for impersistence

> This patch series allows the root to be mounted on tmpfs, and ensures directories that are needed are present. Once the persisting directories needed are configured (with btrfs for instance), one can mount root on tmpfs. These persisting directories should be at least /var/log (not needed, but safety requirement), /var/lib, /var/guix, but they are application-dependent (if using NetworkManager, /etc/NetworkManager; if guix offloading daemon is present, /etc/guix; the same goes for Docker...).

>

> Nicolas Graves (4):

> file-systems: canonicalize-device-name: Ignore the "none" case.

> gnu: services: activation-script: Ensure /var/run existence.

> gnu: build: activate-current-system: Ensure directory existence.

> gnu: system: bootable-kernel-arguments: Ignore the "none" root case.

>

> gnu/build/activation.scm | 1 +

> gnu/build/file-systems.scm | 8 ++++----

> gnu/services.scm | 1 +

> gnu/system.scm | 21 ++++++++++++---------

> 4 files changed, 18 insertions(+), 13 deletions(-)

>

> ---

> gnu/build/file-systems.scm | 8 ++++----

> 1 file changed, 4 insertions(+), 4 deletions(-)

>

> diff --git a/gnu/build/file-systems.scm b/gnu/build/file-systems.scm

> index 36a59f5f5c..e39bd31c3b 100644

> --- a/gnu/build/file-systems.scm

> +++ b/gnu/build/file-systems.scm

> @@ -1021,8 +1021,8 @@ (define find-partition-by-luks-uuid

> 

> (define (canonicalize-device-spec spec)

> "Return the device name corresponding to SPEC, which can be a <uuid>, a

> -<file-system-label>, or a string (typically a /dev file name or an nfs-root

> -containing ':/')."

> +<file-system-label>, the string 'none' or another string (typically a /dev

> +file name or an nfs-root containing ':/')."

> ---

> gnu/services.scm | 1 +

> 1 file changed, 1 insertion(+)

>

> diff --git a/gnu/services.scm b/gnu/services.scm

> index eb9258977e..73235ac2b8 100644

> --- a/gnu/services.scm

> +++ b/gnu/services.scm

> @@ -674,6 +674,7 @@ (define (activation-script gexps)

> ;; Make sure the user accounting database exists. If it

> ;; does not exist, 'setutxent' does not create it and

> ;; thus there is no accounting at all.

> + (mkdir-p "/var/run")

> (close-port (open-file "/var/run/utmpx" "a0"))

> ---

> gnu/build/activation.scm | 1 +

> 1 file changed, 1 insertion(+)

>

> diff --git a/gnu/build/activation.scm b/gnu/build/activation.scm

> index eea2233563..759704ae03 100644

> --- a/gnu/build/activation.scm

> +++ b/gnu/build/activation.scm

> @@ -408,6 +408,7 @@ (define* (activate-current-system

>

> (format #t "making '~a' the current system...~%" system)

>

> + (mkdir-p "/run")

> + (if (string=? root "none")

> + '() ;; Ignore the case where the root is "none" (typically tmpfs).

> ---

> gnu/home/services/symlink-manager.scm | 5 +++++

> 1 file changed, 5 insertions(+)

>

> diff --git a/gnu/home/services/symlink-manager.scm b/gnu/home/services/symlink-manager.scm

> index e4c931fbee..d3c4d01db4 100644

> --- a/gnu/home/services/symlink-manager.scm

> +++ b/gnu/home/services/symlink-manager.scm

> @@ -157,6 +157,11 @@ (define (update-symlinks-script)

> #t

> (G_ "Skipping ~a (not an empty directory)... done\n")

> directory))

> + ((= EBUSY errno)

> + (format

> + #t

> + (G_ "Skipping ~a (dedicated device)... done\n")

> + directory))

> Nicolas Graves <ngraves@ngraves.fr> skribis:

>

>> ---

>> gnu/home/services/symlink-manager.scm | 5 +++++

>> 1 file changed, 5 insertions(+)

>>

>> diff --git a/gnu/home/services/symlink-manager.scm b/gnu/home/services/symlink-manager.scm

>> index e4c931fbee..d3c4d01db4 100644

>> --- a/gnu/home/services/symlink-manager.scm

>> +++ b/gnu/home/services/symlink-manager.scm

>> @@ -157,6 +157,11 @@ (define (update-symlinks-script)

>> #t

>> (G_ "Skipping ~a (not an empty directory)... done\n")

>> directory))

>> + ((= EBUSY errno)

>> + (format

>> + #t

>> + (G_ "Skipping ~a (dedicated device)... done\n")

>> + directory))

>

> How does that relate to the rest of the patch series? What does

> “dedicate device” mean here?

> Hi,

>

> (You forgot to Cc the bug in your reply, so our conversation went off

> the record…)

>

> Ludovic Courtès <ludo@gnu.org> skribis:

>

>> Nicolas Graves <ngraves@ngraves.fr> skribis:

>>

>>> ---

>>> gnu/home/services/symlink-manager.scm | 5 +++++

>>> 1 file changed, 5 insertions(+)

>>>

>>> diff --git a/gnu/home/services/symlink-manager.scm b/gnu/home/services/symlink-manager.scm

>>> index e4c931fbee..d3c4d01db4 100644

>>> --- a/gnu/home/services/symlink-manager.scm

>>> +++ b/gnu/home/services/symlink-manager.scm

>>> @@ -157,6 +157,11 @@ (define (update-symlinks-script)

>>> #t

>>> (G_ "Skipping ~a (not an empty directory)... done\n")

>>> directory))

>>> + ((= EBUSY errno)

>>> + (format

>>> + #t

>>> + (G_ "Skipping ~a (dedicated device)... done\n")

>>> + directory))

>>

>> How does that relate to the rest of the patch series? What does

>> “dedicate device” mean here?

>

> Your use case is for when ~/.local (for instance) is a separate mount

> point (an uncommon use case because mounting file systems on Linux

> usually requires root privileges).

>

> I’d suggested “dedicated device” → “underlying device is busy”, and a

> short comment above explaining the kind of situation where this can

> occur.

>

> Thanks,

> Ludo’.

> Hi Nicolas,

>

> Did you have a chance to work on a revised version of this patch set?

>

> https://issues.guix.gnu.org/65335

>

> Seems to me that little was missing.

> Should /var/run be removed (in the same commit) from ‘directives’ in

> (gnu build install)?

>

> Thanks in advance!

>

> Ludo’.

>

> Ludovic Courtès <ludo@gnu.org> skribis:

>

>> Hi,

>>

>> (You forgot to Cc the bug in your reply, so our conversation went off

>> the record…)

>>

>> Ludovic Courtès <ludo@gnu.org> skribis:

>>

>>> Nicolas Graves <ngraves@ngraves.fr> skribis:

>>>

>>>> ---

>>>> gnu/home/services/symlink-manager.scm | 5 +++++

>>>> 1 file changed, 5 insertions(+)

>>>>

>>>> diff --git a/gnu/home/services/symlink-manager.scm b/gnu/home/services/symlink-manager.scm

>>>> index e4c931fbee..d3c4d01db4 100644

>>>> --- a/gnu/home/services/symlink-manager.scm

>>>> +++ b/gnu/home/services/symlink-manager.scm

>>>> @@ -157,6 +157,11 @@ (define (update-symlinks-script)

>>>> #t

>>>> (G_ "Skipping ~a (not an empty directory)... done\n")

>>>> directory))

>>>> + ((= EBUSY errno)

>>>> + (format

>>>> + #t

>>>> + (G_ "Skipping ~a (dedicated device)... done\n")

>>>> + directory))

>>>

>>> How does that relate to the rest of the patch series? What does

>>> “dedicate device” mean here?

>>

>> Your use case is for when ~/.local (for instance) is a separate mount

>> point (an uncommon use case because mounting file systems on Linux

>> usually requires root privileges).

>>

>> I’d suggested “dedicated device” → “underlying device is busy”, and a

>> short comment above explaining the kind of situation where this can

>> occur.

>>

>> Thanks,

>> Ludo’.

> On 2023-10-05 15:07, Ludovic Courtès wrote:

>> Should /var/run be removed (in the same commit) from ‘directives’ in

>> (gnu build install)?

>

> Maybe you can help a bit if you have worked on this, so that I

> understand correctly if something should be done here or not.

> Hi,

>

> Nicolas Graves <ngraves@ngraves.fr> skribis:

>

>> On 2023-10-05 15:07, Ludovic Courtès wrote:

>

> [...]

>

>>> Should /var/run be removed (in the same commit) from ‘directives’ in

>>> (gnu build install)?

>>

>> Maybe you can help a bit if you have worked on this, so that I

>> understand correctly if something should be done here or not.

>

> The directives in (gnu build install) specify files and directories that

> should be created upfront when initializing a system—meaning when

> running ‘guix system init’, or when populating a VM image. The

> boot/activation code assumes that these files already exist when the

> system boots.

>

> If we’re now creating /var/run in an activation snippet, then it no

> longer needs to be here.

>

> When doing such changes, I would suggest to at least run:

>

> make check-system TESTS=basic

>

> to detect obvious mistakes, and perhaps run the “installed-os” test as

> well (it’s much more expensive, so run it once you’re essentially done).

>

> HTH!

>

> Ludo’.

>

>

>

> Just to tell that the v3 I sent a few weeks ago is indeed fully tested,

> and that commit messages are properly written.

> Should we also think about a blog post on impermanence like nix did?

> Hi Nicolas,

>

> Nicolas Graves <ngraves@ngraves.fr> skribis:

>

>> Just to tell that the v3 I sent a few weeks ago is indeed fully tested,

>> and that commit messages are properly written.

>

> Just pushed v3 as b1ec85533a63c10616d9260f90411ca6f362de00!

>

> I did tweak the commit messages but that’s because I’m too picky :-) and

> the changes were minor.

>

>> Should we also think about a blog post on impermanence like nix did?

>

> Yes, definitely!

>

> There could also be a section in the manual with a complete OS

> definition, to show how one can set up a volatile image.

>

> Last, I realize there’s no system tests. I’m not sure how to do that

> (because system tests already run on a tmpfs overlay) but it’d be nice

> to make sure “root=none” handling keeps working.

>

> Thoughts?

>

> Thanks for all the work!

>

> Ludo’.