bug#22588: root: UNEXPECTED INCONSISTENCY; RUN fsck MANUALLY

[Ludovic Courtès]

Mark H Weaver <address@hidden> skribis:

> I've been bitten by this once before, and a user on #guix did as well.
> When there are problems in the root filesystem that fsck doesn't want to
> fix automatically, the user is dumped into a guile prompt where PATH is
> not set, and it's very inconvenient to run fsck manually.

AFAICS, ‘PATH’ is set in ‘base-initrd’ in (gnu system linux-initrd), and
‘check-file-system’ in (gnu build file-systems) indeed expects it to be
set.

> This is what I just suggested that the user type, with apologies:
>
>   (use-modules (ice-9 ftw) (srfi srfi-26))
>   (define dirs (scandir "/gnu/store" (cut string-suffix? "e2fsprogs-1.42.13" 
> <>)))
>   (define e2fsck (string-append "/gnu/store/" (car dirs) "/sbin/e2fsck"))
>   (system* e2fsck "/dev/XXX")
>
> Is there a better way?

I think one can run:

  (system* "fsck.ext4" "/foo/bar")

What about changing the message to explicitly mention this command?

> Speaking from personal experience, it's very painful to do anything
> non-trivial in that REPL.  Even just adding readline would help a lot.

The statically-linked Guile in the initrd lacks Readline support.  We
could maybe work around that, but the initrd would become much larger.

> Maybe we should at least set PATH to include the available /bin and
> /sbin directories before entering the REPL.

Done, AFAICS.

> We should probably also handle errors from fsck specially.

Currently there’s no Bash in the initrd.  Should we add one?  Our
‘bash-static’ package takes 1.4 MiB (I don’t think we can make it
smaller.)

Thanks,
Ludo’.