bug#71352: branch master updated: services: nix: Mount Nix store read only.

bug#71352: branch master updated: services: nix: Mount Nix store read only.

[Maxim Cournoyer]

Hello,

guix-commits@gnu.org writes:

>     services: nix: Mount Nix store read only.
>     
>     * gnu/services/nix.scm (nix-shepherd-service): Add requirements.
>     (%nix-store-directory): New variable.
>     (nix-service-type): Add file-system-service-type extension.
>     
>     Change-Id: I18a5d58c92c1f2b5b6dcecc3d5b439cc15bf4e49

This commit unfortunately appears to introduce a regression where
reconfiguring a system with the read-only /nix/store causes the
following error:

--8<---------------cut here---------------start------------->8---
guix system: error: chown: Système de fichiers accessible en lecture seulement
--8<---------------cut here---------------end--------------->8---

With the accompanying strace output:

--8<---------------cut here---------------start------------->8---
20261 close(17)                         = 0
20261 chown("/nix/store", 0, 981)       = -1 EROFS (Système de fichiers accessible en lecture seulement)
20261 close(13)                         = 0
20261 write(2, "guix system: \33[1;31merror: \33[0m\33[1mchown\33[0m: Syst\303\250me de fichiers accessible en lecture seulement\n", 99) = 99
--8<---------------cut here---------------end--------------->8---

Are these chown still useful in the activation snippet?

--8<---------------cut here---------------start------------->8---
(define (nix-activation _)
  ;; Return the activation gexp.
  #~(begin
      (use-modules (guix build utils)
                   (srfi srfi-26))
      (for-each (cut mkdir-p <>) '("/nix/store" "/nix/var/log"
                                   "/nix/var/nix/gcroots/per-user"
                                   "/nix/var/nix/profiles/per-user"))
      (chown "/nix/store"
             (passwd:uid (getpw "root")) (group:gid (getpw "nixbld01")))
      (chmod "/nix/store" #o775)
      (for-each (cut chmod <> #o777) '("/nix/var/nix/profiles"
                                       "/nix/var/nix/profiles/per-user"))))
--8<---------------cut here---------------end--------------->8---

If they are useful only on the first time, perhaps we could catch the
exceptions for when it runs on an already read-only mounted /nix/store?

-- 
Thanks,
Maxim

bug#71352: branch master updated: services: nix: Mount Nix store read only.

[Oleg Pykhalov]

[-- Attachment #1: Type: text/plain, Size: 2526 bytes --]

Hello Maxim,

Thank you for your report.  Apologize for any inconvenience caused by
the unexpected breakage.

Maxim Cournoyer <maxim.cournoyer@gmail.com> writes:

> Hello,
>
> guix-commits@gnu.org writes:
>
>>     services: nix: Mount Nix store read only.
>>     
>>     * gnu/services/nix.scm (nix-shepherd-service): Add requirements.
>>     (%nix-store-directory): New variable.
>>     (nix-service-type): Add file-system-service-type extension.
>>     
>>     Change-Id: I18a5d58c92c1f2b5b6dcecc3d5b439cc15bf4e49
>
> This commit unfortunately appears to introduce a regression where
> reconfiguring a system with the read-only /nix/store causes the
> following error:
>
> guix system: error: chown: Système de fichiers accessible en lecture seulement
>
>
> With the accompanying strace output:
>
> 20261 close(17)                         = 0
> 20261 chown("/nix/store", 0, 981)       = -1 EROFS (Système de fichiers accessible en lecture seulement)
> 20261 close(13)                         = 0
> 20261 write(2, "guix system: \33[1;31merror: \33[0m\33[1mchown\33[0m: Syst\303\250me de fichiers accessible en lecture seulement\n", 99) = 99
>
>
> Are these chown still useful in the activation snippet?
>
> (define (nix-activation _)
>   ;; Return the activation gexp.
>   #~(begin
>       (use-modules (guix build utils)
>                    (srfi srfi-26))
>       (for-each (cut mkdir-p <>) '("/nix/store" "/nix/var/log"
>                                    "/nix/var/nix/gcroots/per-user"
>                                    "/nix/var/nix/profiles/per-user"))
>       (chown "/nix/store"
>              (passwd:uid (getpw "root")) (group:gid (getpw "nixbld01")))
>       (chmod "/nix/store" #o775)
>       (for-each (cut chmod <> #o777) '("/nix/var/nix/profiles"
>                                        "/nix/var/nix/profiles/per-user"))))
>
> If they are useful only on the first time, perhaps we could catch the
> exceptions for when it runs on an already read-only mounted /nix/store?

Indeed, it is a good idea.

A hotfix for the issue was discussed and implemented. It has already
been pushed to the master branch. The fix involves a simple
'file-exists?' check. You can find more details in the discussion at
https://debbugs.gnu.org/cgi/bugreport.cgi?bug=71320

What do you think is preferable in this scenario – catching exceptions
or sticking with '(unless (file-exists? ...))'?  Your thoughts on the
best approach here?


Regards,
Oleg.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 861 bytes --]

bug#71352: branch master updated: services: nix: Mount Nix store read only.

[Maxim Cournoyer]

Hi Oleg,

[...]

>> Are these chown still useful in the activation snippet?
>>
>> (define (nix-activation _)
>>   ;; Return the activation gexp.
>>   #~(begin
>>       (use-modules (guix build utils)
>>                    (srfi srfi-26))
>>       (for-each (cut mkdir-p <>) '("/nix/store" "/nix/var/log"
>>                                    "/nix/var/nix/gcroots/per-user"
>>                                    "/nix/var/nix/profiles/per-user"))
>>       (chown "/nix/store"
>>              (passwd:uid (getpw "root")) (group:gid (getpw "nixbld01")))
>>       (chmod "/nix/store" #o775)
>>       (for-each (cut chmod <> #o777) '("/nix/var/nix/profiles"
>>                                        "/nix/var/nix/profiles/per-user"))))
>>
>> If they are useful only on the first time, perhaps we could catch the
>> exceptions for when it runs on an already read-only mounted /nix/store?
>
> Indeed, it is a good idea.
>
> A hotfix for the issue was discussed and implemented. It has already
> been pushed to the master branch. The fix involves a simple
> 'file-exists?' check. You can find more details in the discussion at
> https://debbugs.gnu.org/cgi/bugreport.cgi?bug=71320
>
> What do you think is preferable in this scenario – catching exceptions
> or sticking with '(unless (file-exists? ...))'?  Your thoughts on the
> best approach here?

Exceptions are usually better than 'check then do' as they avoid the
TOCTTOU (time-of-check to time-of-use) class of bugs/vulnerabilities.

By the way, 'Reported-by:' is a fine git trailer to use :-).  I also use
'Fixes:' as a git trailer (trailer means they should be found at the
bottom of the commit message -- these can be parsed with the 'git
interpret-trailers' command)

-- 
Thanks,
Maxim

bug#71352: branch master updated: services: nix: Mount Nix store read only.

[Maxim Cournoyer]

Hi Oleg,

Maxim Cournoyer <maxim.cournoyer@gmail.com> writes:

> Hi Oleg,
>
> [...]
>
>>> Are these chown still useful in the activation snippet?
>>>
>>> (define (nix-activation _)
>>>   ;; Return the activation gexp.
>>>   #~(begin
>>>       (use-modules (guix build utils)
>>>                    (srfi srfi-26))
>>>       (for-each (cut mkdir-p <>) '("/nix/store" "/nix/var/log"
>>>                                    "/nix/var/nix/gcroots/per-user"
>>>                                    "/nix/var/nix/profiles/per-user"))
>>>       (chown "/nix/store"
>>>              (passwd:uid (getpw "root")) (group:gid (getpw "nixbld01")))
>>>       (chmod "/nix/store" #o775)
>>>       (for-each (cut chmod <> #o777) '("/nix/var/nix/profiles"
>>>                                        "/nix/var/nix/profiles/per-user"))))
>>>
>>> If they are useful only on the first time, perhaps we could catch the
>>> exceptions for when it runs on an already read-only mounted /nix/store?
>>
>> Indeed, it is a good idea.
>>
>> A hotfix for the issue was discussed and implemented. It has already
>> been pushed to the master branch. The fix involves a simple
>> 'file-exists?' check. You can find more details in the discussion at
>> https://debbugs.gnu.org/cgi/bugreport.cgi?bug=71320
>>
>> What do you think is preferable in this scenario – catching exceptions
>> or sticking with '(unless (file-exists? ...))'?  Your thoughts on the
>> best approach here?
>
> Exceptions are usually better than 'check then do' as they avoid the
> TOCTTOU (time-of-check to time-of-use) class of bugs/vulnerabilities.

I'm closing this for now; I'm satisfied that working order has been
restored :-).

-- 
Thanks,
Maxim