bug#24659: 'mkdir' procedure is neither thread-safe nor safe (2.0.12)

bug#24659: 'mkdir' procedure is neither thread-safe nor safe (2.0.12)

[Ludovic Courtès]

Hello!

When the optional ‘mode’ argument to the ‘mkdir’ procedure is omitted,
umask(2) is used to find out what the current mask is and to compute the
mode argument to mkdir(2):

--8<---------------cut here---------------start------------->8---
  if (SCM_UNBNDP (mode))
    {
      mask = umask (0);
      umask (mask);
      STRING_SYSCALL (path, c_path, rv = mkdir (c_path, 0777 ^ mask));
    }
--8<---------------cut here---------------end--------------->8---

The problem is that this changes the process’ umask globally for a short
duration, during which other threads will get 0 as the mask…

Furthermore, AFAICS, the above logic is redundant with what the kernel
does anyway.  That is, in a single-threaded program,

  mask = umask (0);
  umask (mask);
  mkdir (file, 0777 ^ mask);

is equivalent to:

  mkdir (file, 0777);

Am I right that we should just remove these two ‘umask’ calls?

Ludo’.

PS: This code has been there since 1996 or earlier.

bug#24659: 'mkdir' procedure is neither thread-safe nor safe (2.0.12)

[tomas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Mon, Oct 10, 2016 at 06:39:47PM +0200, Ludovic Courtès wrote:
> Hello!

[...]

> Furthermore, AFAICS, the above logic is redundant with what the kernel
> does anyway.  That is, in a single-threaded program,
> 
>   mask = umask (0);
>   umask (mask);
>   mkdir (file, 0777 ^ mask);
> 
> is equivalent to:
> 
>   mkdir (file, 0777);
> 
> Am I right that we should just remove these two ‘umask’ calls?

According to umask(2), yes, you are right. Unless someone has been
trying to work around some platform-specific bug/idiosyncracy.

Regards
- -- t
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)

iEYEARECAAYFAlf73SEACgkQBcgs9XrR2kZehwCfQ+P2CMZt8doHe2tKi0u1Yc7f
RU0An0WY1oEq/b3vu/X1rNWjTP7ZxUsY
=7Jhs
-----END PGP SIGNATURE-----

bug#24659: 'mkdir' procedure is neither thread-safe nor safe (2.0.12)

[Ludovic Courtès]

Hi,

<tomas@tuxteam.de> skribis:

> On Mon, Oct 10, 2016 at 06:39:47PM +0200, Ludovic Courtès wrote:
>> Hello!
>
> [...]
>
>> Furthermore, AFAICS, the above logic is redundant with what the kernel
>> does anyway.  That is, in a single-threaded program,
>>
>>   mask = umask (0);
>>   umask (mask);
>>   mkdir (file, 0777 ^ mask);
>>
>> is equivalent to:
>>
>>   mkdir (file, 0777);
>>
>> Am I right that we should just remove these two ‘umask’ calls?
>
> According to umask(2), yes, you are right. Unless someone has been
> trying to work around some platform-specific bug/idiosyncracy.

I don’t think there was ever a good reason for this.

Fixed in 245608911698adb3472803856019bdd5670b6614.

Thanks for your feedback,
Ludo’.