cgroup: errno can be clobbered by free()

[eriksjolund]

Inspired by @jnovy comment quote:
"errno can be clobbered by free()"
I searched for some more

% git grep -B1 -P 'crun_make_error.*errno'  | grep -A1 release
libcrun/cgroup.c-          crun_error_release (err);
libcrun/cgroup.c:          return crun_make_error (err, errno, "error when using statfs on `%s`", CGROUP_ROOT "/freezer");

In other words, here is another one:

crun/src/libcrun/cgroup.c

Lines 164 to 165 in 96c7f15

	crun_error_release (err);
	return crun_make_error (err, errno, "error when using statfs on `%s`", CGROUP_ROOT "/freezer");

[giuseppe]

would it be safer if we just change crun_error_release to restore errno while we adjust callers?

[eriksjolund]

That would lead to restoring errno in many places where it's not needed, but yes it would be safer.

To get a very rough estimations of the proportions, I ran these commands

$ git grep crun_error_release remotes/origin/main | wc -l
     242
$ git grep "saved_errno = " remotes/origin/main | wc -l
      18
$ git grep -B10 saved_errno remotes/origin/main | grep error_release | wc -l
       4

Very unscientific, but maybe the proportion is something like 4/242?

In any case, I'm fine with either way.

Side note:

There is also the variable error_saved that I didn't count.

$ git grep " = errno;"  remotes/origin/main  | grep -v saved_errno
remotes/origin/main:src/libcrun/linux.c:          int errno_saved = errno;

[giuseppe]

my suggestion was just to add it as a safety net (which will prevent future issues too), but in the meanwhile convert all the existing users to use the correct value.