Re: efivarfs: Bad directory entry when variable has / in name

[shea-yfkUTty7RcRWk0Htik3J/w@public.gmane.org]

On 2013-03-20 04:41, Matt Fleming wrote:
> On 03/19/2013 08:36 PM, shea-yfkUTty7RcRWk0Htik3J/w@public.gmane.org wrote:
>> Hello,
>>
>> When there is an EFI variable with a forward slash in its name,
>> /sys/firmware/efi/efivars contains a directory entry with a forward
>> slash, which of course causes all sorts of problems (e.g. EINVAL 
>> from
>> stat(2)). Off the top of my head, I can think of two ways to handle
>> this:
>>
>> 1. Simply skip such variables when making the sysfs entries
>> 2. Come up with an escaping scheme, e.g. "a single backslash is
>> actually a forward slash, two backslashes are actually a single
>> backslash" or some such.
>>
>> Thoughts?
>
> Right. Have you seen machines with such variable names in the wild?
>

No, I just thought of testing it when I saw that variable names only 
require null termination.

>
> There's nothing in the specification that prohibits the use of 
> slashes
> in variable names, so I'm not in favour of option 1 above and 
> mangling
> the variable name would likely cause too much confusion.
>
> The best solution I can come up with is to create directories for any
> variable names that contain "/" (which would also make it possible to
> build filesystem hierarchies with EFI variables). The user could 
> choose
> to ignore the actual layout of the filesystem since opening the 
> variable
> "adirectory/file" with,
>
> 	open("adirectory/file-<guid>");
>
> would work without it being necessary for the user to handle slashes 
> in
> variable names.
>
> What do you think?
>

What happens when I create a variable starting with a slash or ending 
with a slash, or something like "dir/../../../../root/private-file"? 
Also, how would you create such a variable from linux, mkdir followed by 
creating the file?

Looks like an escaping scheme (though of course my proposed one was 
braindead and wouldn't work) might be the best option here, I might spin 
up a patch this weekend.

>
> Presumably this is also an issue with the old EFI vars sysfs code?

Never used that directly myself (only through efibootmgr).

Cheers,
Shea Levy