Re: [PATCH v2 1/2] EFI: avoid OOB config file reads

[Jan Beulich <jbeulich@xxxxxxxx>]

On 16.09.2026 17:05, Jan Beulich wrote:
> On 16.09.2026 16:37, George Dunlap wrote:
>> On Wed, Mar 25, 2026 at 2:24 PM Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>> @@ -878,6 +882,23 @@ static bool __init read_section(const EF
>>>
>>>      file->ptr = ptr;
>>>
>>> +    /* For cfg file, if necessary allocate space to put an extra NUL 
>>> there. */
>>> +    if ( file == &cfg && file->size && !iscntrl(file->str[file->size - 1]) 
>>> )
>>> +    {
>>> +        EFI_PHYSICAL_ADDRESS addr;
>>> +        EFI_STATUS ret = efi_bs->AllocatePages(AllocateMaxAddress,
>>> +                                               EfiLoaderData,
>>> +                                               PFN_UP(file->size + 1), 
>>> &addr);
>>
>> efiapi.h lists the Memory parameter as OUT, but that seems to be a
>> mistake; the UEFI spec [1] says this is IN OUT; and the other two
>> callers of AllocatePages in Xen initialize the value before passing it
>> in.  So we should probably fix both efiapi.h, and initialize this to
>> an appropriate value, rather than passing in stack rubble.
> 
> Oh, yes, we definitely need to init the variable. I'm less certain about
> efiapi.h, though, as that's an imported header. We'd need to check
> gnuefi, and import a possible update from there. I guess I'll leave that
> part to the maintainers...

Actually no. There's no need to use AllocateMaxAddress here, at which point
the need to init addr will go away.

Jan