Re: [PATCH v2 5/7] efi/capsule: Prepare for loading images with security header

[Ard Biesheuvel]

On 28 March 2017 at 17:18, Jan Kiszka <jan.kiszka@xxxxxxxxxxx> wrote:
> On 2017-03-28 17:52, Ard Biesheuvel wrote:
>> On 28 March 2017 at 16:43, Jan Kiszka <jan.kiszka@xxxxxxxxxxx> wrote:
>>> On 2017-03-28 17:13, Jan Kiszka wrote:
>>>> On 2017-03-28 15:49, Ard Biesheuvel wrote:
>> [..]
>>>>> Could you please have a look at
>>>>>
>>>>> https://git.kernel.org/pub/scm/linux/kernel/git/ardb/linux.git/log/?h=quark-capsule
>>>>>
>>>>> and tell me if that would work for you? I will send them out for
>>>>> proper review in any case, but to avoid confusion (if I missed
>>>>> something obvious), I don't want to send them out just yet.
>>>>
>>>> There is more needed to make things work again, maybe around passing the
>>>> right image size. I'm looking into this.
>>>>
>>>
>>> This makes CSH images being accepted again:
>>>
>>> diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c
>>> index 4b6f93f..a4e2311 100644
>>> --- a/arch/x86/platform/efi/quirks.c
>>> +++ b/arch/x86/platform/efi/quirks.c
>>> @@ -562,6 +562,8 @@ int efi_capsule_setup_info(struct capsule_info *cap_info, void *kbuff,
>>>  {
>>>         struct quark_security_header *csh = kbuff;
>>>
>>> +       cap_info->total_size = 0;
>>> +
>>>         if (!x86_match_cpu(quark_ids))
>>>                 goto fallback;
>>>
>>> @@ -587,12 +589,16 @@ int efi_capsule_setup_info(struct capsule_info *cap_info, void *kbuff,
>>>
>>>         kbuff += csh->headersize;
>>>
>>> +       cap_info->total_size = csh->headersize;
>>> +
>>>  fallback:
>>>         if (hdr_bytes < sizeof(efi_capsule_header_t))
>>>                 return 0;
>>>
>>>         memcpy(&cap_info->header, kbuff, sizeof(cap_info->header));
>>>
>>> +       cap_info->total_size += cap_info->header.imagesize;
>>> +
>>>         return __efi_capsule_setup_info(cap_info);
>>>  }
>>>
>>> diff --git a/drivers/firmware/efi/capsule-loader.c b/drivers/firmware/efi/capsule-loader.c
>>> index e851951..40dc354 100644
>>> --- a/drivers/firmware/efi/capsule-loader.c
>>> +++ b/drivers/firmware/efi/capsule-loader.c
>>> @@ -42,7 +42,7 @@ int __efi_capsule_setup_info(struct capsule_info *cap_info)
>>>         int ret;
>>>         void *temp_page;
>>>
>>> -       pages_needed = ALIGN(cap_info->header.imagesize, PAGE_SIZE) / PAGE_SIZE;
>>> +       pages_needed = ALIGN(cap_info->total_size, PAGE_SIZE) / PAGE_SIZE;
>>>
>>>         if (pages_needed == 0) {
>>>                 pr_err("invalid capsule size");
>>> @@ -59,7 +59,6 @@ int __efi_capsule_setup_info(struct capsule_info *cap_info)
>>>                 return ret;
>>>         }
>>>
>>> -       cap_info->total_size = cap_info->header.imagesize;
>>>         temp_page = krealloc(cap_info->pages,
>>>                              pages_needed * sizeof(void *),
>>>                              GFP_KERNEL | __GFP_ZERO);
>>> @@ -87,6 +86,8 @@ int __weak efi_capsule_setup_info(struct capsule_info *cap_info, void *kbuff,
>>>
>>>         memcpy(&cap_info->header, kbuff, sizeof(cap_info->header));
>>>
>>> +       cap_info->total_size = cap_info->header.imagesize;
>>> +
>>>         return __efi_capsule_setup_info(cap_info);
>>>  }
>>>
>>
>> OK, thanks for debugging that.
>>
>>> But then my changes to efi_capsule_update are missing the present the
>>> right format to the loader. As efi_capsule_update needs to lay out the
>>> sg-list in as special way, excluding the CSH on the first page, it needs
>>> to know about the displacement.
>>>
>>> Any suggestions how to address that without rolling back to my aproach?
>>>
>>
>> OK, I'm a bit lost now: for my understanding, could you please
>> reiterate how the CSH image deviates from the ordinary one? Or more
>> specifically, what exactly is preventing us from simply chopping off
>> the CSH header and pushing the capsule header + payload into
>> /dev/capsule_loader?
>>
>
> Devices that mandate a signed capsule for updates expect the CSH in
> front of the regular capsule image. The interface to UEFI remains
> unchanged, i.e. you pass the virtually chopped off capsule, but you have
> to leave the CSH in RAM right in front of that very same image. It's a
> "nice" side-channel API.
>

Wow, that is worse than I thought.

So my suggestion (which I coded up, please pull again), is to replace
the array of struct page pointers with an array of physical addresses
in the capsule_info struct. This way, your special version of
efi_capsule_setup_info() can advance the first one by the size of the
header.

I hope this works for you. I am not sure whether imagesize needs to be
modified, so I left it alone for now (but I suspect it should be).