[PATCH v2 1/1] efi_leader: delete rng-seed if having EFI RNG protocol

[Heinrich Schuchardt]

On 9/17/24 08:38, Ilias Apalodimas wrote:
> Hi Heinrich,
> 
> On Sat, Sep 14, 2024 at 06:08:12PM +0200, Heinrich Schuchardt wrote:
>> For measured be boot we must avoid any volatile values in the device-tree.
>> We already delete /chosen/kaslr-seed if we provide and EFI RNG protocol.
>>
>> Additionally remove /chosen/rng-seed provided by QEMU or U-Boot.
>>
>> Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt at canonical.com>
>> ---
>> v2:
>> 	put log_debug() in else branch
>> ---
>>   include/efi_loader.h          |  2 +-
>>   lib/efi_loader/efi_dt_fixup.c | 16 +++++++++++-----
>>   lib/efi_loader/efi_helper.c   |  2 +-
>>   3 files changed, 13 insertions(+), 7 deletions(-)
>>
>> diff --git a/include/efi_loader.h b/include/efi_loader.h
>> index f84852e384f..511281e150e 100644
>> --- a/include/efi_loader.h
>> +++ b/include/efi_loader.h
>> @@ -567,7 +567,7 @@ efi_status_t EFIAPI efi_convert_pointer(efi_uintn_t debug_disposition,
>>   /* Carve out DT reserved memory ranges */
>>   void efi_carve_out_dt_rsv(void *fdt);
>>   /* Purge unused kaslr-seed */
>> -void efi_try_purge_kaslr_seed(void *fdt);
>> +void efi_try_purge_rng_seed(void *fdt);
>>   /* Called by bootefi to make console interface available */
>>   efi_status_t efi_console_register(void);
>>   /* Called by efi_init_obj_list() to proble all block devices */
>> diff --git a/lib/efi_loader/efi_dt_fixup.c b/lib/efi_loader/efi_dt_fixup.c
>> index 9d017804eea..b97758d1305 100644
>> --- a/lib/efi_loader/efi_dt_fixup.c
>> +++ b/lib/efi_loader/efi_dt_fixup.c
>> @@ -41,7 +41,7 @@ static void efi_reserve_memory(u64 addr, u64 size, bool nomap)
>>   }
>>
>>   /**
>> - * efi_try_purge_kaslr_seed() - Remove unused kaslr-seed
>> + * efi_try_purge_rng_seed() - Remove unused kaslr-seed, rng-seed
>>    *
>>    * Kernel's EFI STUB only relies on EFI_RNG_PROTOCOL for randomization
>>    * and completely ignores the kaslr-seed for its own randomness needs
>> @@ -51,8 +51,9 @@ static void efi_reserve_memory(u64 addr, u64 size, bool nomap)
>>    *
>>    * @fdt: Pointer to device tree
>>    */
>> -void efi_try_purge_kaslr_seed(void *fdt)
>> +void efi_try_purge_rng_seed(void *fdt)
>>   {
>> +	const char * const prop[] = {"kaslr-seed", "rng-seed"};
>>   	const efi_guid_t efi_guid_rng_protocol = EFI_RNG_PROTOCOL_GUID;
>>   	struct efi_handler *handler;
>>   	efi_status_t ret;
>> @@ -67,9 +68,14 @@ void efi_try_purge_kaslr_seed(void *fdt)
>>   	if (nodeoff < 0)
>>   		return;
>>
>> -	err = fdt_delprop(fdt, nodeoff, "kaslr-seed");
>> -	if (err < 0 && err != -FDT_ERR_NOTFOUND)
>> -		log_err("Error deleting kaslr-seed\n");
>> +	for (const char * const *pos = prop; pos < &prop[ARRAY_SIZE(prop)];
> 
> I think
> for (int i = 0; i < ARRAY_SIZE(prop); i++)
>      fdt_delprop(fdt, nodeoff, prop[i]);
> 
> is a lot easier to read

On RISC-V using an index needs 10 bytes more of binary code.
Easy to read and efficient are conflicting here.

I will respin the patch.

Best regards

Heinrich

> 
> Other than that, the patch looks fine
> 
> Thanks
> /Ilias
> 
>> +	     ++pos) {
>> +		err = fdt_delprop(fdt, nodeoff, *pos);
>> +		if (err < 0 && err != -FDT_ERR_NOTFOUND)
>> +			log_err("Error deleting %s\n", *pos);
>> +		else
>> +			log_debug("Deleted /chosen/%s\n", *pos);
>> +	}
>>   }
>>
>>   /**
>> diff --git a/lib/efi_loader/efi_helper.c b/lib/efi_loader/efi_helper.c
>> index 96f847652ec..a481eb4b7e3 100644
>> --- a/lib/efi_loader/efi_helper.c
>> +++ b/lib/efi_loader/efi_helper.c
>> @@ -522,7 +522,7 @@ efi_status_t efi_install_fdt(void *fdt)
>>   	/* Create memory reservations as indicated by the device tree */
>>   	efi_carve_out_dt_rsv(fdt);
>>
>> -	efi_try_purge_kaslr_seed(fdt);
>> +	efi_try_purge_rng_seed(fdt);
>>
>>   	if (CONFIG_IS_ENABLED(EFI_TCG2_PROTOCOL_MEASURE_DTB)) {
>>   		ret = efi_tcg2_measure_dtb(fdt);
>> --
>> 2.45.2
>>