Re: [PATCH v2] efi_loader: Get rid of kaslr-seed

[Ilias Apalodimas <ilias.apalodimas@linaro.org>]

Hi Mark, 

On Fri, Dec 17, 2021 at 12:13:20PM +0100, Mark Kettenis wrote:
> > From: Ilias Apalodimas <ilias.apalodimas@linaro.org>
> > Date: Fri, 17 Dec 2021 09:06:44 +0200
> > 
> > Right now we unconditionally pass a 'kaslr-seed' property to the kernel
> > if the DTB we ended up in EFI includes the entry.  However the kernel
> > EFI stub completely ignores it and only relies on EFI_RNG_PROTOCOL for
> > it's own randomness needs (i.e the randomization of the physical
> > placement of the kernel).
> > So let's get rid of it if EFI_RNG_PPROTOCOL is installed.
> > 
> > It's worth noting that TPMs also provide an RNG.  So if we tweak our
> > EFI_RNG_PROTOCOL slightly and install the protocol when a TPM device
> > is present the 'kaslr-seed' property will always be removed, allowing
> > us to reliably measure our DTB as well.
> 
> That looks good to me.  But I think you should be honest here.  You're
> removing the kaslr-seed property here because it is a part of the DTB
> that changes on every boot, which interferes with the measurability.
> Not because there is something wrong with passing along the
> kaslr-seed.  

Yep never tried to hide that :)

> In the absence of EFI_RNG_PROTOCOL support passing
> kaslr-seed still has a benefit as it will still be used for
> randomization of the virtual placement of the kernel.  So I think the
> comment could use some rewording (but the bit on randomization of the
> physical placement is certainly accurate and worth keeping).
> 
> Oh, and please say "The Linux kernel's EFI STUB".

Sure something along the lines of:

"Right now we unconditionally keep a 'kaslr-seed' property in the DTB,
if what we ended up in EFI includes the entry.  However the Linux kernel
EFI-stub completely ignores it and only relies on EFI_RNG_PROTOCOL for
it's own randomness needs (i.e the randomization of the physical
placement of the kernel).  In fact it (blindly) overwrites the existing 
seed if the EFI_RNG_PROTOCOL protocol is installed.  However it still uses
it for randomization of the virtual placement of the kernel if the EFI
protocol is not installed. 
So let's get rid of it if EFI_RNG_PPROTOCOL is installed.

It's worth noting that TPMs also provide an RNG.  So if we tweak our
EFI_RNG_PROTOCOL slightly and install the protocol when a TPM device
is present the 'kaslr-seed' property will always be removed, allowing
us to reliably measure our DTB as well."

Cheers
/Ilias
> 
> Cheers,
> 
> Mark
> 
> > Acked-by: Ard Biesheuvel <ardb@kernel.org>
> > Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
> > ---
> > changes since v1:
> > - Only removing the property if EFI_RNG_PROTOCOL is installed, since some
> >   OS'es rely on kaslr-seed
> > - Only display an error message if the kaslr-seed entry was found but not
> >   removed
> >  cmd/bootefi.c                 |  2 ++
> >  include/efi_loader.h          |  2 ++
> >  lib/efi_loader/efi_dt_fixup.c | 33 +++++++++++++++++++++++++++++++++
> >  3 files changed, 37 insertions(+)
> > 
> > diff --git a/cmd/bootefi.c b/cmd/bootefi.c
> > index d77d3b6e943d..57f13ce701ec 100644
> > --- a/cmd/bootefi.c
> > +++ b/cmd/bootefi.c
> > @@ -310,6 +310,8 @@ 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);
> > +
> >  	/* Install device tree as UEFI table */
> >  	ret = efi_install_configuration_table(&efi_guid_fdt, fdt);
> >  	if (ret != EFI_SUCCESS) {
> > diff --git a/include/efi_loader.h b/include/efi_loader.h
> > index 9dd6c2033634..1fe003db69e0 100644
> > --- a/include/efi_loader.h
> > +++ b/include/efi_loader.h
> > @@ -519,6 +519,8 @@ efi_status_t EFIAPI efi_convert_pointer(efi_uintn_t debug_disposition,
> >  					void **address);
> >  /* 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);
> >  /* Called by bootefi to make console interface available */
> >  efi_status_t efi_console_register(void);
> >  /* Called by bootefi to make all disk storage accessible as EFI objects */
> > diff --git a/lib/efi_loader/efi_dt_fixup.c b/lib/efi_loader/efi_dt_fixup.c
> > index b6fe5d2e5a34..d3923e5dba1b 100644
> > --- a/lib/efi_loader/efi_dt_fixup.c
> > +++ b/lib/efi_loader/efi_dt_fixup.c
> > @@ -8,6 +8,7 @@
> >  #include <common.h>
> >  #include <efi_dt_fixup.h>
> >  #include <efi_loader.h>
> > +#include <efi_rng.h>
> >  #include <fdtdec.h>
> >  #include <mapmem.h>
> >  
> > @@ -40,6 +41,38 @@ static void efi_reserve_memory(u64 addr, u64 size, bool nomap)
> >  			addr, size);
> >  }
> >  
> > +/**
> > + * efi_try_purge_kaslr_seed() - Remove unused kaslr-seed
> > + *
> > + * Kernel's EFI STUB only relies on EFI_RNG_PROTOCOL for randomization
> > + * and completely ignores the kaslr-seed for its own randomness needs
> > + * (i.e the randomization of the physical placement of the kernel).
> > + * Weed it out from the DTB we hand over, which would mess up our DTB
> > + * TPM measurements as well.
> > + *
> > + * @fdt: Pointer to device tree
> > + */
> > +void efi_try_purge_kaslr_seed(void *fdt)
> > +{
> > +	const efi_guid_t efi_guid_rng_protocol = EFI_RNG_PROTOCOL_GUID;
> > +	struct efi_handler *handler;
> > +	efi_status_t ret;
> > +	int nodeoff = 0;
> > +	int err = 0;
> > +
> > +	ret = efi_search_protocol(efi_root, &efi_guid_rng_protocol, &handler);
> > +	if (ret != EFI_SUCCESS)
> > +		return;
> > +
> > +	nodeoff = fdt_path_offset(fdt, "/chosen");
> > +	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");
> > +}
> > +
> >  /**
> >   * efi_carve_out_dt_rsv() - Carve out DT reserved memory ranges
> >   *
> > -- 
> > 2.30.2
> > 
> >