On 2 September 2017 at 09:26, Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote: > On Sat, Sep 02, 2017 at 09:18:34AM +0100, Ard Biesheuvel wrote: >> On 2 September 2017 at 07:45, Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote: >> > On Sat, Aug 19, 2017 at 04:17:40PM +0100, Ard Biesheuvel wrote: >> >> Early entropy is hard to come by, especially on non-x86 systems that >> >> lack an architected instruction and are not as uniform as PCs. >> >> Fortunately, on UEFI systems, we can invoke the EFI_RNG_PROTOCOL, >> >> which exposes the platform specific entropy source in a generic way. >> >> We use this protocol to fill the RNG UEFI config table (which is used >> >> during early boot to seed the kernel's entropy pool), and on ARM and >> >> arm64, we invoke it to seed KASLR as well. >> >> >> >> Sadly, EFI_RNG_PROTOCOL is not widely implemented, and so it would be >> >> nice to have a fallback for UEFI systems that lack it. So implement >> >> this fallback based on the Chaoskey RNG USB stick, which should be >> >> exposed using the standard UEFI USB I/O protocol if the firmware has >> >> USB support. >> >> >> >> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx> >> >> --- >> >> drivers/firmware/efi/Kconfig | 7 ++ >> >> drivers/firmware/efi/libstub/Makefile | 2 +- >> >> drivers/firmware/efi/libstub/efistub.h | 3 + >> >> drivers/firmware/efi/libstub/random-chaoskey.c | 126 +++++++++++++++++++++++++ >> >> 4 files changed, 137 insertions(+), 1 deletion(-) >> >> create mode 100644 drivers/firmware/efi/libstub/random-chaoskey.c >> >> >> >> diff --git a/drivers/firmware/efi/Kconfig b/drivers/firmware/efi/Kconfig >> >> index 2b4c39fdfa91..448a6186d9fb 100644 >> >> --- a/drivers/firmware/efi/Kconfig >> >> +++ b/drivers/firmware/efi/Kconfig >> >> @@ -87,6 +87,13 @@ config EFI_RUNTIME_WRAPPERS >> >> config EFI_ARMSTUB >> >> bool >> >> >> >> +config EFI_RNG_CHAOSKEY >> >> + bool "Chaoskey USB RNG support in the UEFI stub" >> >> + help >> >> + Fall back to a Chaoskey RNG USB stick in case EFI_RNG_PROTOCOL >> >> + is not implemented by the firmware. This is used for early >> >> + seeding of the entropy pool, and for KASLR on ARM and arm64. >> >> + >> >> config EFI_BOOTLOADER_CONTROL >> >> tristate "EFI Bootloader Control" >> >> depends on EFI_VARS >> >> diff --git a/drivers/firmware/efi/libstub/Makefile b/drivers/firmware/efi/libstub/Makefile >> >> index dedf9bde44db..fad6bc1d0e2a 100644 >> >> --- a/drivers/firmware/efi/libstub/Makefile >> >> +++ b/drivers/firmware/efi/libstub/Makefile >> >> @@ -40,7 +40,7 @@ $(obj)/lib-%.o: $(srctree)/lib/%.c FORCE >> >> >> >> lib-$(CONFIG_EFI_ARMSTUB) += arm-stub.o fdt.o string.o random.o \ >> >> $(patsubst %.c,lib-%.o,$(arm-deps)) >> >> - >> >> +lib-$(CONFIG_EFI_RNG_CHAOSKEY) += random-chaoskey.o >> >> lib-$(CONFIG_ARM) += arm32-stub.o >> >> lib-$(CONFIG_ARM64) += arm64-stub.o >> >> CFLAGS_arm64-stub.o := -DTEXT_OFFSET=$(TEXT_OFFSET) >> >> diff --git a/drivers/firmware/efi/libstub/efistub.h b/drivers/firmware/efi/libstub/efistub.h >> >> index 3afff5facd3b..9579ddb5937d 100644 >> >> --- a/drivers/firmware/efi/libstub/efistub.h >> >> +++ b/drivers/firmware/efi/libstub/efistub.h >> >> @@ -62,6 +62,9 @@ efi_status_t efi_random_alloc(efi_system_table_t *sys_table_arg, >> >> unsigned long size, unsigned long align, >> >> unsigned long *addr, unsigned long random_seed); >> >> >> >> +efi_status_t efi_get_random_chaoskey(efi_system_table_t *sys_table_arg, >> >> + unsigned long size, u8 *out); >> >> + >> >> efi_status_t check_platform_features(efi_system_table_t *sys_table_arg); >> >> >> >> efi_status_t efi_random_get_seed(efi_system_table_t *sys_table_arg); >> >> diff --git a/drivers/firmware/efi/libstub/random-chaoskey.c b/drivers/firmware/efi/libstub/random-chaoskey.c >> >> new file mode 100644 >> >> index 000000000000..3f94ec3df9ff >> >> --- /dev/null >> >> +++ b/drivers/firmware/efi/libstub/random-chaoskey.c >> >> @@ -0,0 +1,126 @@ >> >> +/* >> >> + * Copyright (C) 2017 Linaro Ltd; <ard.biesheuvel@xxxxxxxxxx> >> >> + * >> >> + * This program is free software; you can redistribute it and/or modify >> >> + * it under the terms of the GNU General Public License version 2 as >> >> + * published by the Free Software Foundation. >> >> + * >> >> + */ >> >> + >> >> +#include <linux/efi.h> >> >> +#include <asm/efi.h> >> >> + >> >> +#include "efistub.h" >> >> + >> >> +#define CHAOSKEY_VENDOR_ID 0x1d50 /* OpenMoko */ >> >> +#define CHAOSKEY_PRODUCT_ID 0x60c6 /* ChaosKey */ >> >> + >> >> +static efi_usb_io_protocol_t *chaoskey_usb_io; >> >> +static unsigned long chaoskey_usb_ep_addr, chaoskey_usb_ep_size; >> >> + >> >> +static void locate_chaoskey(efi_system_table_t *sys_table_arg) >> >> +{ >> >> + efi_guid_t usb_io_guid = EFI_USB_IO_PROTOCOL_GUID; >> >> + unsigned long bufsize = 0, i, j; >> >> + efi_handle_t *handles; >> >> + efi_status_t status; >> >> + >> >> + chaoskey_usb_io = (void *)-1; >> >> + >> >> + /* find all USB devices that UEFI knows about */ >> >> + status = efi_call_early(locate_handle, EFI_LOCATE_BY_PROTOCOL, >> >> + &usb_io_guid, NULL, &bufsize, NULL); >> >> + if (status != EFI_BUFFER_TOO_SMALL) >> >> + return; >> >> + >> >> + status = efi_call_early(allocate_pool, EFI_LOADER_DATA, bufsize, >> >> + (void **)&handles); >> >> + if (status != EFI_SUCCESS) >> >> + return; >> >> + >> >> + status = efi_call_early(locate_handle, EFI_LOCATE_BY_PROTOCOL, >> >> + &usb_io_guid, NULL, &bufsize, handles); >> >> + if (status != EFI_SUCCESS) >> >> + goto out; >> >> + >> >> + for (i = 0; i < bufsize / sizeof(efi_handle_t); i++) { >> >> + struct usb_device_descriptor dev; >> >> + struct usb_interface_descriptor iface; >> >> + efi_usb_io_protocol_t *p; >> >> + >> >> + status = efi_call_early(handle_protocol, handles[i], >> >> + &usb_io_guid, (void **)&p); >> >> + if (status != EFI_SUCCESS) >> >> + continue; /* shouldn't happen */ >> >> + >> >> + /* get the device descriptor so we can check the vid/pid */ >> >> + status = p->get_device_descriptor(p, &dev); >> >> + if (status != EFI_SUCCESS) >> >> + continue; /* shouldn't happen either */ >> >> + >> >> + if (dev.idVendor != CHAOSKEY_VENDOR_ID || >> >> + dev.idProduct != CHAOSKEY_PRODUCT_ID) >> > >> > idVendor and idProduct are little endian types. >> > >> > You have run sparse on this code, right? >> > >> > >> >> + continue; >> >> + >> >> + /* get the number of endpoints from the interface descriptor */ >> >> + status = p->get_iface_descriptor(p, &iface); >> >> + if (status != EFI_SUCCESS) >> >> + continue; >> >> + >> >> + /* locate the first BULK IN endpoint */ >> >> + for (j = 0; j < iface.bNumEndpoints; j++) { >> >> + struct usb_endpoint_descriptor ep; >> > >> > Don't we have built-in usb functions for this now? Why not use them >> > instead of hand-rolling your own logic? >> > >> >> This is UEFI stub code, which is executed with the kernel image or >> decompressor loaded at a completely different address than it was >> linked at. Reusing inline functions is usually fine, but anything that >> requires linking kernel objects into the decompressor is a bit >> cumbersome. > > So this is not Linux kernel code? Or is it? I'm totally confused now. > All code that lives under drivers/firmware/efi/libstub is linked into a PE/COFF binary that is either embedded into the decompressor (x86, ARM) or the kernel proper (arm64) but with a separate __efistub_ namespace. It implements the so-called OS loader (in UEFI speak) that loads the images and interfaces with the boot environment in other ways to prepare for OS boot. At this point, we are still running in the 1:1 mapping set up by UEFI, and the PE/COFF image (which contains the entire kernel image as a payload, in one way or another) is loaded at an arbitrary offset by the firmware, and invoked from there. >> In any case, these patches are RFC only, and it is quite obvious that >> having a Chaoskey USB driver in the firmware that produces the >> EFI_RNG_PROTOCOL is much better, and the UEFI stub code can just call >> into the protocol to get entropy. > > Ok, why not write that and not have to deal with any Linux code at all? > Oh, we did write it. And we are leaning towards withdrawing these patches, but only because adding support for individual devices is a bit icky. The way these patches interface with the firmware is no different from all the other Linux code that does so. -- To unsubscribe from this list: send the line "unsubscribe linux-efi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html