efi: random: combine bootloader provided RNG seed with RNG protocol output
PATCH 1/1 efi: random: combine bootloader provided RNG seed with RNG protocol output
From: Ard Biesheuvel <[email protected]>
commit 196dff2712ca5a2e651977bb2fe6b05474111a83 upstream.
Instead of blindly creating the EFI random seed configuration table if the RNG protocol is implemented and works, check whether such a EFI configuration table was provided by an earlier boot stage and if so, concatenate the existing and the new seeds, leaving it up to the core code to mix it in and credit it the way it sees fit.
This can be used for, e.g., systemd-boot, to pass an additional seed to Linux in a way that can be consumed by the kernel very early. In that case, the following definitions should be used to pass the seed to the EFI stub:
struct linux_efi_random_seed {
u32 size; // of the 'seed' array in bytes
u8 seed[];
};
The memory for the struct must be allocated as EFI_ACPI_RECLAIM_MEMORY pool memory, and the address of the struct in memory should be installed as a EFI configuration table using the following GUID:
LINUX_EFI_RANDOM_SEED_TABLE_GUID 1ce1e5bc-7ceb-42f2-81e5-8aadf180f57b
Note that doing so is safe even on kernels that were built without this patch applied, but the seed will simply be overwritten with a seed derived from the EFI RNG protocol, if available. The recommended seed size is 32 bytes, and seeds larger than 512 bytes are considered corrupted and ignored entirely.
In order to preserve forward secrecy, seeds from previous bootloaders are memzero'd out, and in order to preserve memory, those older seeds are also freed from memory. Freeing from memory without first memzeroing is not safe to do, as it's possible that nothing else will ever overwrite those pages used by EFI.
[ardb: incorporate Jason's followup changes to extend the maximum seed
size on the consumer end, memzero() it and drop a needless printk]
--- drivers/firmware/efi/efi.c | 4 +-- drivers/firmware/efi/libstub/efistub.h | 2 ++ drivers/firmware/efi/libstub/random.c | 42 ++++++++++++++++++++++---- include/linux/efi.h | 2 -- 4 files changed, 40 insertions(+), 10 deletions(-)
drivers/firmware/efi/efi.c +2 -2
@@ -608,7 +608,7 @@ int __init efi_config_parse_tables(const efi_config_table_t *config_tables, seed =early_memremap (efi_rng_seed,sizeof (*seed));if (seed !=NULL ) { - size =min (seed->size,EFI_RANDOM_SEED_SIZE ); + size =min_t (u32, seed->size,SZ_1K );// sanity check early_memunmap (seed,sizeof (*seed)); }else {pr_err ("Could not map UEFI random seed!\n" );
@@ -617,8 +617,8 @@ int __init efi_config_parse_tables(const efi_config_table_t *config_tables, seed =early_memremap (efi_rng_seed,sizeof (*seed) + size);if (seed !=NULL ) { -pr_notice ("seeding entropy pool\n" );add_bootloader_randomness (seed->bits, size); +memzero_explicit (seed->bits, size);early_memunmap (seed,sizeof (*seed) + size); }else {pr_err ("Could not map UEFI random seed!\n" );
drivers/firmware/efi/libstub/efistub.h +2 -0
@@ -873,6 +873,8 @@ efi_status_t efi_get_random_bytes(unsigned long size, u8 *out);efi_status_t efi_random_alloc (unsigned long size,unsigned long align,unsigned long *addr,unsigned long random_seed); +efi_status_t efi_random_get_seed (void ); +efi_status_t check_platform_features (void );void *get_efi_config_table (efi_guid_t guid);
drivers/firmware/efi/libstub/random.c +36 -6
77 lines not on this page · raw message
include/linux/efi.h +0 -2
@@ -1139,8 +1139,6 @@ void efi_check_for_embedded_firmwares(void);static inline void efi_check_for_embedded_firmwares (void ) { }#endif -efi_status_t efi_random_get_seed (void ); - #define arch_efi_call_virt(p, f, args...) ((p)->f(args))/*
Signature · 1 line
2.39.0
Greg KH
maintainerQuoted 41 lines · Jason A. Donenfeld
From: Ard Biesheuvel <[email protected]>
commit
196dff2712ca5a2e651977bb2fe6b05474111a83upstream.Instead of blindly creating the EFI random seed configuration table if the RNG protocol is implemented and works, check whether such a EFI configuration table was provided by an earlier boot stage and if so, concatenate the existing and the new seeds, leaving it up to the core code to mix it in and credit it the way it sees fit.
This can be used for, e.g., systemd-boot, to pass an additional seed to Linux in a way that can be consumed by the kernel very early. In that case, the following definitions should be used to pass the seed to the EFI stub:
struct
linux_efi_random_seed{
u32 size; // of the 'seed' array in bytes
u8 seed[];
};The memory for the struct must be allocated as
EFI_ACPI_RECLAIM_MEMORYpool memory, and the address of the struct in memory should be installed as a EFI configuration table using the following GUID:
LINUX_EFI_RANDOM_SEED_TABLE_GUID1ce1e5bc-7ceb-42f2-81e5-8aadf180f57bNote that doing so is safe even on kernels that were built without this patch applied, but the seed will simply be overwritten with a seed derived from the EFI RNG protocol, if available. The recommended seed size is 32 bytes, and seeds larger than 512 bytes are considered corrupted and ignored entirely.
In order to preserve forward secrecy, seeds from previous bootloaders are memzero'd out, and in order to preserve memory, those older seeds are also freed from memory. Freeing from memory without first memzeroing is not safe to do, as it's possible that nothing else will ever overwrite those pages used by EFI.
Reviewed-by Jason A. Donenfeld <[email protected]>[ardb: incorporate Jason's followup changes to extend the maximum seed
size on the consumer end,memzero()it and drop a needless printk]
Quoted 8 lines
Signed-off-by Ard Biesheuvel <[email protected]>Signed-off-by Jason A. Donenfeld <[email protected]>--- drivers/firmware/efi/efi.c | 4 +-- drivers/firmware/efi/libstub/efistub.h | 2 ++ drivers/firmware/efi/libstub/random.c | 42 ++++++++++++++++++++++---- include/linux/efi.h | 2 -- 4 files changed, 40 insertions(+), 10 deletions(-)
Now queued up, thanks.
greg k-h
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
Greg KH
maintainerQuoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
thanks,
greg k-h
Quoted 7 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
Oh, darn. I thought it would for some reason. Okay, lemme get cranking on that.
Quoted 11 lines · Jason A. Donenfeld
Quoted 7 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
Oh, darn. I thought it would for some reason. Okay, lemme get cranking on that.
Should we bother? Isn't v5.10 far enough back for this? This is not a bugfix after all.
Quoted 9 lines · Ard Biesheuvel
Quoted 7 lines · Jason A. Donenfeld
Quoted 5 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
Oh, darn. I thought it would for some reason. Okay, lemme get cranking on that.
Should we bother? Isn't v5.10 far enough back for this? This is not a bugfix after all.
This *is* a bug fix. And not just because we used to clobber that configuration table unnecessarily, but moreover because of the forward secrecy issues due to the missing memzero. We did all that in a single patch under the assumption that this would be backported as a unit.
Anyway, don't sweat it - I'm working on the backport now. Seems straightforward enough.
Jason
Quoted 19 lines · Jason A. Donenfeld
Quoted 9 lines · Ard Biesheuvel
Quoted 7 lines · Jason A. Donenfeld
Quoted 5 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
Oh, darn. I thought it would for some reason. Okay, lemme get cranking on that.
Should we bother? Isn't v5.10 far enough back for this? This is not a bugfix after all.
Quoted 8 lines
This *is* a bug fix. And not just because we used to clobber that configuration table unnecessarily, but moreover because of the forward secrecy issues due to the missing memzero. We did all that in a single patch under the assumption that this would be backported as a unit.
Anyway, don't sweat it - I'm working on the backport now. Seems straightforward enough.
Fair enough.
Quoted 7 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
I did 5.4.y, which turned out to be hairy than I wanted. You and Ard can decide if you want it or not. I'll leave 4.19 and 4.14 for another day.
PATCH 1/1 efi: random: combine bootloader provided RNG seed with RNG protocol output
From: Ard Biesheuvel <[email protected]>
commit 196dff2712ca5a2e651977bb2fe6b05474111a83 upstream.
Instead of blindly creating the EFI random seed configuration table if the RNG protocol is implemented and works, check whether such a EFI configuration table was provided by an earlier boot stage and if so, concatenate the existing and the new seeds, leaving it up to the core code to mix it in and credit it the way it sees fit.
This can be used for, e.g., systemd-boot, to pass an additional seed to Linux in a way that can be consumed by the kernel very early. In that case, the following definitions should be used to pass the seed to the EFI stub:
struct linux_efi_random_seed {
u32 size; // of the 'seed' array in bytes
u8 seed[];
};
The memory for the struct must be allocated as EFI_ACPI_RECLAIM_MEMORY pool memory, and the address of the struct in memory should be installed as a EFI configuration table using the following GUID:
LINUX_EFI_RANDOM_SEED_TABLE_GUID 1ce1e5bc-7ceb-42f2-81e5-8aadf180f57b
Note that doing so is safe even on kernels that were built without this patch applied, but the seed will simply be overwritten with a seed derived from the EFI RNG protocol, if available. The recommended seed size is 32 bytes, and seeds larger than 512 bytes are considered corrupted and ignored entirely.
In order to preserve forward secrecy, seeds from previous bootloaders are memzero'd out, and in order to preserve memory, those older seeds are also freed from memory. Freeing from memory without first memzeroing is not safe to do, as it's possible that nothing else will ever overwrite those pages used by EFI.
[ardb: incorporate Jason's followup changes to extend the maximum seed
size on the consumer end, memzero() it and drop a needless printk]
--- arch/x86/boot/compressed/eboot.c | 3 + drivers/firmware/efi/efi.c | 4 +- drivers/firmware/efi/libstub/Makefile | 5 +- drivers/firmware/efi/libstub/efistub.h | 3 +- drivers/firmware/efi/libstub/random.c | 86 +++++++++++++++++++++----- include/linux/efi.h | 2 + 6 files changed, 83 insertions(+), 20 deletions(-)
arch/x86/boot/compressed/eboot.c +3 -0
@@ -782,6 +782,9 @@ efi_main(struct efi_config *c, struct boot_params *boot_params)/* Ask the firmware to clear memory on unclean shutdown */ efi_enable_reset_attack_mitigation (sys_table); + +efi_random_get_seed (sys_table); +efi_retrieve_tpm2_eventlog (sys_table);setup_graphics (boot_params);
drivers/firmware/efi/efi.c +2 -2
@@ -546,7 +546,7 @@ int __init efi_config_parse_tables(void *config_tables, int count, int sz, seed =early_memremap (efi.rng_seed,sizeof (*seed));if (seed !=NULL ) { - size =min (seed->size,EFI_RANDOM_SEED_SIZE ); + size =min_t (u32, seed->size,SZ_1K );// sanity check early_memunmap (seed,sizeof (*seed)); }else {pr_err ("Could not map UEFI random seed!\n" );
@@ -555,8 +555,8 @@ int __init efi_config_parse_tables(void *config_tables, int count, int sz, seed =early_memremap (efi.rng_seed,sizeof (*seed) + size);if (seed !=NULL ) { -pr_notice ("seeding entropy pool\n" );add_bootloader_randomness (seed->bits, size); +memzero_explicit (seed->bits, size);early_memunmap (seed,sizeof (*seed) + size); }else {pr_err ("Could not map UEFI random seed!\n" );
drivers/firmware/efi/libstub/Makefile +3 -2
@@ -39,7 +39,8 @@ OBJECT_FILES_NON_STANDARD := y # Prevents link failures: __sanitizer_cov_trace_pc() is not linked in. KCOV_INSTRUMENT := n -lib-y := efi-stub-helper.o gop.o secureboot.o tpm.o +lib-y := efi-stub-helper.o gop.o secureboot.o tpm.o \ + random.o # include the stub's generic dependencies from lib/ when building for ARM/arm64 arm-deps-y := fdt_rw.c fdt_ro.c fdt_wip.c fdt.c fdt_empty_tree.c fdt_sw.c
@@ -48,7 +49,7 @@ arm-deps-$(CONFIG_ARM64) += sort.c $(obj)/lib-%.o: $(srctree)/lib/%.c FORCE $(call if_changed_rule,cc_o_c) -lib-$(CONFIG_EFI_ARMSTUB) += arm-stub.o fdt.o string.o random.o \ +lib-$(CONFIG_EFI_ARMSTUB) += arm-stub.o fdt.o string.o \ $(patsubst %.c,lib-%.o,$(arm-deps-y)) lib-$(CONFIG_ARM) += arm32-stub.o
drivers/firmware/efi/libstub/efistub.h +1 -2
@@ -34,6 +34,7 @@ extern int __pure novamap(void); } while (0)#define pr_efi_err (sys_table, msg) efi_printk(sys_table, "EFI stub: ERROR: "msg) +#define pr_efi_warn (sys_table, msg) efi_printk(sys_table, "EFI stub: WARNING: "msg)void efi_char16_printk (efi_system_table_t *,efi_char16_t *);
@@ -63,8 +64,6 @@ efi_status_t efi_random_alloc(efi_system_table_t *sys_table_arg,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); -void *get_efi_config_table (efi_system_table_t *sys_table,efi_guid_t guid);/* Helper macros for the usual case of using simple C variables: */
drivers/firmware/efi/libstub/random.c +72 -14
138 lines not on this page · raw message
include/linux/efi.h +2 -0
@@ -1653,6 +1653,8 @@ static inline void efi_enable_reset_attack_mitigation(efi_system_table_t *sys_table_arg) { }#endif +efi_status_t efi_random_get_seed (efi_system_table_t *sys_table_arg); +void efi_retrieve_tpm2_eventlog (efi_system_table_t *sys_table);/*
Signature · 1 line
2.39.0
Greg KH
maintainerQuoted 4 lines · Jason A. Donenfeld
From: Ard Biesheuvel <[email protected]>
commit
196dff2712ca5a2e651977bb2fe6b05474111a83upstream.
Now queued up, thanks.
greg k-h
Quoted 9 lines · Greg KH
Quoted 4 lines · Jason A. Donenfeld
From: Ard Biesheuvel <[email protected]>
commit
196dff2712ca5a2e651977bb2fe6b05474111a83upstream.Now queued up, thanks.
Queued up where? Not v5.4 I hope?
Greg KH
maintainerQuoted 5 lines · Ard Biesheuvel
Quoted 4 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
From: Ard Biesheuvel <[email protected]>
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
commit
196dff2712ca5a2e651977bb2fe6b05474111a83upstream.Now queued up, thanks.
Queued up where? Not v5.4 I hope?
Ick, yes, I did that, I'll go drop this now, sorry for the noise.
greg "burried in patches" k-h
Quoted 7 lines · Jason A. Donenfeld
Quoted 5 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
Quoted 6 lines
Quoted 2 lines
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
I did 5.4.y, which turned out to be hairy than I wanted. You and Ard can decide if you want it or not. I'll leave 4.19 and 4.14 for another day.
I appreciate you spending the effort, but I'm not convinced this is worth the risk. You are backporting new functionality (invoking the firmware's RNG protocol at boot on x86), and we might end up regressing on systems where the firmware's implementation is problematic, even if the patch by itself is correct. This applies to mixed mode especially, as the conversion between Win64 and i386 calling conventions has kicked up some very surprising issues in the past.
Greg KH
maintainerQuoted 15 lines · Ard Biesheuvel
Quoted 7 lines · Jason A. Donenfeld
Quoted 5 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
Quoted 6 lines
Quoted 2 lines
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
I did 5.4.y, which turned out to be hairy than I wanted. You and Ard can decide if you want it or not. I'll leave 4.19 and 4.14 for another day.
Quoted 8 lines
I appreciate you spending the effort, but I'm not convinced this is worth the risk. You are backporting new functionality (invoking the firmware's RNG protocol at boot on x86), and we might end up regressing on systems where the firmware's implementation is problematic, even if the patch by itself is correct. This applies to mixed mode especially, as the conversion between Win64 and i386 calling conventions has kicked up some very surprising issues in the past.
Yeah, I'll leave this alone on those old kernel trees. They are primarily only used in Android (i.e. arm64) systems and shouldn't be messing with efi. Those x86 systems still stuck using these old kernels are fragile enough, and should have moved to newer kernels anyway...
thanks,
greg k-h
Quoted 15 lines · Ard Biesheuvel
Quoted 7 lines · Jason A. Donenfeld
Quoted 5 lines · Greg KH
Quoted 2 lines · Jason A. Donenfeld
Thanks! IIRC, this applies to all current stable kernels (now that you've sunsetted 4.9).
Quoted 6 lines
Quoted 2 lines
It does not apply cleanly to 5.4.y or 4.19.y or 4.14.y so can you provide working backports for them?
I did 5.4.y, which turned out to be hairy than I wanted. You and Ard can decide if you want it or not. I'll leave 4.19 and 4.14 for another day.
Quoted 8 lines
I appreciate you spending the effort, but I'm not convinced this is worth the risk. You are backporting new functionality (invoking the firmware's RNG protocol at boot on x86), and we might end up regressing on systems where the firmware's implementation is problematic, even if the patch by itself is correct. This applies to mixed mode especially, as the conversion between Win64 and i386 calling conventions has kicked up some very surprising issues in the past.
Alright, yea, I was afraid that might be the case indeed. Oh well.
So this means that for the purposes of systemd's usage of this, 5.10+ is the relevant cut-off. I'm noting it here because I'm sure I'll forget, and the question is bound to come up down the road.
Jason