The current EFI implementation confuses pointers and addresses. Normally
we can get away with this but in the case of sandbox it causes failures.
Despite the fact that efi_allocate_pages() returns a u64, it is actually
a pointer, not an address. Add special handling to avoid a crash when
running 'bootefi hello'.
Signed-off-by: Simon Glass <sjg@chromium.org>
Documentation:
* describe printf() format codes
UEFI
* enable more algorithms for UEFI image verification, e.g. SHA256-RSA2048
General
* simplify printing short texts for GUIDs
* provide a unit test for printing GUIDs
-----BEGIN PGP SIGNATURE-----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=GoR6
-----END PGP SIGNATURE-----
Merge tag 'efi-2022-04-rc1-2' of https://source.denx.de/u-boot/custodians/u-boot-efi
Pull request for efi-2022-04-rc1-2
Documentation:
* describe printf() format codes
UEFI
* enable more algorithms for UEFI image verification, e.g. SHA256-RSA2048
General
* simplify printing short texts for GUIDs
* provide a unit test for printing GUIDs
For printing GUIDs with macro EFI_ENTRY use %pUs instead of %pUl to provide
readable debug output.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
If variable dfu_alt_info is not defined duplicate messages are displayed.
=> efidebug boot dump
Scanning disk mmc2.blk...
Scanning disk mmc1.blk...
Scanning disk mmc0.blk...
Found 3 disks
No EFI system partition
"dfu_alt_info" env variable not defined!
Probably dfu_alt_info not defined
"dfu_alt_info" env variable not defined!
Probably dfu_alt_info not defined
Remove the 'Probably dfu_alt_info not defined' message.
Instead write a warning if the variable contains no entities.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
U-Boot, in some occasions, injects a 'kaslr-seed' property on the /chosen
node. That would be problematic in case we want to measure the DTB we
install in the configuration table, since it would change across reboots.
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 protocol is installed. However it still uses it
for randomizing it's virtual placement.
So let's get rid of it in the presence of the RNG protocol.
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.
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Mark Kettenis <kettenis@openbsd.org>
Currently we allow and explicitly check a single shared page with
StandAloneMM. This is dictated by OP-TEE which runs the application.
However there's no way for us dynamically discover the number of pages we
are allowed to use. Since writing big EFI signature list variable
requires more than a page, OP-TEE has bumped the number of shared pages to
four.
Let's remove our explicit check and allow the request to reach OP-TEE even
if it's bigger than what it supports. There's no need to sanitize the
number of pages internally. OP-TEE will fail if we try to write more
than it's allowed. The error will just trigger later on, during the
StMM access.
While at it add an error message to help users figure out what failed.
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Tested-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Signed-off-by: Ilias Apalodimas <apalos@gmail.com>
When the TCG2 protocol is installed in efi_tcg2_register(),
TPM2 device must be present.
tcg2_measure_pe_image() expects that TCP2 protocol is installed
and TPM device is available. If TCG2 Protocol is installed but
TPM device is not found, tcg2_measure_pe_image() returns
EFI_SECURITY_VIOLATION and efi_load_image() ends with failure.
The same error handling is applied to
efi_tcg2_measure_efi_app_invocation().
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
There are functions that calls tcg2_agile_log_append() outside
of the TCG protocol invocation (e.g tcg2_measure_pe_image).
These functions must to check that TCG2 protocol is installed.
If not, measurement shall be skipped.
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
This commit modify efi_tcg2_register() to return the
appropriate error.
With this fix, sandbox will not boot because efi_tcg2_register()
fails due to some missing feature in GetCapabilities.
So disable sandbox if EFI_TCG2_PROTOCOL is enabled.
UEFI secure boot variable measurement is not directly related
to TCG2 protocol installation, tcg2_measure_secure_boot_variable()
is moved to the separate function.
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Firmwares before U-Boot may be capable of doing tpm measurements
and passing them to U-Boot in the form of eventlog. However there
may be scenarios where the firmwares don't have TPM driver and
are not capable of extending the measurements in the PCRs.
Based on TCG spec, if previous firnware has extended PCR's, PCR0
would not be 0. So, read the PCR0 to determine if the PCR's need
to be extended as eventlog is parsed or not.
Signed-off-by: Ruchika Gupta <ruchika.gupta@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Tested-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Platforms may have support to measure their initial firmware components
and pass the event log to u-boot. The event log address can be passed
in property tpm_event_log_addr and tpm_event_log_size of the tpm node.
Platforms may choose their own specific mechanism to do so. A weak
function is added to check if even log has been passed to u-boot
from earlier firmware components. If available, the eventlog is parsed
to check for its correctness and further event logs are appended to the
passed log.
Signed-off-by: Ruchika Gupta <ruchika.gupta@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Tested-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
If we call efi_clear_os_indications() before initializing the memory store
for UEFI variables a NULL pointer dereference occurs.
The error was observed on the sandbox with:
usb start
host bind 0 sandbox.img
load host 0:1 $kernel_addr_r helloworld.efi
bootefi $kernel_addr_r
Here efi_resister_disk() failed due to an error in the BTRFS implementation.
Move the logic to clear EFI_OS_INDICATIONS_FILE_CAPSULE_DELIVERY_SUPPORTED
to the rest of the capsule code.
If CONFIG_EFI_IGNORE_OSINDICATIONS=y, we should still clear the flag.
If OsIndications does not exist, we should not create it as it is owned by
the operating system.
Fixes: 149108a3eb ("efi_loader: clear OsIndications")
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Our algorithm for creating USB device paths may lead to duplicate device
paths which result in efi_disk_register() failing. Instead we should just
skip devices that cannot be registered as EFI block devices.
Fix a memory leak in efi_disk_add_dev() caused by the duplicate device
path.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
If we look at the path that bootm/booti take when preparing to boot the
OS, we see that as part of (or prior to calling do_bootm_states,
explicitly) the process, bootm_disable_interrupts() is called prior to
announce_and_cleanup() which is where udc_disconnect() /
board_quiesce_devices() / dm_remove_devices_flags() are called from. In
the EFI path, these are called afterwards. In efi_exit_boot_services()
however we have been calling bootm_disable_interrupts() after the above
functions, as part of ensuring that we disable interrupts as required
by the spec. However, bootm_disable_interrupts() is also where we go
and call usb_stop(). While this has been fine before, on the TI J721E
platform this leads us to an exception. This exception seems likely to
be the case that we're trying to stop devices that we have already
disabled clocks for. The most direct way to handle this particular
problem is to make EFI behave like the do_bootm_states() process and
ensure we call bootm_disable_interrupts() prior to ending up in
usb_stop().
Cc: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>
Cc: Simon Glass <sjg@chromium.org>
Suggested-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Due to U-Boot's lazy binding mentality the TPM is probed but not properly
initialized. The user can startup the device from the command line
e.g 'tpm2 startup TPM2_SU_CLEAR'. However we can initialize the TPM during
the TCG protocol installation, which is easier to use overall.
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
As described in the TCG spec [1] in sections 7.1.1 and 7.1.2 the FinalEvent
table should include events after GetEventLog has been called. This
currently works for us as long as the kernel is the only EFI application
calling that. Specifically we only implement what's described in 7.1.1.
So refactor the code a bit and support EFI application(s) calling
GetEventLog. Events will now be logged in both the EventLog and FinalEvent
table as long as ExitBootServices haven't been invoked.
[1] https://trustedcomputinggroup.org/wp-content/uploads/EFI-Protocol-Specification-rev13-160330final.pdf
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Implement the EFI_EVENT_GROUP_BEFORE_EXIT_BOOT_SERVICES event group
handling.
Add the definition of EFI_EVENT_GROUP_AFTER_READY_TO_BOOT.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Record capsule update result only if the capsule file is
successfully read, because the capsule GUID is not sure when
the file can not be read or the file is not a capsule.
Without this fix, if user puts a dummy (non-capsule) file
under (ESP)EFI/UpdateCapsule, U-Boot causes a synchronous
abort.
This also fixes use-after-free bug of the 'capsule' variable.
Fixes: c74cd8bd08 ("efi_loader: capsule: add capsule_on_disk support")
Signed-off-by: Masami Hiramatsu <masami.hiramatsu@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
log_info() is used for the debug level logging statement
which should use log_debug() instead. Convert it to reduce the
log output.
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
UEFI specification does not require that BootOrder is defined.
In current implementation, boot variable measurement fails and
returns EFI_NOT_FOUND if BootOrder is not defined.
This commit correcly handles this case, skip the boot variable
measurement if BootOrder is not defined.
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Most modern platforms use 64-bit EFI so it is useful to have a U-Boot app
that runs under that. Add a (non-functional) build for this.
Note that --whole-archive causes the gcc 9.2 linker to crash, so disable
this for now. Once this is resolved, things should work.
For now, avoid mentioning the documentation for the 64-bit app, since it
does not work.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
This commit adds the EFI_TCG2_PROTOCOL.SubmitCommand
required in the TCG PC Client PFP spec.
SubmitCommand enables to send the raw command to the TPM device.
To implement this api, tpm2_submit_command() is added
into tpm-v2.c.
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Calling unaligned/access-ok.h version of put_unaligned_le64()
causes data abort in arm 32-bit QEMU.
The similar issue also occurs in linux kernel,
unaligned/access-ok.h is no longer used in linux kernel[1].
This commit uses the unaligned/be_byteshift.h and
unaligned/le_byteshift.h helper instead of unaligned/access-ok.h.
[1]https://lore.kernel.org/all/20210507220813.365382-8-arnd@kernel.org/
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
As we discussed in ML, currently a device tree is the only place
to store public keys for capsule authentication. So __weak is not
necessary for now.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
This commit adds the DeployedMode and AuditMode variable
measurement required in TCG PC Client Platform Firmware
Profile Specification.
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
This commit adds the UEFI GPT disk partition topology
measurement required in TCG PC Client Platform Firmware
Profile Specification
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
TCG PC Client Platform Firmware Profile Specification
requires to measure the SMBIOS table that contains static
configuration information (e.g. Platform Manufacturer
Enterprise Number assigned by IANA, platform model number,
Vendor and Device IDs for each SMBIOS table).
The device- and environment-dependent information such as
serial number is cleared to zero or space character for
the measurement.
Existing smbios_string() function returns pointer to the string
with const qualifier, but exisintg use case is updating version
string and const qualifier must be removed.
This commit removes const qualifier from smbios_string()
return value and reuses to clear the strings for the measurement.
This commit also fixes the following compiler warning:
lib/smbios-parser.c:59:39: warning: cast to pointer from integer of
different size [-Wint-to-pointer-cast]
const struct smbios_header *header = (struct smbios_header *)entry->struct_table_address;
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Simplify efi_sigstore_parse_sigdb() by using existing functions.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
In multiple places we need the default GUID matching a variable name.
The patch provides a library function. For secure boot related variables
like 'PK', 'KEK', 'db' a lookup table is used. For all other variable
names EFI_GLOBAL_VARIABLE is returned.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
UEFI variable names are typically constants and hence should be defined as
const. Unfortunately some of our API functions do not define the parameters
for UEFI variable names as const. This requires unnecessary conversions.
Adjust parameters of several internal functions to tre UEFI variable names
as const.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
The commit 47a25e81d3 ("Revert "efi_capsule: Move signature from DTB to
.rodata"") failed to revert the removal of efi_get_public_key_data().
Add back this function and move it under lib/efi_loader so that other
platforms can utilize it. It is now declared as a weak function so that
it can be replaced with a platform-specific implementation.
Fixes: 47a25e81d3 ("Revert "efi_capsule: Move signature from DTB to
.rodata"")
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
This commit fixes the following compilation warning
of boottime->install_configuration_table() function.
lib/efi_selftest/efi_selftest_tcg2.c:475:46:
warning: passing argument 1 of ‘boottime->install_configuration_table’
discards ‘const’ qualifier from pointer target type [-Wdiscarded-qualifiers]
ret = boottime->install_configuration_table(&smbios_guid, dmi);
Signed-off-by: Masahisa Kojima <masahisa.kojima@linaro.org>
Reviewed-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
The 'efidebug boot dump' command should not display the VenMedia() device
path node preceding the device path of the initial ram disk.
By letting efi_dp_from_lo() skip the VenMedia() device path node we can
simplify the coding.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Create the GUID as a global variable.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Parameter size is never used in function efi_dp_from_lo(). Remove it.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Instead of copying a GUID and then using a pointer to the copy for calling
guidcmp(), just pass the pointer to the orginal GUID.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
We are ignoring the alignment communicated via the PE/COFF header.
Starting 5.10 the Linux kernel will loudly complain about it. For more
details look at [1] (in linux kernel).
So add a function that can allocate aligned EFI memory and use it for our
relocated loaded image.
[1] c32ac11da3f83 ("efi/libstub: arm64: Double check image alignment at entry")
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Tested-by: Vincent Stehlé <vincent.stehle@arm.com>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
When using a file to store UEFI variables we must make sure that secure
boot related variables are not loaded from this file. With commit
9ef82e2947 ("efi_loader: don't load signature database from file")
this has already been implemented for variables defined in the UEFI
specification. As most Linux distributions use Shim we should do the same
for Shim's MOK database.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
This code should never have been added as it builds a new feature on top
of legacy code. This has already been improved with the dependency on BLK.
Add a dependency on DM_ETH also, to avoid needing to deal with this old
code.
Boards which want EFI_LOADER should migrate to driver model first.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Convert pixel values when necessary like we do for 16bpp
framebuffers.
Signed-off-by: Mark Kettenis <kettenis@openbsd.org>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
At present EFI_SECURE BOOT selects RSA but does not necessarily enable
FIT_SIGNATURE. Mostly this is fine, but a few boards do not enable it,
so U-Boot tries to do RSA verification when loading FIT images, but it
is not enabled.
This worked because the condition for checking the RSA signature is
wrong in the fit_image_verify_with_data() function. In order to fix it
we need to fix this dependency. Make sure that FIT_SIGNATURE is enabled
so that RSA can be used.
It might be better to avoid using 'select' in this situation.
Signed-off-by: Simon Glass <sjg@chromium.org>
TCG EFI Protocol Specification defines the number_of_algorithms
field in spec ID event to be equal to the number of active
algorithms supported by the TPM device. In current implementation,
this field is populated with the count of all algorithms supported
by the TPM which leads to incorrect spec ID event creation.
Similarly, the algorithm array in spec ID event should be a variable
length array with length being equal to the number_of_algorithms field.
In current implementation this is defined as a fixed length array
which has been fixed.
Signed-off-by: Ruchika Gupta <ruchika.gupta@linaro.org>
CC: Masahisa Kojima <masahisa.kojima@linaro.org>
CC: Ilias Apalodimas <ilias.apalodimas@linaro.org>
CC: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
This was unfortunately applied despite much discussion about it beiong
the wrong way to implement this feature.
Revert it before too many other things are built on top of it.
This reverts commit ddf67daac3.
Signed-off-by: Simon Glass <sjg@chromium.org>
More C files do not use compile time timestamp macros and do not have to be
recompiled every time when SOURCE_DATE_EPOCH changes.
This patch moves version_string[] from version.h to version_string.h and
updates other C files which only needs version_string[] string to include
version_string.h instead of version.h. After applying this patch these
files are not recompiled every time when SOURCE_DATE_EPOCH changes.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
