When linking vboot_api_kernel4_tests, there are two VbBootNormal()
available, the gcc chooses the one in vboot_api_kernel4_tests.c and
the test passes, the clang chooses the one in vboot_api_kernel.c and
make the unittest fail. This CL makes the one in vboot_api_kernel.c
a weak symbol so that clang can choose the one in
vboot_api_kernel4_tests.c
BUG=chromium:498469
BRANCH=none
TEST=CC=x86_64-cros-linux-gnu-clang FEATURES='test'
emerge-amd64-generic vboot_reference
Change-Id: Ibcb78ee055fc9485dbc2bcc1d1cf98144a1a3b64
Reviewed-on: https://chromium-review.googlesource.com/276504
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Commit-Queue: Yunlian Jiang <yunlian@chromium.org>
Tested-by: Yunlian Jiang <yunlian@chromium.org>
This allows the caller to load the kernel partition and then pass it
to vboot for verification, rather than having vboot assume the kernel
partitions are all on a block storage device.
Next up, APIs for the caller to parse partition information from a GPT
(yes, that's cgptlib, but we'll make it more easily callable by
depthcharge).
BUG=chromium:487699
BRANCH=none
TEST=make -j runtests
Change-Id: I388085c7023f4c76d416f37df0607019bea844ac
Signed-off-by: Randall Spangler <rspangler@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/275646
Reviewed-by: Daisuke Nojiri <dnojiri@chromium.org>
This can be used by implementations that want to request vboot to
favor a particular kernel entry for booting without affecting the
checks for rollback protection and image verification.
CQ-DEPEND=CL:274716, CL:274932, CL:275171
BUG=None
BRANCH=None
TEST=Compiles successfully. make -j runtests successful.
Change-Id: I6a4600020354f5d4118c17f083c353c2585c4181
Signed-off-by: Furquan Shaikh <furquan@google.com>
Reviewed-on: https://chromium-review.googlesource.com/274558
Reviewed-by: Furquan Shaikh <furquan@chromium.org>
Reviewed-by: Stefan Reinauer <reinauer@chromium.org>
Tested-by: Nicolas Boichat <drinkcat@chromium.org>
Commit-Queue: Nicolas Boichat <drinkcat@chromium.org>
Trybot-Ready: Nicolas Boichat <drinkcat@chromium.org>
VbVerifyMemoryBootImage
Do not use values from the header or preamble until it is known to be
good.
BUG=None
BRANCH=None
TEST=Compiles successfully and VbVerifyMemoryBootImage returns early
for images with bad values in header.
Change-Id: Ic026f49292a139e0a04c2556ca9fa62ff277b18f
Signed-off-by: Furquan Shaikh <furquan@google.com>
Reviewed-on: https://chromium-review.googlesource.com/274141
Trybot-Ready: Furquan Shaikh <furquan@chromium.org>
Tested-by: Furquan Shaikh <furquan@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Commit-Queue: Furquan Shaikh <furquan@chromium.org>
This patch reintroduces a vb2_secdata->struct_version check similar to
the one that was removed in CL:244846. The CRC is not a reliable way to
detect zeroed buffers, so this check helps vboot fail earlier and more
clearly in certain situations.
BRANCH=kitty,smaug,storm,veyron
BUG=chrome-os-partner:40778
TEST=make runtests. Rebooted Jerry with 'mem w 0xff7601b0 0xfdb9', saw
that recovery reason was now 0x2b (VBNV_RECOVERY_VB2_SECDATA_INIT).
Change-Id: Ic4376d127e6d14d4ef9c2f53c83090040ca4cb68
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/274138
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Reviewed-by: Vadim Bendebury <vbendeb@chromium.org>
Add support for functions to request unlock and lock of devices in
response to fastboot oem unlock/lock commands. Unlock operation is
equivalent to enabling dev mode and lock operation is equivalent to
leaving dev mode. It is the responsibility of the caller to ensure
that user confirmation is obtained before unlock/lock operations.
BUG=chrome-os-partner:40196
BRANCH=None
TEST=Compiles successfully and fastboot lock/unlock operations work as
expected on smaug. Added tests to ensure lock/unlock operations are
covered. Verified using make -j runtests.
Change-Id: Ibafe75abdd1202473009208a414f3996d537db4f
Signed-off-by: Furquan Shaikh <furquan@google.com>
Reviewed-on: https://chromium-review.googlesource.com/273182
Reviewed-by: Furquan Shaikh <furquan@chromium.org>
Tested-by: Furquan Shaikh <furquan@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Commit-Queue: Furquan Shaikh <furquan@chromium.org>
Trybot-Ready: Furquan Shaikh <furquan@chromium.org>
When the lid is closed and external power is applied
the system may boot and shut down faster than required
for the OS to determine that things were alright.
In timed charging setups this led to systems ending up
to consider the current version broken because it "failed"
repeatedly.
Remain generic about the reason for not counting boots
since there may be more situations in which we want to
handle the situation optimistically.
BRANCH=none
BUG=chromium:446945
TEST=none
Change-Id: Iea350e3c98d5c00156da682e52c90a882ba017c0
Signed-off-by: Patrick Georgi <pgeorgi@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/249150
Reviewed-by: Randall Spangler <rspangler@chromium.org>
This API allows fastboot boot from memory command to verify that the
image loaded in memory is signed properly using recovery keys. Thus,
only officially signed recovery images can be booted using fastboot
boot command in recovery mode.
However, if GBB_FLAG_FORCE_DEV_BOOT_FASTBOOT_FULL_CAP is set, then
this routine will not perform any check and return okay for any image
sent by fastboot boot.
BUG=chrome-os-partner:40196
BRANCH=None
TEST=Compiles successfully. With GBB override for FASTBOOT_FULL_CAP
set any signed image is allowed to boot. With FASTBOOT_FULL_CAP not
set, then only officially signed image is allowed to boot. (make -j
runtests successful)
Change-Id: I78028853bd1ad09d3c610a687f327560557d5681
Signed-off-by: Furquan Shaikh <furquan@google.com>
Reviewed-on: https://chromium-review.googlesource.com/272696
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Commit-Queue: Furquan Shaikh <furquan@chromium.org>
Trybot-Ready: Furquan Shaikh <furquan@chromium.org>
Tested-by: Furquan Shaikh <furquan@chromium.org>
These are slightly more complex than the firmware versions, because
they need to deal with developer-signed keyblocks and keyblock flags.
BUG=chromium:487699
BRANCH=none
TEST=make -j runtests
Change-Id: I682c14ddfe729984f2629dfbe66750e5cd5ab75e
Signed-off-by: Randall Spangler <rspangler@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/272541
Reviewed-by: Daisuke Nojiri <dnojiri@chromium.org>
This is necessary for the next change, which adds keyblock hash checking.
Also clean up some other assorted comments, and move the diagnostic
check of root key to see if it's the checked-in one earlier in
firmware preamble validation so it's closer to where the root key is
loaded.
No functional or higher-level API changes; just shuffling around code
under the covers.
BUG=chromium:487699
BRANCH=none
TEST=make -j runtests
Change-Id: Ibc3960a4d882dc2ad8684e235db4b9d066eac080
Signed-off-by: Randall Spangler <rspangler@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/272223
Reviewed-by: Bill Richardson <wfrichar@chromium.org>
This also checks that the bootloader and vmlinuz headers, if present,
are within the signed part of the kernel blob; the vboot1 routines
didn't do that. That wasn't harmful at firmware boot time because the
vboot1 routines would only load as much data as was signed, but in
vboot2 loading the kernel data is the responsibility of the caller so
we need to check.
BUG=chromium:487699
BRANCH=none
TEST=make -j runtests
Change-Id: I73eb4831e5d3d7a642b6cb85cb55857d87fcc0af
Signed-off-by: Randall Spangler <rspangler@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/270797
Add a new flag to nvstorage for controlling fastboot capabilities
offered in firmware in dev-mode. By default, value of this flag would
be ignored in normal mode. Thus, when fastboot-based recovery is
entered from normal mode, only limited capability would be available
in firmware.
After switching to dev-mode, this flag can be set automatically by
user script after performing the wipe or it can be set manually using
crossystem. When fastboot-based recovery is entered from dev mode and
this flag is set, it will provide full fastboot capability in the
firmware.
BUG=chrome-os-partner:40196
BRANCH=None
TEST=Compiles successfully for smaug. make runalltests successful.
Change-Id: I761a9ab304dd90f0b73081acc9ce1f8d9052325f
Signed-off-by: Furquan Shaikh <furquan@google.com>
Reviewed-on: https://chromium-review.googlesource.com/271369
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Commit-Queue: Furquan Shaikh <furquan@chromium.org>
Trybot-Ready: Furquan Shaikh <furquan@chromium.org>
Tested-by: Furquan Shaikh <furquan@chromium.org>
Looks like the DISABLE_FW_ROLLBACK_CHECK GBB flag (0x200) was forgotten
in the vboot2 implementation. It's too late for Veyron now, but let's at
least fix it for future devices.
BRANCH=none
BUG=None
TEST=make runtests
Change-Id: I867f7aada28be3897efda73a6bdc3b0848c23dca
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/271419
Reviewed-by: Bill Richardson <wfrichar@chromium.org>
Older GBB headers (e.g. 1.0 and 1.1) do not have hwid_digest. In such cases,
PCR1 is currently extended from 0, causing a remote attestation failure.
This change makes all GBB headers older than 1.2 incompatible.
BUG=none
BRANCH=tot
TEST=make -j runtests
Change-Id: I7a3b19c2da325a3fa4b9c1fe06ed6f43cb51fb9e
Signed-off-by: Daisuke Nojiri <dnojiri@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/270796
Reviewed-by: Bill Richardson <wfrichar@chromium.org>
This patch fixes what I think is an inconsistency in the existing legacy
boot behavior: when the GBB flag that defaults to legacy boot is set,
running out the 30 second timer would still boot legacy mode even if
dev_boot_legacy is not actually set (whereas pressing CTRL+L in the
same configuration would beep and refuse).
This patch makes both legacy boot trgiggers check the same condition
before boot. This does not restrict functionality since anyone who sets
the DEFAULT_DEV_BOOT_LEGACY GBB flag could simply set
FORCE_DEV_BOOT_LEGACY at the same time. It does however open up an
interesting new use case of using NVRAM to change back-and-forth between
legacy and normal developer mode (after GBB flags are changed once and
write-protection is enabled again).
If this is updated in the field it might lock existing devices out of
legacy mode... however, since by far the most common GBB flag
combination recommended on the internet seems to be 0x489 (including
FORCE_DEV_BOOT_LEGACY), I doubt this would be a problem in practice.
BRANCH=tbd
BUG=chrome-os-partner:39999
TEST=Booted with GBB flags 0x4b9 and 0x439, observed difference.
Change-Id: If6a6d99ab2cf116db2237fdc3df97fc22a68251c
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/270182
Reviewed-by: Duncan Laurie <dlaurie@chromium.org>
Even though legacy boot is an unsafe mode that has to be manually
initiated by the user, we should still lock the kernel TPM space to be
consistent with existing developer mode practice.
BRANCH=tbd
BUG=chrome-os-partner:39999
TEST=Spent over an hour unsuccessfully trying to get SeaBIOS to boot a
Chromium test image on my Falco. Decided that's not worth it an just
tested the firmware side of this (pressing CTRL+L when legacy mode is
enabled and disabled, multiple times, with and without GBB flag
DEFAULT_DEV_BOOT_LEGACY).
Change-Id: I3b02b59a9055431d222c0c7446de2cd7d2e0bb82
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/270181
Reviewed-by: Duncan Laurie <dlaurie@chromium.org>
And add a vb2_digest_buffer() call which produces the hash of a buffer
all in a single function call. That function actually already
existed, but was in a unit test file rather than in the library
itself. It's a small function, so adding it won't increase the size
of the library significantly - or at all, on platforms which compile
with -ffunction-sections.
This allows coreboot to reuse this SHA library for hashing CBFS
entries and file data. All it has to do is #define
NEED_VB2_SHA_LIBRARY and then #include "vb2_api.h".
BUG=chromium:482652
BRANCH=none
TEST=make -j runtests
Change-Id: Ice2d0929324b58b2665f3989b5b887225f6ef61e
Signed-off-by: Randall Spangler <rspangler@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/269523
Reviewed-by: Julius Werner <jwerner@chromium.org>
This is done to break a circular DEPENDency as we want to
send UMA stats from tcsd. Without this, metrics depends on
vboot_reference which depends on trousers which depends on
metrics. Technically the vboot_reference dependency on trousers
is header-file only, but we can't cope with that.
BUG=chromium:481552
TEST=compiled with emerge-<something> vboot_reference
BRANCH=none
Change-Id: Iea5c0c39bb70977c9d375e63ea607687debe9f9f
Reviewed-on: https://chromium-review.googlesource.com/267744
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Commit-Queue: Luigi Semenzato <semenzato@chromium.org>
Tested-by: Luigi Semenzato <semenzato@chromium.org>
When a read fails in getting the GPT, just zero the contents of the
buffer and carry on.
Some testing changes are required for this. When a read of the GPT
fails, it is no longer fatal, so tests of that have been adjusted.
Tests have been improved to show that the GPT is automatically
repaired when a read error occurs.
There was one test which checked that a zero-sized disk would fail
to load a kernel, but it was surrounded by a number of mocked
functions which normally do that error checking, and it amounted
to the same test as read failure; that test was deleted.
BUG=chrome-os-partner:35440
TEST=vboot tests pass
BRANCH=none
Change-Id: I0c05813e7492920433733947d3fb74a7e4aa66f2
Signed-off-by: Dan Ehrenberg <dehrenberg@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/266882
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Recent experience shows that users often get confused and try running
pre-mp signed images under dev firmware control and vice versa. The
matters are further aggravated by the fact that the signage mismatch
is allowed when the device is in dev mode and not in normal mode.
While the users usually can tell what class of keys the Chrome OS
image is signed with, it is much mode difficult to tell what keys the
firmware was signed with.
This patch, reports in the log if the firmware was signed with dev
keys, by comparing the hash calculated over the packed root public key
body with a precompiled value.
A test tweak was required to avoid using uninitialized data.
BRANCH=none
BUG=none
TEST=booted the new code on storm, observed the following message
included in the log:
VB2:vb2_report_key_class() This is developer signed firmware
- verified that 'make run2tests' succeeds in chroot
Change-Id: I97ed6ba384cee59ff3f42943630e92ebae10dd03
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/264469
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Some Chrome OS devices do not allow to login even in developer mode,
as they do not have display/keyboard and sshd is not part of the
Chrome OS image. Even enabling developer mode on those devices is very
involved (requires taking the device apart and is guaranteed to take
long time).
We still want to allow the end user to control those devices in dev
mode. The solution is enabling the ability to boot from the USB stick
when the device transitions from normal to developer mode.
A simple way to do it is to set the NVRAM flag, which allows USB boot.
The flag is set on normal=>dev transition only, and only on those
devices where it is configured (as discovered by invoking
VbExGetSwitches with the appropriate parameters).
BRANCH=storm
BUG=chrome-os-partner:38303
TEST=tested with the corresponding depthcharge patches
Change-Id: I5fa58963256598cde3b534f5250101fba6042f8c
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/264187
Reviewed-by: Hung-Te Lin <hungte@chromium.org>
Reviewed-by: Bill Richardson <wfrichar@chromium.org>
vboot currently uses the |SHA256_CTX| name, which is claimed by OpenSSL.
To work around this, it defines OPENSSL_NO_SHA, but that can't be done
at compile time:
The OPENSSL_NO_* defines are set by OpenSSL to reflect the configuration
that it was built with so that users of OpenSSL can disable features as
needed. They can affect the contents of structures any thus the ABI of
the library.
If these defines are set outside of OpenSSL, then the library and the
code that uses it will have incompatible ABIs. At that point it's only
functioning by blind luck.
This change renames the name-collisions so that this hack isn't needed.
This is the same change as was made internally in cl/85758149.
BUG=none
BRANCH=none
TEST=emerge-samus coreboot; make runtests
Change-Id: I709da2507f341896d89d50129ce30ffb111a20d1
Signed-off-by: Bill Richardson <wfrichar@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/263506
Reviewed-by: Randall Spangler <rspangler@chromium.org>
For test purposes it should be possible to clear the wipeout request
raised by firmware.
BRANCH=none
BUG=chrome-os-partner:36059
TEST=verified that crossystem wipeout_request=0 changes the bit from 1
to 0, and wipeout_request=1 does not change it from 0 to 1.
Change-Id: Ic45ec03ed3e40e6fee4244804b8c231ee88af95b
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/262466
Reviewed-by: Randall Spangler <rspangler@chromium.org>
If so desired by the firmware, disable developer mode each time the
recovery mode is entered.
BRANCH=storm
BUG=chrome-os-partner:36059
TEST=with the rest of the patches applied observed desired behavior on
an SP5 (developer mode state wiped out on entering recovery)
Change-Id: If08dc517363bcc36fcc8b0b875a8700bbcefde4c
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/261630
Reviewed-by: Randall Spangler <rspangler@chromium.org>
It has become necessary to be able to "factory reset" certain devices
on firmware request. The best mechanism for this is NVRAM, as the
request needs to be detected very early in the boot process, before
other means of communications with the upper layers are available.
A previously unused NVRAM bit (bit 0x08 at offset zero) is taken for
this purpose.
A new flag is introduced to allow the firmware to signal the need to
assert this bit.
A new variable name/parameter ('wipeout_request') added to crossystem
to provide user space access to the setting of the dedicated NVRAM
bit.
BRANCH=storm
BUG=chrome-os-partner:37219
TEST=with all the patches applied, on storm, holding the recovery
button at startup for 10 seconds, causes 'crossystem
wipeout_request' to report '1'.
Change-Id: If1f6f061ce5b3f357b92aaa74cb129671dc30446
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/259857
Reviewed-by: Bill Richardson <wfrichar@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
We want a quick and human-friendly way to match keys with
signatures, so we decided to give each key a unique GUID and
carry that ID around when signing things.
But then we realized that we could autogenerate a unique
identifier from the .pem file itself, which is even better
because then we can match our binary keypair structs with the
openssl file used to generate them.
This change replaces the GUID id with a sha1sum calculated from
the public key's "keyb" blob.
BUG=none
BRANCH=none
TEST=make runtests
Also:
futility show tests/testkeys/key_rsa4096.pem
futility create tests/testkeys/key_rsa4096.pem foo
futility show foo.vbp*
Note that the GUID is the same for all files.
Change-Id: Ie44e46c83433718b1ff0163c1e7c51ec331b99f9
Signed-off-by: Bill Richardson <wfrichar@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/256181
Reviewed-by: Randall Spangler <rspangler@chromium.org>
This command reads a single .pem file and emits the public and
private keys generated from it. It can produce both the old-style
vboot 1.0 keys (.vbpubk and .vbprivk), or the new vboot 2.1
format keys (.vbpubk2 and .vbprik2). The default is the new
format, but you can give futility the --vb1 arg to force the old
format.
A test is included.
BUG=chromium:231547
BRANCH=ToT
TEST=make runtests
Change-Id: I4713dc5bf34151052870f88ba52ddccf9d4dab50
Signed-off-by: Bill Richardson <wfrichar@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/246766
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Kernel preamble flags are set by the signer for passing hints about
the image. Read these flags from the preamble and pass it back to the
caller in kparams structure.
BUG=chrome-os-partner:35861
BRANCH=None
TEST=Compiles and boots to kernel prompt for both CrOS image and bootimg.
Change-Id: I07a8b974dcf3ab5cd93d26a752c989d268c8da99
Signed-off-by: Furquan Shaikh <furquan@google.com>
Reviewed-on: https://chromium-review.googlesource.com/245951
Reviewed-by: Bill Richardson <wfrichar@chromium.org>
Tested-by: Furquan Shaikh <furquan@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Commit-Queue: Furquan Shaikh <furquan@chromium.org>
1. Increase kernel preamble revision from 2.1 to 2.2.
2. Add flags field to kernel preamble.
3. Update futility to accept flags parameter for vbutil_kernel and
cmd_sign for kernel.
4. Pass in an extra flags field to SignKernelBlob and
CreateKernelPreamble.
BUG=chrome-os-partner:35861
BRANCH=None
TEST=1) "make runalltests" completes successfully. 2) vboot_reference
compiles successfully for ryu. 3) Verified flags field in header using
futility show.
Change-Id: If9f06f98778a7339194c77090cbef4807d5e34e2
Signed-off-by: Furquan Shaikh <furquan@google.com>
Reviewed-on: https://chromium-review.googlesource.com/245950
Tested-by: Furquan Shaikh <furquan@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Commit-Queue: Furquan Shaikh <furquan@chromium.org>
vboot1 kept track of an internal "LoadFirmware() check" value for both
firmware slots and encoded the value for the slot that managed to go
further in the verification flow into a special range of recovery
reasons. vboot2 instead uses the generic "invalid RW" reason for all
firmware verification failures and communicates further information
through the subcode.
While the subcode may be good enough for developers, it's difficult to
communicate failure reasons to "normal" users (like non-firmware
developers) on the TAB screen. Currently we just display a couple of
numbers that people won't know how to interpret and "RW firmware failed
signature check" for any verification error (including rollback, which
might be the most commonly encountered in practice).
Since our recovery reason space is big enough (and we don't reuse old
numbers anyway), we might as well reuse the more precise numbers (and
strings) from vboot1 to communicate the failure reason, even if we don't
implement its "which slot came further" algorithm. This patch translates
the most common/useful VBSD_LF_CHECK numbers into plain VB2_RECOVERY
reasons and uses them where appropriate.
CQ-DEPEND=CL:248400
BRANCH=veyron
BUG=None
TEST=make runtests VBOOT2=1
test_that my_jerry firmware_CorruptBothFwSigAB
firmware_CorruptBothFwBodyAB firmware_RollbackFirmware
(Confirmed that matched recovery reasons are the more precise ones in
the 0x10-0x1F range.)
Change-Id: I51ecf1b820d1faa40405cb84377380d6f3f6ca1d
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/248392
Reviewed-by: Bill Richardson <wfrichar@chromium.org>
vboot2 added a few new recovery reasons (and abolished many old ones).
In the current vboot2/vboot1 hybrid architecture used on Veyron, the
vboot1 kernel verification part controls the status display when
pressing the TAB key, which may try to show recovery reasons set by the
vboot2 firmware verification part. These currently result in the not
very helpful "We have no idea what this means", so lets hack a few more
strings into vboot1 which will be otherwise harmless. Also add the
recovery_subcode field to the display, which is used much more
extensively by vboot2 and often very useful in firguring out what really
went wrong.
BRANCH=veyron
BUG=None
TEST=Manually set a few recovery reasons and subcodes through crossystem
and made sure they get displayed correctly on my Jerry.
Change-Id: I3f3e6c6ae6e7981337841c0c5e3cd767628472c3
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/248391
Reviewed-by: Bill Richardson <wfrichar@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
The length of the signature is 8 bytes. We've been checking 9
bytes instead, pretty much forever. All the tests have passed
because although the signature we're looking for is an 8-byte
string followed by a '\0', the next field in the header contains
the revision number 0x00010000, so the 9th byte is always zero.
We should follow the spec, though.
BUG=none
BRANCH=none
TEST=make runtests
Change-Id: I7cc6370250fa36a193f4a9fa5bc0099aea465618
Signed-off-by: Bill Richardson <wfrichar@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/247331
Reviewed-by: Randall Spangler <rspangler@chromium.org>
This patch adds a check to vboot2 secdata accessor functions that
returns an error if vb2_secdata_init() has not yet been called or
failed for some reason. This avoids a problem where vboot may
misinterpret random garbage (e.g. from transient read failures) as
valid secdata in recovery mode and write it back to the TPM (bricking
the device in a way that requires manual repair).
Also removes VB2_ERROR_SECDATA_VERSION check. This check was not
terribly useful since there should be no way a vboot2 device could ever
have secdata version 1 (and if it did, it should still fail CRC checks).
This error can trigger for cases when secdata contains random garbage
(e.g. all zeroes) and prevent the much more appropriate
VB2_ERROR_SECDATA_CRC error from even being checked for, which just
creates confusion and makes it harder to determine the real problem.
BRANCH=veyron
BUG=chrome-os-partner:34871
TEST=Emulated TPM read errors by just manually memset()ing secdata to 0
in coreboot, verified that vboot does not write back to the TPM and the
device will start working fine again once the disruption is removed.
Change-Id: I76bcbdbcd8106a0d34717cc91a8f2d7cda303c3f
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/244846
This patchs adds a new vb2_shared_data field to store the current
rollback prevention version number stored in secdata (TPM). This
information needs to be retrieved from there by coreboot (current
hack) or vboot2 kernel verification (bright shiny future) so it can be
passed along to the operating system and user space.
BRANCH=veyron
BUG=chrome-os-partner:35941
TEST=make runtests. Booted Jerry in recovery mode (with corresponding
coreboot patch), ensured that crossystem tpm_fwver still shows the
correct value.
Change-Id: I2a0c3e51b158a35ac129d2abce19b40c6c6381a6
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/244601
Reviewed-by: Randall Spangler <rspangler@chromium.org>
this api allows firmware to get the digest indicating boot mode status.
BUG=chromium:451609
TEST=VBOOT2=1 make run2tests
BRANCH=tot
Signed-off-by: Daisuke Nojiri <dnojiri@chromium.org>
Change-Id: Idca7bc5f6aed947689ad7cf219805aad35047c7d
Reviewed-on: https://chromium-review.googlesource.com/244542
This accurately reflects what's really happening. Vboot 2.0 is
backwards-compatible with the binary structs used in vboot 1.0,
while vboot 2.1 will not be.
When building firmware, vboot_reference should be invoked in one
of three ways:
TARGET OUTPUT VERSION
fwlib vboot_fw.a 1.0
fwlib20 vboot_fw20.a 2.0
fwlib21 vboot_fw21.a 2.1
BUG=chromium:228932
BRANCH=ToT
CQ-DEPEND=CL:243981
TEST=manual
emerge-veyron_pinky coreboot
emerge-samus coreboot
emerge-daisy_spring chromeos-u-boot
make runtests
Change-Id: I98d8ea6b48e5922a470e744d56699cad43eabb3d
Signed-off-by: Bill Richardson <wfrichar@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/243980
Reviewed-by: Randall Spangler <rspangler@chromium.org>
We were assuming 8-byte alignment for buffers. That's not true on
32-bit architectures. We should make the alignment requirements
explicit (and correct) for all architectures.
BUG=chromium:452179
BRANCH=ToT
CQ-DEPEND=CL:243380
TEST=manual
USE=vboot2 FEATURES=test emerge-x86-alex vboot_reference
Change-Id: I120f23e9c5312d7c21ff9ebb6eea2bac1e430e37
Signed-off-by: Bill Richardson <wfrichar@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/243362
Reviewed-by: Randall Spangler <rspangler@chromium.org>
