vboot_kernel: Validate GPT header before using
In AllocAndReadGptData() the code was changed to use the GPT header
to determine the LBA of the GPT entries. This change did not account
for devices that have an invalid header and it can attempt to read
from invalid block addresses on a device.
This commit happened here:
a2d72f7 vboot: cgpt: Refer to partition entries by entries_lba.
The subsequent steps in vboot, LoadKernel->GptInit->GptRepair will
fix a missing header and entries, so it is only necessary for one of
the headers to be valid.
This is commonly the case with a new USB stick that has an image
written to it as only the primary header will be valid in this case.
However it is also true if the primary header has been corrupted and
the secondary header is still valid.
The code has been changed to call CheckHeader() on the primary and
secondary headers before attempting to use the 'entries_lba' field
to read the entries from the device. AllocAndReadGptData() now only
fails if both headers are invalid.
A number of new unit tests are created to check for these failure
conditions. In order to support this I had to extend the vboot_kernel
test infrastructure to have a buffer for the mocked disk data instead
of just ignoring reads and writes. This is because many of the existing
tests assumed they could have an invalid GPT header and still pass.
Now that the header is checked it is necessary for a valid header to
be created before the tests can pass.
TEST=All unit tests pass when running 'make runtests'
In addition real-world testing was done by corrupting the primary
and/or secondary headers of USB stick to ensure that it will
successfully boot if one of the headers is valid.
Signed-off-by: Duncan Laurie <firstname.lastname@example.org>
Reviewed-by: Bill Richardson <email@example.com>
Reviewed-by: Nam Nguyen <firstname.lastname@example.org>
2 files changed