The calculation of the size of the table was incorrect (copy/pasta from
grub_acpi_rsdt_find_table() I assume...). The entries are 64-bit long.
This causes us to access beyond the end of the table which is causing
crashes during boot on some systems. Typically this is causing a crash
on VMWare when using UEFI and enabling serial autodetection, as
grub_acpi_find_table (GRUB_ACPI_SPCR_SIGNATURE);
will goes past the end of the table (the SPCR table doesn't exits).
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Reviewed-by: Ross Philipson <ross.philipson@oracle.com>
Tested-by: Renata Ravanelli <rravanel@redhat.com>
For NX we need to set write and executable permissions on the sections
of GRUB modules when we load them. All allocatable sections are marked
readable. In addition:
- SHF_WRITE sections are marked as writable,
- and SHF_EXECINSTR sections are marked as executable.
Where relevant for the platform the tramp and GOT areas are marked non-writable.
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Jan Setje-Eilers <jan.setjeeilers@oracle.com>
Signed-off-by: Mate Kukri <mate.kukri@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
For NX we need to set the page access permission attributes for write
and execute permissions. This patch adds two new primitives, grub_set_mem_attrs()
and grub_clear_mem_attrs(), and associated constants definitions used
for that purpose. For most platforms it adds a dummy implementation.
On EFI platforms it implements the primitives using the EFI Memory
Attribute Protocol, defined in UEFI 2.10 specification.
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Jan Setje-Eilers <jan.setjeeilers@oracle.com>
Signed-off-by: Mate Kukri <mate.kukri@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Currently we load module sections at whatever alignment gcc+ld happened
to dump into the ELF section header which is often less then the page
size. Since NX protections are page based this alignment must be rounded
up to page size on platforms supporting NX protections. This patch
switches EFI platforms to load module sections at 4 KiB page-aligned
addresses. It then changes the allocation size computation and the
loader code in grub_dl_load_segments() to align the locations and sizes
up to these boundaries and fills any added padding with zeros. All of
this happens before relocations are applied, so the relocations factor
that in with no change.
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Jan Setje-Eilers <jan.setjeeilers@oracle.com>
Signed-off-by: Mate Kukri <mate.kukri@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Currently when loading GRUB modules we allocate space for all sections
including those without SHF_ALLOC set. We then copy the sections that
/do/ have SHF_ALLOC set into the allocated memory leaving some of our
allocation untouched forever. Additionally, on platforms with GOT fixups
and trampolines we currently compute alignment round-ups for the
sections and sections with sh_size = 0. This patch removes the extra
space from the allocation computation and makes the allocation
computation loop skip empty sections as the loading loop does.
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Jan Setje-Eilers <jan.setjeeilers@oracle.com>
Signed-off-by: Mate Kukri <mate.kukri@canonical.com>
Reviewed-By: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Currently GRUB modules built with Clang or GCC have several sections
which we don't actually need or support. We already have a list of
sections to skip in genmod.sh and this patch adds the following
sections to that list (as well as a few newlines):
- .note.gnu.property
- .llvm*
Note that the glob there won't work without a new enough linker but the
failure is just reversion to the status quo. So, that's not a big problem.
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Jan Setje-Eilers <jan.setjeeilers@oracle.com>
Signed-off-by: Mate Kukri <mate.kukri@canonical.com>
Reviewed-By: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
This fixes naming inconsistency that goes against coding style as well
as helps to avoid potential conflicts and confusion as this constant is
used in multiple places.
Signed-off-by: Daniel Kiper <daniel.kiper@oracle.com>
Signed-off-by: Sergii Dmytruk <sergii.dmytruk@3mdeb.com>
Reviewed-by: Ross Philipson <ross.philipson@oracle.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Currently rdmsr and wrmsr commands have own MSR support detection code.
This code is the same. So, it is duplicated. Additionally, this code
cannot be reused by others. Hence, extract this code to a function and
make it public. By the way, improve a code a bit.
Additionally, use GRUB_ERR_BAD_DEVICE instead of GRUB_ERR_BUG to signal
an error because errors encountered by this new routine are not bugs.
Signed-off-by: Daniel Kiper <daniel.kiper@oracle.com>
Signed-off-by: Sergii Dmytruk <sergii.dmytruk@3mdeb.com>
Reviewed-by: Ross Philipson <ross.philipson@oracle.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The loopback image is configured to function as a disk by being mapped
as a block device. Instead of measuring the entire block device we
should focus on tracking the individual files accessed from it. For
example, we do not directly measure block devices like hd0 disk but the
files opened from it.
This method is important to avoid running out of memory since loopback
images can be very large. Trying to read and measure the whole image at
once could cause out of memory errors and disrupt the boot process.
Signed-off-by: Michael Chang <mchang@suse.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Similarly to the issue described in commit c52ae4057 (efinet: skip
virtual IPv4 and IPv6 devices during card enumeration) the UEFI PXE
driver creates additional VLAN child devices when a VLAN ID is
configured on a network interface associated with a physical NIC. These
virtual VLAN devices must be skipped during card enumeration to ensure
that the subsequent SNP exclusive open operation targets the correct
physical card instances. Otherwise packet transfer would fail.
A device path example with VLAN nodes:
/MAC(123456789ABC,0x1)/Vlan(20)/IPv4(0.0.0.0,0x0,DHCP,0.0.0.0,0.0.0.0,0.0.0.0)
Signed-off-by: Michael Chang <mchang@suse.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
A regression in GRUB 2.12 causes the GRUB screen to become cluttered
with artifacts from the previous screen whether it's the UEFI post UI,
UEFI shell or any graphical UI running before GRUB. This issue occurs
in situations like booting GRUB from the UEFI shell and going straight
to the rescue or command shell causing visual discomfort.
The regression was introduced by commit 2d7c3abd8 (efi/console: Do not
set text-mode until it is actually needed). To address the screen
flickering issue this commit suppresses the text-mode setting until the
first output is requested. Before text-mode is set any attempt to clear
the screen has no effect. This inactive period renders the clear screen
ineffective in early boot stages, potentially leaving leftover artifacts
that will clutter the GRUB console display, as there is no guarantee
there will always be a clear screen after the first output.
The issue is fixed by ensuring grub_console_cls() to work through lazy
mode-setting, while also avoiding screen clearing for the hidden menu
which the flicker-free patch aims to improve.
Fixes: 2d7c3abd8 (efi/console: Do not set text-mode until we actually need it)
Signed-off-by: Michael Chang <mchang@suse.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The cycle register is not guaranteed to count at constant frequency.
If it is counting at all depends on the state the performance monitoring
unit. Use the time register to measure time.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Otherwise it breaks the decompressors for MIPS targets.
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
BCJ is not available for all platforms hence arguments may end up unused.
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Otherwise depending on compiler we end up with umoddi3 reference and
failed module dependency resolution.
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The get_part_uuid() function made an assumption that the target GRUB
device is a partition device and accessed device->disk->partition
without checking for NULL. There are four situations where this
assumption is problematic:
1. The device is a net device instead of a disk.
2. The device is an abstraction device, like LVM, RAID, or CRYPTO, which
is mostly logical "disk" ((lvmid/<UUID>) and so on).
3. Firmware RAID may present the ESP to GRUB as an EFI disk (hd0) device
if it is contained within a Linux software RAID.
4. When booting from a CD-ROM, the ESP is a VFAT image indexed by the El
Torito boot catalog. The boot device is set to (cd0), corresponding
to the CD-ROM image mounted as an ISO 9660 filesystem.
As a result, get_part_uuid() could lead to a NULL pointer dereference
and trigger a synchronous exception during boot if the ESP falls into
one of these categories. This patch fixes the problem by adding the
necessary checks to handle cases where the ESP is not a partition device.
Additionally, to avoid disrupting the boot process, this patch relaxes
the severity of the errors in this context to non-critical. Errors will
be logged, but they will not prevent the boot process from continuing.
Fixes: e0fa7dc84 (bli: Add a module for the Boot Loader Interface)
Signed-off-by: Michael Chang <mchang@suse.com>
Reviewed-By: Oliver Steffen <osteffen@redhat.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The fdtdump command allows dumping arbitrary device tree properties
and saving them to a variable similar to the smbios command.
This is useful in scripts where further actions such as selecting
a kernel or loading another device tree depend on the compatible
or model values of the device tree provided by the firmware.
For now only the root level properties of the dtb are exposed.
Signed-off-by: Tobias Heider <tobias.heider@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
First they're use macros so they can't be translated as-is.
Second there is no point in translating them as they're too technical.
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Silently keeping entries in the list if the address matches, but the
page count doesn't is a bad idea, and can lead to double frees.
grub_efi_free_pages() have already freed parts of this block by this
point, and thus keeping the whole block in the list and freeing it again
at exit can lead to double frees.
Signed-off-by: Mate Kukri <mate.kukri@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
If the firmware happens to return 0 as an address of allocated pages,
grub_efi_allocate_pages_real() tries to allocate a new set of pages,
and then free the ones at address 0.
However at that point grub_efi_store_alloc() wasn't yet called, so
freeing the pages at 0 using grub_efi_free_pages() which calls
grub_efi_drop_alloc() isn't necessary, so let's call b->free_pages()
instead.
The call to grub_efi_drop_alloc() doesn't seem particularly harmful,
because it seems to do nothing if the allocation it is asked to drop
isn't on the list, but the call to it is obviously unnecessary here.
Signed-off-by: Mate Kukri <mate.kukri@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
If the map was too big for the initial allocation, it was freed and replaced
with a bigger one, but the free call still used the hard-coded size.
Seems like this wasn't hit for a long time, because most firmware maps
fit into 12K.
This bug was triggered on Project Mu firmware with a big memory map, and
results in the heap getting trashed and the firmware ASSERTING on
corrupted heap guard values when GRUB exits.
Signed-off-by: Mate Kukri <mate.kukri@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Type 0x01 was introduced with the ACPI DBGP table and type 0x12 was introduced
with the ACPI DBG2 table. Type 0x12 is used by the ACPI SPCR table on recent
AWS bare-metal instances (c6i/c7i). Also give each debug type a proper name.
Signed-off-by: Udo Steinberg <udo@hypervisor.org>
Reviewed-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Current code in some codepaths neither discards nor reports errors.
Properly surface the error.
While on it split 2 cases of unrelated variables both named err.
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
When GRUB image is netbooted on ppc64le, the keyboard input exhibits
significant latency, reports even say that characters are processed
about once per second. This issue makes interactively trying to debug
a ppc64le config very difficult.
It seems that the latency is largely caused by a 200 ms timeout in the
idle event loop, during which the network card interface is consistently
polled for incoming packets. Often, no packets arrive during this
period, so the timeout nearly always expires, which blocks the response
to key inputs.
Furthermore, this 200 ms timeout might not need to be enforced at this
basic layer, considering that GRUB performs synchronous reads and its
timeout management is actually handled by higher layers, not directly in
the card instance. Additionally, the idle polling, which reacts to
unsolicited packets like ICMP and SLAAC, would be fine at a less frequent
polling interval, rather than needing a timeout for receiving a response.
For these reasons, we believe the timeout in get_card_packet() should be
effectively removed. According to test results, the delay has disappeared,
and it is now much easier to use interactively.
Signed-Off-by: Michael Chang <mchang@suse.com>
Tested-by: Tony Jones <tonyj@suse.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The measurements for confidential computing has been introduced in the
commit 4c76565b6 (efi/tpm: Add EFI_CC_MEASUREMENT_PROTOCOL support).
Recently the patch 30708dfe3 (tpm: Disable the tpm verifier if the TPM
device is not present) has been introduced to optimize the memory usage
when a TPM device is not available on platforms. This fix prevents the
tpm module to be loaded on confidential computing platforms, e.g. Intel
machines with TDX enabled, where the TPM device is not available.
In this patch, we propose to load the tpm module for this use case by
generalizing the tpm feature detection in order to cover CC platforms.
Basically, we do it by detecting the availability of the
EFI_CC_MEASUREMENT_PROTOCOL EFI protocol.
Fixes: https://savannah.gnu.org/bugs/?65821
Fixes: 30708dfe3 (tpm: Disable the tpm verifier if the TPM device is not present)
Signed-off-by: Hector Cao <hector.cao@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Reviewed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
This patch adds support for Radix, Xive and Radix_gtse in Options
vector5 which is required for KVM LPARs. KVM LPARs ONLY support
Radix and not the Hash. Not enabling Radix on any PowerVM KVM LPARs
will result in boot failure.
Signed-off-by: Avnish Chouhan <avnish@linux.ibm.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
We don't need any actual adjustments as we don't use the affected structures.
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
ARRAY_SIZE() is the count of elements, but the element size is 4 bytes, so
this was only initing the first 1/4th of the table. Detected with valgrind.
This should only matter in error paths, and I've not been able to identify
any actual misbehaviour that results from reading in-bounds but uninited data.
Signed-off-by: Daniel Axtens <dja@axtens.net>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
This isn't intended to be a functional change, but it makes a lot of failures a lot
faster, which is extremely helpful for fuzzing.
Without this change, we keep trying and trying to read more bytes into our buffer,
never being able to (read always returns 0) and so we just return old buffer contents
over and over until the decompression process fails some other way.
Signed-off-by: Daniel Axtens <dja@axtens.net>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Add functionality to disable command line interface access and editing of GRUB
menu entries if GRUB image is built with --disable-cli.
Signed-off-by: Alec Brown <alec.r.brown@oracle.com>
Reviewed-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The EROFS [1] is a lightweight read-only filesystem designed for performance
which has already been shipped in most Linux distributions as well as widely
used in several scenarios, such as Android system partitions, container
images and rootfs for embedded devices.
This patch brings in the EROFS uncompressed support. Now, it's possible to
boot directly through GRUB with an EROFS rootfs.
Support for the EROFS compressed files will be added later.
[1] https://erofs.docs.kernel.org
Signed-off-by: Yifan Zhao <zhaoyifan@sjtu.edu.cn>
Tested-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
When handling a regular LVM volume, GRUB can fail with the message:
error: disk `lvmid/******-****-****-****-****-****-****/******-****-****-****-****-****-******' not found.
If the condition which triggers this exists, grub-probe will report the
error mentioned above. Similarly, the GRUB boot code will fail to detect
LVM volumes, resulting in a failure to boot off of LVM disks/partitions.
The condition can be created on any LVM VG by an LVM configuration change,
so any system with /boot on LVM can become unbootable at "any" time (after
any LVM configuration change).
The problem is caused by an incorrect computation of mda_end in disk/lvm.c,
when the metadata area wraps around. Apparently, this can start happening at
around 220 metadata changes to the VG.
Fixes: 879c4a834 (lvm: Fix two more potential data-dependent alloc overflows)
Fixes: https://savannah.gnu.org/bugs/?61620
Signed-off-by: Rogier <rogier777@gmail.com>
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Tested-By: Michael Chang <mchang@suse.com>
Give the user a chance to re-enter their cryptodisk passphrase after a typo,
rather than immediately failing (and likely dumping them into a GRUB shell).
By default, we allow 3 tries before giving up. A value in the
cryptodisk_passphrase_tries environment variable will override this default.
The user can give up early by entering an empty passphrase, just as they
could before this patch.
Signed-off-by: Forest <forestix@nom.one>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The test corpus for version-1 RAID generated an infinite recursion
in grub_partition_iterate() while attempting to read the superblock.
The reason for the issue was that the data region overlapped with
the superblock.
The infinite call loop looks like this:
grub_partition_iterate() -> partmap->iterate() ->
-> grub_disk_read() -> grub_disk_read_small() ->
-> grub_disk_read_small_real() -> grub_diskfilter_read() ->
-> read_lv() -> read_segment() -> grub_diskfilter_read_node() ->
-> grub_disk_read() -> grub_disk_read_small() -> ...
The fix adds checks for both the superblock region and the data
region when parsing the superblock metadata in grub_mdraid_detect().
Signed-off-by: Lidong Chen <lidong.chen@oracle.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The "ground truth" stack protector cookie value is kept in a global
variable, and loaded in every function prologue and epilogue to store
it into resp. compare it with the stack slot holding the cookie.
If the comparison fails, the program aborts, and this might occur
spuriously when the global variable changes values between the entry and
exit of a function. This implies that assigning the global variable at
boot should not involve any instrumented function calls, unless special
care is taken to ensure that the live call stack is synchronized, which
is non-trivial.
So avoid any function calls, including grub_memcpy(), which is
unnecessary given that the stack cookie is always a suitably aligned
variable of the native word size.
While at it, leave the last byte 0x0 to avoid inadvertent unbounded
strings on the stack.
Note that the use of __attribute__((optimize)) is described as
unsuitable for production use in the GCC documentation, so let's drop
this as well now that it is no longer needed.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Leo Sandoval