Move some calls used to access NTFS attribute header fields into
functions with human-readable names.
Suggested-by: Daniel Kiper <daniel.kiper@oracle.com>
Signed-off-by: Maxim Suhanov <dfirblog@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
This fix introduces checks to ensure that an NTFS volume label is always
read from the corresponding file record segment.
The current NTFS code allows the volume label string to be read from an
arbitrary, attacker-chosen memory location. However, the bytes read are
always treated as UTF-16LE. So, the final string displayed is mostly
unreadable and it can't be easily converted back to raw bytes.
The lack of this check is a minor issue, likely not causing a significant
data leak.
Reported-by: Maxim Suhanov <dfirblog@gmail.com>
Signed-off-by: Maxim Suhanov <dfirblog@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
This fix introduces checks to ensure that bitmaps for directory indices
are never read beyond their actual sizes.
The lack of this check is a minor issue, likely not exploitable in any way.
Reported-by: Maxim Suhanov <dfirblog@gmail.com>
Signed-off-by: Maxim Suhanov <dfirblog@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
This fix introduces checks to ensure that index entries are never read
beyond the corresponding directory index.
The lack of this check is a minor issue, likely not exploitable in any way.
Reported-by: Maxim Suhanov <dfirblog@gmail.com>
Signed-off-by: Maxim Suhanov <dfirblog@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
When reading a file containing resident data, i.e., the file data is stored in
the $DATA attribute within the NTFS file record, not in external clusters,
there are no checks that this resident data actually fits the corresponding
file record segment.
When parsing a specially-crafted file system image, the current NTFS code will
read the file data from an arbitrary, attacker-chosen memory offset and of
arbitrary, attacker-chosen length.
This allows an attacker to display arbitrary chunks of memory, which could
contain sensitive information like password hashes or even plain-text,
obfuscated passwords from BS EFI variables.
This fix implements a check to ensure that resident data is read from the
corresponding file record segment only.
Fixes: CVE-2023-4693
Reported-by: Maxim Suhanov <dfirblog@gmail.com>
Signed-off-by: Maxim Suhanov <dfirblog@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
When parsing an extremely fragmented $MFT file, i.e., the file described
using the $ATTRIBUTE_LIST attribute, current NTFS code will reuse a buffer
containing bytes read from the underlying drive to store sector numbers,
which are consumed later to read data from these sectors into another buffer.
These sectors numbers, two 32-bit integers, are always stored at predefined
offsets, 0x10 and 0x14, relative to first byte of the selected entry within
the $ATTRIBUTE_LIST attribute. Usually, this won't cause any problem.
However, when parsing a specially-crafted file system image, this may cause
the NTFS code to write these integers beyond the buffer boundary, likely
causing the GRUB memory allocator to misbehave or fail. These integers contain
values which are controlled by on-disk structures of the NTFS file system.
Such modification and resulting misbehavior may touch a memory range not
assigned to the GRUB and owned by firmware or another EFI application/driver.
This fix introduces checks to ensure that these sector numbers are never
written beyond the boundary.
Fixes: CVE-2023-4692
Reported-by: Maxim Suhanov <dfirblog@gmail.com>
Signed-off-by: Maxim Suhanov <dfirblog@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
During attempts to configure a serial console, a Page Fault Exception
and system reset were encountered, specifically on release 2.12~rc1.
This issue was not present in prior versions and seemed to affect only
a specific machine, potentially pointing to hardware or firmware flaw.
After investigation, it was discovered that the invalid page access
occurred during the discovery of serial MMIO ports as specified by
ACPI's SPCR table [1]. The recent change uncovered an issue in GRUB's
ACPI driver.
In certain cases, the XSDT/RSDT root table might contain a NULL entry as
a terminator, depending on how the tables are assembled. GRUB cannot
blindly trust the address in the root table to be valid and should
perform a sanity check for NULL entries. This patch introduces this
simple check.
This fix is also inspired by a related Linux kernel fix [2].
[1] 7b192ec4c term/ns8250: Use ACPI SPCR table when available to configure serial
[2] 0f929fbf0 ACPICA: Tables: Add new mechanism to skip NULL entries in RSDT and XSDT.
Signed-off-by: Michael Chang <mchang@suse.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The commit 80948f532d (lib/i386/relocator64: Build fixes for i386) has
broken 64-bit FreeBSD boot on BIOS. This patch fixes the issue.
Fixes: 80948f532d (lib/i386/relocator64: Build fixes for i386)
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Recent mixed-mode Linux kernels, i.e., v4.0 or newer, can access EFI
runtime services at OS runtime even when the OS was not entered via the
EFI stub. This is because, instead of reverting back to the firmware's
segment selectors, GDTs and IDTs, the 64-bit kernel simply calls 32-bit
runtime services using compatibility mode, i.e., the same mode used for
32-bit user space, without taking down all interrupt handling, exception
handling, etc.
This means that GRUB's legacy x86 boot mode is sufficient to make use of
this: 32-bit i686 builds of GRUB can already boot 64-bit kernels in EFI
enlightened mode, but without going via the EFI stub, and provide all
the metadata that the OS needs to map the EFI runtime regions and call
EFI runtime services successfully.
It does mean that GRUB should not attempt to invoke the firmware's
LoadImage()/StartImage() methods on kernel builds that it knows cannot
be started natively. So, add a check for this in the native EFI boot
path and fall back to legacy x86 mode in such cases.
Note that in the general case, booting non-native images of the same
native word size, e.g., x64 EFI apps on arm64 firmware, might be
supported by means of emulation. So, let's only disallow images that use
a non-native word size. This will also permit booting i686 kernels on
x86_64 builds, although without access to runtime services, as this is
not supported by Linux.
This change on top of 2.12-rc1 is sufficient to boot ordinary Linux
mixed mode builds and get full access to the EFI runtime services.
Cc: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Steve McIntyre <steve@einval.com>
Cc: Julian Andres Klode <julian.klode@canonical.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Dimitri John Ledkov <dimitri.ledkov@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The x86_64 Linux kernel can be booted in 32-bit mode, in which case the
startup code creates a set of preliminary page tables that map the first
4 GiB of physical memory 1:1 and enables paging. This is a prerequisite
for 64-bit execution and can therefore only be implemented in 32-bit code.
The x86_64 Linux kernel can also be booted in 64-bit mode directly: this
implies that paging is already enabled and it is the responsibility of
the bootloader to ensure that the active page tables cover the entire
loaded image, including its BSS space, the size of which is described in
the image's setup header.
Given that the EFI spec mandates execution in long mode for x86_64 and
stipulates that all system memory is mapped 1:1, the Linux/x86
requirements for 64-bit entry can be met trivially when booting on
x86_64 via EFI. So, enter via the 64-bit entry point in this case.
This involves inspecting the xloadflags field in the setup header to
check whether the 64-bit entry point is supported. This field was
introduced in Linux version v3.8 (early 2013).
This change ensures that all EFI firmware tables and other assets passed
by the firmware or bootloader in memory remain mapped and accessible
throughout the early startup code.
Avoiding the drop out of long mode will also be needed to support
upcoming CPU designs that no longer implement 32-bit mode at all
(as recently announced by Intel [0]).
[0] https://www.intel.com/content/www/us/en/developer/articles/technical/envisioning-future-simplified-architecture.html
Cc: Daniel Kiper <daniel.kiper@oracle.com>
Cc: Julian Andres Klode <julian.klode@canonical.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Some dnodes are shared with properties zap. This is used
e.g. for quotas. Then dnode type is 0xc4 and GRUB stumbles on
this. Check bonus type and if it's ok then ignore dnode type mismatch
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Reading them is harmless but useless as they are empty by definition
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
We ended up comparing over unset values as we had dnode_phys on one side
and dnode on another
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
This is a speedup used in some ZFS version. This trips GRUB and makes it
unable to access directories. Just skip it for now and revisit
if we ever need this speedup.
Signed-off-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
If the EFI graphics pixel format is PixelBltOnly, we cannot write directly
to the frame buffer. We need the shadow frame buffer which we copy via
the BitBlt operation to the hardware.
If the pixel format is PixelBltOnly and allocation of the shadow frame
buffer fails, we must raise an error to signal that the EFI GOP protocol
is not usable.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
This was causing the cryptomount command to return failure even though
the crypto device was successfully added. Of course, this meant that any
script using the return code would behave unexpectedly.
Fixes: 3cf2e848bc (disk/cryptodisk: Allows UUIDs to be compared in a dash-insensitive manner)
Suggested-by: Olaf Hering <olaf@aepfle.de>
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Patrich Steinhardt <ps@pks.im>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
To comply with C99 and POSIX standards, snprintf() should return the
number of bytes that would be written to the string (excluding the
terminating NUL byte) if the buffer size was big enough. Before this
change, the return value was the minimum of the standard return and the
length of the buffer. Rarely is the return value of grub_snprintf() or
grub_vsnprintf() used with current code, and the few places where it is
used do not need to be changed.
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
In long list mode, if the file can not be opened, the file is not printed.
Instead, print the file but print the size as "????????????".
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
For each non-directory path argument to the ls command, the full path was
being sent to the print functions, instead of the dirname. The long output
print function expected dirname to be the directory containing the file
and so could not open the file to get the file size because the generated
path was incorrect. This caused the output to be a blank line.
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Specifically, return GRUB_ERR_BAD_FILE_TYPE because this is what is
expected by the ls command when it is given a path to a non-directory.
This fixes a bug where calling ls with a list of non-directory paths
outputs a blank line for each such argument.
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The videoinfo command will initialize all non-active video adapters. Video
drivers tend to zero out the global framebuffer object on initialization.
This is not a problem when there is no active video adapter. However, when
there is, then outputting to the video adapter will cause a crash because
methods in the framebuffer object are reinitialized. For example, this
command sequence will cause a crash.
terminal_output --append gfxterm; videoinfo
When running in a QEMU headless with GRUB built for the x86_64-efi target,
the first command initializes the Bochs video adapter, which, among
other things, sets the set_page() member function. Then when videoinfo is
run, all non-Bochs video adapters will be initialized, each one wiping
the framebuffer and thus setting set_page to NULL. Soon after the videoinfo
command finishes there will be a call to grub_refresh(), which will
ultimately call the framebuffer's set_page which will be NULL and cause
a crash when called.
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
According to commit 0231d00082 (ACPI: SPCR: Make SPCR available to x86)
to the Linux kernel, "On x86, many systems have a valid SPCR table but the
table version is not 2 so the table version check must be a warning."
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The LoadImage() provided by the shim does not consult MOK when loading
an image. So, simply signature verification fails when it should not.
This means we cannot use Linux EFI stub to start the kernel when the
shim is loaded. We have to fallback to legacy mode on x86 architectures.
This is not possible on other architectures due to lack of legacy mode.
This is workaround which should disappear when the shim provides
LoadImage() which looks up MOK during signature verification.
On the occasion align constants in include/grub/efi/sb.h.
Signed-off-by: Daniel Kiper <daniel.kiper@oracle.com>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Using grub_strncpy() instead of grub_snprintf() is less overhead and
indicates clearly that the dest should be the same string as the source.
Also fix indentation.
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
When creating at runtime a newc initrd via arguments to initrd with "newc:"
prefixes, only emit a directory path record once. The original code
intended to do that by bailing out of emitting the record when the record
to be created matches an existing record. However, this does not happen
because grub_memcmp() is improperly checked.
Generating duplicate newc directory records does not cause any problems
because the Linux unpacker will skip it once it sees the directory already
exists. This fix saves a little processing and makes the generated newc
cpio archive a little smaller.
Fixes: 92750e4c60 (Add ability to generate newc additions on runtime.)
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
EFI status codes are of different classes depending on the first byte and
all error status codes defined in appendix D of the main spec start from
1 and have the high bit set. When printing as a uint, the decimal is a very
large number that needs have the high bit cleared get the spec error code.
This can be easily visually done by a human if the number is printed as hex.
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Most of leftover code blindly assumes GRUB_RELOCATOR_FIRMWARE_REQUESTS_QUANT
divisibility by 8. So, enforce this at compile time.
Signed-off-by: Daniel Kiper <daniel.kiper@oracle.com>
Reviewed-by: Vladimir Serbinenko <phcoder@gmail.com>
So all we did with the surface in SDL1 was split into window,
surface, renderer and texture. Instead of drawing into the
surface and then flipping, you build your pixels, then update
a texture and then copy the texture to the renderer.
Here we use an empty RGB surface to hold our pixels, which enables
us to keep most of the code the same. The SDL1 code has been adjusted
to refer to "surface" instead of "window" when trying to access the
properties of the surface.
This approaches the configuration by adding a new --enable-grub-emu-sdl2
argument. If set to yes, or auto detected, it disables SDL1 support
automatically.
This duplicates the sdl module block in Makefile.core.def which may
be something to be aware of, but we also don't want to build separate
module.
Fixes: https://bugs.debian.org/1038035
Signed-off-by: Julian Andres Klode <julian.klode@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
These should be quite obvious and will make the SDL2 patch easier
to read then doing it inline there.
Signed-off-by: Julian Andres Klode <julian.klode@canonical.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Currently booting the system is prevented when call to EFI firmware
hash_log_extend_event() returns unknown error. Solve this by following
convention used in commit a4356538d (commands/tpm: Don't propagate
measurement failures to the verifiers layer).
Let the system to be bootable by default when unknown TPM error is
encountered. Check environment variable tpm_fail_fatal to fallback to
previous behaviour.
Signed-off-by: Michał Grzelak <mchl.grzlk@gmail.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Currently bootstrap complains in the following way when
patching gnulib files:
patching file argp-help.c
Hunk #1 succeeded at 52 (offset 1 line).
Hunk #2 succeeded at 1548 (offset 115 lines).
patching file mbswidth.c
patching file mbswidth.h
Hunk #1 succeeded at 40 (offset -5 lines).
Let's fix it by amending line numbers in the patch.
Signed-off-by: Daniel Kiper <daniel.kiper@oracle.com>
Reviewed-by: Alec Brown <alec.r.brown@oracle.com>
Move the constant from grub-core/osdep/linux/getroot.c to
include/grub/disk.h and then reuse it in place of the
hardcoded 1024 limit in diskfilter.
Fixes: 2a5e3c1f2 (disk/diskfilter: Don't make a RAID array with more than 1024 disks)
Cc: Daniel Axtens <dja@axtens.net>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: Julian Andres Klode <julian.klode@canonical.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
According to the relocation documentation, the following function names are
renamed to show their exact meaning:
- from grub_loongarch64_xxx64_hi12() to grub_loongarch64_abs64_hi12(),
- from grub_loongarch64_xxx64_hi12() to grub_loongarch64_abs64_lo20().
Signed-off-by: Xiaotian Wu <wuxiaotian@loongson.cn>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Use PRIxGRUB_INT64_T format specifier for grub_int64_t type
and drop redundant casts.
Signed-off-by: Xiaotian Wu <wuxiaotian@loongson.cn>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
According to the ACPI specification, in ACPI 2.0 or later, an
ACPI-compatible OS must use the XSDT if present. So, we should
use xsdt_addr instead of rsdt_addr if xsdt_addr is valid.
Signed-off-by: Qiumiao Zhang <zhangqiumiao1@huawei.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
According to the ACPI specification, in ACPI 2.0 or later, an
ACPI-compatible OS must use the XSDT if present. So, we should
use xsdt_addr instead of rsdt_addr if xsdt_addr is valid.
Signed-off-by: Qiumiao Zhang <zhangqiumiao1@huawei.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
While performing fuzz testing with XFS filesystem images with ASAN
enabled, several issues were found where the memory accesses are made
beyond the data that is allocated into the struct grub_xfs_data
structure's data field.
The existing structure didn't store the size of the memory allocated into
the buffer in the data field and had no way to check it. To resolve these
issues, the data size is stored to enable checks into the data buffer.
With these checks in place, the fuzzing corpus no longer cause any crashes.
Signed-off-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Marta Lewandowska <mlewando@redhat.com>
Signed-off-by: Lidong Chen <lidong.chen@oracle.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The output in moddep.lst generated from syminfo.lst using genmoddep.awk
is not deterministic since the order of the dependencies on each line
can vary depending on how awk sorts the values in the array.
Be deterministic in the output by sorting the dependencies on each line.
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
EFI firmware determines where to load the GRUB EFI at runtime, and so the
addresses of debug symbols are not known ahead of time. There is a command
defined in the gdb_grub script which will load the debug symbols at the
appropriate addresses, if given the application load address for GRUB.
So add a command named "gdbinfo" to allow the user to print this GDB command
string with the application load address on-demand. For the outputted GDB
command to have any effect when entered into a GDB session, GDB should have
been started with the script as an argument to the -x option or sourced into
an active GDB session before running the outputted command.
Documentation for the gdbinfo command is also added.
Co-developed-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
The EFI chainloader checks that a device path can be created for the $root
device before allowing chainloading to a given file. This is probably to
ensure that the given file can be accessed and loaded by the firmware.
However, since GRUB is loading the image itself, the firmware need not
be able to access the file location of the image. So remove this check.
Also, this fixes an issue where chainloading an image file on a location
that is accessible by the firmware, e.g. (hd0,1)/efi/boot.efi, would
fail when root is a location inaccessible by the firmware, e.g. memdisk.
Use GRUB_EFI_BYTES_TO_PAGES() instead of doing the calculation explicitly.
Add comment noting the section where the load options for the chainloaded
EFI application is constructed.
Signed-off-by: Glenn Washburn <development@efficientek.com>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Add a new module named bli. It implements a small but quite useful part
of the Boot Loader Interface [0]. This interface uses EFI variables for
communication between the boot loader and the operating system.
When loaded, this module sets two EFI variables under the vendor GUID
4a67b082-0a4c-41cf-b6c7-440b29bb8c4f:
- LoaderInfo: contains GRUB + <version number>.
This allows the running operating system to identify the boot loader
used during boot.
- LoaderDevicePartUUID: contains the partition UUID of the EFI System
Partition (ESP). This is used by systemd-gpt-auto-generator [1] to
find the root partitions (and others too), via partition type IDs [2].
This module is available on EFI platforms only. The bli module relies on
the part_gpt module which has to be loaded beforehand to make the GPT
partitions discoverable.
Update the documentation, add a new chapter "Modules" and describe the
bli module there.
[0] https://systemd.io/BOOT_LOADER_INTERFACE/
[1] https://www.freedesktop.org/software/systemd/man/systemd-gpt-auto-generator.html
[2] https://uapi-group.org/specifications/specs/discoverable_partitions_specification/
Signed-off-by: Oliver Steffen <osteffen@redhat.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Add checks for NULL pointers to grub_device_close() and
grub_disk_close() to make these functions more robust.
Signed-off-by: Oliver Steffen <osteffen@redhat.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Maxim Suhanov