zstd: Require at least 8 byte buffer in entropy_common

This fixes the build on s390x which was rightfully complaining that
iend - 7 = buffer + 4 - 7 = buffer -3 is outside the array bounds.

../../grub-core/lib/zstd/entropy_common.c: In function ‘FSE_readNCount’:
../../grub-core/lib/zstd/entropy_common.c:121:28: error: array subscript -3 is outside array bounds of ‘char[4]’ [-Werror=array-bounds]
  121 |             if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
      |                        ~~~~^~
../../grub-core/lib/zstd/entropy_common.c:77:14: note: while referencing ‘buffer’
   77 |         char buffer[4];
      |              ^~~~~~
../../grub-core/lib/zstd/entropy_common.c:105:30: error: array subscript -1 is outside array bounds of ‘char[4]’ [-Werror=array-bounds]
  105 |                 if (ip < iend-5) {
      |                          ~~~~^~
../../grub-core/lib/zstd/entropy_common.c:77:14: note: while referencing ‘buffer’
   77 |         char buffer[4];
      |              ^~~~~~
../../grub-core/lib/zstd/entropy_common.c:150:28: error: array subscript -3 is outside array bounds of ‘char[4]’ [-Werror=array-bounds]
  150 |             if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
      |                        ~~~~^~
../../grub-core/lib/zstd/entropy_common.c:77:14: note: while referencing ‘buffer’
   77 |         char buffer[4];
      |              ^~~~~~

This is fixed in more recent zstd versions in basically the same way,
but the new versions needs more work to import.

Patch-Name: zstd-require-8-byte-buffer.patch

Gbp-Pq: Name zstd-require-8-byte-buffer.patch

efi: Provide a shim for load_image, start_image, unload_image

Provide custom implementations of load_image(), start_image(),
and unload_image() to workaround shim just forwarding those
calls to the firmware.

The code consumes a PE-COFF image loaded into memory. The functions

* check validity of header
* copy the sections
* relocate the code
* invalidate the instruction cache
* execute the image
* return to caller

This was previously in use in Ubuntu on riscv64 and arm64 only,
exposed as a single function  grub_efi_run_image(). It was
originally written by Heinrich and split up into 3 functions
by Julian to integrate with the upstream boot loader.

Caveats:

- We do not always check for over and underflows, but at the
  point we reach this loader, the file has been verified by
  shim already, so this is not much of a concern.

Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Signed-off-by: Julian Andres Klode <julian.klode@canonical.com>
Gbp-Pq: Topic secure-boot
Gbp-Pq: Name efi-use-peimage-shim.patch

efi: Provide wrappers for load_image, start_image, unload_image

These can be used to register a different implementation later,
for example, when shim provides a protocol with those functions.

Gbp-Pq: Topic secure-boot
Gbp-Pq: Name loader-framework.patch

Disable fallback to legacy mode if shim is loaded on x86 archs

This reverts commits
- 6425c12cd77ad51ad24be84c092aefacf0875089:
  this originally adds the fallback
- e60015f574024584e43d1b3b245551e864aa8c4d:
  this triggers it in another case

Gbp-Pq: Topic secure-boot
Gbp-Pq: Name revert-efi-fallback-to-legacy.patch

grub_os-prober

Gbp-Pq: Name grub_os-prober.patch

fix renamed LV detection

Origin: other
Bug: https://savannah.gnu.org/bugs/index.php?61620
Bug-Debian: https://bugs.debian.org/987008
Forwarded: https://savannah.gnu.org/bugs/index.php?61620
Reviewed-By: Antoine Beaupré
Last-Update: 2023-02-25

 It looks like the detection of the LVM logical volumes fails in
 certain edge conditions. In particular, it was reported that
 renaming an LV will make grub fail to boot from the system as it
 cannot properly detect it anymore.
 .
 I have looked at the code surrounding the patch and cannot claim to
 understand the entire function here, as it is huge and quite
 cryptic. But it seems sane: the `ptr` we're inspecting here starts
 at the `rlocn->offset`, but we were adding `mda_size` to the
 (somewhat) unrelated metadatabuf instead. Now we're marking the
 `mda_end` correctly, based on the rlocn->offsite and ->size.
 .
 I have not tested this myself as the test setup is quite involved,
 but it seems others (e.g. "Hoyer, David" <David.Hoyer@netapp.com>)
 have tested the patch and confirmed it worked.

Gbp-Pq: Name 987008-lvrename-boot-fail.patch

util/mkimage: Some fixes to PE binaries section size calculation

Commit f60ba9e5945 (util/mkimage: Refactor section setup to use a helper)
added a helper function to setup PE sections, but it caused regressions
in some arches where the natural alignment lead to wrong section sizes.

This patch fixes a few things that were caused the section sizes to be
calculated wrongly. These fixes are:

 * Only align the virtual memory addresses but not the raw data offsets.
 * Use aligned sizes for virtual memory sizes but not for raw data sizes.
 * Always align the sizes to set the virtual memory sizes.

These seems to not cause problems for x64 and aa64 EFI platforms but was
a problem for ia64. Because the size of the ".data" and "mods" sections
were wrong and didn't have the correct content. Which lead to GRUB not
being able to load any built-in module.

Reported-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Bug-Debian: https://bugs.debian.org/987103

Patch-Name: mkimage-fix-section-sizes.patch

Gbp-Pq: Name mkimage-fix-section-sizes.patch

Add debug to display what's going on with verifiers

Patch-Name: debug_verifiers.patch

Gbp-Pq: Name debug_verifiers.patch

i386-pc: build verifiers API as module

Given no core functions on i386-pc would require verifiers to work and
the only consumer of the verifier API is the pgp module, it looks good
to me that we can move the verifiers out of the kernel image and let
moddep.lst to auto-load it when pgp is loaded on i386-pc platform.

This helps to reduce the size of core image and thus can relax the
tension of exploding on some i386-pc system with very short MBR gap
size. See also a very comprehensive summary from Colin [1] about the
details.

[1] https://lists.gnu.org/archive/html/grub-devel/2021-03/msg00240.html

V2:
Drop COND_NOT_i386_pc and use !COND_i386_pc.
Add comment in kern/verifiers.c to help understanding what's going on
without digging into the commit history.

Reported-by: Colin Watson <cjwatson@debian.org>
Reviewed-by: Colin Watson <cjwatson@debian.org>
Signed-off-by: Michael Chang <mchang@suse.com>
Origin: other, https://lists.gnu.org/archive/html/grub-devel/2021-03/msg00251.html
Bug-Debian: https://bugs.debian.org/984488
Bug-Debian: https://bugs.debian.org/985374
Last-Update: 2021-09-24

Patch-Name: pc-verifiers-module.patch

Gbp-Pq: Name pc-verifiers-module.patch

20_linux_xen: Do not load XSM policy in non-XSM options

For complicated reasons, even if you have XSM/FLASK disabled (as is
the default) the Xen build system still builds a policy file and puts
it in /boot.

Even so, we shouldn't be loading this in the usual non-"XSM enabled"
entries.  It doesn't do any particular harm but it is quite confusing.

Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Bug-Debian: https://bugs.debian.org/961673
Last-Update: 2020-05-29

Patch-Name: xen-no-xsm-policy-in-non-xsm-options.patch

Gbp-Pq: Name xen-no-xsm-policy-in-non-xsm-options.patch

Minimise writes to EFI variable storage

Some UEFI firmware is easily provoked into running out of space in its
variable storage.  This is usually due to certain kernel drivers (e.g.
pstore), but regardless of the cause it can cause grub-install to fail
because it currently asks efibootmgr to delete and re-add entries, and
the deletion often doesn't result in an immediate garbage collection.
Writing variables frequently also increases wear on the NVRAM which may
have limited write cycles.  For these reasons, it's desirable to find a
way to minimise writes while still allowing grub-install to ensure that
a suitable boot entry exists.

Unfortunately, efibootmgr doesn't offer an interface that would let
grub-install do this.  It doesn't in general make very much effort to
minimise writes; it doesn't allow modifying an existing Boot* variable
entry, except in certain limited ways; and current versions don't have a
way to export the expected variable data so that grub-install can
compare it to the current data.  While it would be possible (and perhaps
desirable?) to add at least some of this to efibootmgr, that would still
leave the problem that there isn't a good upstreamable way for
grub-install to guarantee that it has a new enough version of
efibootmgr.  In any case, it's cumbersome and slow for grub-install to
have to fork efibootmgr to get things done.

Fortunately, a few years ago Peter Jones helpfully factored out a
substantial part of efibootmgr to the efivar and efiboot libraries, and
so it's now possible to have grub-install use those directly.  We still
have to use some code from efibootmgr, but much less than would
previously have been necessary.

grub-install now reuses existing boot entries where possible, and avoids
writing to variables when the new contents are the same as the old
contents.  In the common upgrade case where nothing needs to change, it
no longer writes to NVRAM at all.  It's also now slightly faster, since
using libefivar is faster than forking efibootmgr.

Fixes Debian bug #891434.

Signed-off-by: Colin Watson <cjwatson@ubuntu.com>
Bug-Debian: https://bugs.debian.org/891434
Forwarded: https://lists.gnu.org/archive/html/grub-devel/2019-03/msg00119.html
Last-Update: 2019-03-23

Patch-Name: efi-variable-storage-minimise-writes.patch

Gbp-Pq: Name efi-variable-storage-minimise-writes.patch

Fix setup on Secure Boot systems where cryptodisk is in use

On full-encrypted systems, including /boot, the current code omits
cryptodisk commands needed to open the drives if Secure Boot is enabled.
This prevents grub2 from reading any further configuration residing on
the encrypted disk.
This patch fixes this issue by adding the needed "cryptomount" commands in
the load.cfg file that is then copied in the EFI partition.

Bug-Debian: https://bugs.debian.org/917117
Last-Update: 2019-02-10

Patch-Name: uefi-secure-boot-cryptomount.patch

Gbp-Pq: Name uefi-secure-boot-cryptomount.patch

at_keyboard: initialize keyboard in module init if keyboard is ready

The change in 0c62a5b2 caused at_keyboard to fail on some
machines. Immediately initializing the keyboard in the module init if
the keyboard is ready makes the problem go away.

Bug-Debian: https://bugs.debian.org/741464
Last-Update: 2019-02-09

Patch-Name: at_keyboard-module-init.patch

Gbp-Pq: Name at_keyboard-module-init.patch

bash-completion: Drop "have" checks

These don't work with and aren't needed by dynamically-loaded
completions.

Bug-Debian: https://bugs.debian.org/912852
Forwarded: no
Last-Update: 2018-11-16

Patch-Name: bash-completion-drop-have-checks.patch

Gbp-Pq: Name bash-completion-drop-have-checks.patch

Skip flaky grub_cmd_set_date test

Bug-Debian: https://bugs.debian.org/906470
Last-Update: 2018-10-28

Patch-Name: skip-grub_cmd_set_date.patch

Gbp-Pq: Name skip-grub_cmd_set_date.patch

efi/http: change uint32_t to uintn_t

Modify UINT32 to UINTN in EFI_HTTP_MESSAGE to
be UEFI 2.9 compliant.

Signed-off-by: Keng-Yu Lin <kengyu@hpe.com>
Signed-off-by: Nicolas Frayer <nfrayer@redhat.com>
Ubuntu-Bug: https://bugs.launchpad.net/bugs/2043084

Gbp-Pq: Topic network
Gbp-Pq: Name rhboot-http-message-field-size.patch

efinet: Add DHCP proxy support

If a proxyDHCP configuration is used, the server name, server IP and
boot file values should be taken from the DHCP proxy offer instead of
the DHCP server ack packet.

Signed-off-by: Ian Page Hands <iphands@gmail.com>
Co-authored-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name efinet-add-dhcp-proxy-support.patch

normal/main: Discover the device to read the config from as a fallback

When core.img is generated locally, the grub2-probe tool figures out the
device and partition that needs to be read to parse the GRUB
configuration file.

But in some cases the core.img can't be generated on the host and
instead has to be done at package build time.  In particular, this will
be true when it needs to be signed with a key that's only available on
the package building infrastructure.

In that case, the prefix variable won't have a device and partition but
only a directory path.  So there's no way for GRUB to know from which
device has to read the configuration file.

To allow GRUB to continue working on that scenario, fallback to
iterating over all the available devices if reading the config failed
when using the prefix and fw_path variables.

Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Co-authored-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name discover-the-device-to-read-the-config-from-as-fallback.patch

http: Prepend prefix when the HTTP path is relative

There are two different HTTP drivers that can be used when requesting an
HTTP resource: the efi/http that uses the EFI_HTTP_PROTOCOL and the http
that uses GRUB's HTTP and TCP/IP implementation.

The efi/http driver appends a prefix that is defined in the variable
http_path, but the http driver doesn't.  So using this driver and
attempting to fetch a resource using a relative path fails.  Match the
behavior of efi/http.

Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Co-authored-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name http-prepend-prefix-when-the-http-path-is-relative.patch

efi/http: Enclose literal IPv6 addresses in square brackets

According to RFC 2732 (https://www.ietf.org/rfc/rfc2732.txt), literal IPv6
addresses must be enclosed in square brackets. But GRUB currently does not
do this and is causing HTTP servers to send Bad Request (400) responses.

For example, the following is the HTTP stream when fetching a config file:

HEAD /EFI/BOOT/grub.cfg HTTP/1.1
Host: 2000:dead:beef:a::1
Accept: */*
User-Agent: UefiHttpBoot/1.0

HTTP/1.1 400 Bad Request
Date: Thu, 05 Mar 2020 14:46:02 GMT
Server: Apache/2.4.41 (Fedora) OpenSSL/1.1.1d
Connection: close
Content-Type: text/html; charset=iso-8859-1

and after enclosing the IPv6 address the HTTP request is successful:

HEAD /EFI/BOOT/grub.cfg HTTP/1.1
Host: [2000:dead:beef:a::1]
Accept: */*
User-Agent: UefiHttpBoot/1.0

HTTP/1.1 200 OK
Date: Thu, 05 Mar 2020 14:48:04 GMT
Server: Apache/2.4.41 (Fedora) OpenSSL/1.1.1d
Last-Modified: Thu, 27 Feb 2020 17:45:58 GMT
ETag: "206-59f924b24b1da"
Accept-Ranges: bytes
Content-Length: 518

Resolves: rhbz#1732765

Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name efi-http-enclose-literal-ipv6-addresses-in-square-br.patch

Prepend prefix when HTTP path is relative

This sets a couple of variables.  With the url http://www.example.com/foo/bar :
http_path: /foo/bar
http_url: http://www.example.com/foo/bar

Resolves: rhbz#1616395
Co-authored-by: Javier Martinez Canillas <javierm@redhat.com>
Co-authored-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Stephen Benjamin <stephen@redhat.com>
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name prepend-prefix-when-http-path-is-relative.patch

Try mac/guid/etc before grub.cfg on tftp config files

Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name try-prefixes-for-tftp-config-file.patch

use fw_path prefix when fallback searching for grub config

When PXE booting via UEFI firmware, grub was searching for grub.cfg in
the fw_path directory where the grub application was found.  If that
didn't exist, a fallback search would look for config file names based
on MAC and IP address.  However, the search would look in the prefix
directory which may not be the same fw_path.  This patch changes that
behavior to use the fw_path directory for the fallback search.  Only if
fw_path is NULL will the prefix directory be searched.

Signed-off-by: Mark Salter <msalter@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name use-fw_path-prefix-when-fallback-searching-for-grub-config.patch

Add fw_path variable to detect config file on efi

This patch makes grub look for its config file on efi where the app was
found.

Resolves: rhbz#857936, rhbz#1616395
Co-authored-by: Matthew Garrett
Co-authored-by: Javier Martinez Canillas <javierm@redhat.com>
Co-authored-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Paulo Flabiano Smorigo <pfsmorigo@br.ibm.com>
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name add-fw_path-variable-to-detect-config-file-on-efi.patch

efi/http: match protocol+hostname of boot url in root_url

This lets you write config files that don't know urls.

Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name efi-http-match-protocol-hostname-of-boot-url-in-root.patch

efinet: also use the firmware acceleration for http

Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name efinet-also-use-the-firmware-acceleration-for-http.patch

Support UEFI networking protocols

References: fate#320130, bsc#1015589, bsc#1076132, rhbz#1732765
Co-authored-by: Peter Jones <pjones@redhat.com>
Co-authored-by: Sebastian Krahmer <krahmer@suse.com>
Co-authored-by: Javier Martinez Canillas <javierm@redhat.com>
Co-authored-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Michael Chang <mchang@suse.com>
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name support-uefi-networking-protocols.patch

efinet: set DNS server from UEFI protocol

In the URI device path node, any name rather than address can be used
for looking up the resources so that DNS service become needed to get
answer of the name's address.  Unfortunately, DNS is not defined in any
of the device path nodes so that we use the EFI_IP4_CONFIG2_PROTOCOL and
EFI_IP6_CONFIG_PROTOCOL to obtain it.

These two protcols are defined the sections of UEFI specification.

    27.5 EFI IPv4 Configuration II Protocol
    27.7 EFI IPv6 Configuration Protocol

include/grub/efi/api.h:
Add new structure and protocol UUID of EFI_IP4_CONFIG2_PROTOCOL and
EFI_IP6_CONFIG_PROTOCOL.

grub-core/net/drivers/efi/efinet.c:
Use the EFI_IP4_CONFIG2_PROTOCOL and EFI_IP6_CONFIG_PROTOCOL to obtain
the list of DNS server address for IPv4 and IPv6 respectively.  The
address of DNS servers is structured into DHCPACK packet and feed into
the same DHCP packet processing functions to ensure the network
interface is setting up the same way it used to be.

Signed-off-by: Michael Chang <mchang@suse.com>
Signed-off-by: Ken Lin <ken.lin@hpe.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Julian Andres Klode <julian.klode@canonical.com>
(rebased against 2.12)

Gbp-Pq: Topic network
Gbp-Pq: Name efinet-set-dns-from-uefi-proto.patch

efinet Configure network from UEFI device path

The PXE Base Code protocol used to obtain cached PXE DHCPACK packet is
no longer provided for HTTP Boot.  Instead, we have to get the HTTP boot
information from the device path nodes defined in following UEFI
Specification sections.

    9.3.5.12 IPv4 Device Path
    9.3.5.13 IPv6 Device Path
    9.3.5.23 Uniform Resource Identifiers (URI) Device Path

This patch basically does:

include/grub/efi/api.h:
Add new structure for Uniform Resource Identifiers (URI) Device Path

grub-core/net/drivers/efi/efinet.c:

Check if PXE Base Code is available.  If not, try to obtain the netboot
information from the device path where the image booted from.  The
DHCPACK packet is recoverd from the information in device patch and fed
into the same DHCP packet processing functions to ensure the network
interface is set up the same way it used to be.

Signed-off-by: Michael Chang <mchang@suse.com>
Signed-off-by: Ken Lin <ken.lin@hpe.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name efinet-Configure-network-from-UEFI-device-path.patch

bootp: Process DHCPACK packet during HTTP Boot

The vendor class identifier with the string "HTTPClient" is used to
denote the packet as responding to HTTP boot request.  In DHCP4 config,
the filename for HTTP boot is the URL of the boot file, while for PXE
boot it is the path to the boot file.  As a consequence, the next-server
becomes obselete because the HTTP URL already contains the server
address for the boot file.  For DHCP6 config, there's no difference
definition in existing config as dhcp6.bootfile-url can be used to
specify URL for both HTTP and PXE boot file.

Add processing for "HTTPClient" vendor class identifier in DHCPACK
packet by treating it as HTTP format, not as the PXE format.

Signed-off-by: Michael Chang <mchang@suse.com>
Signed-off-by: Ken Lin <ken.lin@hpe.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name bootp-process-dhcpack-http-boot.patch

efinet: add structures for PXE messages

When grub2 image is booted from UEFI IPv6 PXE, the DHCPv6 Reply packet
is cached in firmware buffer which can be obtained by PXE Base Code
protocol.  The network interface can be setup through the parameters in
that obtained packet.

Augment existing structures to represent this, and make them agnostic
between ipv4 and ipv6.

Signed-off-by: Michael Chang <mchang@suse.com>
Signed-off-by: Ken Lin <ken.lin@hpe.com>
Co-authored-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name efinet-add-structures-for-PXE-messages.patch

efinet + bootp: add net_bootp6 command supporting dhcpv6

Implement new net_bootp6 command for IPv6 network auto configuration via
the DHCPv6 protocol (RFC3315).

Signed-off-by: Peter Jones <pjones@redhat.com>
Co-authored-by: Michael Chang <mchang@suse.com>
Signed-off-by: Michael Chang <mchang@suse.com>
Signed-off-by: Ken Lin <ken.lin@hpe.com>
Co-authored-by: Robbie Harwood <rharwood@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name bootp-new-net_bootp6-command.patch

net/http: check result of grub_netbuff_put() in http_receive()

Co-authored-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Peter Jones <pjones@redhat.com>
Signed-off-by: Robbie Harwood <rharwood@redhat.com>
Gbp-Pq: Topic network
Gbp-Pq: Name net-http-check-result-of-grub_netbuff_put-in-http_receive.patch

Tell zpool to emit full device names

zfs-initramfs currently provides extraneous, undesired symlinks to
devices directly underneath /dev/ to satisfy zpool's historical output
of unqualified device names. By including this environment variable to
signal our intent to zpool, zfs-linux packages can drop the symlink
behavior when updating to its upstream or backported output behavior.

Bug: https://savannah.gnu.org/bugs/?43653
Bug-Debian: https://bugs.debian.org/824974
Bug-Ubuntu: https://bugs.launchpad.net/bugs/1527727
Last-Update: 2016-11-01

Patch-Name: zpool-full-device-name.patch

Gbp-Pq: Name zpool-full-device-name.patch

Arrange to insmod xzio and lzopio when booting a kernel as a Xen guest

This is needed in case the Linux kernel is compiled with CONFIG_KERNEL_XZ or
CONFIG_KERNEL_LZO rather than CONFIG_KERNEL_GZ (gzio is already loaded by
grub.cfg today).

Signed-off-by: Ian Campbell <ijc@debian.org>
Bug-Debian: https://bugs.debian.org/755256
Forwarded: http://lists.gnu.org/archive/html/grub-devel/2014-11/msg00091.html
Last-Update: 2014-11-30

Patch-Name: insmod-xzio-and-lzopio-on-xen.patch

Gbp-Pq: Name insmod-xzio-and-lzopio-on-xen.patch

grub-install: Install PV Xen binaries into the upstream specified path

Upstream have defined a specification for where guests ought to place their
xenpv grub binaries in order to facilitate chainloading from a stage 1 grub
loaded from dom0.

http://xenbits.xen.org/docs/unstable-staging/misc/x86-xenpv-bootloader.html

The spec calls for installation into /boot/xen/pvboot-i386.elf or
/boot/xen/pvboot-x86_64.elf.

Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Bug-Debian: https://bugs.debian.org/762307
Forwarded: http://lists.gnu.org/archive/html/grub-devel/2014-10/msg00041.html
Last-Update: 2014-10-24

Patch-Name: grub-install-pvxen-paths.patch

Gbp-Pq: Name grub-install-pvxen-paths.patch

Disable VSX instruction

VSX bit is enabled by default for Power7 and Power8 CPU models,
so we need to disable them in order to avoid instruction exceptions.
Kernel will activate it when necessary.

* grub-core/kern/powerpc/ieee1275/startup.S: Disable VSX.

Also-By: Adhemerval Zanella <azanella@linux.vnet.ibm.com>
Also-By: Colin Watson <cjwatson@debian.org>
Origin: other, https://lists.gnu.org/archive/html/grub-devel/2014-09/msg00078.html
Last-Update: 2015-01-27

Patch-Name: ppc64el-disable-vsx.patch

Gbp-Pq: Name ppc64el-disable-vsx.patch

Include a text attribute reset in the clear command for ppc

Always clear text attribute for clear command in order to avoid problems
after it boots.

* grub-core/term/terminfo.c: Add escape for text attribute reset

Bug-Ubuntu: https://bugs.launchpad.net/bugs/1295255
Origin: other, https://lists.gnu.org/archive/html/grub-devel/2014-09/msg00076.html
Last-Update: 2014-09-26

Patch-Name: ieee1275-clear-reset.patch

Gbp-Pq: Name ieee1275-clear-reset.patch

Port yaboot logic for various powerpc machine types

Some powerpc machines require not updating the NVRAM.  This can be handled
by existing grub-install command-line options, but it's friendlier to detect
this automatically.

On chrp_ibm machines, use the nvram utility rather than nvsetenv.  (This
is possibly suitable for other machines too, but that needs to be
verified.)

Forwarded: no
Last-Update: 2014-10-15

Patch-Name: install-powerpc-machtypes.patch

Gbp-Pq: Name install-powerpc-machtypes.patch

Add GRUB_RECOVERY_TITLE option

This allows the controversial "recovery mode" text to be customised.

Bug-Ubuntu: https://bugs.launchpad.net/bugs/1240360
Forwarded: no
Last-Update: 2013-12-25

Patch-Name: mkconfig-recovery-title.patch

Gbp-Pq: Name mkconfig-recovery-title.patch

Ignore functional test failures for now as they are broken

See: https://lists.gnu.org/archive/html/grub-devel/2013-11/msg00242.html

Forwarded: not-needed
Last-Update: 2013-11-19

Patch-Name: ignore-grub_func_test-failures.patch

Gbp-Pq: Name ignore-grub_func_test-failures.patch

Probe FusionIO devices

Bug-Ubuntu: https://bugs.launchpad.net/bugs/1237519
Forwarded: no
Last-Update: 2016-09-18

Patch-Name: probe-fusionio.patch

Gbp-Pq: Name probe-fusionio.patch

Add configure option to use vt.handoff=7

This is used for non-recovery Linux entries only; it enables
flicker-free booting if gfxpayload=keep is in use and a suitable kernel
is present.

Author: Andy Whitcroft <apw@canonical.com>
Forwarded: not-needed
Last-Update: 2013-12-25

Patch-Name: vt-handoff.patch

Gbp-Pq: Name vt-handoff.patch

Add configure option to enable gfxpayload=keep dynamically

Set GRUB_GFXPAYLOAD_LINUX=keep unless it's known to be unsupported on
the current hardware.  See
https://blueprints.launchpad.net/ubuntu/+spec/packageselection-foundations-n-grub2-boot-framebuffer.

Author: Colin Watson <cjwatson@ubuntu.com>
Forwarded: no
Last-Update: 2019-05-25

Patch-Name: gfxpayload-dynamic.patch

Gbp-Pq: Name gfxpayload-dynamic.patch

If we don't have writable grubenv and we're on EFI, always show the menu

If we don't have writable grubenv, recordfail doesn't work, which means our
quickboot behavior - with a timeout of 0 - leaves the user without a
reliable way to access the boot menu if they're on UEFI, because unlike
BIOS, UEFI does not support checking the state of modifier keys (i.e.
holding down shift at boot is not detectable).

Handle this corner case by always using a non-zero timeout on EFI when
save_env doesn't work.

Reuse GRUB_RECORDFAIL_TIMEOUT to avoid introducing another variable.

Signed-off-by: Steve Langasek <steve.langasek@canonical.com>
Bug-Ubuntu: https://bugs.launchpad.net/bugs/1800722
Last-Update: 2019-06-24

Patch-Name: quick-boot-lvm.patch

Gbp-Pq: Name quick-boot-lvm.patch

Add configure option to bypass boot menu if possible

If other operating systems are installed, then automatically unhide the
menu.  Otherwise, if GRUB_HIDDEN_TIMEOUT is 0, then use keystatus if
available to check whether Shift is pressed.  If it is, show the menu,
otherwise boot immediately.  If keystatus is not available, then fall
back to a short delay interruptible with Escape.

This may or may not remain Ubuntu-specific, although it's not obviously
wanted upstream.  It implements a requirement of
https://wiki.ubuntu.com/DesktopExperienceTeam/KarmicBootExperienceDesignSpec#Bootloader.

If the previous boot failed (defined as failing to get to the end of one
of the normal runlevels), then show the boot menu regardless.

Author: Richard Laager <rlaager@wiktel.com>
Author: Robie Basak <robie.basak@ubuntu.com>
Forwarded: no
Last-Update: 2015-09-04

Patch-Name: quick-boot.patch

Gbp-Pq: Name quick-boot.patch

Adjust efi_distributor for some distributions

This is not a very good approach, and certainly not sanely upstreamable;
we probably need to split GRUB_DISTRIBUTOR into a couple of different
variables.

Bug-Ubuntu: https://bugs.launchpad.net/bugs/1242417
Bug-Debian: https://bugs.debian.org/932966
Forwarded: not-needed
Last-Update: 2019-08-06

Patch-Name: install-efi-adjust-distributor.patch

Gbp-Pq: Name install-efi-adjust-distributor.patch

Add configure option to reduce visual clutter at boot time

If this option is enabled, then do all of the following:

Don't display introductory message about line editing unless we're
actually offering a shell prompt.  (This is believed to be a workaround
for a different bug.  We'll go with this for now, but will drop this in
favour of a better fix upstream if somebody figures out what that is.)

Don't clear the screen just before booting if we never drew the menu in
the first place.

Remove verbose messages printed before reading configuration.  In some
ways this is awkward because it makes debugging harder, but it's a
requirement for a smooth-looking boot process; we may be able to do
better in future.  Upstream doesn't want this, though.

Disable the cursor as well, for similar reasons of tidiness.

Suppress kernel/initrd progress messages, except in recovery mode.

Suppress "GRUB loading" message unless Shift is held down.  Upstream
doesn't want this, as it makes debugging harder.  Ubuntu wants it to
provide a cleaner boot experience.

Author: Will Thompson <will@willthompson.co.uk>
Bug-Ubuntu: https://bugs.launchpad.net/bugs/386922
Bug-Ubuntu: https://bugs.launchpad.net/bugs/861048
Forwarded: (partial) http://lists.gnu.org/archive/html/grub-devel/2009-09/msg00056.html
Last-Update: 2021-09-24

Patch-Name: maybe-quiet.patch

Gbp-Pq: Name maybe-quiet.patch

Don't add suffix to the distributor string

- Ubuntu is called "Ubuntu", not "Ubuntu GNU/Linux"
- Debian already has the suffix in `/etc/os-release`

Author: Colin Watson <cjwatson@debian.org>
Author: Harald Sitter <apachelogger@kubuntu.org>
Author: Mate Kukri <mate.kukri@canonical.com>
Forwarded: not-needed
Last-Update: 2024-11-05

Patch-Name: mkconfig-distributor.patch

Gbp-Pq: Name mkconfig-distributor.patch

Blacklist 1440x900x32 from VBE preferred mode handling

Bug-Ubuntu: https://bugs.launchpad.net/bugs/701111
Forwarded: no
Last-Update: 2013-11-14

Patch-Name: blacklist-1440x900x32.patch

Gbp-Pq: Name blacklist-1440x900x32.patch

Read /etc/default/grub.d/*.cfg after /etc/default/grub

Bug-Ubuntu: https://bugs.launchpad.net/bugs/901600
Forwarded: no
Last-Update: 2021-09-24

Patch-Name: default-grub-d.patch

Gbp-Pq: Name default-grub-d.patch

Avoid getting confused by inaccessible loop device backing paths

Bug-Ubuntu: https://bugs.launchpad.net/bugs/938724
Forwarded: no
Last-Update: 2021-09-24

Patch-Name: mkconfig-nonexistent-loopback.patch

Gbp-Pq: Name mkconfig-nonexistent-loopback.patch

Prefer translations from Ubuntu language packs if available

Bug-Ubuntu: https://bugs.launchpad.net/bugs/537998
Forwarded: not-needed
Last-Update: 2013-12-25

Patch-Name: install-locale-langpack.patch

Gbp-Pq: Name install-locale-langpack.patch

"single" -> "recovery" when friendly-recovery is installed

If configured with --enable-ubuntu-recovery, also set nomodeset for
recovery mode, and disable 'set gfxpayload=keep' even if the system
normally supports it.  See
https://launchpad.net/ubuntu/+spec/desktop-o-xorg-tools-and-processes.

Author: Stéphane Graber <stgraber@ubuntu.com>
Forwarded: no
Last-Update: 2013-12-25

Patch-Name: mkconfig-ubuntu-recovery.patch

Gbp-Pq: Name mkconfig-ubuntu-recovery.patch

Fall back to non-EFI if booted using EFI but -efi is missing

It may be possible, particularly in recovery situations, to be booted
using EFI on x86 when only the i386-pc target is installed, or on ARM
when only the arm-uboot target is installed.  There's nothing actually
stopping us installing i386-pc or arm-uboot from an EFI environment, and
it's better than returning a confusing error.

Author: Steve McIntyre <93sam@debian.org>
Forwarded: no
Last-Update: 2019-05-24

Patch-Name: install-efi-fallback.patch

Gbp-Pq: Name install-efi-fallback.patch

Silence error messages when translations are unavailable

Bug: https://savannah.gnu.org/bugs/?35880
Forwarded: https://savannah.gnu.org/bugs/?35880
Last-Update: 2013-11-14

Patch-Name: gettext-quiet.patch

Gbp-Pq: Name gettext-quiet.patch

Restore grub-mkdevicemap

This is kind of a mess, requiring lots of OS-specific code to iterate
over all possible devices.  However, we use it in a number of scripts to
discover devices and reimplementing those in terms of something else
would be very complicated.

Author: Dimitri John Ledkov <dimitri.ledkov@canonical.com>
Forwarded: no
Last-Update: 2021-09-24

Patch-Name: restore-mkdevicemap.patch

Gbp-Pq: Name restore-mkdevicemap.patch

Handle filesystems loop-mounted on file images

Improve prepare_grub_to_access_device to emit appropriate commands for
such filesystems, and ignore them in Linux grub.d scripts.

This is needed for Ubuntu's Wubi installation method.

This patch isn't inherently Debian/Ubuntu-specific.  losetup and
/proc/mounts are Linux-specific, though, so we might need to refine this
before sending it upstream.  The changes to the Linux grub.d scripts
might be better handled by integrating 10_lupin properly instead.

Patch-Name: mkconfig-loopback.patch

Gbp-Pq: Name mkconfig-loopback.patch

If GRUB Legacy is still around, tell packaging to ignore it

Bug-Debian: http://bugs.debian.org/586143
Forwarded: not-needed
Last-Update: 2021-09-24

Patch-Name: install-stage2-confusion.patch

Gbp-Pq: Name install-stage2-confusion.patch

Disable gfxpayload=keep by default

Setting gfxpayload=keep has been known to cause efifb to be
inappropriately enabled.  In any case, with the current Linux kernel the
result of this option is that early kernelspace will be unable to print
anything to the console, so (for example) if boot fails and you end up
dumped to an initramfs prompt, you won't be able to see anything on the
screen.  As such it shouldn't be enabled by default in Debian, no matter
what kernel options are enabled.

gfxpayload=keep is a good idea but rather ahead of its time ...

Bug-Debian: http://bugs.debian.org/567245
Forwarded: no
Last-Update: 2013-12-25

Patch-Name: gfxpayload-keep-default.patch

Gbp-Pq: Name gfxpayload-keep-default.patch

Disable use of floppy devices

An ugly kludge.  Should this be merged upstream?

Author: Robert Millan

Patch-Name: disable-floppies.patch

Gbp-Pq: Name disable-floppies.patch

Support running grub-probe in grub-legacy's update-grub

Author: Colin Watson <cjwatson@debian.org>
Forwarded: not-needed
Last-Update: 2013-12-25

Patch-Name: grub-legacy-0-based-partitions.patch

Gbp-Pq: Name grub-legacy-0-based-partitions.patch

Write marker if core.img was written to filesystem

The Debian bug reporting script includes a warning in this case.

Patch-Name: core-in-fs.patch

Gbp-Pq: Name core-in-fs.patch

Hack prefix for OLPC

This sucks, but it's better than what OFW was giving us.

Patch-Name: olpc-prefix-hack.patch

Gbp-Pq: Name olpc-prefix-hack.patch

grub2 (2.12-8) unstable; urgency=medium

  [ Mate Kukri ]
  * d/default/grub: Always get distributor string from `/etc/os-release`
  * Avoid adding extra GNU/Linux suffix to menu entries (Closes: #1076723)

[dgit import unpatched grub2 2.12-8]

Import grub2_2.12-8.debian.tar.xz

[dgit import tarball grub2 2.12-8 grub2_2.12-8.debian.tar.xz]

Import grub2_2.12.orig.tar.xz

[dgit import orig grub2_2.12.orig.tar.xz]