Skip to contentSkip to conversation

EFI envblk

PATCH In review Glenn Washburn 4 participants 10 messages Jul 26, 2023 – Sep 14, 2023 Kbuild

lore
Share & tools
lore
#0 lore raw

cover EFI envblk

This patch series (for me) was motivated by the "gdb: Add more support for debugging on EFI platforms" patch series, which allowed printing in early EFI init of the GDB command for properly loading symbols. That approach of having the functionality be included at compile time was ultimately rejected. During the discussion of that series, Robbie suggested[1] using patches by Peter and in the Redhat downstream repo which uses an EFI variable to store a GRUB env block. Using this, a user could store a variable in the env block stored in the EFI variable and then have GRUB load that env block in early init as a way to enable the printing of the GDB command.

I've taken the original patches by Peter, made more suitable for including in GRUB, fixed some bugs, and added a minor feature. The first patch, adds env block loading in early EFI init from the GRUB_ENV EFI variable. The second patch is included to provide tools for GRUB to set this EFI env block itself, as opposed to needing to use the method suggested by Robbie[1], which requires GNU/Linux and the user-space grub2-editenv utility. Of course, if GRUB is crashing before one can run the efi-export-env command, then other ways of creating and setting the EFI envblk variable are necessary. The third patch adds a test which checks the usability of the commands and the early init loading. And the last patch, whichvmotivated this series, prints the GDB command string if the GRUB variable named "enable_earlyinit_gdbinfo" is present and is set to "1", after the env block is loaded from the GRUB_ENV EFI variable. We might want a different name for the GRUB variable, I don't really have a preference.

Glenn

[1] https://mail.gnu.org/archive/html/grub-devel/2023-03/msg00072.html

Glenn Washburn (2):
tests: Add efienv_test for testing efi-export-env and efi-load-env
efi: Print GDB command for loading symbols if variable exists

Peter Jones (2):
efi: Load env block from GRUB_ENV EFI variable in early init
efi: Add efi-export-env and efi-load-env commands

 Makefile.util.def            |   6 ++
 docs/grub.texi               |  24 +++++
 grub-core/Makefile.core.def  |   7 ++
 grub-core/commands/efi/env.c | 182 +++++++++++++++++++++++++++++++++++
 grub-core/kern/efi/efi.c     |   3 +
 grub-core/kern/efi/init.c    |  37 +++++++
 grub-core/lib/envblk.c       |  43 +++++++++
 include/grub/efi/efi.h       |   5 +
 include/grub/lib/envblk.h    |   3 +
 tests/efienv_test.in         |  57 +++++++++++
 10 files changed, 367 insertions(+)
 create mode 100644 grub-core/commands/efi/env.c
 create mode 100644 tests/efienv_test.in
Signature · 1 line
2.34.1
#1 lore raw

PATCH 1/4 efi: Load env block from GRUB_ENV EFI variable in early init

From: Peter Jones <[email protected]>

This allows setting GRUB variables before any user interaction or GRUB script can run. It is primarily intended to be used to turn on early debugging.

Replace grub_efi_guid_t -> grub_guid_t

---
 grub-core/Makefile.core.def |  1 +
 grub-core/kern/efi/init.c   | 31 +++++++++++++++++++++++++++++++
 include/grub/efi/efi.h      |  5 +++++
 3 files changed, 37 insertions(+)
grub-core/Makefile.core.def +1 -0
index d2cf29584a1a..0b447dd6dc67 100644
@@ -219,6 +219,7 @@ kernel = {
   efi = kern/efi/acpi.c;
   efi = kern/efi/sb.c;
   efi = kern/lockdown.c;
+  efi = lib/envblk.c;
   i386_coreboot = kern/i386/pc/acpi.c;
   i386_multiboot = kern/i386/pc/acpi.c;
   i386_coreboot = kern/acpi.c;
grub-core/kern/efi/init.c +31 -0
index 9cdda3f5ca05..18329b1eacd7 100644
@@ -28,8 +28,11 @@
 #include <grub/env.h>
 #include <grub/mm.h>
 #include <grub/kernel.h>
+
 #include <grub/stack_protector.h>
 
+#include <grub/lib/envblk.h>
+
 #ifdef GRUB_STACK_PROTECTOR
 
 static grub_efi_char16_t stack_chk_fail_msg[] =
@@ -111,6 +114,31 @@ stack_protector_init (void)
 
 grub_addr_t grub_modbase;
 
+/* Helper for grub_efi_env_init */
+static int
+set_var (const char *name, const char *value,
+	 void *whitelist __attribute__((__unused__)))
+{
+  grub_env_set (name, value);
+  return 0;
+}
+
+static void
+grub_efi_env_init (void)
+{
+  grub_guid_t efi_grub_guid = GRUB_EFI_GRUB_VARIABLE_GUID;
+  struct grub_envblk envblk_s = { NULL, 0 };
+  grub_envblk_t envblk = &envblk_s;
+
+  grub_efi_get_variable ("GRUB_ENV", &efi_grub_guid, &envblk_s.size,
+                         (void **) &envblk_s.buf);
+  if (envblk_s.buf == NULL || envblk_s.size < 1)
+    return;
+
+  grub_envblk_iterate (envblk, NULL, set_var);
+  grub_free (envblk_s.buf);
+}
+
 void
 grub_efi_init (void)
 {
@@ -124,6 +152,9 @@ grub_efi_init (void)
   /* Initialize the memory management system.  */
   grub_efi_mm_init ();
 
+  /* Populate environment with variables from EFI envblk, if it exists.  */
+  grub_efi_env_init ();
+
   /*
    * Lockdown the GRUB and register the shim_lock verifier
    * if the UEFI Secure Boot is enabled.
include/grub/efi/efi.h +5 -0
index 564c2f62045b..aa6a360a4999 100644
@@ -25,6 +25,11 @@
 #include <grub/efi/api.h>
 #include <grub/efi/pe32.h>
 
+#define GRUB_EFI_GRUB_VARIABLE_GUID \
+  { 0x91376aff, 0xcba6, 0x42be, \
+    { 0x94, 0x9d, 0x06, 0xfd, 0xe8, 0x11, 0x28, 0xe8 } \
+  }
+
 #define GRUB_LINUX_ARM_MAGIC_SIGNATURE 0x016f2818
 
 struct linux_arch_kernel_header {
Signature · 1 line
2.34.1
#2 lore raw

PATCH 2/4 efi: Add efi-export-env and efi-load-env commands

In reply to Glenn Washburn

From: Peter Jones <[email protected]>

This adds "efi-export-env VARIABLES" and "efi-load-env", which manipulate the environment block stored in the EFI variable GRUB_ENV-91376aff-cba6-42be-949d-06fde81128e8.

Fix rebase, cherry-pick issues, grub_efi_guid_t -> grub_guid_t,
module name bug, missing return bug, memory leaks, formatting
Add support for specifying multiple variables at once for efi-export-env.

---
 docs/grub.texi               |  24 +++++
 grub-core/Makefile.core.def  |   6 ++
 grub-core/commands/efi/env.c | 182 +++++++++++++++++++++++++++++++++++
 grub-core/kern/efi/efi.c     |   3 +
 grub-core/lib/envblk.c       |  43 +++++++++
 include/grub/lib/envblk.h    |   3 +
 6 files changed, 261 insertions(+)
 create mode 100644 grub-core/commands/efi/env.c
docs/grub.texi +24 -0
index b39b72230c6f..730e8c8d75a4 100644
@@ -4324,6 +4324,8 @@ you forget a command, you can run the command @command{help}
 * distrust::                    Remove a pubkey from trusted keys
 * drivemap::                    Map a drive to another
 * echo::                        Display a line of text
+* efi-export-env::              Export environment variable to UEFI
+* efi-load-env::                Load the grub environment from UEFI
 * efitextmode::                 Set/Get text output mode resolution
 * eval::                        Evaluate agruments as GRUB commands
 * export::                      Export an environment variable
@@ -4784,6 +4786,28 @@ character will print that character.
 @end deffn
 
 
+@node efi-export-env
+@subsection efi-export-env
+
+@deffn Command efi-export-env varname [varname @dots{}]
+Export the given GRUB variables to the GRUB_ENV UEFI variable as a GRUB
+environment block. The contents of an existing environment block in GRUB_ENV
+will be merged. So variables existing in GRUB_ENV but not given as arguements
+will be unaffected by this command.
+@end deffn
+
+
+@node efi-load-env
+@subsection efi-load-env
+
+@deffn Command efi-load-env
+Load the grub environment from the GRUB_ENV UEFI variable. This will overwrite
+existing variables if they exist in the environment block present in the UEFI
+variable. Other existing GRUB variables will be unchanged. This can allow for
+the persistence of state across boots.
+@end deffn
+
+
 @node efitextmode
 @subsection efitextmode
 
grub-core/Makefile.core.def +6 -0
index 0b447dd6dc67..a6e6b8cb135a 100644
@@ -825,6 +825,12 @@ module = {
   enable = efi;
 };
 
+module = {
+  name = efienv;
+  common = commands/efi/env.c;
+  enable = efi;
+};
+
 module = {
   name = efifwsetup;
   efi = commands/efi/efifwsetup.c;
grub-core/commands/efi/env.c new file +182 -0

182 lines not on this page · raw message

grub-core/kern/efi/efi.c +3 -0
index 4b60495333e0..2b6bd7e2413d 100644
@@ -224,6 +224,9 @@ grub_efi_set_variable_with_attributes (const char *var, const grub_guid_t *guid,
   if (status == GRUB_EFI_SUCCESS)
     return GRUB_ERR_NONE;
 
+  if (status == GRUB_EFI_NOT_FOUND && datasize == 0)
+    return GRUB_ERR_NONE;
+
   return grub_error (GRUB_ERR_IO, "could not set EFI variable `%s'", var);
 }
 
grub-core/lib/envblk.c +43 -0

49 lines not on this page · raw message

include/grub/lib/envblk.h +3 -0
index c3e655921709..ab969af24612 100644
@@ -22,6 +22,8 @@
 #define GRUB_ENVBLK_SIGNATURE	"# GRUB Environment Block\n"
 #define GRUB_ENVBLK_DEFCFG	"grubenv"
 
+#define DEFAULT_ENVBLK_SIZE	1024
+
 #ifndef ASM_FILE
 
 struct grub_envblk
@@ -33,6 +35,7 @@ typedef struct grub_envblk *grub_envblk_t;
 
 grub_envblk_t grub_envblk_open (char *buf, grub_size_t size);
 int grub_envblk_set (grub_envblk_t envblk, const char *name, const char *value);
+grub_err_t grub_envblk_get (grub_envblk_t envblk, const char * const name, char ** const value);
 void grub_envblk_delete (grub_envblk_t envblk, const char *name);
 void grub_envblk_iterate (grub_envblk_t envblk,
                           void *hook_data,
Signature · 1 line
2.34.1
#3 lore raw

PATCH 3/4 tests: Add efienv_test for testing efi-export-env and efi-load-env

In reply to Glenn Washburn
---
 Makefile.util.def    |  6 +++++
 tests/efienv_test.in | 57 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 63 insertions(+)
 create mode 100644 tests/efienv_test.in
Makefile.util.def +6 -0
index 1e9a13d3e348..a7cf53111221 100644
@@ -1172,6 +1172,12 @@ script = {
   common = tests/help_test.in;
 };
 
+script = {
+  testcase = nonnative;
+  name = efienv_test;
+  common = tests/efienv_test.in;
+};
+
 script = {
   testcase = nonnative;
   name = grub_script_gettext;
tests/efienv_test.in new file +57 -0

57 lines not on this page · raw message

Signature · 1 line
2.34.1
#4 lore raw

PATCH 4/4 efi: Print GDB command for loading symbols if variable exists

In reply to Glenn Washburn

If the variable "enable_earlyinit_gdbinfo" exists, print the GDB command needed to properly load symbols in an attached GDB session. This variable must come from the environment block stored in the GRUB_ENV EFI variable, which allows this to get printed as early as possible.

---
 grub-core/kern/efi/init.c | 6 ++++++
 1 file changed, 6 insertions(+)
grub-core/kern/efi/init.c +6 -0
index 18329b1eacd7..3dce0308cce8 100644
@@ -142,6 +142,8 @@ grub_efi_env_init (void)
 void
 grub_efi_init (void)
 {
+  const char *value;
+
   grub_modbase = grub_efi_section_addr ("mods");
   /* First of all, initialize the console so that GRUB can display
      messages.  */
@@ -155,6 +157,10 @@ grub_efi_init (void)
   /* Populate environment with variables from EFI envblk, if it exists.  */
   grub_efi_env_init ();
 
+  value = grub_env_get ("enable_earlyinit_gdbinfo");
+  if (value != NULL && !grub_strcmp (value, "1"))
+    grub_efi_print_gdb_info ();
+
   /*
    * Lockdown the GRUB and register the shim_lock verifier
    * if the UEFI Secure Boot is enabled.
Signature · 1 line
2.34.1
#5 lore raw
In reply to Glenn Washburn

Have you considered that some firmwares brick if EFI storage is over 50% full? Why not having a file on ESP instead?

Quoted 62 lines · Glenn Washburn

This patch series (for me) was motivated by the "gdb: Add more support for debugging on EFI platforms" patch series, which allowed printing in early EFI init of the GDB command for properly loading symbols. That approach of having the functionality be included at compile time was ultimately rejected. During the discussion of that series, Robbie suggested[1] using patches by Peter and in the Redhat downstream repo which uses an EFI variable to store a GRUB env block. Using this, a user could store a variable in the env block stored in the EFI variable and then have GRUB load that env block in early init as a way to enable the printing of the GDB command.

I've taken the original patches by Peter, made more suitable for including in GRUB, fixed some bugs, and added a minor feature. The first patch, adds env block loading in early EFI init from the GRUB_ENV EFI variable. The second patch is included to provide tools for GRUB to set this EFI env block itself, as opposed to needing to use the method suggested by Robbie[1], which requires GNU/Linux and the user-space grub2-editenv utility. Of course, if GRUB is crashing before one can run the efi-export-env command, then other ways of creating and setting the EFI envblk variable are necessary. The third patch adds a test which checks the usability of the commands and the early init loading. And the last patch, whichvmotivated this series, prints the GDB command string if the GRUB variable named "enable_earlyinit_gdbinfo" is present and is set to "1", after
the env block is loaded from the GRUB_ENV EFI variable. We might want a different name for the GRUB variable, I don't really have a preference.

Glenn

[1] https://mail.gnu.org/archive/html/grub-devel/2023-03/msg00072.html

Glenn Washburn (2):
tests: Add efienv_test for testing efi-export-env and efi-load-env
efi: Print GDB command for loading symbols if variable exists

Peter Jones (2):
efi: Load env block from GRUB_ENV EFI variable in early init
efi: Add efi-export-env and efi-load-env commands

 Makefile.util.def            |   6 ++
 docs/grub.texi               |  24 +++++
 grub-core/Makefile.core.def  |   7 ++
 grub-core/commands/efi/env.c | 182 +++++++++++++++++++++++++++++++++++
 grub-core/kern/efi/efi.c     |   3 +
 grub-core/kern/efi/init.c    |  37 +++++++
 grub-core/lib/envblk.c       |  43 +++++++++
 include/grub/efi/efi.h       |   5 +
 include/grub/lib/envblk.h    |   3 +
 tests/efienv_test.in         |  57 +++++++++++
 10 files changed, 367 insertions(+)
 create mode 100644 grub-core/commands/efi/env.c
 create mode 100644 tests/efienv_test.in

--
2.34.1

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel

#6 lore raw
Quoted 2 lines · On Thu, 27 Jul 2023 00:49:06 +0300 "Vladimir 'phcoder' Serbinenko"…

Have you considered that some firmwares brick if EFI storage is over 50% full? Why not having a file on ESP instead?

What if the ESP is readonly? What if there is no ESP? I take your point about crappy firmwares. As mentioned these patches come from RedHat's downstream GRUB and I believe they are using them in production. So apparently this hasn't been an issue for their users. Now I suspect its probably almost never getting used, so perhaps its a low probability that the issue you mention is seen.

Should there be giant warnings in the usage, documentation, or command execution (with confirmation)? How big of a problem is it really (considering the likelihood that this might be used on one of those machines)?

Glenn

Quoted 66 lines
Quoted 62 lines · Glenn Washburn

This patch series (for me) was motivated by the "gdb: Add more support for debugging on EFI platforms" patch series, which allowed printing in early EFI init of the GDB command for properly loading symbols. That approach of having the functionality be included at compile time was ultimately rejected. During the discussion of that series, Robbie suggested[1] using patches by Peter and in the Redhat downstream repo which uses an EFI variable to store a GRUB env block. Using this, a user could store a variable in the env block stored in the EFI variable and then have GRUB load that env block in early init as a way to enable the printing of the GDB command.

I've taken the original patches by Peter, made more suitable for including in GRUB, fixed some bugs, and added a minor feature. The first patch, adds env block loading in early EFI init from the GRUB_ENV EFI variable. The second patch is included to provide tools for GRUB to set this EFI env block itself, as opposed to needing to use the method suggested by Robbie[1], which requires GNU/Linux and the user-space grub2-editenv utility. Of course, if GRUB is crashing before one can run the efi-export-env command, then other ways of creating and setting the EFI envblk variable are necessary. The third patch adds a test which checks the usability of the commands and the early init loading. And the last patch, whichvmotivated this series, prints the GDB command string if the GRUB variable named "enable_earlyinit_gdbinfo" is present and is set to "1", after
the env block is loaded from the GRUB_ENV EFI variable. We might want a different name for the GRUB variable, I don't really have a preference.

Glenn

[1] https://mail.gnu.org/archive/html/grub-devel/2023-03/msg00072.html

Glenn Washburn (2):
tests: Add efienv_test for testing efi-export-env and efi-load-env
efi: Print GDB command for loading symbols if variable exists

Peter Jones (2):
efi: Load env block from GRUB_ENV EFI variable in early init
efi: Add efi-export-env and efi-load-env commands

 Makefile.util.def            |   6 ++
 docs/grub.texi               |  24 +++++
 grub-core/Makefile.core.def  |   7 ++
 grub-core/commands/efi/env.c | 182 +++++++++++++++++++++++++++++++++++
 grub-core/kern/efi/efi.c     |   3 +
 grub-core/kern/efi/init.c    |  37 +++++++
 grub-core/lib/envblk.c       |  43 +++++++++
 include/grub/efi/efi.h       |   5 +
 include/grub/lib/envblk.h    |   3 +
 tests/efienv_test.in         |  57 +++++++++++
 10 files changed, 367 insertions(+)
 create mode 100644 grub-core/commands/efi/env.c
 create mode 100644 tests/efienv_test.in

--
2.34.1

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel

#7 lore raw
In reply to Vladimir 'phcoder' Serbinenko
Quoted 3 lines · On Wed, 26 Jul 2023, 23:03 Vladimir 'phcoder' Serbinenko, <[email protected]>…

Have you considered that some firmwares brick if EFI storage is over 50% full? Why not having a file on ESP instead?

Yes bricking did happen before. It was triggered by excessive runtime updates of bootloader settings. It did surface bugs of reclaiming variable storage which got fixed and released by most manufacturers. Due to recent revocations EFI storage is now routinely excercises meaning it is now reliable to use. EFI storage is used by kernel pstore too. Unlike read-only ESP EFI storage can be ephemeral r/w through the lifecycle of VM across reboots without a shutdown. Which is better than ESP.

Quoted 71 lines
Quoted 54 lines · Glenn Washburn

This patch series (for me) was motivated by the "gdb: Add more support for debugging on EFI platforms" patch series, which allowed printing in early EFI init of the GDB command for properly loading symbols. That approach of having the functionality be included at compile time was ultimately rejected. During the discussion of that series, Robbie suggested[1] using patches by Peter and in the Redhat downstream repo which uses an EFI variable to store a GRUB env block. Using this, a user could store a variable in the env block stored in the EFI variable and then have GRUB load that env block in early init as a way to enable the printing of the GDB command.

I've taken the original patches by Peter, made more suitable for including in GRUB, fixed some bugs, and added a minor feature. The first patch, adds env block loading in early EFI init from the GRUB_ENV EFI variable. The second patch is included to provide tools for GRUB to set this EFI env block itself, as opposed to needing to use the method suggested by Robbie[1], which requires GNU/Linux and the user-space grub2-editenv utility. Of course, if GRUB is crashing before one can run the efi-export-env command, then other ways of creating and setting the EFI envblk variable are necessary. The third patch adds a test which checks the
usability of the commands and the early init loading. And the last patch, whichvmotivated this series, prints the GDB command string if the GRUB variable named "enable_earlyinit_gdbinfo" is present and is set to "1", after
the env block is loaded from the GRUB_ENV EFI variable. We might want a different name for the GRUB variable, I don't really have a preference.

Glenn

[1] https://mail.gnu.org/archive/html/grub-devel/2023-03/msg00072.html

Glenn Washburn (2):
tests: Add efienv_test for testing efi-export-env and efi-load-env
efi: Print GDB command for loading symbols if variable exists

Peter Jones (2):
efi: Load env block from GRUB_ENV EFI variable in early init
efi: Add efi-export-env and efi-load-env commands

 Makefile.util.def            |   6 ++
 docs/grub.texi               |  24 +++++
 grub-core/Makefile.core.def  |   7 ++
 grub-core/commands/efi/env.c | 182 +++++++++++++++++++++++++++++++++++
 grub-core/kern/efi/efi.c     |   3 +
 grub-core/kern/efi/init.c    |  37 +++++++
 grub-core/lib/envblk.c       |  43 +++++++++
 include/grub/efi/efi.h       |   5 +
 include/grub/lib/envblk.h    |   3 +
 tests/efienv_test.in         |  57 +++++++++++
 10 files changed, 367 insertions(+)
 create mode 100644 grub-core/commands/efi/env.c
 create mode 100644 tests/efienv_test.in
Quoted 9 lines

--
2.34.1

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel

#8 lore raw
In reply to Glenn Washburn
Quoted 11 lines · Glenn Washburn

This patch series (for me) was motivated by the "gdb: Add more support for debugging on EFI platforms" patch series, which allowed printing in early EFI init of the GDB command for properly loading symbols. That approach of having the functionality be included at compile time was ultimately rejected. During the discussion of that series, Robbie suggested[1] using patches by Peter and in the Redhat downstream repo which uses an EFI variable to store a GRUB env block. Using this, a user could store a variable in the env block stored in the EFI variable and then have GRUB load that env block in early init as a way to enable the printing of the GDB command.

Quoted 14 lines

I've taken the original patches by Peter, made more suitable for including in GRUB, fixed some bugs, and added a minor feature. The first patch, adds env block loading in early EFI init from the GRUB_ENV EFI variable. The second patch is included to provide tools for GRUB to set this EFI env block itself, as opposed to needing to use the method suggested by Robbie[1], which requires GNU/Linux and the user-space grub2-editenv utility. Of course, if GRUB is crashing before one can run the efi-export-env command, then other ways of creating and setting the EFI envblk variable are necessary. The third patch adds a test which checks the usability of the commands and the early init loading. And the last patch, whichvmotivated this series, prints the GDB command string if the GRUB variable named "enable_earlyinit_gdbinfo" is present and is set to "1", after the env block is loaded from the GRUB_ENV EFI variable. We might want a different name for the GRUB variable, I don't really have a preference.

I am torn apart. On one hand it is nice feature. On the other hand it requires a lot of changes. So, I would prefer to defer merging this thing after the release. Until you convince me I should do otherwise...

Daniel

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel

#9 lore raw

Adding a few others in case interested...

Quoted 27 lines · On Thu, 14 Sep 2023 18:47:13 +0200 Daniel Kiper <[email protected]> wrote:
Quoted 11 lines · Glenn Washburn

This patch series (for me) was motivated by the "gdb: Add more support for debugging on EFI platforms" patch series, which allowed printing in early EFI init of the GDB command for properly loading symbols. That approach of having the functionality be included at compile time was ultimately rejected. During the discussion of that series, Robbie suggested[1] using patches by Peter and in the Redhat downstream repo which uses an EFI variable to store a GRUB env block. Using this, a user could store a variable in the env block stored in the EFI variable and then have GRUB load that env block in early init as a way to enable the printing of the GDB command.

Quoted 14 lines

I've taken the original patches by Peter, made more suitable for including in GRUB, fixed some bugs, and added a minor feature. The first patch, adds env block loading in early EFI init from the GRUB_ENV EFI variable. The second patch is included to provide tools for GRUB to set this EFI env block itself, as opposed to needing to use the method suggested by Robbie[1], which requires GNU/Linux and the user-space grub2-editenv utility. Of course, if GRUB is crashing before one can run the efi-export-env command, then other ways of creating and setting the EFI envblk variable are necessary. The third patch adds a test which checks the usability of the commands and the early init loading. And the last patch, whichvmotivated this series, prints the GDB command string if the GRUB variable named "enable_earlyinit_gdbinfo" is present and is set to "1", after the env block is loaded from the GRUB_ENV EFI variable. We might want a different name for the GRUB variable, I don't really have a preference.

Quoted 3 lines

I am torn apart. On one hand it is nice feature. On the other hand it requires a lot of changes. So, I would prefer to defer merging this thing after the release. Until you convince me I should do otherwise...

I would contend that there's only few small changes that matter. Most of the changes are only executed if the user explicitly does something (ie set an EFI variable), so its opt-in mostly. And it would be nearly impossible to accidentally opt-in (especially considering there's no documentation on the name of the special variable "enable_earlyinit_gdbinfo", which I'm still not convinced is the right name). And the changes that do matter are in old code that I'd have expected to shown an issue by now. Here's my analysis of those.

Patch 1:
* grub_efi_set_variable() is added to grub_efi_init(). This looks scary, but all its doing is calling grub_efi_get_variable() and returning (assuming the user has not done the manual opt in previously). grub_efi_get_variable() is small and well-used. I would say this is low risk.

Patch 2:
* grub_efi_set_variable_with_attributes() is changed to have a different return value (ie to succeed) when the datasize is 0 and the variable does not exist. There are very few code paths to this function. I've checked the existing ones and none present a problem.

Patch 3:
* No changes that matter.

Patch 4:
* grub_env_get() is called in grub_efi_init(). The initial envblk is setup at load time to have no vars (in the BSS), so this call will fail quickly if using the initial envblk, which should be the case, unless the user has explicitly already run the "efi-export-env" command (which means there wasn't a failure on a previous run up to this point with these changes). The "if" block after will never be executed nor will with grub_strcmp(), unless there was explicit interaction as stated above.

So the vast majority of new/"untested" code (though I have tested with the test in patch 3), never gets run unless the user opts in. In that case, we do not want to make the situation worse, but its already in a state if they are wanting to enable this feature (ie can't get to shell to run gdbinfo). Supposing that there is a bug in the "opt-in" code paths that cause GRUB to not reach the shell, the user then needs to run either the EFI shell or some other bootloader that can delete UEFI environment variables. While this could be a pain, I would expect this to be a reasonable ask when GRUB is already in a very messed up state (and a very low probability of this situation ever even occurring).

These don't appear to be the kind of changes that would have issue with non-x86 architectures (for which it hasn't been tested on _baremetal_, with QEMU it has). And considering that UEFI variable getting/setting has been around a long time, I'd expect that to be well tested and not a source of risk here.

I'll admit that including these changes, will probably be low impact as this is for really bad situations. However, that's precisely when you want these kinds of changes. And I think its worth considering that there's a large base of users using similar code via RedHat's patched GRUB. So I think its a pretty low risk overall. Since this is something I expect to be used more by the distros (I doubt we'll see anyone needing this kind of help here), I wonder if they have any thoughts on this issue?

Glenn

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel