efivarfs: fix abnormal GUID in variable name by using strcpy to replace null with dash
PATCH 1/1 efivarfs: fix abnormal GUID in variable name by using strcpy to replace null with dash
From: Michael Schroeder <mls-IBi9RG/[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d /sys/firmware/efi/vars/*8be4d* | grep -v -- -8be returns /sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8c
That causes by the following statement in efivar_create_sysfs_entry function:
*(short_name + strlen(short_name)) = '-';efi_guid_unparse(vendor_guid, short_name + strlen(short_name));
The trailing \0 is overwritten with '-', but the next char doesn't seem to be a \0 as well for HP. So, the second strlen return the point of next '\0', causes there have garbage string attached before GUID.
Tested on On HP z220.
--- drivers/firmware/efivars.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/firmware/efivars.c +1 -1
@@ -1708,7 +1708,7 @@ efivar_create_sysfs_entry(struct efivars *efivars,/* This is ugly, but necessary to separate one vendor's private variables from another's. */ - *(short_name +strlen (short_name)) ='-' ; +strcpy (short_name +strlen (short_name),"-" );efi_guid_unparse (vendor_guid, short_name +strlen (short_name)); new_efivar->kobj.kset = efivars->kset;
Signature · 1 line
1.6.0.2
Quoted 13 lines · Chun-Yi
From: Michael Schroeder <[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cThat causes by the following statement in
efivar_create_sysfs_entryfunction:
Quoted 8 lines
*(short_name +
strlen(short_name)) = '-';efi_guid_unparse(vendor_guid, short_name + strlen(short_name));The trailing \0 is overwritten with '-', but the next char doesn't seem to be a \0 as well for HP. So, the second strlen return the point of next '\0', causes there have garbage string attached before GUID.
Tested on On HP z220.
So short_name has trailing garbage, or rather, variable_name_size is larger than variable_name actually is, wouldn't new_efivar->var.VariableName also gets filled with trailing garbage?
In efivar_store_raw the VariableName's trailing garbage can cause problem for comparison. You might want to also cover that or fix variable_name_size.
Lingzhu Xiang
於 五,2013-03-01 於 17:31 +0800,Lingzhu Xiang 提到:
Quoted 27 lines · Chun-Yi
Quoted 13 lines
From: Michael Schroeder <mls-IBi9RG/[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cThat causes by the following statement in
efivar_create_sysfs_entryfunction:Quoted 8 lines
*(short_name +
strlen(short_name)) = '-';efi_guid_unparse(vendor_guid, short_name + strlen(short_name));The trailing \0 is overwritten with '-', but the next char doesn't seem to be a \0 as well for HP. So, the second strlen return the point of next '\0', causes there have garbage string attached before GUID.
Tested on On HP z220.
So
short_namehas trailing garbage, or rather,variable_name_sizeis larger thanvariable_nameactually is, wouldn'tnew_efivar->var.VariableNamealso gets filled with trailing garbage?
Quoted 7 lines
In
efivar_store_rawthe VariableName's trailing garbage can cause problem for comparison. You might want to also cover that or fixvariable_name_size.Lingzhu Xiang
Yes, it's possible! I will add patch for print more debug messages to bug reporter for check it.
Thanks for your review!
Joey Lee
Quoted 13 lines · Chun-Yi
From: Michael Schroeder <mls-IBi9RG/[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cThat causes by the following statement in
efivar_create_sysfs_entryfunction:
Quoted 8 lines
*(short_name +
strlen(short_name)) = '-';efi_guid_unparse(vendor_guid, short_name + strlen(short_name));The trailing \0 is overwritten with '-', but the next char doesn't seem to be a \0 as well for HP. So, the second strlen return the point of next '\0', causes there have garbage string attached before GUID.
Tested on On HP z220.
What's more likely happening here is that GetNextVariable() is broken on this HP firmware and variable_name_size is too big for the given variable in variable_name. We've seen other reports of similar bugs,
https://bugzilla.kernel.org/show_bug.cgi?id=47631
Could someone try this patch against Linus' tree?
---
Quoted 1 line
From
fe38d6f69c889b0f95c5548c724633aa58c2c99fMon Sep 17 00:00:00 2001
From: Matt Fleming <[email protected]>
Date: Fri, 1 Mar 2013 14:49:12 +0000
Subject: [PATCH] efivars: Sanitise string length returned by
GetNextVariableName()
Some buggy firmware implementations return a string length from GetNextVariableName() that is actually larger than the string in 'variable_name', as Michael Schroeder writes,
Quoted 8 lines
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8c
Since 'variable_name' is a string, we can validate its size by searching for the terminating NULL character.
--- drivers/firmware/efivars.c | 35 ++++++++++++++++++++++++++++++++++- 1 file changed, 34 insertions(+), 1 deletion(-)
drivers/firmware/efivars.c +34 -1
@@ -1653,6 +1653,33 @@ static bool variable_is_present(efi_char16_t *variable_name, efi_guid_t *vendor)return found; } +/* +* Sanity check string length of a variable name. +*/ +static unsigned long sanity_check_strlen (efi_char16_t *variable_name, +unsigned long variable_name_size) +{ +efi_char16_t c; +unsigned long len; + +/* +* The variable name is, by definition, a NULL-terminated +* string, so make absolutely sure that variable_name_size is +* the value we expect it to be. If not, use whichever size is +* smaller. +*/ +for (len =1 ; len <= variable_name_size; len++) { + c = variable_name[len -1 ]; +if (!c) +break ; + } + +if (len != variable_name_size) +printk (KERN_WARNING "efivars: bogus variable_name_size\n" ); + +return min (len, variable_name_size); +} +static void efivar_update_sysfs_entries (struct work_struct *work) {struct efivars *efivars = &__efivars;
@@ -1693,10 +1720,13 @@ static void efivar_update_sysfs_entries(struct work_struct *work)if (!found) {kfree (variable_name);break ; - }else + }else { + variable_name_size =sanity_check_strlen (variable_name, + variable_name_size);efivar_create_sysfs_entry (efivars, variable_name_size, variable_name, &vendor); + } } }
@@ -1941,8 +1971,11 @@ int register_efivars(struct efivars *efivars, status = ops->get_next_variable (&variable_name_size, variable_name, &vendor_guid); +switch (status) {case EFI_SUCCESS : + variable_name_size =sanity_check_strlen (variable_name, + variable_name_size); efivar_create_sysfs_entry(efivars, variable_name_size, variable_name,
Signature · 1 line
1.7.11.7
Quoted 23 lines · Matt Fleming
Quoted 13 lines · Chun-Yi
From: Michael Schroeder <mls-IBi9RG/[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cThat causes by the following statement in
efivar_create_sysfs_entryfunction:Quoted 8 lines
*(short_name +
strlen(short_name)) = '-';efi_guid_unparse(vendor_guid, short_name + strlen(short_name));The trailing \0 is overwritten with '-', but the next char doesn't seem to be a \0 as well for HP. So, the second strlen return the point of next '\0', causes there have garbage string attached before GUID.
Tested on On HP z220.
Quoted 8 lines
What's more likely happening here is that
GetNextVariable()is broken on this HP firmware andvariable_name_sizeis too big for the given variable invariable_name. We've seen other reports of similar bugs,https://bugzilla.kernel.org/show_bug.cgi?id=47631Could someone try this patch against Linus' tree?
Urgh, and here's a version that isn't utterly, utterly broken...
---
Quoted 1 line
From
4b2ef72bca72039717efe4570ec858a86d565b34Mon Sep 17 00:00:00 2001
From: Matt Fleming <[email protected]>
Date: Fri, 1 Mar 2013 14:49:12 +0000
Subject: [PATCH] efivars: Sanitise string length returned by
GetNextVariableName()
Some buggy firmware implementations return a string length from GetNextVariableName() that is actually larger than the string in 'variable_name', as Michael Schroeder writes,
Quoted 8 lines
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8c
Since 'variable_name' is a string, we can validate its size by searching for the terminating NULL character.
--- drivers/firmware/efivars.c | 30 ++++++++++++++++++++++++++++++ 1 file changed, 30 insertions(+)
drivers/firmware/efivars.c +30 -0
@@ -1895,6 +1895,33 @@ void unregister_efivars(struct efivars *efivars) }EXPORT_SYMBOL_GPL (unregister_efivars); +/* +* Sanity check size of a variable name. +*/ +static unsigned long sanity_check_size (efi_char16_t *variable_name, +unsigned long variable_name_size) +{ +unsigned long len; +efi_char16_t c; + +/* +* The variable name is, by definition, a NULL-terminated +* string, so make absolutely sure that variable_name_size is +* the value we expect it to be. If not, return the real size. +*/ +for (len =2 ; len <= variable_name_size; len +=sizeof (c)) { + c = variable_name[(len /sizeof (c)) -1 ]; +if (!c) +break ; + } + + +if (len != variable_name_size) +printk (KERN_WARNING "efivars: bogus variable_name_size: %lu %lu\n" , len, variable_name_size); + +return min (len, variable_name_size); +} +int register_efivars (struct efivars *efivars,const struct efivar_operations *ops,struct kobject *parent_kobj)
@@ -1941,8 +1968,11 @@ int register_efivars(struct efivars *efivars, status = ops->get_next_variable (&variable_name_size, variable_name, &vendor_guid); +switch (status) {case EFI_SUCCESS : + variable_name_size =sanity_check_size (variable_name, + variable_name_size); efivar_create_sysfs_entry(efivars, variable_name_size, variable_name,
Signature · 1 line
1.7.11.7
於 五,2013-03-01 於 16:31 +0000,Matt Fleming 提到:
Quoted 28 lines · Matt Fleming
Quoted 27 lines · Chun-Yi
Quoted 13 lines
From: Michael Schroeder <mls-IBi9RG/[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cThat causes by the following statement in
efivar_create_sysfs_entryfunction:Quoted 8 lines
*(short_name +
strlen(short_name)) = '-';efi_guid_unparse(vendor_guid, short_name + strlen(short_name));The trailing \0 is overwritten with '-', but the next char doesn't seem to be a \0 as well for HP. So, the second strlen return the point of next '\0', causes there have garbage string attached before GUID.
Tested on On HP z220.
What's more likely happening here is that
GetNextVariable()is broken on this HP firmware andvariable_name_sizeis too big for the given variable invariable_name. We've seen other reports of similar bugs,
Quoted 6 lines
Quoted 4 lines
https://bugzilla.kernel.org/show_bug.cgi?id=47631Could someone try this patch against Linus' tree?
Urgh, and here's a version that isn't utterly, utterly broken...
Thanks for Matt's patch, we will try it!
Joey Lee
Quoted 86 lines
---
Quoted 1 line
From
4b2ef72bca72039717efe4570ec858a86d565b34Mon Sep 17 00:00:00 2001From: Matt Fleming <[email protected]>
Date: Fri, 1 Mar 2013 14:49:12 +0000
Subject: [PATCH] efivars: Sanitise string length returned by
GetNextVariableName()Some buggy firmware implementations return a string length from
GetNextVariableName()that is actually larger than the string in 'variable_name', as Michael Schroeder writes,Quoted 8 lines
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cSince '
variable_name' is a string, we can validate its size by searching for the terminating NULL character.Cc Matthew Garrett <[email protected]>Cc Josh Boyer <[email protected]>Cc Michael Schroeder <mls-IBi9RG/[email protected]>Cc Lee, Chun-Yi <jlee-IBi9RG/[email protected]>Cc Lingzhu Xiang <[email protected]>Signed-off-by Matt Fleming <[email protected]>--- drivers/firmware/efivars.c | 30 ++++++++++++++++++++++++++++++ 1 file changed, 30 insertions(+)drivers/firmware/efivars.c +30
-0@@ -1895,6 +1895,33 @@ void unregister_efivars(struct efivars *efivars) }EXPORT_SYMBOL_GPL (unregister_efivars); +/* +* Sanity check size of a variable name. +*/ +static unsigned long sanity_check_size (efi_char16_t *variable_name, +unsigned long variable_name_size) +{ +unsigned long len; +efi_char16_t c; + +/* +* The variable name is, by definition, a NULL-terminated +* string, so make absolutely sure that variable_name_size is +* the value we expect it to be. If not, return the real size. +*/ +for (len =2 ; len <= variable_name_size; len +=sizeof (c)) { + c = variable_name[(len /sizeof (c)) -1 ]; +if (!c) +break ; + } + + +if (len != variable_name_size) +printk (KERN_WARNING "efivars: bogus variable_name_size: %lu %lu\n" , len, variable_name_size); + +return min (len, variable_name_size); +} +int register_efivars (struct efivars *efivars,const struct efivar_operations *ops,struct kobject *parent_kobj)@@ -1941,8 +1968,11 @@ int register_efivars(struct efivars *efivars, status = ops->get_next_variable (&variable_name_size, variable_name, &vendor_guid); +switch (status) {case EFI_SUCCESS : + variable_name_size =sanity_check_size (variable_name, + variable_name_size); efivar_create_sysfs_entry(efivars, variable_name_size, variable_name,
joeyli
Tested-by Frederic CrozatHi Matt,
於 六,2013-03-02 於 07:41 +0800,joeyli 提到:
Quoted 32 lines
於 五,2013-03-01 於 16:31 +0000,Matt Fleming 提到:
Quoted 31 lines · Matt Fleming
Quoted 30 lines · Chun-Yi
Quoted 13 lines
From: Michael Schroeder <mls-IBi9RG/[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cThat causes by the following statement in
efivar_create_sysfs_entryfunction:Quoted 8 lines
*(short_name +
strlen(short_name)) = '-';efi_guid_unparse(vendor_guid, short_name + strlen(short_name));The trailing \0 is overwritten with '-', but the next char doesn't seem to be a \0 as well for HP. So, the second strlen return the point of next '\0', causes there have garbage string attached before GUID.
Tested on On HP z220.
What's more likely happening here is that
GetNextVariable()is broken on this HP firmware andvariable_name_sizeis too big for the given variable invariable_name. We've seen other reports of similar bugs,https://bugzilla.kernel.org/show_bug.cgi?id=47631
Quoted 7 lines
Quoted 3 lines
Quoted 1 line
Could someone try this patch against Linus' tree?
Urgh, and here's a version that isn't utterly, utterly broken...
Thanks for Matt's patch, we will try it!
Joey Lee
Frederic confirmed your patch works to him for fix issue on HP machine.
But, I suggest we just use the length with NULL in variable_name but not VariableNameSize that was returned by GetNextVariableName.
My thinking is following...
Quoted 57 lines
Quoted 57 lines
---
Quoted 1 line
From
4b2ef72bca72039717efe4570ec858a86d565b34Mon Sep 17 00:00:00 2001From: Matt Fleming <[email protected]>
Date: Fri, 1 Mar 2013 14:49:12 +0000
Subject: [PATCH] efivars: Sanitise string length returned by
GetNextVariableName()Some buggy firmware implementations return a string length from
GetNextVariableName()that is actually larger than the string in 'variable_name', as Michael Schroeder writes,Quoted 8 lines
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cSince '
variable_name' is a string, we can validate its size by searching for the terminating NULL character.Cc Matthew Garrett <[email protected]>Cc Josh Boyer <[email protected]>Cc Michael Schroeder <mls-IBi9RG/[email protected]>Cc Lee, Chun-Yi <jlee-IBi9RG/[email protected]>Cc Lingzhu Xiang <[email protected]>Signed-off-by Matt Fleming <[email protected]>--- drivers/firmware/efivars.c | 30 ++++++++++++++++++++++++++++++ 1 file changed, 30 insertions(+)drivers/firmware/efivars.c +12
-0@@ -1895,6 +1895,33 @@ void unregister_efivars(struct efivars *efivars) }EXPORT_SYMBOL_GPL (unregister_efivars); +/* +* Sanity check size of a variable name. +*/ +static unsigned long sanity_check_size (efi_char16_t *variable_name, +unsigned long variable_name_size) +{ +unsigned long len; +efi_char16_t c; + +/* +* The variable name is, by definition, a NULL-terminated +* string, so make absolutely sure that variable_name_size is
Quoted 8 lines
drivers/firmware/efivars.c void unregister_efivars(struct efivars *efivars) · line 1917
Quoted 8 lines
drivers/firmware/efivars.c +8
-0+* the value we expect it to be. If not, return the real size. +*/ +for (len =2 ; len <= variable_name_size; len +=sizeof (c)) { + c = variable_name[(len /sizeof (c)) -1 ]; +if (!c) +break ; + } +
We just direct return the len here but don't need compare with variable_name_size.
return len;
Quoted 7 lines
drivers/firmware/efivars.c void unregister_efivars(struct efivars *efivars) · line 1924
Quoted 7 lines
drivers/firmware/efivars.c +7
-0+ +if (len != variable_name_size) +printk (KERN_WARNING "efivars: bogus variable_name_size: %lu %lu\n" , len, variable_name_size); + +return min (len, variable_name_size); +} +
In UEFI 2.3.1 spec:
VariableNameSize The size of the VariableName buffer
Note that if EFI_BUFFER_TOO_SMALL is returned, the VariableName buffer was too small for the next variable. When such an error occurs, the VariableNameSize is updated to reflect the size of buffer needed. In all cases when calling GetNextVariableName() the VariableNameSize must not exceed the actual buffer size that was allocated for VariableName.
Per spec, VariableNameSize will updated to 'the size of buffer needed' when EFI_BUFFER_TOO_SMALL, but spec doesn't mention VariableNameSize will also updated when EFI_SUCCESS. That means 'VariableNameSize = 1024' when EFI_SUCCESS is not a bogus value.
Quoted 7 lines
Quoted 7 lines
drivers/firmware/efivars.c +0
-0int register_efivars (struct efivars *efivars,const struct efivar_operations *ops,struct kobject *parent_kobj)@@ -1941,8 +1968,11 @@ int register_efivars(struct efivars *efivars, status = ops->get_next_variable (&variable_name_size, variable_name, &vendor_guid);
Quoted 9 lines
drivers/firmware/efivars.c int register_efivars(struct efivars *efivars, · line 1978
Quoted 8 lines
drivers/firmware/efivars.c +3
-0+switch (status) {case EFI_SUCCESS : + variable_name_size =sanity_check_size (variable_name, + variable_name_size); efivar_create_sysfs_entry(efivars, variable_name_size, variable_name,
Due to variable_name_size could be 1024, so I suggest direct feed the length of variable_name to efivar_create_sysfs_entry since we are need to count the length.
This is a simply diff for my think:
drivers/firmware/efivars.c +16 -1
@@ -1895,6 +1895,21 @@ void unregister_efivars(struct efivars *efivars) }EXPORT_SYMBOL_GPL (unregister_efivars); +static unsigned long variable_name_length (efi_char16_t *variable_name) +{ +unsigned long len; +efi_char16_t c; + + len =2 ; +do { + c = variable_name[len /sizeof (c) -1 ]; +if (c) + len +=sizeof (c); + }while (c); + +return len; +} +int register_efivars (struct efivars *efivars,const struct efivar_operations *ops,struct kobject *parent_kobj)
@@ -1944,7 +1959,7 @@ int register_efivars(struct efivars *efivars,switch (status) {case EFI_SUCCESS :efivar_create_sysfs_entry (efivars, - variable_name_size, +variable_name_length (variable_name), variable_name, &vendor_guid);break ;
於 三,2013-03-06 於 15:34 +0800,joeyli 提到:
Quoted 50 lines
Hi Matt,
於 六,2013-03-02 於 07:41 +0800,joeyli 提到:
Quoted 6 lines
於 五,2013-03-01 於 16:31 +0000,Matt Fleming 提到:
Quoted 5 lines · Matt Fleming
Quoted 4 lines · Chun-Yi
Quoted 3 lines
From: Michael Schroeder <mls-IBi9RG/[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables
are incorrectly
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returnsQuoted 9 lines
/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8cQuoted 1 line
Quoted 1 line
/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8cQuoted 1 line
Quoted 1 line
/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8cQuoted 1 line
Quoted 1 line
/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cQuoted 2 lines
Quoted 2 lines
Quoted 2 lines
That causes by the following statement in
efivar_create_sysfs_entryfunction:Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
*(short_name +
strlen(short_name)) = '-';
efi_guid_unparse(vendor_guid,short_name+
strlen(short_name));Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
The trailing \0 is overwritten with '-', but the next char
doesn't seem to be a \0
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
as well for HP. So, the second strlen return the point of next
'\0', causes there
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 3 lines
have garbage string attached before GUID.
Tested on On HP z220.
What's more likely happening here is that
GetNextVariable()isbroken on
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
this HP firmware and
variable_name_sizeis too big for the given variable invariable_name. We've seen other reports of similarbugs,
Quoted 8 lines
Quoted 7 lines
Quoted 5 lines
https://bugzilla.kernel.org/show_bug.cgi?id=47631
Could someone try this patch against Linus' tree?
Urgh, and here's a version that isn't utterly, utterly broken...
Quoted 8 lines
Quoted 3 lines
Thanks for Matt's patch, we will try it!
Joey Lee
Frederic confirmed your patch works to him for fix issue on HP machine.
Sorry forgot paste variable_name_size from log:
efivars: bogus variable_name_size: 34 1024
efivars: bogus variable_name_size: 22 1024
The GetNextVariable() always return variable_name_size is 1024.
Thanks a lot!
Joey Lee
Quoted 5 lines · joeyli
drivers/firmware/efivars.c +5
-0+static unsigned long variable_name_length (efi_char16_t *variable_name) +{ +unsigned long len; +efi_char16_t c; +
drivers/firmware/efivars.c void unregister_efivars(struct efivars *efivars) · line 1912
Quoted 10 lines
drivers/firmware/efivars.c +10
-0+ len =2 ; +do { + c = variable_name[len /sizeof (c) -1 ]; +if (c) + len +=sizeof (c); + }while (c); + +return len; +} +
It looks like this function can be replaced with utf16_strsize.
Quoted 10 lines · Lingzhu Xiang
Quoted 9 lines · joeyli
drivers/firmware/efivars.c +9
-0+static unsigned long variable_name_length (efi_char16_t *variable_name) +{ +unsigned long len; +efi_char16_t c; + + len =2 ; + do { + c = variable_name[len / sizeof(c) - 1]; + if (c)
Quoted 8 lines
drivers/firmware/efivars.c void unregister_efivars(struct efivars *efivars) · line 1912
Quoted 6 lines
drivers/firmware/efivars.c +6
-0+ len +=sizeof (c); + }while (c); + +return len; +} +It looks like this function can be replaced with
utf16_strsize.
Nearly. For this particular case we also need to count the terminating NULL.
Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
於 三,2013-03-06 於 17:20 +0800,Lingzhu Xiang 提到:
Quoted 10 lines · joeyli
Quoted 9 lines
drivers/firmware/efivars.c +9
-0+static unsigned long variable_name_length (efi_char16_t *variable_name) +{ +unsigned long len; +efi_char16_t c; + + len =2 ; + do { + c = variable_name[len / sizeof(c) - 1]; + if (c)
Quoted 9 lines
drivers/firmware/efivars.c void unregister_efivars(struct efivars *efivars) · line 1912
Quoted 6 lines
drivers/firmware/efivars.c +6
-0+ len +=sizeof (c); + }while (c); + +return len; +} +It looks like this function can be replaced with
utf16_strsize.
Ah! thanks for you point out, I will use utf16_strsize.
Thanks a lot!
Joey Lee
Quoted 199 lines · joeyli
Hi Matt,
於 六,2013-03-02 於 07:41 +0800,joeyli 提到:
Quoted 32 lines
於 五,2013-03-01 於 16:31 +0000,Matt Fleming 提到:
Quoted 31 lines · Matt Fleming
Quoted 30 lines · Chun-Yi
Quoted 13 lines
From: Michael Schroeder <mls-IBi9RG/[email protected]>
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cThat causes by the following statement in
efivar_create_sysfs_entryfunction:Quoted 8 lines
*(short_name +
strlen(short_name)) = '-';efi_guid_unparse(vendor_guid, short_name + strlen(short_name));The trailing \0 is overwritten with '-', but the next char doesn't seem to be a \0 as well for HP. So, the second strlen return the point of next '\0', causes there have garbage string attached before GUID.
Tested on On HP z220.
What's more likely happening here is that
GetNextVariable()is broken on this HP firmware andvariable_name_sizeis too big for the given variable invariable_name. We've seen other reports of similar bugs,https://bugzilla.kernel.org/show_bug.cgi?id=47631Quoted 7 lines
Quoted 3 lines
Quoted 1 line
Could someone try this patch against Linus' tree?
Urgh, and here's a version that isn't utterly, utterly broken...
Thanks for Matt's patch, we will try it!
Joey Lee
Frederic confirmed your patch works to him for fix issue on HP machine.
But, I suggest we just use the length with NULL in
variable_namebut not VariableNameSize that was returned by GetNextVariableName.My thinking is following...
Quoted 57 lines
Quoted 57 lines
---
Quoted 1 line
From
4b2ef72bca72039717efe4570ec858a86d565b34Mon Sep 17 00:00:00 2001From: Matt Fleming <[email protected]>
Date: Fri, 1 Mar 2013 14:49:12 +0000
Subject: [PATCH] efivars: Sanitise string length returned by
GetNextVariableName()Some buggy firmware implementations return a string length from
GetNextVariableName()that is actually larger than the string in 'variable_name', as Michael Schroeder writes,Quoted 8 lines
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8cSince '
variable_name' is a string, we can validate its size by searching for the terminating NULL character.Cc Matthew Garrett <[email protected]>Cc Josh Boyer <[email protected]>Cc Michael Schroeder <mls-IBi9RG/[email protected]>Cc Lee, Chun-Yi <jlee-IBi9RG/[email protected]>Cc Lingzhu Xiang <[email protected]>Signed-off-by Matt Fleming <[email protected]>--- drivers/firmware/efivars.c | 30 ++++++++++++++++++++++++++++++ 1 file changed, 30 insertions(+)drivers/firmware/efivars.c +12
-0@@ -1895,6 +1895,33 @@ void unregister_efivars(struct efivars *efivars) }EXPORT_SYMBOL_GPL (unregister_efivars); +/* +* Sanity check size of a variable name. +*/ +static unsigned long sanity_check_size (efi_char16_t *variable_name, +unsigned long variable_name_size) +{ +unsigned long len; +efi_char16_t c; + +/* +* The variable name is, by definition, a NULL-terminated +* string, so make absolutely sure that variable_name_size is Quoted 8 lines
drivers/firmware/efivars.c void unregister_efivars(struct efivars *efivars) · line 1917
Quoted 8 lines
drivers/firmware/efivars.c +8
-0+* the value we expect it to be. If not, return the real size. +*/ +for (len =2 ; len <= variable_name_size; len +=sizeof (c)) { + c = variable_name[(len /sizeof (c)) -1 ]; +if (!c) +break ; + } +We just direct return the len here but don't need compare with
variable_name_size.return len;Quoted 7 lines
drivers/firmware/efivars.c void unregister_efivars(struct efivars *efivars) · line 1912
Quoted 7 lines
drivers/firmware/efivars.c +7
-0+ +if (len != variable_name_size) +printk (KERN_WARNING "efivars: bogus variable_name_size: %lu %lu\n" , len, variable_name_size); + +return min (len, variable_name_size); +} +In UEFI 2.3.1 spec:
VariableNameSize The size of the VariableName buffer
Note that if
EFI_BUFFER_TOO_SMALLis returned, the VariableName buffer was too small for the next variable. When such an error occurs, the VariableNameSize is updated to reflect the size of buffer needed. In all cases when callingGetNextVariableName()the VariableNameSize must not exceed the actual buffer size that was allocated for VariableName.Per spec, VariableNameSize will updated to 'the size of buffer needed' when
EFI_BUFFER_TOO_SMALL, but spec doesn't mention VariableNameSize will also updated whenEFI_SUCCESS. That means 'VariableNameSize = 1024' whenEFI_SUCCESSis not a bogus value.Quoted 7 lines
Quoted 7 lines
drivers/firmware/efivars.c +0
-0int register_efivars (struct efivars *efivars,const struct efivar_operations *ops,struct kobject *parent_kobj)@@ -1941,8 +1968,11 @@ int register_efivars(struct efivars *efivars, status = ops->get_next_variable (&variable_name_size, variable_name, &vendor_guid);Quoted 9 lines
drivers/firmware/efivars.c int register_efivars(struct efivars *efivars, · line 1963
Quoted 8 lines
drivers/firmware/efivars.c +3
-0+switch (status) {case EFI_SUCCESS : + variable_name_size =sanity_check_size (variable_name, + variable_name_size); efivar_create_sysfs_entry(efivars, variable_name_size, variable_name,Due to
variable_name_sizecould be 1024, so I suggest direct feed the length ofvariable_nametoefivar_create_sysfs_entrysince we are need to count the length.This is a simply diff for my think:
drivers/firmware/efivars.c +15
-0@@ -1895,6 +1895,21 @@ void unregister_efivars(struct efivars *efivars) }EXPORT_SYMBOL_GPL (unregister_efivars); +static unsigned long variable_name_length (efi_char16_t *variable_name) +{ +unsigned long len; +efi_char16_t c; + + len =2 ; +do { + c = variable_name[len /sizeof (c) -1 ]; +if (c) + len +=sizeof (c); + }while (c); + +return len; +} +int register_efivars (struct efivars *efivars,const struct efivar_operations *ops,struct kobject *parent_kobj)
drivers/firmware/efivars.c int register_efivars(struct efivars *efivars, · line 1965
Quoted 8 lines
drivers/firmware/efivars.c +1
-1@@ -1944,7 +1959,7 @@ int register_efivars(struct efivars *efivars,switch (status) {case EFI_SUCCESS :efivar_create_sysfs_entry (efivars, - variable_name_size, +variable_name_length (variable_name), variable_name, &vendor_guid);break ;
Hmm.. the reason I didn't implement the patch this way is because I do think it's important to make sure we don't go out of bounds looking for the terminating NULL, i.e. you need a 'len < variable_name_size' check somewhere.
Care to update and resend your patch, ensuring we don't inspect more than variable_name_size characters?
Also, which machine did you see this behaviour on?
Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
Quoted 1 line · Matt Fleming
Also, which machine did you see this behaviour on?
HP z220 desktop, System BIOS K51 v 1.54
this issue is only the "Secure Boot" related variables, which make me think HP incorrectly created them in their latest firmware (where they added Windows 8 support).
Signature · 2 lines
Frederic Crozat <fcrozat-IBi9RG/[email protected]> SUSE
Quoted 9 lines · Frederic Crozat
Quoted 1 line · Matt Fleming
Also, which machine did you see this behaviour on?
HP z220 desktop, System BIOS K51 v 1.54
this issue is only the "Secure Boot" related variables, which make me think HP incorrectly created them in their latest firmware (where they added Windows 8 support).
Are you saying that the behaviour regarding VariableNameSize changes depending on which variable is returned?
Mercy.....
Thanks for the info.
Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
Quoted 7 lines · Matt Fleming
Quoted 6 lines · Frederic Crozat
Quoted 1 line · Matt Fleming
Also, which machine did you see this behaviour on?
HP z220 desktop, System BIOS K51 v 1.54
Quoted 8 lines
Quoted 3 lines
this issue is only the "Secure Boot" related variables, which make me think HP incorrectly created them in their latest firmware (where they added Windows 8 support).
Are you saying that the behaviour regarding VariableNameSize changes depending on which variable is returned?
Mercy.....
Well, I only get the error message for some variables (8), not all of them.
It was quite visible on unpatched kernel, with ls
-d /sys/firmware/efi/vars.
ls -d /sys/firmware/efi/vars/*8be4d* | grep -v -- -8be returns /sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/SignatureSupport-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/VendorKeys-pport8be4df61-93ca-11d2-aa0d-00e098032b8c
All other variable names were fine.
Signature · 2 lines
Frederic Crozat <fcrozat-IBi9RG/[email protected]> SUSE
Quoted 31 lines · Frederic Crozat
Quoted 7 lines · Matt Fleming
Quoted 6 lines · Frederic Crozat
Quoted 1 line · Matt Fleming
Also, which machine did you see this behaviour on?
HP z220 desktop, System BIOS K51 v 1.54
Quoted 8 lines
Quoted 3 lines
this issue is only the "Secure Boot" related variables, which make me think HP incorrectly created them in their latest firmware (where they added Windows 8 support).
Are you saying that the behaviour regarding VariableNameSize changes depending on which variable is returned?
Mercy.....
Well, I only get the error message for some variables (8), not all of them.
It was quite visible on unpatched kernel, with ls
-d/sys/firmware/efi/vars.ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SignatureSupport-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/VendorKeys-pport8be4df61-93ca-11d2-aa0d-00e098032b8c
Quoted 1 line
All other variable names were fine.
Interesting. Thanks.
Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
於 三,2013-03-06 於 11:19 +0000,Matt Fleming 提到:
Quoted 1 line · joeyli
...
Quoted 34 lines
Quoted 20 lines
(quoted diff) +15
-0+static unsigned long variable_name_length(efi_char16_t *variable_name) +{ + unsigned long len; + efi_char16_t c; + + len = 2; + do { + c = variable_name[len / sizeof(c) - 1]; + if (c) + len += sizeof(c); + } while (c); + + return len; +} + int register_efivars(struct efivars *efivars, const struct efivar_operations *ops, struct kobject *parent_kobj)Review int register_efivars(struct efivars *efivars, · line 1965
Quoted 8 lines
(quoted diff) +1
-1@@ -1944,7 +1959,7 @@ int register_efivars(struct efivars *efivars, switch (status) { case EFI_SUCCESS: efivar_create_sysfs_entry(efivars, - variable_name_size, + variable_name_length(variable_name), variable_name, &vendor_guid); break;Hmm.. the reason I didn't implement the patch this way is because I do think it's important to make sure we don't go out of bounds looking for the terminating NULL, i.e. you need a 'len <
variable_name_size' check somewhere.
Quoted 2 lines
Care to update and resend your patch, ensuring we don't inspect more than
variable_name_sizecharacters?
The VariableNameSize is not reliable when EFI_SUCCESS is returned because UEFI 2.3.1 spec only mention VariableNameSize should updated when EFI_BUFFER_TOO_SMALL is returned. And, the 1024 bytes of buffer is from old UEFI spec. There doesn't have any size condition of variable data or variable name in 2.3.1 spec.
I modified the patch to grab VariableNameSize when EFI_BUFFER_TOO_SMALL, the behavior like what we do in efivarfs_file_read().
This patch works on a normal UEFI machine, we will test it on HP z220. I will send out it formally after test success.
Thanks a lot!
Joey Lee
Quoted 1 line
From
1f88fab2bdf07da51975d31c20ee66415c51e14eMon Sep 17 00:00:00 2001
From: Lee, Chun-Yi <jlee-IBi9RG/[email protected]>
Date: Thu, 7 Mar 2013 18:14:25 +0800
Subject: [PATCH] efivars: Sanitise length of variable name for register
On HP z220 system (firmware version 1.54), some EFI variables have incorrectly named :
/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c /sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8c
Matt Fleming and Lingzhu Xiang pointed out the reason is the variable_name_size longer then the name string in variable_name when we feed them to efivar_create_sysfs_entry().
Per UEFI 2.3.1 spec, the VariableNameSize is updated to reflect the size of buffer that needed by VariableName when EFI_BUFFER_TOO_SMALL is returned. This behavior is the same with the DataSize for GetVariable().
This patch do the same thing like efivarfs_file_read(), it feed a zero VariableNameSize to GetNextVariableName() for grab the variable name size from UEFI BIOS. When VariableNameSize larger then the buffer size of variable name, we will reallocate more buffer to handle it.
The default buffer size is 1024, this number is from old UEFI spec.
In addition, we calculate the length of VariableName by using utf16_strsize() but not direct feed VariableNameSize to efivar_create_sysfs_entry() because the value of VariableNameSize is not reliable when EFI_SUCCESS is returned. UEFI spec only claim the VariableNameSize updated on EFI_BUFFER_TOO_SMALL but not on EFI_SUCCESS.
--- drivers/firmware/efivars.c | 35 +++++++++++++++++++++++++++-------- 1 files changed, 27 insertions(+), 8 deletions(-)
drivers/firmware/efivars.c +27 -8
@@ -1688,10 +1688,11 @@ int register_efivars(struct efivars *efivars,efi_status_t status =EFI_NOT_FOUND ;efi_guid_t vendor_guid;efi_char16_t *variable_name; -unsigned long variable_name_size =1024 ; +unsigned long variable_name_size; +unsigned long variable_name_buff_size =1024 ;int error =0 ; - variable_name =kzalloc (variable_name_size,GFP_KERNEL ); + variable_name =kzalloc (variable_name_buff_size,GFP_KERNEL );if (!variable_name) {printk (KERN_ERR "efivars: Memory allocation failed.\n" );return -ENOMEM ;
@@ -1722,17 +1723,35 @@ int register_efivars(struct efivars *efivars,*/ do { - variable_name_size = 1024; + variable_name_size = 0; status = ops->get_next_variable(&variable_name_size, variable_name, &vendor_guid); switch (status) { - case EFI_SUCCESS: - efivar_create_sysfs_entry(efivars, - variable_name_size, - variable_name, - &vendor_guid); + case EFI_BUFFER_TOO_SMALL: + if (variable_name_size < 2) { + /* set variable_name_size to buffer size when it's too small */ + variable_name_size = variable_name_buff_size; + } else if (variable_name_size > variable_name_buff_size) { + /* re-allocate more buffer when size doesn't enough */ + kfree(variable_name); + variable_name = kzalloc(variable_name_size, GFP_KERNEL); + if (!variable_name) { + printk(KERN_ERR "efivars: Memory allocation failed.\n"); + return -ENOMEM; + } + variable_name_buff_size = variable_name_size; + } + status = ops->get_next_variable(&variable_name_size, + variable_name, + &vendor_guid); + variable_name_size = utf16_strsize(variable_name, variable_name_buff_size)+2; + if (status == EFI_SUCCESS) + efivar_create_sysfs_entry(efivars, + variable_name_size, + variable_name, + &vendor_guid); break; case EFI_NOT_FOUND: break;
Signature · 1 line
1.6.4.2
Quoted 5 lines · joeyli
The VariableNameSize is not reliable when
EFI_SUCCESSis returned because UEFI 2.3.1 spec only mention VariableNameSize should updated whenEFI_BUFFER_TOO_SMALLis returned. And, the 1024 bytes of buffer is from old UEFI spec. There doesn't have any size condition of variable data or variable name in 2.3.1 spec.
The spec may only mention what happens in the EFI_BUFFER_TOO_SMALL case, but for EFI_SUCCESS, any behaviour other than leaving VariableNameSize alone or updating it with the required size of the buffer is just completely insane.
Quoted 2 lines
I modified the patch to grab VariableNameSize when
EFI_BUFFER_TOO_SMALL, the behavior like what we do inefivarfs_file_read().
Thanks, this does seem like the most robust solution.
Quoted 2 lines
This patch works on a normal UEFI machine, we will test it on HP z220. I will send out it formally after test success.
Has anyone tried contacting HP to tell them their firmware is doing bizarre things?
[...]
Quoted 9 lines
drivers/firmware/efivars.c +1
-1@@ -1722,17 +1723,35 @@ int register_efivars(struct efivars *efivars,*/ do { - variable_name_size = 1024; + variable_name_size = 0; status = ops->get_next_variable(&variable_name_size, variable_name,
drivers/firmware/efivars.c int register_efivars(struct efivars *efivars, · line 1734
Quoted 10 lines
drivers/firmware/efivars.c +3
-5&vendor_guid);switch (status) { -case EFI_SUCCESS : -efivar_create_sysfs_entry (efivars, - variable_name_size, - variable_name, - &vendor_guid); +case EFI_BUFFER_TOO_SMALL : +if (variable_name_size <2 ) { +/* set variable_name_size to buffer size when it's too small */
This hunk would be better written like,
if (variable_name_size < sizeof(efi_char16_t) * 2) {
/* Bogus size - expect at least one char + NULL */
variable_name_size = variable_name_buff_size;
}A variable name containing only '\0' is bogus. We need at least one unicode character + '\0' for a valid variable name.
drivers/firmware/efivars.c int register_efivars(struct efivars *efivars, · line 1737
Quoted 3 lines
drivers/firmware/efivars.c +3
-0+ variable_name_size = variable_name_buff_size; + }else if (variable_name_size > variable_name_buff_size) { +/* re-allocate more buffer when size doesn't enough */
This comment is redundant. Just delete it.
drivers/firmware/efivars.c int register_efivars(struct efivars *efivars, · line 1749
Quoted 12 lines
drivers/firmware/efivars.c +12
-0+kfree (variable_name); + variable_name =kzalloc (variable_name_size,GFP_KERNEL ); +if (!variable_name) { +printk (KERN_ERR "efivars: Memory allocation failed.\n" ); +return -ENOMEM ; + } + variable_name_buff_size = variable_name_size; + } + status = ops->get_next_variable (&variable_name_size, + variable_name, + &vendor_guid); + variable_name_size =utf16_strsize (variable_name, variable_name_buff_size)+2 ;
Please document what this +2 represents and why we need it. Better yet, use sizeof(efi_char16_t), and still document why it's needed, e.g. "Add terminating NULL".
Quoted 8 lines
drivers/firmware/efivars.c +5
-0+if (status ==EFI_SUCCESS ) +efivar_create_sysfs_entry (efivars, + variable_name_size, + variable_name, + &vendor_guid);break ;case EFI_NOT_FOUND :break ;
Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
Quoted 12 lines · Matt Fleming
Quoted 5 lines · joeyli
The VariableNameSize is not reliable when
EFI_SUCCESSis returned because UEFI 2.3.1 spec only mention VariableNameSize should updated whenEFI_BUFFER_TOO_SMALLis returned. And, the 1024 bytes of buffer is from old UEFI spec. There doesn't have any size condition of variable data or variable name in 2.3.1 spec.The spec may only mention what happens in the
EFI_BUFFER_TOO_SMALLcase, but forEFI_SUCCESS, any behaviour other than leaving VariableNameSize alone or updating it with the required size of the buffer is just completely insane.
Quoted 4 lines
Quoted 2 lines
I modified the patch to grab VariableNameSize when
EFI_BUFFER_TOO_SMALL, the behavior like what we do inefivarfs_file_read().Thanks, this does seem like the most robust solution.
Also, you're probably going to need to updateefivar_update_sysfs_entries() too.
Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
Hi Matt,
於 四,2013-03-07 於 13:57 +0000,Matt Fleming 提到:
Quoted 13 lines · Matt Fleming
Quoted 12 lines · joeyli
Quoted 5 lines
The VariableNameSize is not reliable when
EFI_SUCCESSis returned because UEFI 2.3.1 spec only mention VariableNameSize should updated whenEFI_BUFFER_TOO_SMALLis returned. And, the 1024 bytes of buffer is from old UEFI spec. There doesn't have any size condition of variable data or variable name in 2.3.1 spec.The spec may only mention what happens in the
EFI_BUFFER_TOO_SMALLcase, but forEFI_SUCCESS, any behaviour other than leaving VariableNameSize alone or updating it with the required size of the buffer is just completely insane.
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
I modified the patch to grab VariableNameSize when
EFI_BUFFER_TOO_SMALL, the behavior like what we do inefivarfs_file_read().Thanks, this does seem like the most robust solution.
Also, you're probably going to need to update
efivar_update_sysfs_entries()too.
Thanks for your review! I will send out v2 patch after modify and test on issue machine.
Thanks a lot!
Joey Lee
Hi Matt,
於 四,2013-03-07 於 13:57 +0000,Matt Fleming 提到:
Quoted 13 lines · Matt Fleming
Quoted 12 lines · joeyli
Quoted 5 lines
The VariableNameSize is not reliable when
EFI_SUCCESSis returned because UEFI 2.3.1 spec only mention VariableNameSize should updated whenEFI_BUFFER_TOO_SMALLis returned. And, the 1024 bytes of buffer is from old UEFI spec. There doesn't have any size condition of variable data or variable name in 2.3.1 spec.The spec may only mention what happens in the
EFI_BUFFER_TOO_SMALLcase, but forEFI_SUCCESS, any behaviour other than leaving VariableNameSize alone or updating it with the required size of the buffer is just completely insane.
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
I modified the patch to grab VariableNameSize when
EFI_BUFFER_TOO_SMALL, the behavior like what we do inefivarfs_file_read().Thanks, this does seem like the most robust solution.
Also, you're probably going to need to update
efivar_update_sysfs_entries()too.
Sorry for after I wrote patch, I think it's better we still use your original patch to fix this bug, because I found the efi_variable->VariableName allocated 1024 size and it also used by old vars system.
The following is my patch for reference, but I think your original patch is better for backward compatible on variable name.
Please consider to merge your original patch!
Thanks a lot!
Joey Lee
Quoted 1 line
From
c067288dbbb963b9cf9be4c5f59f5e39e88361adMon Sep 17 00:00:00 2001
From: Lee, Chun-Yi <jlee-IBi9RG/[email protected]>
Date: Mon, 11 Mar 2013 18:26:04 +0800
Subject: [PATCH] efivars: Sanitise length of variable name for register
--- drivers/firmware/efivars.c | 39 ++++++++++++++++++++++++++++++--------- 1 files changed, 30 insertions(+), 9 deletions(-)
drivers/firmware/efivars.c +30 -9
@@ -111,7 +111,7 @@ MODULE_VERSION(EFIVARS_VERSION);*/ struct efi_variable { -efi_char16_t VariableName[1024 /sizeof (efi_char16_t)]; +efi_char16_t VariableName[1024 /sizeof (efi_char16_t)];/* PROBLEM: 1024 size, need backward compatible to old vars */ efi_guid_t VendorGuid;unsigned long DataSize;__u8 Data[1024 ];
@@ -1903,10 +1903,11 @@ int register_efivars(struct efivars *efivars,efi_status_t status =EFI_NOT_FOUND ;efi_guid_t vendor_guid;efi_char16_t *variable_name; -unsigned long variable_name_size =1024 ; +unsigned long variable_name_size; +unsigned long variable_name_buff_size =1024 ;int error =0 ; - variable_name =kzalloc (variable_name_size,GFP_KERNEL ); + variable_name =kzalloc (variable_name_buff_size,GFP_KERNEL );if (!variable_name) {printk (KERN_ERR "efivars: Memory allocation failed.\n" );return -ENOMEM ;
@@ -1937,17 +1938,37 @@ int register_efivars(struct efivars *efivars,*/ do { - variable_name_size = 1024; + variable_name_size = 0; status = ops->get_next_variable(&variable_name_size, variable_name, &vendor_guid); switch (status) { - case EFI_SUCCESS: - efivar_create_sysfs_entry(efivars, - variable_name_size, - variable_name, - &vendor_guid); + case EFI_BUFFER_TOO_SMALL: + if (variable_name_size < sizeof(efi_char16_t) * 2) { + /* Bogus size - expect at least one char + NULL */ + variable_name_size = variable_name_buff_size; + } else if (variable_name_size > variable_name_buff_size) { + kfree(variable_name); + variable_name = kzalloc(variable_name_size, GFP_KERNEL); + if (!variable_name) { + printk(KERN_ERR "efivars: Memory allocation failed.\n"); + return -ENOMEM; + } + variable_name_buff_size = variable_name_size; + } + status = ops->get_next_variable(&variable_name_size, + variable_name, + &vendor_guid); + /* Length of the variable_name, plus terminating NULL */ + variable_name_size = utf16_strsize( + variable_name, variable_name_buff_size) + + sizeof(efi_char16_t); + if (status == EFI_SUCCESS) + efivar_create_sysfs_entry(efivars, + variable_name_size, /* PROBLEM: variable_name_size could not larger then new_efivar->var.VariableName = 1024 */ + variable_name, + &vendor_guid); break; case EFI_NOT_FOUND: break;
Signature · 1 line
1.6.4.2
Quoted 9 lines · joeyli
Sorry for after I wrote patch, I think it's better we still use your original patch to fix this bug, because I found the
efi_variable->VariableNameallocated 1024 size and it also used by old vars system.The following is my patch for reference, but I think your original patch is better for backward compatible on variable name.
Please consider to merge your original patch!
OK, this is what I've got queued up (note I removed the warning).
---
Quoted 1 line
From
afa9ae7bf47145d661487f88f2ec67b062ca98bcMon Sep 17 00:00:00 2001
From: Matt Fleming <[email protected]>
Date: Fri, 1 Mar 2013 14:49:12 +0000
Subject: [PATCH] efivars: explicitly calculate length of VariableName
It's not wise to assume VariableNameSize represents the length of VariableName, as not all firmware updates VariableNameSize in the same way (some don't update it at all if EFI_SUCCESS is returned). There are even implementations out there that update VariableNameSize with values that are both larger than the string returned in VariableName and smaller than the buffer passed to GetNextVariableName(), which resulted in the following bug report from Michael Schroeder,
Quoted 8 lines
On HP z220 system (firmware version 1.54), some EFI variables are incorrectly named :
ls -d
/sys/firmware/efi/vars/*8be4d*| grep -v -- -8be returns/sys/firmware/efi/vars/dbxDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/KEKDefault-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SecureBoot-pport8be4df61-93ca-11d2-aa0d-00e098032b8c/sys/firmware/efi/vars/SetupMode-Information8be4df61-93ca-11d2-aa0d-00e098032b8c
The issue here is that because we blindly use VariableNameSize without verifying its value, we can potentially read garbage values from the buffer containing VariableName if VariableNameSize is larger than the length of VariableName.
Since VariableName is a string, we can calculate its size by searching for the terminating NULL character.
--- drivers/firmware/efivars.c | 32 +++++++++++++++++++++++++++++++- 1 file changed, 31 insertions(+), 1 deletion(-)
drivers/firmware/efivars.c +31 -1
@@ -1704,6 +1704,31 @@ static bool variable_is_present(efi_char16_t *variable_name, efi_guid_t *vendor)return found; } +/* +* Returns the size of variable_name, in bytes, including the +* terminating NULL character, or variable_name_size if no NULL +* character is found among the first variable_name_size bytes. +*/ +static unsigned long var_name_strnsize (efi_char16_t *variable_name, +unsigned long variable_name_size) +{ +unsigned long len; +efi_char16_t c; + +/* +* The variable name is, by definition, a NULL-terminated +* string, so make absolutely sure that variable_name_size is +* the value we expect it to be. If not, return the real size. +*/ +for (len =2 ; len <= variable_name_size; len +=sizeof (c)) { + c = variable_name[(len /sizeof (c)) -1 ]; +if (!c) +break ; + } + +return min (len, variable_name_size); +} +static void efivar_update_sysfs_entries (struct work_struct *work) {struct efivars *efivars = &__efivars;
@@ -1744,10 +1769,13 @@ static void efivar_update_sysfs_entries(struct work_struct *work)if (!found) {kfree (variable_name);break ; - }else + }else { + variable_name_size =var_name_strnsize (variable_name, + variable_name_size);efivar_create_sysfs_entry (efivars, variable_name_size, variable_name, &vendor); + } } }
@@ -1994,6 +2022,8 @@ int register_efivars(struct efivars *efivars, &vendor_guid);switch (status) {case EFI_SUCCESS : + variable_name_size =var_name_strnsize (variable_name, + variable_name_size); efivar_create_sysfs_entry(efivars, variable_name_size, variable_name,
Signature · 6 lines
1.7.11.7 -- Matt Fleming, Intel Open Source Technology Center
於 四,2013-03-07 於 11:39 +0000,Matt Fleming 提到:
Quoted 6 lines
Quoted 1 line
This patch works on a normal UEFI machine, we will test it on HP
z220. I
Quoted 1 line
will send out it formally after test success.
Has anyone tried contacting HP to tell them their firmware is doing bizarre things?
We will try to contact with HP workstation department. Another good chance is in UEFI Plugfest Taipei at next next week, I will discuss with HP engineer for this issue.
Thanks a lot!
Joey Lee