[PATCH] selinux: use explicit 64-bit division

[PATCH] selinux: use explicit 64-bit division

[Arnd Bergmann]

From: Arnd Bergmann <arnd@arndb.de>

On 32-bit targets, the division of a 64-bit integer by 33 causes a
function call:

arm-linux-gnueabi-ld: security/selinux/selinuxfs.o: in function `sel_read_class':
selinuxfs.c:(.text+0xdcc): undefined reference to `__aeabi_uldivmod'
arm-linux-gnueabi-ld: security/selinux/selinuxfs.o: in function `sel_read_perm':
selinuxfs.c:(.text+0xed0): undefined reference to `__aeabi_uldivmod'

Replace this with an explicit call to div_u64() and div_u64_rem() to
annotate that these are potentially very slow.

Fixes: fc68b6a45f16 ("selinux: convert selinuxfs inode numbers from unsigned long to u64")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 security/selinux/selinuxfs.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/security/selinux/selinuxfs.c b/security/selinux/selinuxfs.c
index f60f92e2ac34..7894a2f480f7 100644
--- a/security/selinux/selinuxfs.c
+++ b/security/selinux/selinuxfs.c
@@ -1573,7 +1573,7 @@ static inline u64 sel_class_to_ino(u16 class)
 
 static inline u16 sel_ino_to_class(u64 ino)
 {
-	return (ino & SEL_INO_MASK) / (SEL_VEC_MAX + 1);
+	return div_u64(ino & SEL_INO_MASK, SEL_VEC_MAX + 1);
 }
 
 static inline u64 sel_perm_to_ino(u16 class, u32 perm)
@@ -1583,7 +1583,11 @@ static inline u64 sel_perm_to_ino(u16 class, u32 perm)
 
 static inline u32 sel_ino_to_perm(u64 ino)
 {
-	return (ino & SEL_INO_MASK) % (SEL_VEC_MAX + 1);
+	u32 rem;
+
+	div_u64_rem(ino & SEL_INO_MASK, SEL_VEC_MAX + 1, &rem);
+
+	return rem;
 }
 
 static ssize_t sel_read_class(struct file *file, char __user *buf,
-- 
2.53.0

Re: [PATCH] selinux: use explicit 64-bit division

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] selinux: unnecessary 64-bit division for 24-bit bounded value
--

commit 7622bbc45f42944796ad657eff450b2c4b7eb729
Author: Arnd Bergmann <arnd@arndb.de>

selinux: use explicit 64-bit division

The patch addresses a 32-bit build error where dividing a 64-bit integer
by 33 caused an undefined reference to __aeabi_uldivmod. It fixes this
by replacing standard division and modulo with explicit 64-bit division
helpers.

> diff --git a/security/selinux/selinuxfs.c b/security/selinux/selinuxfs.c
> index f60f92e2ac341..7894a2f480f74 100644
> --- a/security/selinux/selinuxfs.c
> +++ b/security/selinux/selinuxfs.c
> @@ -1573,7 +1573,7 @@ static inline u64 sel_class_to_ino(u16 class)
>  
>  static inline u16 sel_ino_to_class(u64 ino)
>  {
> -	return (ino & SEL_INO_MASK) / (SEL_VEC_MAX + 1);
> +	return div_u64(ino & SEL_INO_MASK, SEL_VEC_MAX + 1);
>  }

[Severity: Low]
Since SEL_INO_MASK restricts the dividend to 24 bits, the masked value fits
comfortably in a standard 32-bit unsigned integer.

Would it be better to downcast the masked value to u32 before division?

This would avoid the 64-bit division overhead entirely and allow the
compiler to use fast 32-bit math or optimize the constant division into a
multiply-and-shift.

>  static inline u64 sel_perm_to_ino(u16 class, u32 perm)
> @@ -1583,7 +1583,11 @@ static inline u64 sel_perm_to_ino(u16 class, u32 perm)
>  
>  static inline u32 sel_ino_to_perm(u64 ino)
>  {
> -	return (ino & SEL_INO_MASK) % (SEL_VEC_MAX + 1);
> +	u32 rem;
> +
> +	div_u64_rem(ino & SEL_INO_MASK, SEL_VEC_MAX + 1, &rem);
> +
> +	return rem;
>  }

[Severity: Low]
Similar to sel_ino_to_class(), does this need a full 64-bit quotient and
remainder calculation?

Given the dividend is also bounded to 24 bits by SEL_INO_MASK here, could
this be replaced with a 32-bit downcast and a standard modulo operation to be
more efficient while remaining mathematically identical?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260915195728.3519469-1-arnd@kernel.org?part=1

Re: [PATCH] selinux: use explicit 64-bit division

[Paul Moore]

On Tue, Sep 15, 2026 at 3:57 PM Arnd Bergmann <arnd@kernel.org> wrote:
>
> From: Arnd Bergmann <arnd@arndb.de>
>
> On 32-bit targets, the division of a 64-bit integer by 33 causes a
> function call:
>
> arm-linux-gnueabi-ld: security/selinux/selinuxfs.o: in function `sel_read_class':
> selinuxfs.c:(.text+0xdcc): undefined reference to `__aeabi_uldivmod'
> arm-linux-gnueabi-ld: security/selinux/selinuxfs.o: in function `sel_read_perm':
> selinuxfs.c:(.text+0xed0): undefined reference to `__aeabi_uldivmod'
>
> Replace this with an explicit call to div_u64() and div_u64_rem() to
> annotate that these are potentially very slow.
>
> Fixes: fc68b6a45f16 ("selinux: convert selinuxfs inode numbers from unsigned long to u64")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  security/selinux/selinuxfs.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)

Thanks Arnd.

As sashiko pointed out[1], since we don't really use the full 64-bit
inode space, what do you think about the following fix instead?

diff --git a/security/selinux/selinuxfs.c b/security/selinux/selinuxfs.c
index 292302eb60f3..d864c314c873 100644
--- a/security/selinux/selinuxfs.c
+++ b/security/selinux/selinuxfs.c
@@ -1574,7 +1574,8 @@ static inline u64 sel_class_to_ino(u16 class)

 static inline u16 sel_ino_to_class(u64 ino)
 {
-       return (ino & SEL_INO_MASK) / (SEL_VEC_MAX + 1);
+       u32 ino_masked = ino & SEL_INO_MASK;
+       return ino_masked / (SEL_VEC_MAX + 1);
 }

 static inline u64 sel_perm_to_ino(u16 class, u32 perm)
@@ -1584,7 +1585,8 @@ static inline u64 sel_perm_to_ino(u16 class, u32 perm)

 static inline u32 sel_ino_to_perm(u64 ino)
 {
-       return (ino & SEL_INO_MASK) % (SEL_VEC_MAX + 1);
+       u32 ino_masked = ino & SEL_INO_MASK;
+       return ino_masked % (SEL_VEC_MAX + 1);
 }

[1] https://lore.kernel.org/selinux/20260915200522.BBC8A1F000FF@smtp.kernel.org/

-- 
paul-moore.com

Re: [PATCH] selinux: use explicit 64-bit division

[Arnd Bergmann]

On Tue, Sep 15, 2026, at 22:36, Paul Moore wrote:
> On Tue, Sep 15, 2026 at 3:57 PM Arnd Bergmann <arnd@kernel.org> wrote:
>>
>> From: Arnd Bergmann <arnd@arndb.de>
>>
>> On 32-bit targets, the division of a 64-bit integer by 33 causes a
>> function call:
>>
>> arm-linux-gnueabi-ld: security/selinux/selinuxfs.o: in function `sel_read_class':
>> selinuxfs.c:(.text+0xdcc): undefined reference to `__aeabi_uldivmod'
>> arm-linux-gnueabi-ld: security/selinux/selinuxfs.o: in function `sel_read_perm':
>> selinuxfs.c:(.text+0xed0): undefined reference to `__aeabi_uldivmod'
>>
>> Replace this with an explicit call to div_u64() and div_u64_rem() to
>> annotate that these are potentially very slow.
>>
>> Fixes: fc68b6a45f16 ("selinux: convert selinuxfs inode numbers from unsigned long to u64")
>> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
>> ---
>>  security/selinux/selinuxfs.c | 8 ++++++--
>>  1 file changed, 6 insertions(+), 2 deletions(-)
>
> Thanks Arnd.
>
> As sashiko pointed out[1], since we don't really use the full 64-bit
> inode space, what do you think about the following fix instead?

Ah right, I hadn't checked the actual value of SEL_INO_MASK, and the
description of your patch made it sound like the 64-bit space
was actually required.

Your patch is better then, you can treat my mail as Reported-by.

     Arnd

Re: [PATCH] selinux: use explicit 64-bit division

[Paul Moore]

On Tue, Sep 15, 2026 at 4:43 PM Arnd Bergmann <arnd@arndb.de> wrote:
> On Tue, Sep 15, 2026, at 22:36, Paul Moore wrote:
> > On Tue, Sep 15, 2026 at 3:57 PM Arnd Bergmann <arnd@kernel.org> wrote:
> >>
> >> From: Arnd Bergmann <arnd@arndb.de>
> >>
> >> On 32-bit targets, the division of a 64-bit integer by 33 causes a
> >> function call:
> >>
> >> arm-linux-gnueabi-ld: security/selinux/selinuxfs.o: in function `sel_read_class':
> >> selinuxfs.c:(.text+0xdcc): undefined reference to `__aeabi_uldivmod'
> >> arm-linux-gnueabi-ld: security/selinux/selinuxfs.o: in function `sel_read_perm':
> >> selinuxfs.c:(.text+0xed0): undefined reference to `__aeabi_uldivmod'
> >>
> >> Replace this with an explicit call to div_u64() and div_u64_rem() to
> >> annotate that these are potentially very slow.
> >>
> >> Fixes: fc68b6a45f16 ("selinux: convert selinuxfs inode numbers from unsigned long to u64")
> >> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> >> ---
> >>  security/selinux/selinuxfs.c | 8 ++++++--
> >>  1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > Thanks Arnd.
> >
> > As sashiko pointed out[1], since we don't really use the full 64-bit
> > inode space, what do you think about the following fix instead?
>
> Ah right, I hadn't checked the actual value of SEL_INO_MASK, and the
> description of your patch made it sound like the 64-bit space
> was actually required.

No worries, selinuxfs is a bit odd sometimes.

> Your patch is better then, you can treat my mail as Reported-by.

Done, patch link below.  Thanks again!

https://lore.kernel.org/selinux/20260915213807.481355-2-paul@paul-moore.com

-- 
paul-moore.com