[PATCH v2 0/2] vfs: follow-on fixes for i_ino widening

[PATCH v2 0/2] vfs: follow-on fixes for i_ino widening

[Jeff Layton]

Just some patches to fix follow-on issues reported after the
inode->i_ino widening series. Christian, could you toss these
onto the vfs-7.1.kino branch?

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
Changes in v2:
- rename variable in nilfs2 patch from "rem" to "index"
- reword comment and commit log for better accuracy in EVM patch

---
Jeff Layton (2):
      nilfs2: fix 64-bit division operations in nilfs_bmap_find_target_in_group()
      EVM: add comment describing why ino field is still unsigned long

 fs/nilfs2/bmap.c                    | 9 ++++++---
 security/integrity/evm/evm_crypto.c | 6 ++++++
 2 files changed, 12 insertions(+), 3 deletions(-)
---
base-commit: 9840bb66e7e5dffd72b03201318f154a10b06b4a
change-id: 20260310-iino-u64-424fa570d850

Best regards,
-- 
Jeff Layton <jlayton@kernel.org>

[PATCH v2 1/2] nilfs2: fix 64-bit division operations in nilfs_bmap_find_target_in_group()

[Jeff Layton]

With the change to make inode->i_ino a u64, the build started failing on
32-bit ARM with:

    ERROR: modpost: "__aeabi_uldivmod" [fs/nilfs2/nilfs2.ko] undefined!

Fix this by using the 64-bit division interfaces in
nilfs_bmap_find_target_in_group().

Fixes: 998a59d371c2 ("treewide: fix missed i_ino format specifier conversions")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202603100602.KPxiClIO-lkp@intel.com/
Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>
Acked-by: Ryusuke Konishi <konishi.ryusuke@gmail.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 fs/nilfs2/bmap.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/fs/nilfs2/bmap.c b/fs/nilfs2/bmap.c
index 824f2bd91c167965ec3a660202b6e6c5f1fe007e..abcf5252578ad24f694bfccf525893674bfcb4bc 100644
--- a/fs/nilfs2/bmap.c
+++ b/fs/nilfs2/bmap.c
@@ -455,11 +455,14 @@ __u64 nilfs_bmap_find_target_in_group(const struct nilfs_bmap *bmap)
 {
 	struct inode *dat = nilfs_bmap_get_dat(bmap);
 	unsigned long entries_per_group = nilfs_palloc_entries_per_group(dat);
-	unsigned long group = bmap->b_inode->i_ino / entries_per_group;
+	unsigned long group;
+	u32 index;
+
+	group = div_u64(bmap->b_inode->i_ino, entries_per_group);
+	div_u64_rem(bmap->b_inode->i_ino, NILFS_BMAP_GROUP_DIV, &index);
 
 	return group * entries_per_group +
-		(bmap->b_inode->i_ino % NILFS_BMAP_GROUP_DIV) *
-		(entries_per_group / NILFS_BMAP_GROUP_DIV);
+	       index * (entries_per_group / NILFS_BMAP_GROUP_DIV);
 }
 
 static struct lock_class_key nilfs_bmap_dat_lock_key;

-- 
2.53.0

[PATCH v2 2/2] EVM: add comment describing why ino field is still unsigned long

[Jeff Layton]

Mimi pointed out that we didn't widen the inode number field in struct
h_misc alongside the inode->i_ino widening. While we could make an
equivalent change there, that would require EVM resigning on all 32-bit
hosts.

Instead, leave the field as an unsigned long. This should have no effect
on 64-bit hosts, and allow things to continue working on 32-bit hosts in
the cases where the i_ino fits in 32-bits.

Add a comment explaining why it's being left as unsigned long.

Cc: Mimi Zohar <zohar@linux.ibm.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 security/integrity/evm/evm_crypto.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/security/integrity/evm/evm_crypto.c b/security/integrity/evm/evm_crypto.c
index c0ca4eedb0fe5d5c30f45f515a4bc90248ec64ea..1c41af2f91a60a714878ff93b554c90e45546503 100644
--- a/security/integrity/evm/evm_crypto.c
+++ b/security/integrity/evm/evm_crypto.c
@@ -144,6 +144,12 @@ static void hmac_add_misc(struct shash_desc *desc, struct inode *inode,
 			  char type, char *digest)
 {
 	struct h_misc {
+		/*
+		 * Although inode->i_ino is now u64, this field remains
+		 * unsigned long to allow existing HMAC and signatures from
+		 * 32-bit hosts to continue working when i_ino hasn't changed
+		 * and fits in a u32.
+		 */
 		unsigned long ino;
 		__u32 generation;
 		uid_t uid;

-- 
2.53.0

Re: [PATCH v2 2/2] EVM: add comment describing why ino field is still unsigned long

[Mimi Zohar]

On Fri, 2026-03-13 at 14:45 -0400, Jeff Layton wrote:
> Mimi pointed out that we didn't widen the inode number field in struct
> h_misc alongside the inode->i_ino widening. While we could make an
> equivalent change there, that would require EVM resigning on all 32-bit
> hosts.
> 
> Instead, leave the field as an unsigned long. This should have no effect
> on 64-bit hosts, and allow things to continue working on 32-bit hosts in
> the cases where the i_ino fits in 32-bits.
> 
> Add a comment explaining why it's being left as unsigned long.
> 
> Cc: Mimi Zohar <zohar@linux.ibm.com>
> Signed-off-by: Jeff Layton <jlayton@kernel.org>

Thanks, Jeff.

Reviewed-by: Mimi Zohar <zohar@linux.ibm.com>


> ---
>  security/integrity/evm/evm_crypto.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/security/integrity/evm/evm_crypto.c b/security/integrity/evm/evm_crypto.c
> index c0ca4eedb0fe5d5c30f45f515a4bc90248ec64ea..1c41af2f91a60a714878ff93b554c90e45546503 100644
> --- a/security/integrity/evm/evm_crypto.c
> +++ b/security/integrity/evm/evm_crypto.c
> @@ -144,6 +144,12 @@ static void hmac_add_misc(struct shash_desc *desc, struct inode *inode,
>  			  char type, char *digest)
>  {
>  	struct h_misc {
> +		/*
> +		 * Although inode->i_ino is now u64, this field remains
> +		 * unsigned long to allow existing HMAC and signatures from
> +		 * 32-bit hosts to continue working when i_ino hasn't changed
> +		 * and fits in a u32.
> +		 */
>  		unsigned long ino;
>  		__u32 generation;
>  		uid_t uid;

Re: [PATCH v2 1/2] nilfs2: fix 64-bit division operations in nilfs_bmap_find_target_in_group()

[David Laight]

On Fri, 13 Mar 2026 14:45:20 -0400
Jeff Layton <jlayton@kernel.org> wrote:

> With the change to make inode->i_ino a u64, the build started failing on
> 32-bit ARM with:
> 
>     ERROR: modpost: "__aeabi_uldivmod" [fs/nilfs2/nilfs2.ko] undefined!
> 
> Fix this by using the 64-bit division interfaces in
> nilfs_bmap_find_target_in_group().
> 
> Fixes: 998a59d371c2 ("treewide: fix missed i_ino format specifier conversions")
> Reported-by: kernel test robot <lkp@intel.com>
> Closes: https://lore.kernel.org/oe-kbuild-all/202603100602.KPxiClIO-lkp@intel.com/
> Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>
> Acked-by: Ryusuke Konishi <konishi.ryusuke@gmail.com>
> Signed-off-by: Jeff Layton <jlayton@kernel.org>
> ---
>  fs/nilfs2/bmap.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/nilfs2/bmap.c b/fs/nilfs2/bmap.c
> index 824f2bd91c167965ec3a660202b6e6c5f1fe007e..abcf5252578ad24f694bfccf525893674bfcb4bc 100644
> --- a/fs/nilfs2/bmap.c
> +++ b/fs/nilfs2/bmap.c
> @@ -455,11 +455,14 @@ __u64 nilfs_bmap_find_target_in_group(const struct nilfs_bmap *bmap)
>  {
>  	struct inode *dat = nilfs_bmap_get_dat(bmap);
>  	unsigned long entries_per_group = nilfs_palloc_entries_per_group(dat);
> -	unsigned long group = bmap->b_inode->i_ino / entries_per_group;

Are you sure entries_per_group can be more than 32 bits?
It looks like something that will be the same size on 32 and 64bit.

> +	unsigned long group;
> +	u32 index;
> +
> +	group = div_u64(bmap->b_inode->i_ino, entries_per_group);

You don't need the full 64 by 64 divide.
IIRC there are both div_u64_u32() and div_u64_ulong().

> +	div_u64_rem(bmap->b_inode->i_ino, NILFS_BMAP_GROUP_DIV, &index);

NILFD_BMAP_GROUP_DIV is 8 (and probably has to be a power of 2).
So:
	index = bmap->b_inode->i_ino & (NILFS_BMAP_GROUP_DIV - 1);
is the same and likely much faster to calculate.
(The compiler will have done that optimisation before.)

	David


>  
>  	return group * entries_per_group +
> -		(bmap->b_inode->i_ino % NILFS_BMAP_GROUP_DIV) *
> -		(entries_per_group / NILFS_BMAP_GROUP_DIV);
> +	       index * (entries_per_group / NILFS_BMAP_GROUP_DIV);
>  }
>  
>  static struct lock_class_key nilfs_bmap_dat_lock_key;
> 

Re: [PATCH v2 1/2] nilfs2: fix 64-bit division operations in nilfs_bmap_find_target_in_group()

[Jeff Layton]

On Sat, 2026-03-14 at 12:47 +0000, David Laight wrote:
> On Fri, 13 Mar 2026 14:45:20 -0400
> Jeff Layton <jlayton@kernel.org> wrote:
> 
> > With the change to make inode->i_ino a u64, the build started failing on
> > 32-bit ARM with:
> > 
> >     ERROR: modpost: "__aeabi_uldivmod" [fs/nilfs2/nilfs2.ko] undefined!
> > 
> > Fix this by using the 64-bit division interfaces in
> > nilfs_bmap_find_target_in_group().
> > 
> > Fixes: 998a59d371c2 ("treewide: fix missed i_ino format specifier conversions")
> > Reported-by: kernel test robot <lkp@intel.com>
> > Closes: https://lore.kernel.org/oe-kbuild-all/202603100602.KPxiClIO-lkp@intel.com/
> > Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>
> > Acked-by: Ryusuke Konishi <konishi.ryusuke@gmail.com>
> > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> > ---
> >  fs/nilfs2/bmap.c | 9 ++++++---
> >  1 file changed, 6 insertions(+), 3 deletions(-)
> > 
> > diff --git a/fs/nilfs2/bmap.c b/fs/nilfs2/bmap.c
> > index 824f2bd91c167965ec3a660202b6e6c5f1fe007e..abcf5252578ad24f694bfccf525893674bfcb4bc 100644
> > --- a/fs/nilfs2/bmap.c
> > +++ b/fs/nilfs2/bmap.c
> > @@ -455,11 +455,14 @@ __u64 nilfs_bmap_find_target_in_group(const struct nilfs_bmap *bmap)
> >  {
> >  	struct inode *dat = nilfs_bmap_get_dat(bmap);
> >  	unsigned long entries_per_group = nilfs_palloc_entries_per_group(dat);
> > -	unsigned long group = bmap->b_inode->i_ino / entries_per_group;
> 
> Are you sure entries_per_group can be more than 32 bits?
> It looks like something that will be the same size on 32 and 64bit.
> 

I'm not sure of anything here. I'm just want to get this to compile on
all arches. FWIW, I'm not looking to optimize anything in this patch.

> > +	unsigned long group;
> > +	u32 index;
> > +
> > +	group = div_u64(bmap->b_inode->i_ino, entries_per_group);
> 
> You don't need the full 64 by 64 divide.
> IIRC there are both div_u64_u32() and div_u64_ulong().
>
> > +	div_u64_rem(bmap->b_inode->i_ino, NILFS_BMAP_GROUP_DIV, &index);
> 
> NILFD_BMAP_GROUP_DIV is 8 (and probably has to be a power of 2).
> So:
> 	index = bmap->b_inode->i_ino & (NILFS_BMAP_GROUP_DIV - 1);
> is the same and likely much faster to calculate.
> (The compiler will have done that optimisation before.)
> 
> 

That all sounds reasonable to me. At this point though, it would be
better if the NILFS2 folks stepped in with how they'd prefer this be
done.

> 
> >  
> >  	return group * entries_per_group +
> > -		(bmap->b_inode->i_ino % NILFS_BMAP_GROUP_DIV) *
> > -		(entries_per_group / NILFS_BMAP_GROUP_DIV);
> > +	       index * (entries_per_group / NILFS_BMAP_GROUP_DIV);
> >  }
> >  
> >  static struct lock_class_key nilfs_bmap_dat_lock_key;
> > 

-- 
Jeff Layton <jlayton@kernel.org>

Re: [PATCH v2 1/2] nilfs2: fix 64-bit division operations in nilfs_bmap_find_target_in_group()

[Ryusuke Konishi]

On Sat, Mar 14, 2026 at 9:59 PM Jeff Layton wrote:
>
> On Sat, 2026-03-14 at 12:47 +0000, David Laight wrote:
> > On Fri, 13 Mar 2026 14:45:20 -0400
> > Jeff Layton <jlayton@kernel.org> wrote:
> >
> > > With the change to make inode->i_ino a u64, the build started failing on
> > > 32-bit ARM with:
> > >
> > >     ERROR: modpost: "__aeabi_uldivmod" [fs/nilfs2/nilfs2.ko] undefined!
> > >
> > > Fix this by using the 64-bit division interfaces in
> > > nilfs_bmap_find_target_in_group().
> > >
> > > Fixes: 998a59d371c2 ("treewide: fix missed i_ino format specifier conversions")
> > > Reported-by: kernel test robot <lkp@intel.com>
> > > Closes: https://lore.kernel.org/oe-kbuild-all/202603100602.KPxiClIO-lkp@intel.com/
> > > Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>
> > > Acked-by: Ryusuke Konishi <konishi.ryusuke@gmail.com>
> > > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> > > ---
> > >  fs/nilfs2/bmap.c | 9 ++++++---
> > >  1 file changed, 6 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/fs/nilfs2/bmap.c b/fs/nilfs2/bmap.c
> > > index 824f2bd91c167965ec3a660202b6e6c5f1fe007e..abcf5252578ad24f694bfccf525893674bfcb4bc 100644
> > > --- a/fs/nilfs2/bmap.c
> > > +++ b/fs/nilfs2/bmap.c
> > > @@ -455,11 +455,14 @@ __u64 nilfs_bmap_find_target_in_group(const struct nilfs_bmap *bmap)
> > >  {
> > >     struct inode *dat = nilfs_bmap_get_dat(bmap);
> > >     unsigned long entries_per_group = nilfs_palloc_entries_per_group(dat);
> > > -   unsigned long group = bmap->b_inode->i_ino / entries_per_group;
> >
> > Are you sure entries_per_group can be more than 32 bits?
> > It looks like something that will be the same size on 32 and 64bit.
> >
>
> I'm not sure of anything here. I'm just want to get this to compile on
> all arches. FWIW, I'm not looking to optimize anything in this patch.
>
> > > +   unsigned long group;
> > > +   u32 index;
> > > +
> > > +   group = div_u64(bmap->b_inode->i_ino, entries_per_group);
> >
> > You don't need the full 64 by 64 divide.
> > IIRC there are both div_u64_u32() and div_u64_ulong().

Isn't the type of divisor in div_u64() u32?
Since entries_per_group cannot exceed 32 bits according to the current
specification, I think using div_u64() is fine.

> >
> > > +   div_u64_rem(bmap->b_inode->i_ino, NILFS_BMAP_GROUP_DIV, &index);
> >
> > NILFD_BMAP_GROUP_DIV is 8 (and probably has to be a power of 2).
> > So:
> >       index = bmap->b_inode->i_ino & (NILFS_BMAP_GROUP_DIV - 1);
> > is the same and likely much faster to calculate.
> > (The compiler will have done that optimisation before.)
> >
> >
>
> That all sounds reasonable to me. At this point though, it would be
> better if the NILFS2 folks stepped in with how they'd prefer this be
> done.

Yes, indeed.  It seems that the application of optimizations will
change, so this proposed correction is better.

Since NILFS_BMAP_GROUP_DIV is a fixed constant and cannot be anything
other than a power of 2, could you please adopt this proposed
correction with the following comment?

#define NILFS_BMAP_GROUP_DIV    8  /* must be a power of 2 */

Thanks,
Ryusuke Konishi

>
> >
> > >
> > >     return group * entries_per_group +
> > > -           (bmap->b_inode->i_ino % NILFS_BMAP_GROUP_DIV) *
> > > -           (entries_per_group / NILFS_BMAP_GROUP_DIV);
> > > +          index * (entries_per_group / NILFS_BMAP_GROUP_DIV);
> > >  }
> > >
> > >  static struct lock_class_key nilfs_bmap_dat_lock_key;
> > >
>
> --
> Jeff Layton <jlayton@kernel.org>

Re: [PATCH v2 1/2] nilfs2: fix 64-bit division operations in nilfs_bmap_find_target_in_group()

[Jeff Layton]

On Tue, 2026-03-17 at 03:06 +0900, Ryusuke Konishi wrote:
> On Sat, Mar 14, 2026 at 9:59 PM Jeff Layton wrote:
> > 
> > On Sat, 2026-03-14 at 12:47 +0000, David Laight wrote:
> > > On Fri, 13 Mar 2026 14:45:20 -0400
> > > Jeff Layton <jlayton@kernel.org> wrote:
> > > 
> > > > With the change to make inode->i_ino a u64, the build started failing on
> > > > 32-bit ARM with:
> > > > 
> > > >     ERROR: modpost: "__aeabi_uldivmod" [fs/nilfs2/nilfs2.ko] undefined!
> > > > 
> > > > Fix this by using the 64-bit division interfaces in
> > > > nilfs_bmap_find_target_in_group().
> > > > 
> > > > Fixes: 998a59d371c2 ("treewide: fix missed i_ino format specifier conversions")
> > > > Reported-by: kernel test robot <lkp@intel.com>
> > > > Closes: https://lore.kernel.org/oe-kbuild-all/202603100602.KPxiClIO-lkp@intel.com/
> > > > Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>
> > > > Acked-by: Ryusuke Konishi <konishi.ryusuke@gmail.com>
> > > > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> > > > ---
> > > >  fs/nilfs2/bmap.c | 9 ++++++---
> > > >  1 file changed, 6 insertions(+), 3 deletions(-)
> > > > 
> > > > diff --git a/fs/nilfs2/bmap.c b/fs/nilfs2/bmap.c
> > > > index 824f2bd91c167965ec3a660202b6e6c5f1fe007e..abcf5252578ad24f694bfccf525893674bfcb4bc 100644
> > > > --- a/fs/nilfs2/bmap.c
> > > > +++ b/fs/nilfs2/bmap.c
> > > > @@ -455,11 +455,14 @@ __u64 nilfs_bmap_find_target_in_group(const struct nilfs_bmap *bmap)
> > > >  {
> > > >     struct inode *dat = nilfs_bmap_get_dat(bmap);
> > > >     unsigned long entries_per_group = nilfs_palloc_entries_per_group(dat);
> > > > -   unsigned long group = bmap->b_inode->i_ino / entries_per_group;
> > > 
> > > Are you sure entries_per_group can be more than 32 bits?
> > > It looks like something that will be the same size on 32 and 64bit.
> > > 
> > 
> > I'm not sure of anything here. I'm just want to get this to compile on
> > all arches. FWIW, I'm not looking to optimize anything in this patch.
> > 
> > > > +   unsigned long group;
> > > > +   u32 index;
> > > > +
> > > > +   group = div_u64(bmap->b_inode->i_ino, entries_per_group);
> > > 
> > > You don't need the full 64 by 64 divide.
> > > IIRC there are both div_u64_u32() and div_u64_ulong().
> 
> Isn't the type of divisor in div_u64() u32?
> Since entries_per_group cannot exceed 32 bits according to the current
> specification, I think using div_u64() is fine.
> 

Yep, it is:

static inline u64 div_u64(u64 dividend, u32 divisor)                  


> > > 
> > > > +   div_u64_rem(bmap->b_inode->i_ino, NILFS_BMAP_GROUP_DIV, &index);
> > > 
> > > NILFD_BMAP_GROUP_DIV is 8 (and probably has to be a power of 2).
> > > So:
> > >       index = bmap->b_inode->i_ino & (NILFS_BMAP_GROUP_DIV - 1);
> > > is the same and likely much faster to calculate.
> > > (The compiler will have done that optimisation before.)
> > > 
> > > 
> > 
> > That all sounds reasonable to me. At this point though, it would be
> > better if the NILFS2 folks stepped in with how they'd prefer this be
> > done.
> 
> Yes, indeed.  It seems that the application of optimizations will
> change, so this proposed correction is better.
> 
> Since NILFS_BMAP_GROUP_DIV is a fixed constant and cannot be anything
> other than a power of 2, could you please adopt this proposed
> correction with the following comment?
> 
> #define NILFS_BMAP_GROUP_DIV    8  /* must be a power of 2 */
> 

Even better, lets use a compile-time assertion:

    BUILD_BUG_ON_NOT_POWER_OF_TWO(NILFS_BMAP_GROUP_DIV);

I'll send a respin.

Thanks!
-- 
Jeff Layton <jlayton@kernel.org>