diff mbox series

[v2,1/6] device-mapper: Check that target specs are sufficiently aligned

Message ID 20230603145244.1538-2-demi@invisiblethingslab.com
State New
Headers
Series Several device-mapper fixes |

Commit Message

Demi Marie Obenour June 3, 2023, 2:52 p.m. UTC 
  Otherwise subsequent code will dereference a misaligned
`struct dm_target_spec *`, which is undefined behavior.

Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
---
 drivers/md/dm-ioctl.c | 7 +++++++
 1 file changed, 7 insertions(+)

Comments

Mike Snitzer June 22, 2023, 4:28 p.m. UTC | #1 
On Sat, Jun 03 2023 at 10:52P -0400,
Demi Marie Obenour <demi@invisiblethingslab.com> wrote:

> Otherwise subsequent code will dereference a misaligned
> `struct dm_target_spec *`, which is undefined behavior.
> 
> Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Cc: stable@vger.kernel.org
> ---
>  drivers/md/dm-ioctl.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/drivers/md/dm-ioctl.c b/drivers/md/dm-ioctl.c
> index cc77cf3d410921432eb0c62cdede7d55b9aa674a..34fa74c6a70db8aa67aaba3f6a2fc4f38ef736bc 100644
> --- a/drivers/md/dm-ioctl.c
> +++ b/drivers/md/dm-ioctl.c
> @@ -1394,6 +1394,13 @@ static inline fmode_t get_mode(struct dm_ioctl *param)
>  static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
>  		       struct dm_target_spec **spec, char **target_params)
>  {
> +	static_assert(_Alignof(struct dm_target_spec) <= 8,
> +		      "struct dm_target_spec has excessive alignment requirements");

Really not sure what you mean by "has excessive alignment requirements"...

> +	if (next % 8) {
> +		DMERR("Next target spec (offset %u) is not 8-byte aligned", next);
> +		return -EINVAL;
> +	}
> +
>  	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
>  	*target_params = (char *) (*spec + 1);
>  

But this patch and patches 2 and 3 need more review. I'd like Mikulas to review.

I did pick up patches 4-6 for the upcoming 6.5 merge window.

Note, please prefix with "dm ioctl" instead of "device-mapper".

(I just switched my "dm" prefix to "dm ioctl" and forced update on the
dm-6.5 branch, so the commit I referenced earlier for your version
copy patch is now here:
https://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm.git/commit/?h=dm-6.5&id=a5a3de762b3ae8959347928843c12502b1b23163
)

Mike
Mikulas Patocka June 22, 2023, 5:29 p.m. UTC | #2 
On Sat, 3 Jun 2023, Demi Marie Obenour wrote:

> Otherwise subsequent code will dereference a misaligned
> `struct dm_target_spec *`, which is undefined behavior.
> 
> Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Cc: stable@vger.kernel.org
> ---
>  drivers/md/dm-ioctl.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/drivers/md/dm-ioctl.c b/drivers/md/dm-ioctl.c
> index cc77cf3d410921432eb0c62cdede7d55b9aa674a..34fa74c6a70db8aa67aaba3f6a2fc4f38ef736bc 100644
> --- a/drivers/md/dm-ioctl.c
> +++ b/drivers/md/dm-ioctl.c
> @@ -1394,6 +1394,13 @@ static inline fmode_t get_mode(struct dm_ioctl *param)
>  static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
>  		       struct dm_target_spec **spec, char **target_params)
>  {
> +	static_assert(_Alignof(struct dm_target_spec) <= 8,
> +		      "struct dm_target_spec has excessive alignment requirements");
> +	if (next % 8) {
> +		DMERR("Next target spec (offset %u) is not 8-byte aligned", next);
> +		return -EINVAL;
> +	}
> +
>  	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
>  	*target_params = (char *) (*spec + 1);
>  
> -- 
> Sincerely,
> Demi Marie Obenour (she/her/hers)
> Invisible Things Lab

Hi

Some architectures (such as 32-bit x86) specify that the alignment of 
64-bit integers is only 4-byte. This could in theory break old userspace 
code that only uses 4-byte alignment. I would change "next % 8" to "next % 
__alignof__(struct dm_target_spec)".

I think that there is no need to backport this patch series to the stable 
kernels because the bugs that it fixes may only be exploited by the user 
with CAP_SYS_ADMIN privilege. So, there is no security or reliability 
problem being fixed.

Mikulas
Demi Marie Obenour June 22, 2023, 7:51 p.m. UTC | #3 
On Thu, Jun 22, 2023 at 12:28:28PM -0400, Mike Snitzer wrote:
> On Sat, Jun 03 2023 at 10:52P -0400,
> Demi Marie Obenour <demi@invisiblethingslab.com> wrote:
> 
> > Otherwise subsequent code will dereference a misaligned
> > `struct dm_target_spec *`, which is undefined behavior.
> > 
> > Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
> > Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> > Cc: stable@vger.kernel.org
> > ---
> >  drivers/md/dm-ioctl.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> > 
> > diff --git a/drivers/md/dm-ioctl.c b/drivers/md/dm-ioctl.c
> > index cc77cf3d410921432eb0c62cdede7d55b9aa674a..34fa74c6a70db8aa67aaba3f6a2fc4f38ef736bc 100644
> > --- a/drivers/md/dm-ioctl.c
> > +++ b/drivers/md/dm-ioctl.c
> > @@ -1394,6 +1394,13 @@ static inline fmode_t get_mode(struct dm_ioctl *param)
> >  static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
> >  		       struct dm_target_spec **spec, char **target_params)
> >  {
> > +	static_assert(_Alignof(struct dm_target_spec) <= 8,
> > +		      "struct dm_target_spec has excessive alignment requirements");
> 
> Really not sure what you mean by "has excessive alignment requirements"...

This patch checks that struct dm_target_spec is 8-byte aligned.  That is
okay if its alignment is 8 or less, but not if is 16 or more, so I added
a static assert to check that struct dm_target_spec indeed requires at
most 8-byte alignment.  That said, “excessive alignment requirements” is
(as shown by you having to ask this question) a bad error message.
Would “must not require more than 8-byte alignment” be better?

> > +	if (next % 8) {
> > +		DMERR("Next target spec (offset %u) is not 8-byte aligned", next);
> > +		return -EINVAL;
> > +	}
> > +
> >  	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
> >  	*target_params = (char *) (*spec + 1);
> >  
> 
> But this patch and patches 2 and 3 need more review. I'd like Mikulas to review.
> 
> I did pick up patches 4-6 for the upcoming 6.5 merge window.

Thanks!

> Note, please prefix with "dm ioctl" instead of "device-mapper".

Good to know, thanks!  I have several additional patches written that
require patch 4.  Should I send patches 1 through 3 in the same series
as well?
Demi Marie Obenour June 22, 2023, 8:27 p.m. UTC | #4 
On Thu, Jun 22, 2023 at 07:29:52PM +0200, Mikulas Patocka wrote:
> 
> 
> On Sat, 3 Jun 2023, Demi Marie Obenour wrote:
> 
> > Otherwise subsequent code will dereference a misaligned
> > `struct dm_target_spec *`, which is undefined behavior.
> > 
> > Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
> > Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> > Cc: stable@vger.kernel.org
> > ---
> >  drivers/md/dm-ioctl.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> > 
> > diff --git a/drivers/md/dm-ioctl.c b/drivers/md/dm-ioctl.c
> > index cc77cf3d410921432eb0c62cdede7d55b9aa674a..34fa74c6a70db8aa67aaba3f6a2fc4f38ef736bc 100644
> > --- a/drivers/md/dm-ioctl.c
> > +++ b/drivers/md/dm-ioctl.c
> > @@ -1394,6 +1394,13 @@ static inline fmode_t get_mode(struct dm_ioctl *param)
> >  static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
> >  		       struct dm_target_spec **spec, char **target_params)
> >  {
> > +	static_assert(_Alignof(struct dm_target_spec) <= 8,
> > +		      "struct dm_target_spec has excessive alignment requirements");
> > +	if (next % 8) {
> > +		DMERR("Next target spec (offset %u) is not 8-byte aligned", next);
> > +		return -EINVAL;
> > +	}
> > +
> >  	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
> >  	*target_params = (char *) (*spec + 1);
> >  
> > -- 
> > Sincerely,
> > Demi Marie Obenour (she/her/hers)
> > Invisible Things Lab
> 
> Hi
> 
> Some architectures (such as 32-bit x86) specify that the alignment of 
> 64-bit integers is only 4-byte. This could in theory break old userspace 
> code that only uses 4-byte alignment. I would change "next % 8" to "next % 
> __alignof__(struct dm_target_spec)".

That’s fine, provided that the rest of the code is okay with 4-byte
alignment.

> I think that there is no need to backport this patch series to the stable 
> kernels because the bugs that it fixes may only be exploited by the user 
> with CAP_SYS_ADMIN privilege. So, there is no security or reliability 
> problem being fixed.

I agree that there is no reliability problem, but with kernel lockdown
root → kernel is a security boundary, so fixes for memory unsafety
problems should still be backported IMO.
Mike Snitzer June 22, 2023, 10:54 p.m. UTC | #5 
On Thu, Jun 22 2023 at  3:51P -0400,
Demi Marie Obenour <demi@invisiblethingslab.com> wrote:

> On Thu, Jun 22, 2023 at 12:28:28PM -0400, Mike Snitzer wrote:
> > On Sat, Jun 03 2023 at 10:52P -0400,
> > Demi Marie Obenour <demi@invisiblethingslab.com> wrote:
> > 
> > > Otherwise subsequent code will dereference a misaligned
> > > `struct dm_target_spec *`, which is undefined behavior.
> > > 
> > > Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
> > > Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> > > Cc: stable@vger.kernel.org
> > > ---
> > >  drivers/md/dm-ioctl.c | 7 +++++++
> > >  1 file changed, 7 insertions(+)
> > > 
> > > diff --git a/drivers/md/dm-ioctl.c b/drivers/md/dm-ioctl.c
> > > index cc77cf3d410921432eb0c62cdede7d55b9aa674a..34fa74c6a70db8aa67aaba3f6a2fc4f38ef736bc 100644
> > > --- a/drivers/md/dm-ioctl.c
> > > +++ b/drivers/md/dm-ioctl.c
> > > @@ -1394,6 +1394,13 @@ static inline fmode_t get_mode(struct dm_ioctl *param)
> > >  static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
> > >  		       struct dm_target_spec **spec, char **target_params)
> > >  {
> > > +	static_assert(_Alignof(struct dm_target_spec) <= 8,
> > > +		      "struct dm_target_spec has excessive alignment requirements");
> > 
> > Really not sure what you mean by "has excessive alignment requirements"...
> 
> This patch checks that struct dm_target_spec is 8-byte aligned.  That is
> okay if its alignment is 8 or less, but not if is 16 or more, so I added
> a static assert to check that struct dm_target_spec indeed requires at
> most 8-byte alignment.  That said, “excessive alignment requirements” is
> (as shown by you having to ask this question) a bad error message.
> Would “must not require more than 8-byte alignment” be better?

Yes, that's better, I've updated it to use that.
 
> > > +	if (next % 8) {
> > > +		DMERR("Next target spec (offset %u) is not 8-byte aligned", next);
> > > +		return -EINVAL;
> > > +	}
> > > +
> > >  	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
> > >  	*target_params = (char *) (*spec + 1);
> > >  
> > 
> > But this patch and patches 2 and 3 need more review. I'd like Mikulas to review.
> > 
> > I did pick up patches 4-6 for the upcoming 6.5 merge window.
> 
> Thanks!
> 
> > Note, please prefix with "dm ioctl" instead of "device-mapper".
> 
> Good to know, thanks!  I have several additional patches written that
> require patch 4.  Should I send patches 1 through 3 in the same series
> as well?

I did end up picking up patches 1-3 and rebased so they are in front
of your patches 4-6 like you intended.

But I agree with Mikulas, I'm not seeing the point in tagging any of
these for stable@.

All commits are currently at the tip of dm-6.5, see:
https://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm.git/log/?h=dm-6.5

Mike
diff mbox series

Patch

diff --git a/drivers/md/dm-ioctl.c b/drivers/md/dm-ioctl.c
index cc77cf3d410921432eb0c62cdede7d55b9aa674a..34fa74c6a70db8aa67aaba3f6a2fc4f38ef736bc 100644
--- a/drivers/md/dm-ioctl.c
+++ b/drivers/md/dm-ioctl.c
@@ -1394,6 +1394,13 @@  static inline fmode_t get_mode(struct dm_ioctl *param)
 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
 		       struct dm_target_spec **spec, char **target_params)
 {
+	static_assert(_Alignof(struct dm_target_spec) <= 8,
+		      "struct dm_target_spec has excessive alignment requirements");
+	if (next % 8) {
+		DMERR("Next target spec (offset %u) is not 8-byte aligned", next);
+		return -EINVAL;
+	}
+
 	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
 	*target_params = (char *) (*spec + 1);