CIPSO: Fix unaligned memory access in cipso_v4_gentag_hdr

Commit Message

Sergey Nazarov March 4, 2021, 9 p.m. UTC

We need to use put_unaligned when writing 32-bit DOI value
in cipso_v4_gentag_hdr to avoid unaligned memory access.

Signed-off-by: Sergey Nazarov <[e-mail]>
---
 net/ipv4/cipso_ipv4.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

 /**

Comments

Ondrej Mosnacek March 4, 2021, 9:17 p.m. UTC | #1

On Thu, Mar 4, 2021 at 10:09 PM Seergey Nazarov <[e-mail]> wrote:
> We need to use put_unaligned when writing 32-bit DOI value
> in cipso_v4_gentag_hdr to avoid unaligned memory access.
>
> Signed-off-by: Sergey Nazarov <[e-mail]>
> ---
>  net/ipv4/cipso_ipv4.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/net/ipv4/cipso_ipv4.c b/net/ipv4/cipso_ipv4.c
> index 471d33a..28dfe40 100644
> --- a/net/ipv4/cipso_ipv4.c
> +++ b/net/ipv4/cipso_ipv4.c
> @@ -1162,7 +1162,7 @@ static void cipso_v4_gentag_hdr(const struct
> cipso_v4_doi *doi_def,
>  {
>         buf[0] = IPOPT_CIPSO;
>         buf[1] = CIPSO_V4_HDR_LEN + len;
> -       *(__be32 *)&buf[2] = htonl(doi_def->doi);
> +       put_unaligned_be32(doi_def->doi, (__be32 *)&buf[2]);

I think you can now also drop the cast, since put_unaligned_be32()
expects a void *.

>  }
>
>  /**
> --
> 1.8.3.1
>
>

Paul Moore March 4, 2021, 9:45 p.m. UTC | #2

On Thu, Mar 4, 2021 at 4:17 PM Ondrej Mosnacek <[e-mail]> wrote:
> On Thu, Mar 4, 2021 at 10:09 PM Seergey Nazarov <[e-mail]> wrote:
> > We need to use put_unaligned when writing 32-bit DOI value
> > in cipso_v4_gentag_hdr to avoid unaligned memory access.
> >
> > Signed-off-by: Sergey Nazarov <[e-mail]>
> > ---
> >  net/ipv4/cipso_ipv4.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/net/ipv4/cipso_ipv4.c b/net/ipv4/cipso_ipv4.c
> > index 471d33a..28dfe40 100644
> > --- a/net/ipv4/cipso_ipv4.c
> > +++ b/net/ipv4/cipso_ipv4.c
> > @@ -1162,7 +1162,7 @@ static void cipso_v4_gentag_hdr(const struct
> > cipso_v4_doi *doi_def,
> >  {
> >         buf[0] = IPOPT_CIPSO;
> >         buf[1] = CIPSO_V4_HDR_LEN + len;
> > -       *(__be32 *)&buf[2] = htonl(doi_def->doi);
> > +       put_unaligned_be32(doi_def->doi, (__be32 *)&buf[2]);
>
> I think you can now also drop the cast, since put_unaligned_be32()
> expects a void *.

Yes, it shouldn't be necessary.

You should also send this to netdev as NetLabel code lives under net/
and goes to Linus via the netdev tree.

Also, looking at the code, there are a few similar spots you should
probably fix while you are revising this patch:

* the short/__be16 assignment in cipso_v4_map_cat_enum_hton()
* the short/__be16 assignments in cipso_v4_map_cat_rng_hton()
* the long/__be32 assignment in calipso_genopt()

Sergey Nazarov March 5, 2021, 7:28 a.m. UTC | #3

В Чт, 04/03/2021 в 22:17 +0100, Ondrej Mosnacek пишет:
> On Thu, Mar 4, 2021 at 10:09 PM Seergey Nazarov <[e-mail]>
> wrote:
> > We need to use put_unaligned when writing 32-bit DOI value
> > in cipso_v4_gentag_hdr to avoid unaligned memory access.
> > 
> > Signed-off-by: Sergey Nazarov <[e-mail]>
> > ---
> >  net/ipv4/cipso_ipv4.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/net/ipv4/cipso_ipv4.c b/net/ipv4/cipso_ipv4.c
> > index 471d33a..28dfe40 100644
> > --- a/net/ipv4/cipso_ipv4.c
> > +++ b/net/ipv4/cipso_ipv4.c
> > @@ -1162,7 +1162,7 @@ static void cipso_v4_gentag_hdr(const struct
> > cipso_v4_doi *doi_def,
> >  {
> >         buf[0] = IPOPT_CIPSO;
> >         buf[1] = CIPSO_V4_HDR_LEN + len;
> > -       *(__be32 *)&buf[2] = htonl(doi_def->doi);
> > +       put_unaligned_be32(doi_def->doi, (__be32 *)&buf[2]);
> 
> I think you can now also drop the cast, since put_unaligned_be32()
> expects a void *.
> 
> >  }
> > 
> >  /**
> > --
> > 1.8.3.1
> > 
> > 
> 
> 
You are right, thanks!

Sergey Nazarov March 5, 2021, 7:40 a.m. UTC | #4

В Чт, 04/03/2021 в 16:45 -0500, Paul Moore пишет:
> On Thu, Mar 4, 2021 at 4:17 PM Ondrej Mosnacek <[e-mail]>
> wrote:
> > On Thu, Mar 4, 2021 at 10:09 PM Seergey Nazarov <
> > [e-mail]> wrote:
> > > We need to use put_unaligned when writing 32-bit DOI value
> > > in cipso_v4_gentag_hdr to avoid unaligned memory access.
> > > 
> > > Signed-off-by: Sergey Nazarov <[e-mail]>
> > > ---
> > >  net/ipv4/cipso_ipv4.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/net/ipv4/cipso_ipv4.c b/net/ipv4/cipso_ipv4.c
> > > index 471d33a..28dfe40 100644
> > > --- a/net/ipv4/cipso_ipv4.c
> > > +++ b/net/ipv4/cipso_ipv4.c
> > > @@ -1162,7 +1162,7 @@ static void cipso_v4_gentag_hdr(const
> > > struct
> > > cipso_v4_doi *doi_def,
> > >  {
> > >         buf[0] = IPOPT_CIPSO;
> > >         buf[1] = CIPSO_V4_HDR_LEN + len;
> > > -       *(__be32 *)&buf[2] = htonl(doi_def->doi);
> > > +       put_unaligned_be32(doi_def->doi, (__be32 *)&buf[2]);
> > 
> > I think you can now also drop the cast, since put_unaligned_be32()
> > expects a void *.
> 
> Yes, it shouldn't be necessary.
> 
> You should also send this to netdev as NetLabel code lives under net/
> and goes to Linus via the netdev tree.
> 
> Also, looking at the code, there are a few similar spots you should
> probably fix while you are revising this patch:
> 
> * the short/__be16 assignment in cipso_v4_map_cat_enum_hton()
> * the short/__be16 assignments in cipso_v4_map_cat_rng_hton()
> * the long/__be32 assignment incalipso_genopt()
> 
Yes, I looked it.
In cipso_v4_map_cat_enum_hton, cipso_v4_map_cat_rng_hton access
address aligned by 2, so it's okay.
In cipso_v4_gentag_loc, cipso_v4_parsetag_loc access address is
CIPSO_V4_HDR_LEN(6) + 2, so it's okay too.
In calipso_genopt access address is 4n + 2 + 2, it's okay too.

Patch

diff --git a/net/ipv4/cipso_ipv4.c b/net/ipv4/cipso_ipv4.c
index 471d33a..28dfe40 100644
--- a/net/ipv4/cipso_ipv4.c
+++ b/net/ipv4/cipso_ipv4.c
@@ -1162,7 +1162,7 @@  static void cipso_v4_gentag_hdr(const struct
cipso_v4_doi *doi_def,
 {
 	buf[0] = IPOPT_CIPSO;
 	buf[1] = CIPSO_V4_HDR_LEN + len;
-	*(__be32 *)&buf[2] = htonl(doi_def->doi);
+	put_unaligned_be32(doi_def->doi, (__be32 *)&buf[2]);
 }