As part of trying to clean up struct sock_addr, add comments about the
sockaddr arguments of dev_[gs]et_mac_address() being actual classic "max
14 bytes in sa_data" sockaddr instances and not struct sockaddr_storage.

Signed-off-by: Kees Cook <kees@kernel.org>
---
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: netdev@vger.kernel.org
---
 net/core/dev.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 45a8c3dd4a64..5abfd29a35bf 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -9183,7 +9183,8 @@ EXPORT_SYMBOL(dev_pre_changeaddr_notify);
 /**
  *	dev_set_mac_address - Change Media Access Control Address
  *	@dev: device
- *	@sa: new address
+ *	@sa: new address in a classic "struct sockaddr", which will never
+ *	     have more than 14 bytes in @sa::sa_data
  *	@extack: netlink extended ack
  *
  *	Change the hardware (MAC) address of the device
@@ -9217,6 +9218,7 @@ EXPORT_SYMBOL(dev_set_mac_address);
 
 DECLARE_RWSEM(dev_addr_sem);
 
+/* "sa" is a classic sockaddr: it will only ever use 14 bytes from sa_data. */
 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
 			     struct netlink_ext_ack *extack)
 {
@@ -9229,6 +9231,7 @@ int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
 }
 EXPORT_SYMBOL(dev_set_mac_address_user);
 
+/* "sa" is a classic sockaddr: it will only ever use 14 bytes from sa_data. */
 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
 {
 	size_t size = sizeof(sa->sa_data_min);
-- 
2.34.1

Kees Cook wrote:
> As part of trying to clean up struct sock_addr, add comments about the
> sockaddr arguments of dev_[gs]et_mac_address() being actual classic "max
> 14 bytes in sa_data" sockaddr instances and not struct sockaddr_storage.

What is this assertion based on?

I see various non-Ethernet .ndo_set_mac_address implementations, which
dev_set_mac_address calls. And dev_set_mac_addr_user is called from
rtnetlink do_setlink. Which kmalloc's sa based on dev->addr_len.

> 
> Signed-off-by: Kees Cook <kees@kernel.org>
> ---
> Cc: "David S. Miller" <davem@davemloft.net>
> Cc: Eric Dumazet <edumazet@google.com>
> Cc: Jakub Kicinski <kuba@kernel.org>
> Cc: Paolo Abeni <pabeni@redhat.com>
> Cc: netdev@vger.kernel.org
> ---
>  net/core/dev.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 45a8c3dd4a64..5abfd29a35bf 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -9183,7 +9183,8 @@ EXPORT_SYMBOL(dev_pre_changeaddr_notify);
>  /**
>   *	dev_set_mac_address - Change Media Access Control Address
>   *	@dev: device
> - *	@sa: new address
> + *	@sa: new address in a classic "struct sockaddr", which will never
> + *	     have more than 14 bytes in @sa::sa_data
>   *	@extack: netlink extended ack
>   *
>   *	Change the hardware (MAC) address of the device
> @@ -9217,6 +9218,7 @@ EXPORT_SYMBOL(dev_set_mac_address);
>  
>  DECLARE_RWSEM(dev_addr_sem);
>  
> +/* "sa" is a classic sockaddr: it will only ever use 14 bytes from sa_data. */
>  int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
>  			     struct netlink_ext_ack *extack)
>  {
> @@ -9229,6 +9231,7 @@ int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
>  }
>  EXPORT_SYMBOL(dev_set_mac_address_user);
>  
> +/* "sa" is a classic sockaddr: it will only ever use 14 bytes from sa_data. */
>  int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
>  {
>  	size_t size = sizeof(sa->sa_data_min);
> -- 
> 2.34.1
>

On Tue, Dec 17, 2024 at 10:35:44AM -0500, Willem de Bruijn wrote:
> Kees Cook wrote:
> > As part of trying to clean up struct sock_addr, add comments about the
> > sockaddr arguments of dev_[gs]et_mac_address() being actual classic "max
> > 14 bytes in sa_data" sockaddr instances and not struct sockaddr_storage.
> 
> What is this assertion based on?
> 
> I see various non-Ethernet .ndo_set_mac_address implementations, which
> dev_set_mac_address calls. And dev_set_mac_addr_user is called from
> rtnetlink do_setlink. Which kmalloc's sa based on dev->addr_len.

Yeah, I was clearly missing several cases. Please ignore this patch. I
will re-examine this.

-- 
Kees Cook

On Tue, Dec 17, 2024 at 12:30:36PM -0800, Kees Cook wrote:
> On Tue, Dec 17, 2024 at 10:35:44AM -0500, Willem de Bruijn wrote:
> > Kees Cook wrote:
> > > As part of trying to clean up struct sock_addr, add comments about the
> > > sockaddr arguments of dev_[gs]et_mac_address() being actual classic "max
> > > 14 bytes in sa_data" sockaddr instances and not struct sockaddr_storage.
> > 
> > What is this assertion based on?
> > 
> > I see various non-Ethernet .ndo_set_mac_address implementations, which
> > dev_set_mac_address calls. And dev_set_mac_addr_user is called from
> > rtnetlink do_setlink. Which kmalloc's sa based on dev->addr_len.
> 
> Yeah, I was clearly missing several cases. Please ignore this patch. I
> will re-examine this.

So, I think I see what happened -- I missed the dev->addr_len in
dev_set_mac_address(), and saw that dev_get_mac_address() caps the
address to 14:

        size_t size = sizeof(sa->sa_data_min);
...
        if (!dev->addr_len)
                memset(sa->sa_data, 0, size);
        else
                memcpy(sa->sa_data, dev->dev_addr,
                       min_t(size_t, size, dev->addr_len));

It seems only tun/tap and SIOCGIFHWADDR use the "get" interface, though.

-- 
Kees Cook