Re: [PATCH] net: fix uninitialized heap leak in multicast address dumps
From: Eric Dumazet
Date: Wed Nov 18 2026 - 05:40:29 EST
Le mer. 18 nov. 2026 à 11:38, Eric Dumazet <edumazet@xxxxxxxxxx> a écrit :
>
> Le mer. 18 nov. 2026 à 11:07, Daniel Machon
> <daniel.machon@xxxxxxxxxxxxx> a écrit :
> >
> > Hi,
> >
> > > __hw_addr_create() allocates entries with kmalloc() and initializes
> > > only addr_len bytes of ha->addr. Dump paths emit entries using the
> > > current dev->addr_len, which can grow after the entry was created:
> > > tun's TUNSETLINK changes a device from ARPHRD_NONE (addr_len 0) to
> > > ARPHRD_IP6GRE (addr_len 16). Both the RTM_GETMULTICAST dump and
> > > /proc/net/dev_mcast then emit up to 16 bytes of never-written slab
> > > memory per entry.
> > >
> > > This discloses leftover kernel heap contents to anyone reading those
> > > interfaces.
> > >
> > > Zero the tail of ha->addr at creation time, so dumps can only emit
> > > initialized bytes.
> > >
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Fixes: a748ee242681 ("net: move address list functions to a separate file")
> > > Link: https://lists.openwall.net/linux-kernel/2026/09/10/320
> > > Signed-off-by: Henry Martin <bsdhenrymartin@xxxxxxxxx>
> > > ---
> > > net/core/dev_addr_lists.c | 5 +++++
> > > 1 file changed, 5 insertions(+)
> > >
> > > diff --git a/net/core/dev_addr_lists.c b/net/core/dev_addr_lists.c
> > > index d827cc21867b..53a2dfa2706c 100644
> > > --- a/net/core/dev_addr_lists.c
> > > +++ b/net/core/dev_addr_lists.c
> > > @@ -100,6 +100,11 @@ __hw_addr_create(const unsigned char *addr, int addr_len,
> > > if (!ha)
> > > return NULL;
> > > memcpy(ha->addr, addr, addr_len);
> > > + /* Entries are dumped with the current dev->addr_len, which can grow
> > > + * after creation (e.g. tun TUNSETLINK), so never leave the tail
> > > + * uninitialized.
> > > + */
> > > + memset(ha->addr + addr_len, 0, MAX_ADDR_LEN - addr_len);
> > > ha->type = addr_type;
> > > ha->refcount = 1;
> > > ha->global_use = global;
> > > --
> > > 2.43.7
> > >
> >
> > Looks right.
> >
> > Just a thought, I think switching to kzalloc, like so:
> >
> > + ha = kzalloc(alloc_size, GFP_ATOMIC);
> >
> > Would give you the same fix, and you would get rid of that extra tail length
> > arithmetic.
> >
> > Reviewed-by: Daniel Machon <daniel.machon@xxxxxxxxxxxxx>
>
> I agree with Daniel, kzalloc() would be simpler here, this is not
> fast path. You can then remove the "ha->sync_cnt = 0;" line.
>
> This is fine as a hardening measure, but the root cause is in tun:
>
> bonding and team both call dev_uc_flush() and dev_mc_flush() when
> they change dev->type and dev->addr_len. TUNSETLINK does not, so
> entries created with the old addr_len stay in dev->uc and dev->mc,
> in an rb-tree which is now searched with a different key length.
>
> I think TUNSETLINK should flush both lists, in a separate patch.
Something like:
diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index d4870bdc156f..f675e5469cfd 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -3387,6 +3387,9 @@ static long __tun_chr_ioctl(struct file *file,
unsigned int cmd,
"Refused to change device type\n");
break;
}
+ /* Addresses were added with the old dev->addr_len. */
+ dev_uc_flush(tun->dev);
+ dev_mc_flush(tun->dev);
tun->dev->type = (int) arg;
tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
netif_info(tun, drv, tun->dev, "linktype set to %d\n",