Vasiliy Kulikov schrieb:
> packet_getname_spkt() doesn't initialize all members of sa_data field of
> sockaddr struct if strlen(dev->name) < 13. This structure is then copied
> to userland. It leads to leaking of contents of kernel stack memory.
> We have to fully fill sa_data with strncpy() instead of strlcpy().
>
> The same with packet_getname(): it doesn't initialize sll_pkttype field of
> sockaddr_ll. Set it to zero.
>
> Signed-off-by: Vasiliy Kulikov <[email protected]>
> ---
> net/packet/af_packet.c | 3 ++-
> 1 files changed, 2 insertions(+), 1 deletions(-)
>
> diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> index 3616f27..0856a13 100644
> --- a/net/packet/af_packet.c
> +++ b/net/packet/af_packet.c
> @@ -1719,7 +1719,7 @@ static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
> rcu_read_lock();
> dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
> if (dev)
> - strlcpy(uaddr->sa_data, dev->name, 15);
> + strncpy(uaddr->sa_data, dev->name, 14);
> else
> memset(uaddr->sa_data, 0, 14);
if i understand the code correcly the max size for dev->name is IFNAMSIZ.
You can simply that part:
memset(uaddr->sa_data, 0, IFNAMSIZ);
dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
if (dev)
strlcpy(uaddr->sa_data, dev->name, IFNAMSIZ);
you should send that as separate patch.
re,
wh
> rcu_read_unlock();
> @@ -1742,6 +1742,7 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
> sll->sll_family = AF_PACKET;
> sll->sll_ifindex = po->ifindex;
> sll->sll_protocol = po->num;
> + sll->sll_pkttype = 0;
> rcu_read_lock();
> dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
> if (dev) {
On Mon, Nov 01, 2010 at 10:14 +0100, walter harms wrote:
>
>
> Vasiliy Kulikov schrieb:
> > packet_getname_spkt() doesn't initialize all members of sa_data field of
> > sockaddr struct if strlen(dev->name) < 13. This structure is then copied
> > to userland. It leads to leaking of contents of kernel stack memory.
> > We have to fully fill sa_data with strncpy() instead of strlcpy().
> >
> > The same with packet_getname(): it doesn't initialize sll_pkttype field of
> > sockaddr_ll. Set it to zero.
> >
> > Signed-off-by: Vasiliy Kulikov <[email protected]>
> > ---
> > net/packet/af_packet.c | 3 ++-
> > 1 files changed, 2 insertions(+), 1 deletions(-)
> >
> > diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> > index 3616f27..0856a13 100644
> > --- a/net/packet/af_packet.c
> > +++ b/net/packet/af_packet.c
> > @@ -1719,7 +1719,7 @@ static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
> > rcu_read_lock();
> > dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
> > if (dev)
> > - strlcpy(uaddr->sa_data, dev->name, 15);
> > + strncpy(uaddr->sa_data, dev->name, 14);
> > else
> > memset(uaddr->sa_data, 0, 14);
>
> if i understand the code correcly the max size for dev->name is IFNAMSIZ.
For dev->name - IFNAMSIZ, for uaddr->sa_data - 14.
> You can simply that part:
>
> memset(uaddr->sa_data, 0, IFNAMSIZ);
> dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
> if (dev)
> strlcpy(uaddr->sa_data, dev->name, IFNAMSIZ);
This will overflow uaddr->sa_data. Also I don't see any difficulty to
fill the array only once.
> you should send that as separate patch.
> re,
> wh
>
>
> > rcu_read_unlock();
> > @@ -1742,6 +1742,7 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
> > sll->sll_family = AF_PACKET;
> > sll->sll_ifindex = po->ifindex;
> > sll->sll_protocol = po->num;
> > + sll->sll_pkttype = 0;
> > rcu_read_lock();
> > dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
> > if (dev) {
Thanks,
--
Vasiliy
Am 06.11.2010 15:39, schrieb Vasiliy Kulikov:
> On Mon, Nov 01, 2010 at 10:14 +0100, walter harms wrote:
>>
>>
>> Vasiliy Kulikov schrieb:
>>> packet_getname_spkt() doesn't initialize all members of sa_data field of
>>> sockaddr struct if strlen(dev->name) < 13. This structure is then copied
>>> to userland. It leads to leaking of contents of kernel stack memory.
>>> We have to fully fill sa_data with strncpy() instead of strlcpy().
>>>
>>> The same with packet_getname(): it doesn't initialize sll_pkttype field of
>>> sockaddr_ll. Set it to zero.
>>>
>>> Signed-off-by: Vasiliy Kulikov <[email protected]>
>>> ---
>>> net/packet/af_packet.c | 3 ++-
>>> 1 files changed, 2 insertions(+), 1 deletions(-)
>>>
>>> diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
>>> index 3616f27..0856a13 100644
>>> --- a/net/packet/af_packet.c
>>> +++ b/net/packet/af_packet.c
>>> @@ -1719,7 +1719,7 @@ static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
>>> rcu_read_lock();
>>> dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
>>> if (dev)
>>> - strlcpy(uaddr->sa_data, dev->name, 15);
>>> + strncpy(uaddr->sa_data, dev->name, 14);
>>> else
>>> memset(uaddr->sa_data, 0, 14);
>>
>> if i understand the code correcly the max size for dev->name is IFNAMSIZ.
>
> For dev->name - IFNAMSIZ, for uaddr->sa_data - 14.
>
did not notice, since uaddr->sa_data should take dev->name this does no look very
clever. How is the size of sa_data defined ? Would it hurt when some uses IFNAMSIZ here ?
Perhaps someone who know more about the network stack can figure out what is actualy done
with uaddr->sa_data.
looks like a can of worms.
In packet_bind_spkt() they will copy a char[15], obviously it is a real problem.
re,
wh
>> You can simply that part:
>>
>> memset(uaddr->sa_data, 0, IFNAMSIZ);
>> dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
>> if (dev)
>> strlcpy(uaddr->sa_data, dev->name, IFNAMSIZ);
>
> This will overflow uaddr->sa_data. Also I don't see any difficulty to
> fill the array only once.
>
>> you should send that as separate patch.
>> re,
>> wh
>>
>>
>>> rcu_read_unlock();
>>> @@ -1742,6 +1742,7 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
>>> sll->sll_family = AF_PACKET;
>>> sll->sll_ifindex = po->ifindex;
>>> sll->sll_protocol = po->num;
>>> + sll->sll_pkttype = 0;
>>> rcu_read_lock();
>>> dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
>>> if (dev) {
>
> Thanks,
>
On Sun, Nov 07, 2010 at 12:37 +0100, walter harms wrote:
> Am 06.11.2010 15:39, schrieb Vasiliy Kulikov:
> > On Mon, Nov 01, 2010 at 10:14 +0100, walter harms wrote:
> >> Vasiliy Kulikov schrieb:
> >>> @@ -1719,7 +1719,7 @@ static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
> >>> rcu_read_lock();
> >>> dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
> >>> if (dev)
> >>> - strlcpy(uaddr->sa_data, dev->name, 15);
> >>> + strncpy(uaddr->sa_data, dev->name, 14);
> >>> else
> >>> memset(uaddr->sa_data, 0, 14);
> >>
> >> if i understand the code correcly the max size for dev->name is IFNAMSIZ.
> >
> > For dev->name - IFNAMSIZ, for uaddr->sa_data - 14.
> >
>
>
> did not notice, since uaddr->sa_data should take dev->name this does no look very
> clever. How is the size of sa_data defined ?
Magic size...
~/linux/include/linux/socket.h:
struct sockaddr {
sa_family_t sa_family; /* address family, AF_xxx */
char sa_data[14]; /* 14 bytes of protocol address */
};
> Would it hurt when some uses IFNAMSIZ here ?
If copy _to_ sa_data string of maximum IFNAMSIZ bytes - yes.
In packet_getname_spkt() the output buffer is 128 bytes, so it doesn't
really overflows anything. I don't think that *_getname() implementations
don't know this.
> Perhaps someone who know more about the network stack can figure out what is actualy done
> with uaddr->sa_data.
Yeah, also I wonder whether it is always NULL-terminated string or not.
> looks like a can of worms.
>
>
> In packet_bind_spkt() they will copy a char[15], obviously it is a real problem.
No, packet_bind_spkt() copies 14-byte string into array of 15 bytes.
The vice versa would be a bug.
> re,
> wh
>
>
> >> You can simply that part:
> >>
> >> memset(uaddr->sa_data, 0, IFNAMSIZ);
> >> dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
> >> if (dev)
> >> strlcpy(uaddr->sa_data, dev->name, IFNAMSIZ);
> >
> > This will overflow uaddr->sa_data. Also I don't see any difficulty to
> > fill the array only once.
> >
> >> you should send that as separate patch.
> >> re,
> >> wh
> >>
> >>
> >>> rcu_read_unlock();
> >>> @@ -1742,6 +1742,7 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
> >>> sll->sll_family = AF_PACKET;
> >>> sll->sll_ifindex = po->ifindex;
> >>> sll->sll_protocol = po->num;
> >>> + sll->sll_pkttype = 0;
> >>> rcu_read_lock();
> >>> dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
> >>> if (dev) {
> >
> > Thanks,
> >
--
Vasiliy
Am 07.11.2010 13:06, schrieb Vasiliy Kulikov:
> On Sun, Nov 07, 2010 at 12:37 +0100, walter harms wrote:
>> Am 06.11.2010 15:39, schrieb Vasiliy Kulikov:
>>> On Mon, Nov 01, 2010 at 10:14 +0100, walter harms wrote:
>>>> Vasiliy Kulikov schrieb:
>>>>> @@ -1719,7 +1719,7 @@ static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
>>>>> rcu_read_lock();
>>>>> dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
>>>>> if (dev)
>>>>> - strlcpy(uaddr->sa_data, dev->name, 15);
>>>>> + strncpy(uaddr->sa_data, dev->name, 14);
>>>>> else
>>>>> memset(uaddr->sa_data, 0, 14);
>>>>
>>>> if i understand the code correcly the max size for dev->name is IFNAMSIZ.
>>>
>>> For dev->name - IFNAMSIZ, for uaddr->sa_data - 14.
>>>
>>
>>
>> did not notice, since uaddr->sa_data should take dev->name this does no look very
>> clever. How is the size of sa_data defined ?
>
> Magic size...
>
> ~/linux/include/linux/socket.h:
>
> struct sockaddr {
> sa_family_t sa_family; /* address family, AF_xxx */
> char sa_data[14]; /* 14 bytes of protocol address */
> };
>
>
>> Would it hurt when some uses IFNAMSIZ here ?
>
so i should be more direct. the idea was :
char sa_data[IFNAMSIZ];
> If copy _to_ sa_data string of maximum IFNAMSIZ bytes - yes.
>
>
> In packet_getname_spkt() the output buffer is 128 bytes, so it doesn't
> really overflows anything. I don't think that *_getname() implementations
> don't know this.
>
>> Perhaps someone who know more about the network stack can figure out what is actualy done
>> with uaddr->sa_data.
>
> Yeah, also I wonder whether it is always NULL-terminated string or not.
>
>> looks like a can of worms.
>>
>>
>> In packet_bind_spkt() they will copy a char[15], obviously it is a real problem.
>
> No, packet_bind_spkt() copies 14-byte string into array of 15 bytes.
> The vice versa would be a bug.
>
ups your are right, wrong way around. it still does not look clever.
I have the feeling that the basic idea what to store the string with out \0.
according to this:
http://www.gnu.org/s/libc/manual/html_node/Address-Formats.html
It should work.
re,
wh
David,
Are there any other suggestions about this patch?
Thanks,
--
Vasiliy
From: Vasiliy Kulikov <[email protected]>
Date: Wed, 10 Nov 2010 21:16:42 +0300
> Are there any other suggestions about this patch?
None from me, so even if there are no changes please just repost
this.
Thanks.