| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Mon, 26 May 2014 21:24:34 -0700
From: ebiederm@...ssion.com (Eric W. Biederman)
To: Andy Lutomirski <luto@...capital.net>
Cc: Linus Torvalds <torvalds@...ux-foundation.org>,
"Jorge Boncompte \[DTI2\]" <jorge@...2.net>,
Jiri Benc <jbenc@...hat.com>,
David Miller <davem@...emloft.net>,
Vivek Goyal <vgoyal@...hat.com>,
Simo Sorce <ssorce@...hat.com>,
"security\@kernel.org" <security@...nel.org>,
Network Development <netdev@...r.kernel.org>,
"Serge E. Hallyn" <serge@...lyn.com>,
Michael Kerrisk-manpages <mtk.manpages@...il.com>
Subject: Re: [RFC][PATCH 2/1] netlink: Use the credential at the time the destination address was set.
Andy Lutomirski <luto@...capital.net> writes:
> On Sun, May 25, 2014 at 10:36 PM, Eric W. Biederman
> <ebiederm@...ssion.com> wrote:
>>
>> When sending a message over a netlink socket and then checking to see if
>> the person is authorized to send the message it is important that we verify
>> both the sender of the message and the whoever it is that set the destination
>> of the message both have permission. Otherwise it becomes possible for an
>> unpriivleged user to set the message destination and trick an suid process
>> to write to the socket and change the network connection.
>>
>> For netlink sockets socket() sets the default destination address to 0 (the kernel)
>> so we need to remember the credentials when a socket is created.
>>
>> For netlink sockets connect() changes the default destination address so
>> we need to remember the credentials of the last changer of the default destination
>> with connect.
>>
>> This results is there always being a valid remembered credential on the socket
>> and so that credential is unconditionally freed in netlink_release().
>>
>> This change makes the semantics of the permission checks of netlink sockets
>> make sense,and removes the possibility of an unprivileged user getting access
>> to a root own socket and changing the destination address with connect.
>>
>> Signed-off-by: "Eric W. Biederman" <ebiederm@...ssion.com>
>> ---
>>
>> This winds up continuig to grab credentials when socket() is called
>> because we actually set the destination address in socket() for netlink
>> sockets, but now we update those credentials when connect() is called.
>>
>> include/linux/netlink.h | 2 +-
>> net/netlink/af_netlink.c | 27 ++++++++++++++++++++++++---
>> net/netlink/af_netlink.h | 1 +
>> 3 files changed, 26 insertions(+), 4 deletions(-)
>>
>> diff --git a/include/linux/netlink.h b/include/linux/netlink.h
>> index f289d085f87f..4f4607c0a1a1 100644
>> --- a/include/linux/netlink.h
>> +++ b/include/linux/netlink.h
>> @@ -19,7 +19,7 @@ enum netlink_skb_flags {
>> NETLINK_SKB_MMAPED = 0x1, /* Packet data is mmaped */
>> NETLINK_SKB_TX = 0x2, /* Packet was sent by userspace */
>> NETLINK_SKB_DELIVERED = 0x4, /* Packet was delivered */
>> - NETLINK_SKB_DST = 0x8, /* Packet not socket destination */
>> + NETLINK_SKB_DST = 0x8, /* Dst set in sendto or sendmsg */
>> };
>>
>> struct netlink_skb_parms {
>> diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
>> index 15c731f03fa6..5b886ed6a648 100644
>> --- a/net/netlink/af_netlink.c
>> +++ b/net/netlink/af_netlink.c
>> @@ -1191,6 +1191,7 @@ static int __netlink_create(struct net *net, struct socket *sock,
>> mutex_init(nlk->cb_mutex);
>> }
>> init_waitqueue_head(&nlk->wait);
>> + nlk->cred = get_current_cred();
>> #ifdef CONFIG_NETLINK_MMAP
>> mutex_init(&nlk->pg_vec_lock);
>> #endif
>> @@ -1291,6 +1292,7 @@ static int netlink_release(struct socket *sock)
>> NETLINK_URELEASE, &n);
>> }
>>
>> + put_cred(nlk->cred);
>> module_put(nlk->module);
>>
>> netlink_table_grab();
>> @@ -1377,9 +1379,17 @@ retry:
>> bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
>> struct user_namespace *user_ns, int cap)
>> {
>> - return ((nsp->flags & NETLINK_SKB_DST) ||
>> - file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
>> - ns_capable(user_ns, cap);
>> + const struct cred *cred;
>> + bool capable;
>> +
>> + rcu_read_lock();
>> + cred = nlk_sk(nsp->sk)->cred;
>> + capable = ((nsp->flags & NETLINK_SKB_DST) ||
>> + security_capable(cred, user_ns, cap)) &&
>> + ns_capable(user_ns, cap);
>> + rcu_read_unlock();
>> +
>> + return capable;
>> }
>> EXPORT_SYMBOL(__netlink_ns_capable);
>>
>> @@ -1569,14 +1579,20 @@ static int netlink_connect(struct socket *sock, struct sockaddr *addr,
>> struct sock *sk = sock->sk;
>> struct netlink_sock *nlk = nlk_sk(sk);
>> struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
>> + const struct cred *old_cred;
>>
>> if (alen < sizeof(addr->sa_family))
>> return -EINVAL;
>>
>> if (addr->sa_family == AF_UNSPEC) {
>> + lock_sock(sk);
>> sk->sk_state = NETLINK_UNCONNECTED;
>> nlk->dst_portid = 0;
>> nlk->dst_group = 0;
>> + old_cred = nlk->cred;
>> + nlk->cred = get_current_cred();
>> + put_cred(old_cred);
>> + release_sock(sk);
>> return 0;
>
> I'm confused about the locking. Either lock_sock prevents concurrent
> sendmsg, in which case the rcu_read_lock should be unnecessary, or it
> doesn't, in which case I think this is racy, because sendmsg could see
> a cred and dst_whatever that don't match. Admittedly, exploiting that
> race is probably extremely difficult.
My intention was to prevent races during connect which can cause the
wrong cred to be freed, which could be pretty nasty.
The intention of the rcu bits was just to guarantee that the cred
doesn't get freed during the write.
As for mismatches. There is only an issue with write/send that does
not specify a destination address. Although I admit sendmsg differing
could be a problem as well.
Getting the races out of the send path without destroying performance
is beyond my imagination today. It would be nice if we could make the
locking simply that connect waits until sendmmsg and similar functions
are complete preventing races.
I hate interfaces where things can be set multiple times on a kernel
data structure the races are absolutely horrible.
> If the RCU protection is needed, shouldn't this use rcu_assign_pointer
> and rcu_dereference?
Yep that would be good.
Any chance you could pick this up. I am going to be busy for the next
couple of days.
Eric
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists