Re: bpf: net/core/filter.c:2115 suspicious rcu_dereference_protected() usage!

From: Daniel Borkmann
Date: Wed Mar 30 2016 - 07:33:56 EST


On 03/30/2016 11:42 AM, Michal Kubecek wrote:
On Tue, Mar 29, 2016 at 04:39:43PM +0200, Daniel Borkmann wrote:

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index afdf950617c3..7417d7c20bab 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1818,11 +1818,13 @@ static int set_offload(struct tun_struct *tun, unsigned long arg)
static void tun_detach_filter(struct tun_struct *tun, int n)
{
int i;
- struct tun_file *tfile;

for (i = 0; i < n; i++) {
- tfile = rtnl_dereference(tun->tfiles[i]);
- sk_detach_filter(tfile->socket.sk);
+ struct sock *sk = rtnl_dereference(tun->tfiles[i])->socket.sk;
+
+ lock_sock(sk);
+ sk_detach_filter(sk);
+ release_sock(sk);
}

tun->filter_attached = false;


In tun case, the control path for tun_attach_filter() and tun_detach_filter()
is under RTNL lock (held in __tun_chr_ioctl()).

So in the BPF core the rcu_dereference_protected(<sk_filter>, sock_owned_by_user(sk))
looks like a false positive in this specific use case to me, that we should probably
just silence.

Running the filter via sk_filter() in tun device happens under rcu_read_lock(),
so the dereference and assignment pair seems okay to me.

Was wondering whether we should convert this to unattached BPF filter, but this
would break with existing expectations from sk_filter() (e.g. security modules).

If we want to silence it, could be something like the below (only compile-tested):

drivers/net/tun.c | 8 +++++---
include/linux/filter.h | 4 ++++
net/core/filter.c | 33 +++++++++++++++++++++------------
3 files changed, 30 insertions(+), 15 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index afdf950..510e90a 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -622,7 +622,8 @@ static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filte

/* Re-attach the filter to persist device */
if (!skip_filter && (tun->filter_attached == true)) {
- err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
+ err = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
+ lockdep_rtnl_is_held());
if (!err)
goto out;
}
@@ -1822,7 +1823,7 @@ static void tun_detach_filter(struct tun_struct *tun, int n)

for (i = 0; i < n; i++) {
tfile = rtnl_dereference(tun->tfiles[i]);
- sk_detach_filter(tfile->socket.sk);
+ __sk_detach_filter(tfile->socket.sk, lockdep_rtnl_is_held());
}

tun->filter_attached = false;
@@ -1835,7 +1836,8 @@ static int tun_attach_filter(struct tun_struct *tun)

for (i = 0; i < tun->numqueues; i++) {
tfile = rtnl_dereference(tun->tfiles[i]);
- ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
+ ret = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
+ lockdep_rtnl_is_held());
if (ret) {
tun_detach_filter(tun, i);
return ret;
diff --git a/include/linux/filter.h b/include/linux/filter.h
index 43aa1f8..a51a536 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -465,10 +465,14 @@ int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
void bpf_prog_destroy(struct bpf_prog *fp);

int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
+int __sk_attach_filter(struct sock_fprog *fprog, struct sock *sk,
+ bool locked);
int sk_attach_bpf(u32 ufd, struct sock *sk);
int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
int sk_detach_filter(struct sock *sk);
+int __sk_detach_filter(struct sock *sk, bool locked);
+
int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
unsigned int len);

diff --git a/net/core/filter.c b/net/core/filter.c
index 2429918..02f2f6c 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -1149,7 +1149,8 @@ void bpf_prog_destroy(struct bpf_prog *fp)
}
EXPORT_SYMBOL_GPL(bpf_prog_destroy);

-static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
+static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk,
+ bool locked)
{
struct sk_filter *fp, *old_fp;

@@ -1165,10 +1166,8 @@ static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
return -ENOMEM;
}

- old_fp = rcu_dereference_protected(sk->sk_filter,
- sock_owned_by_user(sk));
+ old_fp = rcu_dereference_protected(sk->sk_filter, locked);
rcu_assign_pointer(sk->sk_filter, fp);
-
if (old_fp)
sk_filter_uncharge(sk, old_fp);

@@ -1247,7 +1246,8 @@ struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
* occurs or there is insufficient memory for the filter a negative
* errno code is returned. On success the return is zero.
*/
-int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
+int __sk_attach_filter(struct sock_fprog *fprog, struct sock *sk,
+ bool locked)
{
struct bpf_prog *prog = __get_filter(fprog, sk);
int err;
@@ -1255,7 +1255,7 @@ int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
if (IS_ERR(prog))
return PTR_ERR(prog);

- err = __sk_attach_prog(prog, sk);
+ err = __sk_attach_prog(prog, sk, locked);
if (err < 0) {
__bpf_prog_release(prog);
return err;
@@ -1263,7 +1263,12 @@ int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)

return 0;
}
-EXPORT_SYMBOL_GPL(sk_attach_filter);
+EXPORT_SYMBOL_GPL(__sk_attach_filter);
+
+int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
+{
+ return __sk_attach_filter(fprog, sk, sock_owned_by_user(sk));
+}

int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
{
@@ -1309,7 +1314,7 @@ int sk_attach_bpf(u32 ufd, struct sock *sk)
if (IS_ERR(prog))
return PTR_ERR(prog);

- err = __sk_attach_prog(prog, sk);
+ err = __sk_attach_prog(prog, sk, sock_owned_by_user(sk));
if (err < 0) {
bpf_prog_put(prog);
return err;
@@ -2445,7 +2450,7 @@ static int __init register_sk_filter_ops(void)
}
late_initcall(register_sk_filter_ops);

-int sk_detach_filter(struct sock *sk)
+int __sk_detach_filter(struct sock *sk, bool locked)
{
int ret = -ENOENT;
struct sk_filter *filter;
@@ -2453,8 +2458,7 @@ int sk_detach_filter(struct sock *sk)
if (sock_flag(sk, SOCK_FILTER_LOCKED))
return -EPERM;

- filter = rcu_dereference_protected(sk->sk_filter,
- sock_owned_by_user(sk));
+ filter = rcu_dereference_protected(sk->sk_filter, locked);
if (filter) {
RCU_INIT_POINTER(sk->sk_filter, NULL);
sk_filter_uncharge(sk, filter);
@@ -2463,7 +2467,12 @@ int sk_detach_filter(struct sock *sk)

return ret;
}
-EXPORT_SYMBOL_GPL(sk_detach_filter);
+EXPORT_SYMBOL_GPL(__sk_detach_filter);
+
+int sk_detach_filter(struct sock *sk)
+{
+ return __sk_detach_filter(sk, sock_owned_by_user(sk));
+}

int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
unsigned int len)
--
1.9.3

Looks good to me.

I'm just not sure checking if we hold the right lock depending on caller
is worth the extra complexity. After all, what is really needed is to
hold _some_ lock guaranteeing sk_attach_prog() and sk_detach_filter()
are safe so that just changing the condition in both to

sock_owned_by_user(sk) || lockdep_rtnl_is_held()

It would certainly silence it, but would be less accurate in terms of lock
proving as opposed to the diff above. E.g. rntl could be held elsewhere,
while someone attaches a socket filter w/o having locked the socket (currently
not the case, but it would kind of defeat the purpose of rcu_dereference_protected()
here). Was thinking about using a extra socket flag to indicate it's
externally managed, but it's not really worth wasting sk's flags bit
space just for this corner case.

could suffice.

Michal Kubecek