Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 20 Oct 2021 19:24:57 +0000
From:   Sargun Dhillon <sargun@sargun.me>
To:     Sergey Ryazanov <ryazanov.s.a@gmail.com>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        netdev <netdev@vger.kernel.org>,
        Christian Brauner <christian.brauner@ubuntu.com>
Subject: Re: Retrieving the network namespace of a socket
Message-ID: <20211020192456.GA23489@ircssh-2.c.rugged-nimbus-611.internal>
References: <20211020095707.GA16295@ircssh-2.c.rugged-nimbus-611.internal>
 <CAHNKnsRFah6MRxECTLNwu+maN0o9jS9ENzSAiWS4v1247BqYdg@mail.gmail.com>
 <20211020163417.GA21040@ircssh-2.c.rugged-nimbus-611.internal>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20211020163417.GA21040@ircssh-2.c.rugged-nimbus-611.internal>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

On Wed, Oct 20, 2021 at 04:34:18PM +0000, Sargun Dhillon wrote:
> On Wed, Oct 20, 2021 at 05:03:56PM +0300, Sergey Ryazanov wrote:
> > Hello Sargun,
> > 
> > On Wed, Oct 20, 2021 at 12:57 PM Sargun Dhillon <sargun@sargun.me> wrote:
> > > I'm working on a problem where I need to determine which network namespace a
> > > given socket is in. I can currently bruteforce this by using INET_DIAG, and
> > > enumerating namespaces and working backwards.
> > 
> > Namespace is not a per-socket, but a per-process attribute. So each
> > socket of a process belongs to the same namespace.
> > 
> > Could you elaborate what kind of problem you are trying to solve?
> > Maybe there is a more simple solution. for it.
> > 
> > -- 
> > Sergey
> 
> That's not entirely true. See the folowing code:
> 
> int main() {
> 	int fd1, fd2;
> 	fd1 = socket(AF_INET, SOCK_STREAM, 0);
> 	assert(fd1 >= 0);
> 	assert(unshare(CLONE_NEWNET) == 0);
> 	fd2 = socket(AF_INET, SOCK_STREAM, 0);
> 	assert(fd2 >= 0);
> }
> 
> fd1 and fd2 have different sock_net.
> 
> The context for this is:
> https://linuxplumbersconf.org/event/11/contributions/932/
> 
> We need to figure out, for a given socket, if it has reachability to a given IP.

So, I was lazy / misread documentation. It turns out SIOCGSKNS does exactly
what I need.

Nonetheless, it's a little weird and awkward that it is exists. I was wondering
if this functionality made sense as part of kcmp. I wrote up a quick patch
to see if anyone was interested:

diff --git a/include/uapi/linux/kcmp.h b/include/uapi/linux/kcmp.h
index ef1305010925..d6b9c3923d20 100644
--- a/include/uapi/linux/kcmp.h
+++ b/include/uapi/linux/kcmp.h
@@ -14,6 +14,7 @@ enum kcmp_type {
 	KCMP_IO,
 	KCMP_SYSVSEM,
 	KCMP_EPOLL_TFD,
+	KCMP_NETNS,
 
 	KCMP_TYPES,
 };
diff --git a/kernel/kcmp.c b/kernel/kcmp.c
index 5353edfad8e1..8fadae4b588f 100644
--- a/kernel/kcmp.c
+++ b/kernel/kcmp.c
@@ -18,6 +18,8 @@
 #include <linux/file.h>
 
 #include <asm/unistd.h>
+#include <net/net_namespace.h>
+#include <net/sock.h>
 
 /*
  * We don't expose the real in-memory order of objects for security reasons.
@@ -132,6 +134,58 @@ static int kcmp_epoll_target(struct task_struct *task1,
 }
 #endif
 
+#ifdef CONFIG_NET
+static int __kcmp_netns_target(struct task_struct *task1,
+			       struct task_struct *task2,
+			       struct file *filp1,
+			       struct file *filp2)
+{
+	struct socket *sock1, *sock2;
+	struct net *net1, *net2;
+
+	sock1 = sock_from_file(filp1);
+	sock2 = sock_from_file(filp1);
+	if (!sock1 || !sock2)
+		return -ENOTSOCK;
+
+	net1 = sock_net(sock1->sk);
+	net2 = sock_net(sock2->sk);
+
+	return kcmp_ptr(net1, net2, KCMP_NETNS);
+}
+
+static int kcmp_netns_target(struct task_struct *task1,
+			     struct task_struct *task2,
+			     unsigned long idx1,
+			     unsigned long idx2)
+{
+	struct file *filp1, *filp2;
+
+	int ret = -EBADF;
+
+	filp1 = fget_task(task1, idx1);
+	if (filp1) {
+		filp2 = fget_task(task2, idx2);
+		if (filp2) {
+			ret = __kcmp_netns_target(task1, task2, filp1, filp2);
+			fput(filp2);
+		}
+
+		fput(filp1);
+	}
+
+	return ret;
+}
+#else
+static int kcmp_netns_target(struct task_struct *task1,
+			     struct task_struct *task2,
+			     unsigned long idx1,
+			     unsigned long idx2)
+{
+	return -EOPNOTSUPP;
+}
+#endif
+
 SYSCALL_DEFINE5(kcmp, pid_t, pid1, pid_t, pid2, int, type,
 		unsigned long, idx1, unsigned long, idx2)
 {
@@ -206,6 +260,9 @@ SYSCALL_DEFINE5(kcmp, pid_t, pid1, pid_t, pid2, int, type,
 	case KCMP_EPOLL_TFD:
 		ret = kcmp_epoll_target(task1, task2, idx1, (void *)idx2);
 		break;
+	case KCMP_NETNS:
+		ret = kcmp_netns_target(task1, task2, idx1, idx2);
+		break;
 	default:
 		ret = -EINVAL;
 		break;