[PATCH] net: fix oops in socket ioctl cmd SIOCGSKNS when NET_NS is disabled

When NET_NS is not enabled, socket ioctl cmd SIOCGSKNS should do nothing
but acknowledge userspace it is not supported. Otherwise, kernel would
panic wherever nsfs trys to access ns->ops since the proc_ns_operations
is not implemented in this case.

[7.670023] Unable to handle kernel NULL pointer dereference at virtual address 00000010
[7.670268] pgd = 32b54000
[7.670544] [00000010] *pgd=00000000
[7.671861] Internal error: Oops: 5 [#1] SMP ARM
[7.672315] Modules linked in:
[7.672918] CPU: 0 PID: 1 Comm: systemd Not tainted 5.13.0-rc3-00375-g6799d4f2da49 #16
[7.673309] Hardware name: Generic DT based system
[7.673642] PC is at nsfs_evict+0x24/0x30
[7.674486] LR is at clear_inode+0x20/0x9c

Signed-off-by: Changbin Du <redacted>
Cc: <redacted> # v4.9
---
 net/socket.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/net/socket.c b/net/socket.c
index 27e3e7d53f8e..644b46112d35 100644
--- a/net/socket.c
+++ b/net/socket.c

@@ -1149,11 +1149,15 @@ static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 			mutex_unlock(&vlan_ioctl_mutex);
 			break;
 		case SIOCGSKNS:
+#ifdef CONFIG_NET_NS
 			err = -EPERM;
 			if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 				break;
 
 			err = open_related_ns(&net->ns, get_net_ns);
+#else
+			err = -ENOTSUPP;
+#endif
 			break;
 		case SIOCGSTAMP_OLD:
 		case SIOCGSTAMPNS_OLD:

-- 
2.27.0

On Sat, 29 May 2021 14:05:26 +0800 Changbin Du wrote:

When NET_NS is not enabled, socket ioctl cmd SIOCGSKNS should do nothing
but acknowledge userspace it is not supported. Otherwise, kernel would
panic wherever nsfs trys to access ns->ops since the proc_ns_operations
is not implemented in this case.

[7.670023] Unable to handle kernel NULL pointer dereference at virtual address 00000010
[7.670268] pgd = 32b54000
[7.670544] [00000010] *pgd=00000000
[7.671861] Internal error: Oops: 5 [#1] SMP ARM
[7.672315] Modules linked in:
[7.672918] CPU: 0 PID: 1 Comm: systemd Not tainted 5.13.0-rc3-00375-g6799d4f2da49 #16
[7.673309] Hardware name: Generic DT based system
[7.673642] PC is at nsfs_evict+0x24/0x30
[7.674486] LR is at clear_inode+0x20/0x9c

Signed-off-by: Changbin Du <redacted>
Cc: <redacted> # v4.9

Please provide a Fixes tag.

There's a few more places with this exact code. Can we please add the
check in get_net_ns? That should fix all callers.

+#else
+			err = -ENOTSUPP;

EOPNOTSUPP, you shouldn't return ENOTSUPP to user space.

+#endif
 			break;
 		case SIOCGSTAMP_OLD:
 		case SIOCGSTAMPNS_OLD:

On Fri, May 28, 2021 at 11:08 PM Changbin Du [off-list ref] wrote:

I wonder if it is easier if we just reject ns->ops==NULL case
in open_related_ns(). For 1) we can save an ugly #ifdef here;
2) drivers/net/tun.c has the same bugs.

Something like this:

diff --git a/fs/nsfs.c b/fs/nsfs.c
index 800c1d0eb0d0..d63414604e99 100644
--- a/fs/nsfs.c
+++ b/fs/nsfs.c

@@ -152,6 +152,9 @@ int open_related_ns(struct ns_common *ns,
        int err;
        int fd;

+       if (!ns->ops)
+               return -EOPNOTSUPP;
+
        fd = get_unused_fd_flags(O_CLOEXEC);
        if (fd < 0)
                return fd;

From: Cong Wang

Sent: 29 May 2021 20:15

On Fri, May 28, 2021 at 11:08 PM Changbin Du [off-list ref] wrote:

quoted

diff --git a/net/socket.c b/net/socket.c
index 27e3e7d53f8e..644b46112d35 100644
--- a/net/socket.c
+++ b/net/socket.c

@@ -1149,11 +1149,15 @@ static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
                        mutex_unlock(&vlan_ioctl_mutex);
                        break;
                case SIOCGSKNS:
+#ifdef CONFIG_NET_NS
                        err = -EPERM;
                        if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
                                break;

                        err = open_related_ns(&net->ns, get_net_ns);
+#else
+                       err = -ENOTSUPP;
+#endif

I wonder if it is easier if we just reject ns->ops==NULL case
in open_related_ns(). For 1) we can save an ugly #ifdef here;
2) drivers/net/tun.c has the same bugs.

If CONFIG_NET_NS is unset then why not make both ns_capable()
and open_related_ns() unconditionally return -ENOTSUPP?

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

On Sat, May 29, 2021 at 11:27:35AM -0700, Jakub Kicinski wrote:

On Sat, 29 May 2021 14:05:26 +0800 Changbin Du wrote:

quoted

When NET_NS is not enabled, socket ioctl cmd SIOCGSKNS should do nothing
but acknowledge userspace it is not supported. Otherwise, kernel would
panic wherever nsfs trys to access ns->ops since the proc_ns_operations
is not implemented in this case.

[7.670023] Unable to handle kernel NULL pointer dereference at virtual address 00000010
[7.670268] pgd = 32b54000
[7.670544] [00000010] *pgd=00000000
[7.671861] Internal error: Oops: 5 [#1] SMP ARM
[7.672315] Modules linked in:
[7.672918] CPU: 0 PID: 1 Comm: systemd Not tainted 5.13.0-rc3-00375-g6799d4f2da49 #16
[7.673309] Hardware name: Generic DT based system
[7.673642] PC is at nsfs_evict+0x24/0x30
[7.674486] LR is at clear_inode+0x20/0x9c

Signed-off-by: Changbin Du <redacted>
Cc: <redacted> # v4.9

Please provide a Fixes tag.

Now it will be fixed by nsfs side. And the code has been changed to many times..

quoted

diff --git a/net/socket.c b/net/socket.c
index 27e3e7d53f8e..644b46112d35 100644
--- a/net/socket.c
+++ b/net/socket.c

@@ -1149,11 +1149,15 @@ static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 			mutex_unlock(&vlan_ioctl_mutex);
 			break;
 		case SIOCGSKNS:
+#ifdef CONFIG_NET_NS
 			err = -EPERM;
 			if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 				break;
 
 			err = open_related_ns(&net->ns, get_net_ns);

There's a few more places with this exact code. Can we please add the
check in get_net_ns? That should fix all callers.

quoted

+#else
+			err = -ENOTSUPP;

EOPNOTSUPP, you shouldn't return ENOTSUPP to user space.

Thanks for pointing out. Will change it.

quoted

+#endif
 			break;
 		case SIOCGSTAMP_OLD:
 		case SIOCGSTAMPNS_OLD:

-- 
Cheers,
Changbin Du

On Mon, May 31, 2021 at 08:30:58AM +0000, David Laight wrote:

From: Cong Wang

quoted

Sent: 29 May 2021 20:15

On Fri, May 28, 2021 at 11:08 PM Changbin Du [off-list ref] wrote:

quoted

diff --git a/net/socket.c b/net/socket.c
index 27e3e7d53f8e..644b46112d35 100644
--- a/net/socket.c
+++ b/net/socket.c

@@ -1149,11 +1149,15 @@ static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
                        mutex_unlock(&vlan_ioctl_mutex);
                        break;
                case SIOCGSKNS:
+#ifdef CONFIG_NET_NS
                        err = -EPERM;
                        if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
                                break;

                        err = open_related_ns(&net->ns, get_net_ns);
+#else
+                       err = -ENOTSUPP;
+#endif

I wonder if it is easier if we just reject ns->ops==NULL case
in open_related_ns(). For 1) we can save an ugly #ifdef here;
2) drivers/net/tun.c has the same bugs.

If CONFIG_NET_NS is unset then why not make both ns_capable()
and open_related_ns() unconditionally return -ENOTSUPP?

Here is the new fix that reject creating indoe for disabled ns.

--- a/fs/nsfs.c
+++ b/fs/nsfs.c

@@ -62,6 +62,10 @@ static int __ns_get_path(struct path *path, struct ns_common *ns)
        struct inode *inode;
        unsigned long d;
 
+       /* In case the namespace is not actually enabled. */
+       if (!ns->ops)
+               return -EOPNOTSUPP;

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

-- 
Cheers,
Changbin Du