# Some open questions about this patch below.
#
One of the rules regarding capabilities is:
A process that resides in the parent of the user namespace and
whose effective user ID matches the owner of the namespace has
all capabilities in the namespace.
Therefore, in order to write code that discovers whether process X has
capabilities in namespace Y, we need a way to find out who the creator
of a user namespace is. This patch adds an NS_GET_CREATOR_UID ioctl()
that returns the (munged) UID of the creator of the user namespace
referred to by the specified file descriptor.
If the supplied file descriptor does not refer to a user namespace,
the operation fails with the error EINVAL.
Signed-off-by: Michael Kerrisk <[email protected]>
---
fs/nsfs.c | 6 ++++++
include/uapi/linux/nsfs.h | 8 +++++---
2 files changed, 11 insertions(+), 3 deletions(-)
Open questions:
* Would it be preferabe to separate the logic for NS_GET_CREATOR_UID
into a small helper function?
* Is this a correct use of container_of()? I did not immediately
see another way to get to the user_namespace struct, but I
may well have missed something.
diff --git a/fs/nsfs.c b/fs/nsfs.c
index 5d53476..26f6d94 100644
--- a/fs/nsfs.c
+++ b/fs/nsfs.c
@@ -163,6 +163,7 @@ int open_related_ns(struct ns_common *ns,
static long ns_ioctl(struct file *filp, unsigned int ioctl,
unsigned long arg)
{
+ struct user_namespace *user_ns;
struct ns_common *ns = get_proc_ns(file_inode(filp));
switch (ioctl) {
@@ -174,6 +175,11 @@ static long ns_ioctl(struct file *filp, unsigned int ioctl,
return open_related_ns(ns, ns->ops->get_parent);
case NS_GET_NSTYPE:
return ns->ops->type;
+ case NS_GET_CREATOR_UID:
+ if (ns->ops->type != CLONE_NEWUSER)
+ return -EINVAL;
+ user_ns = container_of(ns, struct user_namespace, ns);
+ return from_kuid_munged(current_user_ns(), user_ns->owner);
default:
return -ENOTTY;
}
diff --git a/include/uapi/linux/nsfs.h b/include/uapi/linux/nsfs.h
index 2b48df1..b3c6c78 100644
--- a/include/uapi/linux/nsfs.h
+++ b/include/uapi/linux/nsfs.h
@@ -6,11 +6,13 @@
#define NSIO 0xb7
/* Returns a file descriptor that refers to an owning user namespace */
-#define NS_GET_USERNS _IO(NSIO, 0x1)
+#define NS_GET_USERNS _IO(NSIO, 0x1)
/* Returns a file descriptor that refers to a parent namespace */
-#define NS_GET_PARENT _IO(NSIO, 0x2)
+#define NS_GET_PARENT _IO(NSIO, 0x2)
/* Returns the type of namespace (CLONE_NEW* value) referred to by
file descriptor */
-#define NS_GET_NSTYPE _IO(NSIO, 0x3)
+#define NS_GET_NSTYPE _IO(NSIO, 0x3)
+/* Get creator UID for a user namespace */
+#define NS_GET_CREATOR_UID _IO(NSIO, 0x4)
#endif /* __LINUX_NSFS_H */
--
2.5.5
On Mon, Dec 19, 2016 at 03:38:35PM +0100, Michael Kerrisk (man-pages) wrote:
> # Some open questions about this patch below.
> #
> One of the rules regarding capabilities is:
>
> A process that resides in the parent of the user namespace and
> whose effective user ID matches the owner of the namespace has
> all capabilities in the namespace.
>
> Therefore, in order to write code that discovers whether process X has
> capabilities in namespace Y, we need a way to find out who the creator
> of a user namespace is. This patch adds an NS_GET_CREATOR_UID ioctl()
> that returns the (munged) UID of the creator of the user namespace
> referred to by the specified file descriptor.
>
> If the supplied file descriptor does not refer to a user namespace,
> the operation fails with the error EINVAL.
>
> Signed-off-by: Michael Kerrisk <[email protected]>
> ---
> fs/nsfs.c | 6 ++++++
> include/uapi/linux/nsfs.h | 8 +++++---
> 2 files changed, 11 insertions(+), 3 deletions(-)
>
> Open questions:
>
> * Would it be preferabe to separate the logic for NS_GET_CREATOR_UID
> into a small helper function?
> * Is this a correct use of container_of()? I did not immediately
> see another way to get to the user_namespace struct, but I
> may well have missed something.
>
> diff --git a/fs/nsfs.c b/fs/nsfs.c
> index 5d53476..26f6d94 100644
> --- a/fs/nsfs.c
> +++ b/fs/nsfs.c
> @@ -163,6 +163,7 @@ int open_related_ns(struct ns_common *ns,
> static long ns_ioctl(struct file *filp, unsigned int ioctl,
> unsigned long arg)
> {
> + struct user_namespace *user_ns;
> struct ns_common *ns = get_proc_ns(file_inode(filp));
>
> switch (ioctl) {
> @@ -174,6 +175,11 @@ static long ns_ioctl(struct file *filp, unsigned int ioctl,
> return open_related_ns(ns, ns->ops->get_parent);
> case NS_GET_NSTYPE:
> return ns->ops->type;
> + case NS_GET_CREATOR_UID:
> + if (ns->ops->type != CLONE_NEWUSER)
> + return -EINVAL;
> + user_ns = container_of(ns, struct user_namespace, ns);
> + return from_kuid_munged(current_user_ns(), user_ns->owner);
uid_t is "unsigned int", ioctl() returns long, so it may be hard to
distinguish user id-s from errors on x32.
off-topic: What is about user_ns->group? I can't find where it is used...
> default:
> return -ENOTTY;
> }
> diff --git a/include/uapi/linux/nsfs.h b/include/uapi/linux/nsfs.h
> index 2b48df1..b3c6c78 100644
> --- a/include/uapi/linux/nsfs.h
> +++ b/include/uapi/linux/nsfs.h
> @@ -6,11 +6,13 @@
> #define NSIO 0xb7
>
> /* Returns a file descriptor that refers to an owning user namespace */
> -#define NS_GET_USERNS _IO(NSIO, 0x1)
> +#define NS_GET_USERNS _IO(NSIO, 0x1)
> /* Returns a file descriptor that refers to a parent namespace */
> -#define NS_GET_PARENT _IO(NSIO, 0x2)
> +#define NS_GET_PARENT _IO(NSIO, 0x2)
> /* Returns the type of namespace (CLONE_NEW* value) referred to by
> file descriptor */
> -#define NS_GET_NSTYPE _IO(NSIO, 0x3)
> +#define NS_GET_NSTYPE _IO(NSIO, 0x3)
> +/* Get creator UID for a user namespace */
> +#define NS_GET_CREATOR_UID _IO(NSIO, 0x4)
>
> #endif /* __LINUX_NSFS_H */
> --
> 2.5.5
>
Hi Andrei,
On 12/21/2016 04:13 AM, Andrei Vagin wrote:
> On Mon, Dec 19, 2016 at 03:38:35PM +0100, Michael Kerrisk (man-pages) wrote:
>> # Some open questions about this patch below.
>> #
>> One of the rules regarding capabilities is:
>>
>> A process that resides in the parent of the user namespace and
>> whose effective user ID matches the owner of the namespace has
>> all capabilities in the namespace.
>>
>> Therefore, in order to write code that discovers whether process X has
>> capabilities in namespace Y, we need a way to find out who the creator
>> of a user namespace is. This patch adds an NS_GET_CREATOR_UID ioctl()
>> that returns the (munged) UID of the creator of the user namespace
>> referred to by the specified file descriptor.
>>
>> If the supplied file descriptor does not refer to a user namespace,
>> the operation fails with the error EINVAL.
>>
>> Signed-off-by: Michael Kerrisk <[email protected]>
>> ---
>> fs/nsfs.c | 6 ++++++
>> include/uapi/linux/nsfs.h | 8 +++++---
>> 2 files changed, 11 insertions(+), 3 deletions(-)
>>
>> Open questions:
>>
>> * Would it be preferabe to separate the logic for NS_GET_CREATOR_UID
>> into a small helper function?
>> * Is this a correct use of container_of()? I did not immediately
>> see another way to get to the user_namespace struct, but I
>> may well have missed something.
>>
>> diff --git a/fs/nsfs.c b/fs/nsfs.c
>> index 5d53476..26f6d94 100644
>> --- a/fs/nsfs.c
>> +++ b/fs/nsfs.c
>> @@ -163,6 +163,7 @@ int open_related_ns(struct ns_common *ns,
>> static long ns_ioctl(struct file *filp, unsigned int ioctl,
>> unsigned long arg)
>> {
>> + struct user_namespace *user_ns;
>> struct ns_common *ns = get_proc_ns(file_inode(filp));
>>
>> switch (ioctl) {
>> @@ -174,6 +175,11 @@ static long ns_ioctl(struct file *filp, unsigned int ioctl,
>> return open_related_ns(ns, ns->ops->get_parent);
>> case NS_GET_NSTYPE:
>> return ns->ops->type;
>> + case NS_GET_CREATOR_UID:
>> + if (ns->ops->type != CLONE_NEWUSER)
>> + return -EINVAL;
>> + user_ns = container_of(ns, struct user_namespace, ns);
>> + return from_kuid_munged(current_user_ns(), user_ns->owner);
>
> uid_t is "unsigned int", ioctl() returns long, so it may be hard to
> distinguish user id-s from errors on x32.
Good point. So, we could instead return the UID via a buffer pointed to
by the ioctl() arg. That would seem better, right?
> off-topic: What is about user_ns->group? I can't find where it is used...
I've no idea. Like you, I can't see any place where it's being used.
Cheers,
Michael
>> default:
>> return -ENOTTY;
>> }
>> diff --git a/include/uapi/linux/nsfs.h b/include/uapi/linux/nsfs.h
>> index 2b48df1..b3c6c78 100644
>> --- a/include/uapi/linux/nsfs.h
>> +++ b/include/uapi/linux/nsfs.h
>> @@ -6,11 +6,13 @@
>> #define NSIO 0xb7
>>
>> /* Returns a file descriptor that refers to an owning user namespace */
>> -#define NS_GET_USERNS _IO(NSIO, 0x1)
>> +#define NS_GET_USERNS _IO(NSIO, 0x1)
>> /* Returns a file descriptor that refers to a parent namespace */
>> -#define NS_GET_PARENT _IO(NSIO, 0x2)
>> +#define NS_GET_PARENT _IO(NSIO, 0x2)
>> /* Returns the type of namespace (CLONE_NEW* value) referred to by
>> file descriptor */
>> -#define NS_GET_NSTYPE _IO(NSIO, 0x3)
>> +#define NS_GET_NSTYPE _IO(NSIO, 0x3)
>> +/* Get creator UID for a user namespace */
>> +#define NS_GET_CREATOR_UID _IO(NSIO, 0x4)
>>
>> #endif /* __LINUX_NSFS_H */
>> --
>> 2.5.5
>>
>
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/
Andrei Vagin <[email protected]> writes:
> On Mon, Dec 19, 2016 at 03:38:35PM +0100, Michael Kerrisk (man-pages) wrote:
>> @@ -174,6 +175,11 @@ static long ns_ioctl(struct file *filp, unsigned int ioctl,
>> return open_related_ns(ns, ns->ops->get_parent);
>> case NS_GET_NSTYPE:
>> return ns->ops->type;
>> + case NS_GET_CREATOR_UID:
>> + if (ns->ops->type != CLONE_NEWUSER)
>> + return -EINVAL;
>> + user_ns = container_of(ns, struct user_namespace, ns);
>> + return from_kuid_munged(current_user_ns(), user_ns->owner);
>
> uid_t is "unsigned int", ioctl() returns long, so it may be hard to
> distinguish user id-s from errors on x32.
Very good point.
> off-topic: What is about user_ns->group? I can't find where it is
> used...
Over design. I put it in because I thought it might be useful. It turns
out it never was used so we can clean things up and remove it. The
group has never been exposed to userspace so no one will care.
Eric