11:34 p.m.
On 12/10/2013 01:53 AM, Serge Hallyn wrote:
Quoting Gao feng (gaofeng(a)cn.fujitsu.com):
> On 12/07/2013 06:10 AM, Serge E. Hallyn wrote:
>> Quoting Gao feng (gaofeng(a)cn.fujitsu.com):
>>> Since there is no more place for flags of clone system call.
>>> we need to find a way to create audit namespace.
>>>
>>> this patch add a new type of message AUDIT_CREATE_NS.
>>> user space can create new audit namespace through
>>> netlink.
>>>
>>> Right now, The privileged user in user namespace is allowed
>>> to create audit namespace. it means the unprivileged user can
>>> create an user namespace and then create audit namespace.
>>>
>>> Looks like it is not safe, but even the unprivileged user can
>>> create audit namespace, it can do no harm to the host. un-init
>>> audit namespace cann't effect the host.
>>>
>>> In the follow patches, the audit_backlog_limit will be per
>>> audit namesapace, but only the privileged user has rights to
>>> modify it. and the default value of audit_backlog_limit for
>>> uninit audit namespace will be set to 0.
>>>
>>> And the audit_rate_limit will be limited too.
>>>
>>> Signed-off-by: Gao feng <gaofeng(a)cn.fujitsu.com>
>>> ---
>>> include/linux/audit_namespace.h | 7 +++++++
>>> include/uapi/linux/audit.h | 1 +
>>> kernel/audit.c | 22 ++++++++++++++++++++++
>>> kernel/audit_namespace.c | 29 +++++++++++++++++++++++++++++
>>> 4 files changed, 59 insertions(+)
>>>
>>> diff --git a/include/linux/audit_namespace.h
b/include/linux/audit_namespace.h
>>> index 79a9b78..b17f052 100644
>>> --- a/include/linux/audit_namespace.h
>>> +++ b/include/linux/audit_namespace.h
>>> @@ -54,6 +54,8 @@ void put_audit_ns(struct audit_namespace *ns)
>>> rcu_read_unlock();
>>> }
>>> }
>>> +
>>> +extern int unshare_audit_namespace(void);
>>> #else
>>> static inline
>>> struct audit_namespace *get_audit_ns(struct audit_namespace *ns)
>>> @@ -66,6 +68,11 @@ void put_audit_ns(struct audit_namespace *ns)
>>> {
>>>
>>> }
>>> +
>>> +static inline int unshare_audit_namespace()
>>> +{
>>> + return -EINVAL;
>>> +}
>>> #endif
>>>
>>> static inline struct
>>> diff --git a/include/uapi/linux/audit.h b/include/uapi/linux/audit.h
>>> index 75cef3f..877d509 100644
>>> --- a/include/uapi/linux/audit.h
>>> +++ b/include/uapi/linux/audit.h
>>> @@ -68,6 +68,7 @@
>>> #define AUDIT_MAKE_EQUIV 1015 /* Append to watched tree */
>>> #define AUDIT_TTY_GET 1016 /* Get TTY auditing status */
>>> #define AUDIT_TTY_SET 1017 /* Set TTY auditing status */
>>> +#define AUDIT_CREATE_NS 1018 /* Create new audit namespace */
>>>
>>> #define AUDIT_FIRST_USER_MSG 1100 /* Userspace messages mostly uninteresting
to kernel */
>>> #define AUDIT_USER_AVC 1107 /* We filter this differently */
>>> diff --git a/kernel/audit.c b/kernel/audit.c
>>> index c4d4291..86212d3 100644
>>> --- a/kernel/audit.c
>>> +++ b/kernel/audit.c
>>> @@ -596,6 +596,12 @@ static int audit_netlink_ok(struct sk_buff *skb, u16
msg_type)
>>> !capable(CAP_AUDIT_CONTROL))
>>> err = -EPERM;
>>> break;
>>> + case AUDIT_CREATE_NS:
>>> + /* Allow privileged user in user namespace to
>>> + * create audit namespace */
>>> + if (!ns_capable(current_user_ns(), CAP_AUDIT_CONTROL))
>>> + err = -EPERM;
>>> + break;
>>> case AUDIT_USER:
>>> case AUDIT_FIRST_USER_MSG ... AUDIT_LAST_USER_MSG:
>>> case AUDIT_FIRST_USER_MSG2 ... AUDIT_LAST_USER_MSG2:
>>> @@ -735,6 +741,22 @@ static int audit_receive_msg(struct sk_buff *skb, struct
nlmsghdr *nlh)
>>> if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
>>> err = audit_set_backlog_limit(status_get->backlog_limit);
>>> break;
>>> + case AUDIT_CREATE_NS:
>>> + err = unshare_audit_namespace();
>>> +
>>> + if (audit_enabled == AUDIT_OFF)
>>> + break;
>>> +
>>> + ab = audit_log_start_ns(ns, NULL, GFP_KERNEL, AUDIT_CREATE_NS);
>>> + if (ab) {
>>> + audit_log_format(ab, "Create audit namespace");
>>> + audit_log_session_info(ab);
>>> + audit_log_task_context(ab);
>>> + audit_log_format(ab, "res=%d", err ? 0 : 1);
>>> + audit_log_end_ns(ns, ab);
>>> + }
>>> +
>>> + break;
>>> case AUDIT_USER:
>>> case AUDIT_FIRST_USER_MSG ... AUDIT_LAST_USER_MSG:
>>> case AUDIT_FIRST_USER_MSG2 ... AUDIT_LAST_USER_MSG2:
>>> diff --git a/kernel/audit_namespace.c b/kernel/audit_namespace.c
>>> index 6d9cb8f..28c608e 100644
>>> --- a/kernel/audit_namespace.c
>>> +++ b/kernel/audit_namespace.c
>>> @@ -6,3 +6,32 @@ struct audit_namespace init_audit_ns = {
>>> .user_ns = &init_user_ns,
>>> };
>>> EXPORT_SYMBOL_GPL(init_audit_ns);
>>> +
>>> +int unshare_audit_namespace(void)
>>> +{
>>> + struct task_struct *tsk = current;
>>> + struct audit_namespace *new_audit = NULL;
>>> + struct nsproxy *new_nsp;
>>> +
>>> + new_audit = kzalloc(sizeof(struct audit_namespace), GFP_KERNEL);
>>> + if (!new_audit)
>>> + return -ENOMEM;
>>> +
>>> + skb_queue_head_init(&new_audit->queue);
>>> + skb_queue_head_init(&new_audit->hold_queue);
>>> + init_waitqueue_head(&new_audit->kauditd_wait);
>>> + init_waitqueue_head(&new_audit->backlog_wait);
>>> +
>>> + new_nsp = create_new_namespaces(0, tsk, NULL, NULL);
>>> + if (IS_ERR(new_nsp)) {
>>> + kfree(new_audit);
>>> + return PTR_ERR(new_nsp);
>>> + }
>>> +
>>> + new_audit->user_ns = get_user_ns(current_user_ns());
>>> + new_nsp->audit_ns = get_audit_ns(new_audit);
>>> +
>>> + switch_task_namespaces(current, new_nsp);
>>
>> Do you need to drop the old audit->ns refcount?
>>
>
> task doesn't hold namespace's refcount directly. it hold ns's refcount
No but when you create the new_nsp with create_new_namespaces(),
that new_nsp bumps the refcount on the init_audit_ns. Then you
point new_nsp to a new audit_ns, but you never drop the refcount
on init_audit_ns.
Yes, I got it.
Thanks for your explantion, will fix this problem in next version.
Thanks!