On 2018-05-09 11:28, Paul Moore wrote:
> On Fri, May 4, 2018 at 4:54 PM, Richard Guy Briggs <rgb@xxxxxxxxxx> wrote:
> > Recognizing that the audit context is an internal audit value, use an
> > access function to retrieve the audit context pointer for the task
> > rather than reaching directly into the task struct to get it.
> >
> > Signed-off-by: Richard Guy Briggs <rgb@xxxxxxxxxx>
> > ---
> > include/linux/audit.h | 16 ++++++++---
> > include/net/xfrm.h | 2 +-
> > kernel/audit.c | 4 +--
> > kernel/audit_watch.c | 2 +-
> > kernel/auditsc.c | 52 ++++++++++++++++++------------------
> > net/bridge/netfilter/ebtables.c | 2 +-
> > net/core/dev.c | 2 +-
> > net/netfilter/x_tables.c | 2 +-
> > net/netlabel/netlabel_user.c | 2 +-
> > security/integrity/ima/ima_api.c | 2 +-
> > security/integrity/integrity_audit.c | 2 +-
> > security/lsm_audit.c | 2 +-
> > security/selinux/hooks.c | 4 +--
> > security/selinux/selinuxfs.c | 6 ++---
> > security/selinux/ss/services.c | 12 ++++-----
> > 15 files changed, 60 insertions(+), 52 deletions(-)
> >
> > diff --git a/include/linux/audit.h b/include/linux/audit.h
> > index 5f86f7c..93e4c61 100644
> > --- a/include/linux/audit.h
> > +++ b/include/linux/audit.h
> > @@ -235,26 +235,30 @@ extern void __audit_inode_child(struct inode *parent,
> > extern void __audit_seccomp(unsigned long syscall, long signr, int code);
> > extern void __audit_ptrace(struct task_struct *t);
> >
> > +static inline struct audit_context *audit_context(struct task_struct *task)
> > +{
> > + return task->audit_context;
> > +}
>
> Another case where I think I agree with everything here on principle,
> especially when one considers it in the larger context of the audit
> container ID work. However, I think we might be able to somply this a
> bit by eliminating the parameter to the new audit_context() helper and
> making it always reference the current task_struct. Based on this
> patch it would appear that this change would work for all callers
> except for audit_take_context() and __audit_syscall_entry(), both of
> which are contained within the core audit code and are enough of a
> special case that I think it is acceptable for them to access the
> context directly. I'm trying to think of reasons why a non-audit
> kernel subsystem would ever need to access the audit context of a
> process other than current and I can't think of any ... removing the
> task_struct pointer might help prevent mistakes/abuse in the future.
As for __audit_syscall_{entry,exit}() and audit_signal_info(), they are
using current. current is assigned to local variable tsk only to be
used as the LHS in assignments and for locking.
But, audit_take_context() and audit_log_exit() are both called also from
__audit_free() which can have non-current handed to it by copy_process()
cleaning up, while do_exit() appears to still be in current.
So, Ok, ditch the parameter to audit_context() and use local access when
needed.
> > diff --git a/kernel/auditsc.c b/kernel/auditsc.c
> > index 6e3ceb9..a4bbdcc 100644
> > --- a/kernel/auditsc.c
> > +++ b/kernel/auditsc.c
> > @@ -836,7 +836,7 @@ static inline struct audit_context *audit_take_context(struct task_struct *tsk,
> > int return_valid,
> > long return_code)
> > {
> > - struct audit_context *context = tsk->audit_context;
> > + struct audit_context *context = audit_context(tsk);
> >
> > if (!context)
> > return NULL;
> > @@ -1510,7 +1510,7 @@ void __audit_syscall_entry(int major, unsigned long a1, unsigned long a2,
> > unsigned long a3, unsigned long a4)
> > {
> > struct task_struct *tsk = current;
> > - struct audit_context *context = tsk->audit_context;
> > + struct audit_context *context = audit_context(tsk);
> > enum audit_state state;
> >
> > if (!audit_enabled || !context)
>
> --
> paul moore
> www.paul-moore.com
- RGB
--
Richard Guy Briggs <rgb@xxxxxxxxxx>
Sr. S/W Engineer, Kernel Security, Base Operating Systems
Remote, Ottawa, Red Hat Canada
IRC: rgb, SunRaycer
Voice: +1.647.777.2635, Internal: (81) 32635
