Re: [PATCH ghak90 V7 04/21] audit: convert to contid list to check for orch/engine ownership

On 2019-10-10 20:38, Paul Moore wrote: > On Wed, Sep 18, 2019 at 9:24 PM Richard Guy Briggs <rgb(a)redhat.com&gt; wrote: > > Store the audit container identifier in a refcounted kernel object that > > is added to the master list of audit container identifiers. This will > > allow multiple container orchestrators/engines to work on the same > > machine without danger of inadvertantly re-using an existing identifier. > > It will also allow an orchestrator to inject a process into an existing > > container by checking if the original container owner is the one > > injecting the task. A hash table list is used to optimize searches. > > > > Signed-off-by: Richard Guy Briggs <rgb(a)redhat.com&gt; > > --- > > include/linux/audit.h | 26 ++++++++++++++-- > > kernel/audit.c | 86 ++++++++++++++++++++++++++++++++++++++++++++++++--- > > kernel/audit.h | 8 +++++ > > 3 files changed, 112 insertions(+), 8 deletions(-) > > One general comment before we go off into the weeds on this ... I can > understand why you wanted to keep this patch separate from the earlier > patches, but as we get closer to having mergeable code this should get > folded into the previous patches. For example, there shouldn't be a > change in audit_task_info where you change the contid field from a u64 > to struct pointer, it should be a struct pointer from the start. I should have marked this patchset as RFC even though it was v7 due to a lot of new ideas/code that was added with uncertainties needing comment and direction. > It's also disappointing that idr appears to only be for 32-bit ID > values, if we had a 64-bit idr I think we could simplify this greatly. Perhaps. I do still see value in letting the orchestrator choose the value.

> > diff --git a/include/linux/audit.h b/include/linux/audit.h > > index f2e3b81f2942..e317807cdd3e 100644 > > --- a/include/linux/audit.h > > +++ b/include/linux/audit.h > > @@ -95,10 +95,18 @@ struct audit_ntp_data { > > struct audit_ntp_data {}; > > #endif > > > > +struct audit_cont { > > + struct list_head list; > > + u64 id; > > + struct task_struct *owner; > > + refcount_t refcount; > > + struct rcu_head rcu; > > +}; > > It seems as though in most of the code you are using "contid", any > reason why didn't stick with that naming scheme here, e.g. "struct > audit_contid"? I was using contid to refer to the value itself and cont to refer to the refcounted object. I find cont a bit too terse, so I'm still thinking of changing it. Perhaps contobj?

> > @@ -203,11 +211,15 @@ static inline unsigned int audit_get_sessionid(struct task_struct *tsk) > > > > static inline u64 audit_get_contid(struct task_struct *tsk) > > { > > - if (!tsk->audit) > > + if (!tsk->audit || !tsk->audit->cont) > > return AUDIT_CID_UNSET; > > - return tsk->audit->contid; > > + return tsk->audit->cont->id; > > } > > Assuming for a moment that we implement an audit_contid_get() (see > Neil's comment as well as mine below), we probably need to name this > something different so we don't all lose our minds when we read this > code. On the plus side we can probably preface it with an underscore > since it is a static, in which case _audit_contid_get() might be okay, > but I'm open to suggestions. I'm fine with the "_" prefix, can you point to precedent or convention?

> > @@ -231,7 +235,9 @@ int audit_alloc(struct task_struct *tsk) > > } > > info->loginuid = audit_get_loginuid(current); > > info->sessionid = audit_get_sessionid(current); > > - info->contid = audit_get_contid(current); > > + info->cont = audit_cont(current); > > + if (info->cont) > > + refcount_inc(&info->cont->refcount); > > See the other comments about a "get" function, but I think we need a > RCU read lock around the above, no? The rcu read lock is to protect the list rather than the cont object itself, the latter of which is protected by its refcount.

> > @@ -2397,8 +2438,43 @@ int audit_set_contid(struct task_struct *task, u64 contid) > > else if (audit_contid_set(task)) > > rc = -ECHILD; > > read_unlock(&tasklist_lock); > > - if (!rc) > > - task->audit->contid = contid; > > + if (!rc) { > > + struct audit_cont *oldcont = audit_cont(task); > > Previously we held the tasklist_lock to protect the audit container ID > associated with the struct, should we still be holding it here? We held the tasklist_lock to protect access to the target task's child/parent/thread relationships.

> Regardless, I worry that the lock dependencies between the > tasklist_lock and the audit_contid_list_lock are going to be tricky. > It might be nice to document the relationship in a comment up near > where you declare audit_contid_list_lock. I don't think there should be a conflict between the two. The contid_list_lock doesn't care if the cont object is associated to a particular task.

> > + struct audit_cont *cont = NULL; > > + struct audit_cont *newcont = NULL; > > + int h = audit_hash_contid(contid); > > + > > + spin_lock(&audit_contid_list_lock); > > + list_for_each_entry_rcu(cont, &audit_contid_hash[h], list) > > + if (cont->id == contid) { > > + /* task injection to existing container */ > > + if (current == cont->owner) { > > I understand the desire to limit a given audit container ID to the > orchestrator that created it, but are we certain that we can track > audit container ID "ownership" via a single instance of a task_struct? Are you suggesting that a task_struct representing a task may be replaced for a specific task? I don't believe that will ever happen. > What happens when the orchestrator stops/restarts/crashes? Do we > even care? Reap all of its containers?