On Tue, Nov 28, 2023 at 03:11:06PM +0100, Michal Hocko wrote:

On Wed 22-11-23 16:11:55, Gregory Price wrote:
[...]

quoted

+ * Like get_vma_policy and get_task_policy, must hold alloc/task_lock
+ * while calling this.
+ */
+static struct mempolicy *get_task_vma_policy(struct task_struct *task,
+					     struct vm_area_struct *vma,
+					     unsigned long addr, int order,
+					     pgoff_t *ilx)

[...]

You should add lockdep annotation for alloc_lock/task_lock here for clarity and 
also...  

quoted

@@ -1844,16 +1899,7 @@ struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
 struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
 				 unsigned long addr, int order, pgoff_t *ilx)
 {
-	struct mempolicy *pol;
-
-	pol = __get_vma_policy(vma, addr, ilx);
-	if (!pol)
-		pol = get_task_policy(current);
-	if (pol->mode == MPOL_INTERLEAVE) {
-		*ilx += vma->vm_pgoff >> order;
-		*ilx += (addr - vma->vm_start) >> (PAGE_SHIFT + order);
-	}
-	return pol;
+	return get_task_vma_policy(current, vma, addr, order, ilx);

I do not think that all get_vma_policy take task_lock (just random check
dequeue_hugetlb_folio_vma->huge_node->get_vma_policy AFAICS)

Also I do not see policy_nodemask to be handled anywhere. That one is
used along with get_vma_policy (sometimes hidden like in
alloc_pages_mpol). It has a dependency on
cpuset_nodemask_valid_mems_allowed. That means that e.g. mbind on a
remote task would be constrained by current task cpuset when allocating
migration targets for the target task. I am wondering how many other
dependencies like that are lurking there.

So after further investigation, I'm going to have to back out the
changes that make home_node and mbind modifiable by an external task
and revisit it at a later time.

Right now, there's a very nasty rats nest of entanglement between
mempolicy and vma/shmem that hides a bunch of accesses to current.

It only becomes apparently when you start chasing all the callers of
mpol_dup, which had another silent access to current->cpusets.

mpol_dup calls the following:
	current_cpuset_is_being_rebound
	cpuset_mems_allowed(current)

So we would need to do the following
1) create mpol_dup_task and make current explicit, not implicit
2) chase down all callers to mpol_dup and make sure it isn't generated
   from any of the task interfaces
3) if it is generated from the task interfaces, plumb a reference to
   current down through... somehow... if possible...

Here's a ~1 hour chase that lead me to the conclusion that this will
take considerably more work, and is not to be taken lightly:

do_mbind
	mbind_range
		vma_modify_policy
			split_vma
				__split_vma
					vma_dup_policy
						mpol_dup
		vma_replace_policy
			mpol_dup
			vma->vm_ops->set_policy - see below

__set_mempolicy_home_node
	mbind_range
		... same as above ...

digging into vma->vm_ops->set_policy we end up in mm/shmem.c

shmem_set_policy
	mpol_set_shared_policy
		sp_alloc
			mpol_dup
				current_cpuset_is_being_rebound()
				cpuset_mems_allowed(current)

Who knows what else is burried in the vma stack, but making vma
mempolicies externally modifiable looks to be a much more monumental
task than just simply making the task policy modifiable.

For now i'm going to submit a V2 with home_node and mbind removed from
the proposal.  Those will take far more investigation.

This also means that process_set_mempolicy should not be extended to
allow for vma policy replacements.

~Gregory