```
commit 264db34cd79e7c07b4b1e58e2efae3d632df9cb2
Author: Roman Gushchin <roman.gushchin@linux.dev>

sched: psi: implement bpf_psi struct ops

This patch implements a BPF struct ops-based mechanism to create PSI
triggers, attach them to cgroups or system wide and handle PSI events
in BPF.

> diff --git a/kernel/sched/bpf_psi.c b/kernel/sched/bpf_psi.c
> new file mode 100644
> index 000000000..c383a2011
> --- /dev/null
> +++ b/kernel/sched/bpf_psi.c

[ ... ]

> +void bpf_psi_cgroup_offline(struct cgroup *cgroup)
> +{
> +	struct psi_group *group = cgroup->psi;
> +	u64 cgrp_id = cgroup_id(cgroup);
> +	struct psi_trigger *t, *p;
> +	struct bpf_psi *bpf_psi;
> +	LIST_HEAD(to_destroy);
> +	int idx;
> +
> +	if (!group)
> +		return;
> +
> +	spin_lock(&group->bpf_triggers_lock);
> +	list_for_each_entry_safe(t, p, &group->bpf_triggers, bpf_group_node) {
> +		if (!t->pinned) {
> +			t->pinned = true;
> +			list_move(&t->bpf_group_node, &to_destroy);
> +		}
> +	}
> +	spin_unlock(&group->bpf_triggers_lock);
> +
> +	list_for_each_entry_safe(t, p, &to_destroy, bpf_group_node) {
> +		bpf_psi = READ_ONCE(t->bpf_psi);
> +
> +		idx = srcu_read_lock(&bpf_psi->srcu);
> +		if (bpf_psi->ops->handle_cgroup_offline)
> +			bpf_psi->ops->handle_cgroup_offline(bpf_psi, cgrp_id);
> +		srcu_read_unlock(&bpf_psi->srcu, idx);
> +
> +		spin_lock(&bpf_psi->lock);
> +		list_del(&t->bpf_psi_node);
> +		spin_unlock(&bpf_psi->lock);
> +
> +		WRITE_ONCE(t->bpf_psi, NULL);
> +		flush_workqueue(bpf_psi_wq);
> +		synchronize_srcu(&bpf_psi->srcu);
> +		psi_trigger_destroy(t);
> +	}
> +}

Can bpf_psi_cgroup_offline() race with bpf_psi_ops_unreg() and access
freed memory?

Consider this sequence:

Thread A (bpf_psi_cgroup_offline):
  - Pins triggers under group->bpf_triggers_lock
  - Reads bpf_psi = READ_ONCE(t->bpf_psi) outside any lock

Thread B (bpf_psi_ops_unreg):
  - Iterates bpf_psi->triggers, skips pinned triggers
  - Calls synchronize_srcu(&bpf_psi->srcu)
  - Calls kfree(bpf_psi)

Thread A:
  - Tries srcu_read_lock(&bpf_psi->srcu) on freed memory

The pinning mechanism prevents Thread B from NULLing t->bpf_psi when
Thread A has pinned the trigger. However, Thread B unconditionally
frees the bpf_psi object itself even when it didn't process any
triggers because they were all pinned by Thread A.

After Thread A pins all triggers and releases group->bpf_triggers_lock,
Thread B can complete bpf_psi_ops_unreg(), call synchronize_srcu() (no
readers yet), and free bpf_psi. Then Thread A accesses the freed
bpf_psi via srcu_read_lock() and spin_lock().

> +static void bpf_psi_ops_unreg(void *kdata, struct bpf_link *link)
> +{
> +	struct bpf_psi_ops *ops = kdata;
> +	struct bpf_psi *bpf_psi = ops->bpf_psi;
> +	struct psi_trigger *t, *p;
> +	LIST_HEAD(to_destroy);
> +
> +	spin_lock(&bpf_psi->lock);
> +	list_for_each_entry_safe(t, p, &bpf_psi->triggers, bpf_psi_node) {
> +		spin_lock(&t->group->bpf_triggers_lock);
> +		if (!t->pinned) {
> +			t->pinned = true;
> +			list_move(&t->bpf_group_node, &to_destroy);
> +			list_del(&t->bpf_psi_node);
> +
> +			WRITE_ONCE(t->bpf_psi, NULL);
> +		}
> +		spin_unlock(&t->group->bpf_triggers_lock);
> +	}
> +	spin_unlock(&bpf_psi->lock);
> +
> +	flush_workqueue(bpf_psi_wq);
> +	synchronize_srcu(&bpf_psi->srcu);
> +
> +	list_for_each_entry_safe(t, p, &to_destroy, bpf_group_node)
> +		psi_trigger_destroy(t);
> +
> +	if (!list_empty(&bpf_psi->node)) {
> +		mutex_lock(&bpf_psi_lock);
> +		list_del(&bpf_psi->node);
> +		mutex_unlock(&bpf_psi_lock);
> +		static_branch_dec(&bpf_psi_notify_key);
> +	}
> +
> +	cleanup_srcu_struct(&bpf_psi->srcu);
> +	kfree(bpf_psi);
> +}

When bpf_psi_ops_unreg() skips all triggers because they're pinned
(by bpf_psi_cgroup_offline()), does it still free bpf_psi too early?

The to_destroy list will be empty if all triggers were pinned, but
the function still proceeds to cleanup_srcu_struct() and kfree(bpf_psi)
while bpf_psi_cgroup_offline() may be using that bpf_psi pointer.


```

---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

In-Reply-To-Subject: `sched: psi: implement bpf_psi struct ops`
CI run summary: https://github.com/kernel-patches/bpf/actions/runs/18859027430