Merge tag 'cgroup-for-6.19-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup

6b63f90f · Linus Torvalds <[email protected]> · 2025-12-16 19:21:17 +12:00

Pull cgroup fix from Tejun Heo:

 - Fix a race condition in css_rstat_updated() where CMPXCHG without
   LOCK prefix could cause lnode corruption when the flusher runs
   concurrently on another CPU. The issue was introduced in 6.17 and
   causes memcg stats to become corrupted in production.

* tag 'cgroup-for-6.19-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
  cgroup: rstat: use LOCK CMPXCHG in css_rstat_updated

Commit: 6b63f90fa2afaa901e7edc9403014e46f0da1c69
Parents: 8f0b4cce · 3309b63a

Merge commit. Changes below are compared with the first parent: 8f0b4cce.

Changes

1 files changed · +8 −5

File	Status	Add	Del	Content
kernel/cgroup/rstat.c	modified	+8	−5	View file

kernel/cgroup/rstat.c

View file

diff --git a/kernel/cgroup/rstat.c b/kernel/cgroup/rstat.c
index a198e40..150e587 100644
--- a/kernel/cgroup/rstat.c
+++ b/kernel/cgroup/rstat.c
@@ -71,7 +71,6 @@ __bpf_kfunc void css_rstat_updated(struct cgroup_subsys_state *css, int cpu)
 {
 	struct llist_head *lhead;
 	struct css_rstat_cpu *rstatc;
-	struct css_rstat_cpu __percpu *rstatc_pcpu;
 	struct llist_node *self;
 
 	/*
@@ -104,18 +103,22 @@ __bpf_kfunc void css_rstat_updated(struct cgroup_subsys_state *css, int cpu)
 	/*
 	 * This function can be renentered by irqs and nmis for the same cgroup
 	 * and may try to insert the same per-cpu lnode into the llist. Note
-	 * that llist_add() does not protect against such scenarios.
+	 * that llist_add() does not protect against such scenarios. In addition
+	 * this same per-cpu lnode can be modified through init_llist_node()
+	 * from css_rstat_flush() running on a different CPU.
 	 *
 	 * To protect against such stacked contexts of irqs/nmis, we use the
 	 * fact that lnode points to itself when not on a list and then use
-	 * this_cpu_cmpxchg() to atomically set to NULL to select the winner
+	 * try_cmpxchg() to atomically set to NULL to select the winner
 	 * which will call llist_add(). The losers can assume the insertion is
 	 * successful and the winner will eventually add the per-cpu lnode to
 	 * the llist.
+	 *
+	 * Please note that we can not use this_cpu_cmpxchg() here as on some
+	 * archs it is not safe against modifications from multiple CPUs.
 	 */
 	self = &rstatc->lnode;
-	rstatc_pcpu = css->rstat_cpu;
-	if (this_cpu_cmpxchg(rstatc_pcpu->lnode.next, self, NULL) != self)
+	if (!try_cmpxchg(&rstatc->lnode.next, &self, NULL))
 		return;
 
 	lhead = ss_lhead_cpu(css->ss, cpu);