Re: [RFC PATCH bpf-next v4 08/12] mm: memcontrol: Add BPF struct_ops for memory controller
From: bot+bpf-ci@kernel.org
Date: 2026-01-26 09:30:26
Also in:
bpf, cgroups, linux-kselftest, linux-mm, lkml
quoted hunk ↗ jump to hunk
diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h index f3b8c71870d8..1083be5d0362 100644 --- a/include/linux/memcontrol.h +++ b/include/linux/memcontrol.h
[ ... ]
+#else /* CONFIG_BPF_SYSCALL */
+
+static inline unsigned long
+bpf_memcg_get_high_delay(struct mem_cgroup *memcg) { return 0; }
+static inline bpf_memcg_below_low(struct mem_cgroup *memcg) { return false; }
+static inline bpf_memcg_below_min(struct mem_cgroup *memcg) { return false; }^^^^^^^^^^^^^^^^^^^^ These two stubs are missing the bool return type. This will cause a compilation error when CONFIG_BPF_SYSCALL is not set. Should these be: static inline bool bpf_memcg_below_low(...) static inline bool bpf_memcg_below_min(...)
+static inline void memcontrol_bpf_online(struct mem_cgroup *memcg) { }
+static inline void memcontrol_bpf_offline(struct mem_cgroup *memcg) { }
+
+#endif /* CONFIG_BPF_SYSCALL */[ ... ]
quoted hunk ↗ jump to hunk
diff --git a/mm/bpf_memcontrol.c b/mm/bpf_memcontrol.c index 716df49d7647..20c5c3552ce3 100644 --- a/mm/bpf_memcontrol.c +++ b/mm/bpf_memcontrol.c
[ ... ]
+static int bpf_memcg_ops_reg(void *kdata, struct bpf_link *link)
+{
+ struct bpf_struct_ops_link *ops_link
+ = container_of(link, struct bpf_struct_ops_link, link);
+ struct memcg_bpf_ops *ops = kdata;
+ struct mem_cgroup *memcg, *iter = NULL;
+ int err = 0;
+
+ memcg = mem_cgroup_get_from_ino(ops_link->cgroup_id);
+ if (IS_ERR_OR_NULL(memcg))
+ return PTR_ERR(memcg);
+
+ cgroup_lock();
+ while ((iter = mem_cgroup_iter(memcg, iter, NULL))) {
+ if (READ_ONCE(iter->bpf_ops)) {
+ mem_cgroup_iter_break(memcg, iter);
+ err = -EBUSY;
+ break;
+ }
+ WRITE_ONCE(iter->bpf_ops, ops);
+ }
+ if (err)
+ clean_memcg_bpf_ops(memcg, NULL); ^^^^
When registration fails because a descendant cgroup already has a BPF
program attached, this calls clean_memcg_bpf_ops() with NULL. Looking at
clean_memcg_bpf_ops():
while ((iter = mem_cgroup_iter(memcg, iter, NULL))) {
if (ops) {
if (!WARN_ON(READ_ONCE(iter->bpf_ops) != ops))
WRITE_ONCE(iter->bpf_ops, NULL);
} else
WRITE_ONCE(iter->bpf_ops, NULL); // unconditional clear
}
With ops=NULL, this will unconditionally clear bpf_ops for all
descendants in the hierarchy, including the cgroup that already had a
different BPF program attached (which caused the -EBUSY error). This
would incorrectly detach that existing program.
Should this be clean_memcg_bpf_ops(memcg, ops) instead, so it only clears
the ops that were written during this failed registration attempt?
Note: This issue appears to be addressed in the subsequent commit
5f3bb9a8c33f ("mm/bpf: Add BPF_F_ALLOW_OVERRIDE support for
memcg_bpf_ops") which completely rewrites the registration logic.
+ cgroup_unlock(); + + mem_cgroup_put(memcg); + return err; +}
--- 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 CI run summary: https://github.com/kernel-patches/bpf/actions/runs/21352241149