bpf: Free special fields when update hash and local storage maps

As [lru_,]percpu_hash maps support BPF_KPTR_{REF,PERCPU}, missing calls to 'bpf_obj_free_fields()' in 'pcpu_copy_value()' could cause the memory referenced by BPF_KPTR_{REF,PERCPU} fields to be held until the map gets freed.

Fix this by calling 'bpf_obj_free_fields()' after 'copy_map_value[,_long]()' in 'pcpu_copy_value()'.

Fixes 65334e64a493 ("bpf: Support kptrs in percpu hashmap and percpu LRU hashmap")

Signed-off-by Leon Hwang <[email protected]>

---
 kernel/bpf/hashtab.c | 2 ++
 1 file changed, 2 insertions(+)

kernel/bpf/hashtab.c +0 -0

index c2fcd0cd51e51..26308adc9ccb3 100644

kernel/bpf/hashtab.c +1 -1

        } else {
                u32 size = round_up(htab->map.value_size, 8);
                int off = 0, cpu;
 
                for_each_possible_cpu(cpu) {
                        copy_map_value_long(&htab->map, per_cpu_ptr(pptr, cpu), value + off);
+                       bpf_obj_free_fields(htab->map.record, per_cpu_ptr(pptr, cpu));
                        off += size;
                }
        }
--

2.51.0

When updating local storage maps with BPF_F_LOCK on the fast path, the special fields were not freed after being replaced. This could cause memory referenced by BPF_KPTR_{REF,PERCPU} fields to be held until the map gets freed.

Similarly, on the other path, the old sdata's special fields were never freed regardless of whether BPF_F_LOCK was used, causing the same issue.

Fix this by calling 'bpf_obj_free_fields()' after 'copy_map_value_locked()' to properly release the old fields.

Fixes 9db44fdd8105 ("bpf: Support kptrs in local storage maps")

Signed-off-by Leon Hwang <[email protected]>

---
 kernel/bpf/bpf_local_storage.c | 3 +++
 1 file changed, 3 insertions(+)

kernel/bpf/bpf_local_storage.c +1 -0

index b931fbceb54da..8e3aea4e07c50 100644
@@ -609,6 +609,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
                if (old_sdata && selem_linked_to_storage_lockless(SELEM(old_sdata))) {
                        copy_map_value_locked(&smap->map, old_sdata->data,
                                              value, false);
+                       bpf_obj_free_fields(smap->map.record, old_sdata->data);

kernel/bpf/bpf_local_storage.c +0 -1

                bpf_selem_unlink_map(SELEM(old_sdata));
                bpf_selem_unlink_storage_nolock(local_storage, SELEM(old_sdata),
                                                true, &old_selem_free_list);
--

2.51.0

Quoted 27 lines · Leon Hwang

When updating local storage maps with BPF_F_LOCK on the fast path, the special fields were not freed after being replaced. This could cause memory referenced by BPF_KPTR_{REF,PERCPU} fields to be held until the map gets freed.

Similarly, on the other path, the old sdata's special fields were never freed regardless of whether BPF_F_LOCK was used, causing the same issue.

Fix this by calling 'bpf_obj_free_fields()' after 'copy_map_value_locked()' to properly release the old fields.

Fixes 9db44fdd8105 ("bpf: Support kptrs in local storage maps")

Signed-off-by Leon Hwang <[email protected]>

---
 kernel/bpf/bpf_local_storage.c | 3 +++
 1 file changed, 3 insertions(+)

kernel/bpf/bpf_local_storage.c +1 -0

index b931fbceb54da..8e3aea4e07c50 100644
@@ -609,6 +609,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
                if (old_sdata && selem_linked_to_storage_lockless(SELEM(old_sdata))) {
                        copy_map_value_locked(&smap->map, old_sdata->data,
                                              value, false);
+                       bpf_obj_free_fields(smap->map.record, old_sdata->data);

[ ... ]

kernel/bpf/bpf_local_storage.c bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap, · line 645

Quoted 8 lines

                        return old_sdata;
                }
        }

kernel/bpf/bpf_local_storage.c +1 -0

@@ -641,6 +642,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
        if (old_sdata && (map_flags & BPF_F_LOCK)) {
                copy_map_value_locked(&smap->map, old_sdata->data, value,
                                      false);
+               bpf_obj_free_fields(smap->map.record, old_sdata->data);

The one above and this make sense. Thanks for fixing it.

Quoted 5 lines

kernel/bpf/bpf_local_storage.c +0 -0

                selem = SELEM(old_sdata);
                goto unlock;
        }

@@ -654,6 +656,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
 

kernel/bpf/bpf_local_storage.c bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap, · line 659

Quoted 3 lines

kernel/bpf/bpf_local_storage.c +1 -0

        /* Third, remove old selem, SELEM(old_sdata) */
        if (old_sdata) {
+               bpf_obj_free_fields(smap->map.record, old_sdata->data);

Is this really needed? bpf_selem_free_list() later should free special fields in this selem.

Hi Amery,

Quoted 31 lines · Amery Hung

Quoted 27 lines · Leon Hwang

When updating local storage maps with BPF_F_LOCK on the fast path, the special fields were not freed after being replaced. This could cause memory referenced by BPF_KPTR_{REF,PERCPU} fields to be held until the map gets freed.

Similarly, on the other path, the old sdata's special fields were never freed regardless of whether BPF_F_LOCK was used, causing the same issue.

Fix this by calling 'bpf_obj_free_fields()' after 'copy_map_value_locked()' to properly release the old fields.

Fixes 9db44fdd8105 ("bpf: Support kptrs in local storage maps")

Signed-off-by Leon Hwang <[email protected]>

---
 kernel/bpf/bpf_local_storage.c | 3 +++
 1 file changed, 3 insertions(+)

kernel/bpf/bpf_local_storage.c +1 -0

index b931fbceb54da..8e3aea4e07c50 100644
@@ -609,6 +609,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
                if (old_sdata && selem_linked_to_storage_lockless(SELEM(old_sdata))) {
                        copy_map_value_locked(&smap->map, old_sdata->data,
                                              value, false);
+                       bpf_obj_free_fields(smap->map.record, old_sdata->data);

[ ... ]

Quoted 11 lines

kernel/bpf/bpf_local_storage.c bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap, · line 645

Quoted 8 lines

                        return old_sdata;
                }
        }

kernel/bpf/bpf_local_storage.c +1 -0

@@ -641,6 +642,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
        if (old_sdata && (map_flags & BPF_F_LOCK)) {
                copy_map_value_locked(&smap->map, old_sdata->data, value,
                                      false);
+               bpf_obj_free_fields(smap->map.record, old_sdata->data);

The one above and this make sense. Thanks for fixing it.

Thanks for your review.

Quoted 5 lines

Quoted 5 lines

kernel/bpf/bpf_local_storage.c +0 -0

                selem = SELEM(old_sdata);
                goto unlock;
        }

@@ -654,6 +656,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
 

Quoted 7 lines

kernel/bpf/bpf_local_storage.c bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap, · line 659

Quoted 3 lines

kernel/bpf/bpf_local_storage.c +1 -0

        /* Third, remove old selem, SELEM(old_sdata) */
        if (old_sdata) {
+               bpf_obj_free_fields(smap->map.record, old_sdata->data);

Is this really needed? bpf_selem_free_list() later should free special fields in this selem.

Yes, it’s needed. The new selftest confirms that the special fields are not freed when updating a local storage map.

Also, bpf_selem_unlink_storage_nolock() doesn’t invoke bpf_selem_free_list(), unlike bpf_selem_unlink_storage(). So we need to call bpf_obj_free_fields() here explicitly to free those fields.

Thanks,
Leon

[...]

Quoted 5 lines · Leon Hwang

Hi Amery,

Quoted 1 line · Amery Hung

On Sun, Oct 26, 2025 at 8:41 AM Leon Hwang <[email protected]> wrote:

Quoted 13 lines

Quoted 5 lines

Quoted 5 lines

                selem = SELEM(old_sdata);
                goto unlock;
        }

(quoted diff) +0 -0

@@ -654,6 +656,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
 

Quoted 7 lines

kernel/bpf/bpf_local_storage.c bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap, · line 659

Quoted 3 lines

kernel/bpf/bpf_local_storage.c +1 -0

        /* Third, remove old selem, SELEM(old_sdata) */
        if (old_sdata) {
+               bpf_obj_free_fields(smap->map.record, old_sdata->data);

Is this really needed? bpf_selem_free_list() later should free special fields in this selem.

Quoted 3 lines

Yes, it’s needed. The new selftest confirms that the special fields are not freed when updating a local storage map.

Hmmm. I don't think so.

Quoted 4 lines

Also, bpf_selem_unlink_storage_nolock() doesn’t invoke bpf_selem_free_list(), unlike bpf_selem_unlink_storage(). So we need to call bpf_obj_free_fields() here explicitly to free those fields.

bpf_selem_unlink_storage_nolock() unlinks the old selem and adds it to old_selem_free_list. Later, bpf_selem_free_list() will call bpf_selem_free() to free selem in bpf_selem_free_list, which should also free special fields in the selem.

The selftests may have checked the refcount before an task trace RCU gp and thought it is a leak. I added a 300ms delay before the checking program runs and the test did not detect any leak even without this specific bpf_obj_free_fields().

Quoted 6 lines · Amery Hung

Quoted 5 lines · Leon Hwang

Hi Amery,

Quoted 1 line · Amery Hung

On Sun, Oct 26, 2025 at 8:41 AM Leon Hwang <[email protected]> wrote:

[...]

Quoted 29 lines

Quoted 13 lines

Quoted 5 lines

Quoted 5 lines

                selem = SELEM(old_sdata);
                goto unlock;
        }

(quoted diff) +0 -0

@@ -654,6 +656,7 @@ bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
 

Quoted 7 lines

kernel/bpf/bpf_local_storage.c bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap, · line 659

Quoted 3 lines

kernel/bpf/bpf_local_storage.c +1 -0

        /* Third, remove old selem, SELEM(old_sdata) */
        if (old_sdata) {
+               bpf_obj_free_fields(smap->map.record, old_sdata->data);

Is this really needed? bpf_selem_free_list() later should free special fields in this selem.

Quoted 3 lines

Yes, it’s needed. The new selftest confirms that the special fields are not freed when updating a local storage map.

Hmmm. I don't think so.

Quoted 4 lines

Also, bpf_selem_unlink_storage_nolock() doesn’t invoke bpf_selem_free_list(), unlike bpf_selem_unlink_storage(). So we need to call bpf_obj_free_fields() here explicitly to free those fields.

bpf_selem_unlink_storage_nolock() unlinks the old selem and adds it to old_selem_free_list. Later, bpf_selem_free_list() will call bpf_selem_free() to free selem in bpf_selem_free_list, which should also free special fields in the selem.

Quoted 4 lines

The selftests may have checked the refcount before an task trace RCU gp and thought it is a leak. I added a 300ms delay before the checking program runs and the test did not detect any leak even without this specific bpf_obj_free_fields().

Yeah, you're right. Thanks for the clear explanation.

I also verified it by adding a 300ms delay.

So this bpf_obj_free_fields() call isn't needed — I'll drop it in the next revision.

Thanks,
Leon

Add tests to verify that updating hash and local storage maps decrements refcount when BPF_KPTR_REF objects are involved.

The tests perform the following steps:

1. Call update_elem() to insert an initial value.
2. Use bpf_refcount_acquire() to increment the refcount.
3. Store the node pointer in the map value.
4. Add the node to a linked list.
5. Probe-read the refcount and verify it is *2*.
6. Call update_elem() again to trigger refcount decrement.
7. Probe-read the refcount and verify it is *1*.

Signed-off-by Leon Hwang <[email protected]>

---
 .../bpf/prog_tests/refcounted_kptr.c          | 178 +++++++++++++++++-
 .../selftests/bpf/progs/refcounted_kptr.c     | 160 ++++++++++++++++
 2 files changed, 337 insertions(+), 1 deletion(-)

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +22 -1

index d6bd5e16e6372..0a60330a1f4b3 100644
@@ -3,7 +3,7 @@
 
 #include <test_progs.h>
 #include <network_helpers.h>
-
+#include "cgroup_helpers.h"
 #include "refcounted_kptr.skel.h"
 #include "refcounted_kptr_fail.skel.h"
 

@@ -44,3 +44,179 @@ void test_refcounted_kptr_wrong_owner(void)
        ASSERT_OK(opts.retval, "rbtree_wrong_owner_remove_fail_a2 retval");
        refcounted_kptr__destroy(skel);
 }
+
+static void test_refcnt_leak(void *values, size_t values_sz, u64 flags, struct bpf_map *map,
+                            struct bpf_program *prog_leak, struct bpf_program *prog_check)
+{
+       int ret, fd, key = 0;
+       LIBBPF_OPTS(bpf_test_run_opts, opts,
+                   .data_in = &pkt_v4,
+                   .data_size_in = sizeof(pkt_v4),
+                   .repeat = 1,
+       );
+
+       ret = bpf_map__update_elem(map, &key, sizeof(key), values, values_sz, flags);
+       if (!ASSERT_OK(ret, "bpf_map__update_elem init"))
+               return;
+
+       fd = bpf_program__fd(prog_leak);
+       ret = bpf_prog_test_run_opts(fd, &opts);
+       if (!ASSERT_OK(ret, "test_run_opts"))
+               return;
+       if (!ASSERT_EQ(opts.retval, 2, "retval refcount"))
+               return;

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +113 -0

+
+static void test_percpu_hash_refcount_leak(void)
+{
+       struct refcounted_kptr *skel;
+       size_t values_sz;
+       u64 *values;
+       int cpu_nr;
+
+       cpu_nr = libbpf_num_possible_cpus();
+       if (!ASSERT_GT(cpu_nr, 0, "libbpf_num_possible_cpus"))
+               return;
+
+       values = calloc(cpu_nr, sizeof(u64));
+       if (!ASSERT_OK_PTR(values, "calloc values"))
+               return;
+
+       skel = refcounted_kptr__open_and_load();
+       if (!ASSERT_OK_PTR(skel, "refcounted_kptr__open_and_load")) {
+               free(values);
+               return;
+       }
+
+       values_sz = cpu_nr * sizeof(u64);
+       memset(values, 0, values_sz);
+
+       test_refcnt_leak(values, values_sz, 0, skel->maps.pcpu_hash,
+                        skel->progs.pcpu_hash_refcount_leak,
+                        skel->progs.check_pcpu_hash_refcount);
+
+       refcounted_kptr__destroy(skel);
+       free(values);
+}
+
+struct lock_map_value {
+       u64 kptr;
+       struct bpf_spin_lock lock;
+       int value;
+};
+
+static void test_hash_lock_refcount_leak(void)
+{
+       struct lock_map_value value = {};
+       struct refcounted_kptr *skel;
+
+       skel = refcounted_kptr__open_and_load();
+       if (!ASSERT_OK_PTR(skel, "refcounted_kptr__open_and_load"))
+               return;
+
+       test_refcnt_leak(&value, sizeof(value), BPF_F_LOCK, skel->maps.lock_hash,
+                        skel->progs.hash_lock_refcount_leak,
+                        skel->progs.check_hash_lock_refcount);
+
+       refcounted_kptr__destroy(skel);
+}
+
+static void test_cgrp_storage_refcount_leak(u64 flags)
+{
+       int server_fd = -1, client_fd = -1;
+       struct lock_map_value value = {};
+       struct refcounted_kptr *skel;
+       struct bpf_link *link;
+       struct bpf_map *map;
+       int cgroup, err;
+
+       cgroup = test__join_cgroup("/cg_refcount_leak");
+       if (!ASSERT_GE(cgroup, 0, "test__join_cgroup"))
+               return;
+
+       skel = refcounted_kptr__open_and_load();
+       if (!ASSERT_OK_PTR(skel, "refcounted_kptr__open_and_load"))
+               goto out;
+
+       link = bpf_program__attach_cgroup(skel->progs.cgroup_storage_refcount_leak, cgroup);
+       if (!ASSERT_OK_PTR(link, "bpf_program__attach_cgroup"))
+               goto out;
+       skel->links.cgroup_storage_refcount_leak = link;
+
+       server_fd = start_server(AF_INET6, SOCK_STREAM, "::1", 0, 0);
+       if (!ASSERT_GE(server_fd, 0, "start_server"))
+               goto out;
+
+       client_fd = connect_to_fd(server_fd, 0);
+       if (!ASSERT_GE(client_fd, 0, "connect_to_fd"))
+               goto out;
+
+       map = skel->maps.cgrp_strg;
+       err = bpf_map__lookup_elem(map, &cgroup, sizeof(cgroup), &value, sizeof(value), flags);
+       if (!ASSERT_OK(err, "bpf_map__lookup_elem"))
+               goto out;
+
+       ASSERT_EQ(value.value, 2, "refcount");
+
+       err = bpf_map__update_elem(map, &cgroup, sizeof(cgroup), &value, sizeof(value), flags);
+       if (!ASSERT_OK(err, "bpf_map__update_elem"))
+               goto out;
+
+       err = bpf_link__detach(skel->links.cgroup_storage_refcount_leak);
+       if (!ASSERT_OK(err, "bpf_link__detach"))
+               goto out;
+
+       link = bpf_program__attach(skel->progs.check_cgroup_storage_refcount);
+       if (!ASSERT_OK_PTR(link, "bpf_program__attach"))
+               goto out;
+       skel->links.check_cgroup_storage_refcount = link;
+
+       close(client_fd);
+       client_fd = connect_to_fd(server_fd, 0);
+       if (!ASSERT_GE(client_fd, 0, "connect_to_fd"))
+               goto out;
+
+       err = bpf_map__lookup_elem(map, &cgroup, sizeof(cgroup), &value, sizeof(value), flags);
+       if (!ASSERT_OK(err, "bpf_map__lookup_elem"))
+               goto out;

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +22 -0

+
+static void test_cgroup_storage_refcount_leak(void)
+{
+       test_cgrp_storage_refcount_leak(0);
+}
+
+static void test_cgroup_storage_lock_refcount_leak(void)
+{
+       test_cgrp_storage_refcount_leak(BPF_F_LOCK);
+}
+
+void test_kptr_refcount_leak(void)
+{
+       if (test__start_subtest("percpu_hash_refcount_leak"))
+               test_percpu_hash_refcount_leak();
+       if (test__start_subtest("hash_lock_refcount_leak"))
+               test_hash_lock_refcount_leak();
+       if (test__start_subtest("cgroup_storage_refcount_leak"))
+               test_cgroup_storage_refcount_leak();
+       if (test__start_subtest("cgroup_storage_lock_refcount_leak"))
+               test_cgroup_storage_lock_refcount_leak();
+}

tools/testing/selftests/bpf/progs/refcounted_kptr.c +128 -0

index 893a4fdb4b6e9..09efae9537c9b 100644
@@ -7,6 +7,7 @@
 #include <bpf/bpf_core_read.h>
 #include "bpf_misc.h"
 #include "bpf_experimental.h"
+#include "bpf_tracing_net.h"
 
 extern void bpf_rcu_read_lock(void) __ksym;
 extern void bpf_rcu_read_unlock(void) __ksym;

@@ -568,4 +569,163 @@ int BPF_PROG(rbtree_sleepable_rcu_no_explicit_rcu_lock,
        return 0;
 }
 
+private(leak) u64 ref;
+
+static u32 probe_read_refcount(void)
+{
+       u32 refcnt;
+
+       bpf_probe_read_kernel(&refcnt, sizeof(refcnt), (void *) ref);
+       return refcnt;
+}
+
+static int __insert_in_list(struct bpf_list_head *head, struct bpf_spin_lock *lock,
+                           struct node_data __kptr **node)
+{
+       struct node_data *n, *m;
+
+       n = bpf_obj_new(typeof(*n));
+       if (!n)
+               return -1;
+
+       m = bpf_refcount_acquire(n);
+       n = bpf_kptr_xchg(node, n);
+       if (n) {
+               bpf_obj_drop(n);
+               bpf_obj_drop(m);
+               return -2;
+       }
+
+       bpf_spin_lock(lock);
+       bpf_list_push_front(head, &m->l);
+       ref = (u64)(void *) &m->ref;
+       bpf_spin_unlock(lock);
+       return probe_read_refcount();
+}
+
+static void *__lookup_map(void *map)
+{
+       int key = 0;
+
+       return bpf_map_lookup_elem(map, &key);
+}
+
+struct {
+       __uint(type, BPF_MAP_TYPE_PERCPU_HASH);
+       __type(key, int);
+       __type(value, struct map_value);
+       __uint(max_entries, 1);
+} pcpu_hash SEC(".maps");
+
+SEC("tc")
+int pcpu_hash_refcount_leak(void *ctx)
+{
+       struct map_value *v;
+
+       v = __lookup_map(&pcpu_hash);
+       if (!v)
+               return 0;
+
+       return __insert_in_list(&head, &lock, &v->node);
+}
+
+SEC("tc")
+int check_pcpu_hash_refcount(void *ctx)
+{
+       return probe_read_refcount();
+}
+
+struct lock_map_value {
+       struct node_data __kptr *node;
+       struct bpf_spin_lock lock;
+       int value;
+};
+
+struct {
+       __uint(type, BPF_MAP_TYPE_HASH);
+       __type(key, int);
+       __type(value, struct lock_map_value);
+       __uint(max_entries, 1);
+} lock_hash SEC(".maps");
+
+SEC("tc")
+int hash_lock_refcount_leak(void *ctx)
+{
+       struct lock_map_value *v;
+
+       v = __lookup_map(&lock_hash);
+       if (!v)
+               return 0;
+
+       bpf_spin_lock(&v->lock);
+       v->value = 42;
+       bpf_spin_unlock(&v->lock);
+       return __insert_in_list(&head, &lock, &v->node);
+}
+
+SEC("tc")
+int check_hash_lock_refcount(void *ctx)
+{
+       return probe_read_refcount();
+}
+
+struct {
+       __uint(type, BPF_MAP_TYPE_CGRP_STORAGE);
+       __uint(map_flags, BPF_F_NO_PREALLOC);
+       __type(key, int);
+       __type(value, struct lock_map_value);
+} cgrp_strg SEC(".maps");
+
+SEC("cgroup/connect6")
+int cgroup_storage_refcount_leak(struct bpf_sock_addr *ctx)
+{
+       struct lock_map_value *v;
+       struct tcp_sock *tsk;
+       struct bpf_sock *sk;
+       u32 refcnt;
+
+       if (ctx->family != AF_INET6 || ctx->user_family != AF_INET6)
+               return 1;
+
+       sk = ctx->sk;
+       if (!sk)
+               return 1;
+
+       tsk = bpf_skc_to_tcp_sock(sk);
+       if (!tsk)
+               return 1;
+
+       v = bpf_cgrp_storage_get(&cgrp_strg, tsk->inet_conn.icsk_inet.sk.sk_cgrp_data.cgroup, 0,

tools/testing/selftests/bpf/progs/refcounted_kptr.c +22 -1

+
+SEC("fexit/inet_stream_connect")
+int BPF_PROG(check_cgroup_storage_refcount, struct socket *sock, struct sockaddr *uaddr, int addr_len,
+            int flags)
+{
+       struct lock_map_value *v;
+       u32 refcnt;
+
+       if (uaddr->sa_family != AF_INET6)
+               return 0;
+
+       v = bpf_cgrp_storage_get(&cgrp_strg, sock->sk->sk_cgrp_data.cgroup, 0, 0);
+       if (!v)
+               return 0;
+
+       refcnt = probe_read_refcount();
+       bpf_spin_lock(&v->lock);
+       v->value = refcnt;
+       bpf_spin_unlock(&v->lock);
+       return 0;
+}
+
 char _license[] SEC("license") = "GPL";
--

2.51.0

Quoted 20 lines · Leon Hwang

Add tests to verify that updating hash and local storage maps decrements refcount when BPF_KPTR_REF objects are involved.

The tests perform the following steps:

1. Call update_elem() to insert an initial value.
2. Use bpf_refcount_acquire() to increment the refcount.
3. Store the node pointer in the map value.
4. Add the node to a linked list.
5. Probe-read the refcount and verify it is *2*.
6. Call update_elem() again to trigger refcount decrement.
7. Probe-read the refcount and verify it is *1*.

Signed-off-by Leon Hwang <[email protected]>

---
 .../bpf/prog_tests/refcounted_kptr.c          | 178 +++++++++++++++++-
 .../selftests/bpf/progs/refcounted_kptr.c     | 160 ++++++++++++++++
 2 files changed, 337 insertions(+), 1 deletion(-)

Quoted 8 lines

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +4 -0

@@ -44,3 +44,179 @@ void test_refcounted_kptr_wrong_owner(void)
        ASSERT_OK(opts.retval, "rbtree_wrong_owner_remove_fail_a2 retval");
        refcounted_kptr__destroy(skel);
 }
+
+static void test_refcnt_leak(void *values, size_t values_sz, u64 flags, struct bpf_map *map,
+                            struct bpf_program *prog_leak, struct bpf_program *prog_check)
+{

Quoted 1 line

+}

Just use syscall BPF programs across different subtests, and you can share this test_refcnt_leak() across subtests.

It also saves you some code setting up bpf_test_run_opts. You can just call bpf_prog_test_run_opts(prog_fd, NULL) as you don't pass any input from ctx.

Quoted 3 lines

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +3 -0

+
+static void test_percpu_hash_refcount_leak(void)
+{

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c void test_refcounted_kptr_wrong_owner(void) · line 200

Quoted 8 lines

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +8 -0

+out:
+       close(cgroup);
+       refcounted_kptr__destroy(skel);
+       if (client_fd >= 0)
+               close(client_fd);
+       if (server_fd >= 0)
+               close(server_fd);
+}

Then, you won't need to set up server, connection.... just to read/write cgroup local storage. Just call test_refcnt_leak() that runs the two BPF syscall programs for cgroup local storage.

And in syscall BPF program, you can simply get the cgroup through the current task

SEC("syscall")
int syscall_prog(void *ctx)

{
        struct task_struct *task = bpf_get_current_task_btf();

        v = bpf_cgrp_storage_get(&cgrp_strg, task->cgroups->dfl_cgrp, 0,
                               BPF_LOCAL_STORAGE_GET_F_CREATE);
        ...
}

Quoted 21 lines · Amery Hung

Quoted 20 lines · Leon Hwang

Add tests to verify that updating hash and local storage maps decrements refcount when BPF_KPTR_REF objects are involved.

The tests perform the following steps:

1. Call update_elem() to insert an initial value.
2. Use bpf_refcount_acquire() to increment the refcount.
3. Store the node pointer in the map value.
4. Add the node to a linked list.
5. Probe-read the refcount and verify it is *2*.
6. Call update_elem() again to trigger refcount decrement.
7. Probe-read the refcount and verify it is *1*.

Signed-off-by Leon Hwang <[email protected]>

---
 .../bpf/prog_tests/refcounted_kptr.c          | 178 +++++++++++++++++-
 .../selftests/bpf/progs/refcounted_kptr.c     | 160 ++++++++++++++++
 2 files changed, 337 insertions(+), 1 deletion(-)

[...]

Quoted 8 lines

Quoted 8 lines

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +4 -0

@@ -44,3 +44,179 @@ void test_refcounted_kptr_wrong_owner(void)
        ASSERT_OK(opts.retval, "rbtree_wrong_owner_remove_fail_a2 retval");
        refcounted_kptr__destroy(skel);
 }
+
+static void test_refcnt_leak(void *values, size_t values_sz, u64 flags, struct bpf_map *map,
+                            struct bpf_program *prog_leak, struct bpf_program *prog_check)
+{

[...]

Quoted 12 lines

Quoted 1 line

+}

Just use syscall BPF programs across different subtests, and you can share this test_refcnt_leak() across subtests.

It also saves you some code setting up bpf_test_run_opts. You can just call bpf_prog_test_run_opts(prog_fd, NULL) as you don't pass any input from ctx.

Quoted 3 lines

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +3 -0

+
+static void test_percpu_hash_refcount_leak(void)
+{

[...]

Quoted 13 lines

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c void test_refcounted_kptr_wrong_owner(void) · line 200

Quoted 8 lines

tools/testing/selftests/bpf/prog_tests/refcounted_kptr.c +8 -0

+out:
+       close(cgroup);
+       refcounted_kptr__destroy(skel);
+       if (client_fd >= 0)
+               close(client_fd);
+       if (server_fd >= 0)
+               close(server_fd);
+}

Then, you won't need to set up server, connection.... just to read/write cgroup local storage. Just call test_refcnt_leak() that runs the two BPF syscall programs for cgroup local storage.

[...]

Quoted 7 lines

And in syscall BPF program, you can simply get the cgroup through the current task

SEC("syscall")
int syscall_prog(void *ctx)

Quoted 8 lines

{
        struct task_struct *task = bpf_get_current_task_btf();

        v = bpf_cgrp_storage_get(&cgrp_strg, task->cgroups->dfl_cgrp, 0,
                               BPF_LOCAL_STORAGE_GET_F_CREATE);
        ...
}

Hi Amery,

Thanks for the suggestion.

I tried your approach, but the verifier rejected it with the following error:

0: R1=ctx() R10=fp0
; task = bpf_get_current_task_btf(); @ refcounted_kptr.c:686
0: (85) call bpf_get_current_task_btf#158 ; R0=trusted_ptr_task_struct()
; v = bpf_cgrp_storage_get(&cgrp_strg, task->cgroups->dfl_cgrp, 0, @ refcounted_kptr.c:687
1: (79) r1 = *(u64 *)(r0 +4856) ; R0=trusted_ptr_task_struct()
R1=untrusted_ptr_css_set()
2: (79) r2 = *(u64 *)(r1 +96) ; R1=untrusted_ptr_css_set()
R2=untrusted_ptr_cgroup()

3: (18) r1 = 0xffffa1b442a4b800       ; R1=map_ptr(map=cgrp_strg,ks=4,vs=16)
5: (b7) r3 = 0                        ; R3=0
6: (b7) r4 = 1                        ; R4=1

7: (85) call bpf_cgrp_storage_get#210
R2 type=untrusted_ptr_ expected=ptr_, trusted_ptr_, rcu_ptr_ processed 7 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0

Hello:

This series was applied to bpf/bpf-next.git (master) by Andrii Nakryiko <[email protected]>:

Quoted 10 lines · On Sun, 26 Oct 2025 23:39:56 +0800 you wrote:

In the discussion thread
"[PATCH bpf-next v9 0/7] bpf: Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags for percpu maps"[1], it was pointed out that missing calls to bpf_obj_free_fields() could
lead to memory leaks.

A selftest was added to confirm that this is indeed a real issue - the refcount of BPF_KPTR_REF field is not decremented when bpf_obj_free_fields() is missing after copy_map_value[,_long]().

[...]

Here is the summary with links:
- [bpf,v3,1/4] bpf: Free special fields when update [lru_,]percpu_hash maps

    https://git.kernel.org/bpf/bpf-next/c/f6de8d643ff1
  - [bpf,v3,2/4] bpf: Free special fields when update hash maps with BPF_F_LOCK
    https://git.kernel.org/bpf/bpf-next/c/c7fcb7972196
  - [bpf,v3,3/4] bpf: Free special fields when update local storage maps
    (no matching commit)
  - [bpf,v3,4/4] selftests/bpf: Add tests to verify freeing the special fields when update hash and local storage maps
    https://git.kernel.org/bpf/bpf-next/c/d5a7e7af14cc

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

Quoted 20 lines · On Tue, Oct 28, 2025 at 11:10 AM <[email protected]> wrote:

Hello:

This series was applied to bpf/bpf-next.git (master) by Andrii Nakryiko <[email protected]>:

Quoted 10 lines · On Sun, 26 Oct 2025 23:39:56 +0800 you wrote:

In the discussion thread
"[PATCH bpf-next v9 0/7] bpf: Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags for percpu maps"[1], it was pointed out that missing calls to bpf_obj_free_fields() could
lead to memory leaks.

A selftest was added to confirm that this is indeed a real issue - the refcount of BPF_KPTR_REF field is not decremented when bpf_obj_free_fields() is missing after copy_map_value[,_long]().

[...]

Here is the summary with links:
- [bpf,v3,1/4] bpf: Free special fields when update [lru_,]percpu_hash maps

Quoted 8 lines

    https://git.kernel.org/bpf/bpf-next/c/f6de8d643ff1
  - [bpf,v3,2/4] bpf: Free special fields when update hash maps with BPF_F_LOCK
    https://git.kernel.org/bpf/bpf-next/c/c7fcb7972196
  - [bpf,v3,3/4] bpf: Free special fields when update local storage maps
    (no matching commit)
  - [bpf,v3,4/4] selftests/bpf: Add tests to verify freeing the special fields when update hash and local storage maps
    https://git.kernel.org/bpf/bpf-next/c/d5a7e7af14cc

Ok, I had to drop this from bpf-next after all. First, kptr_refcount_leak/cgroup_storage_refcount_leak needs to be adjusted due to that one line removal in patch 3.

But what's worse, we started getting deadlock warning when running one of the tests, see [0]:

[  418.260323] bpf_testmod: oh no, recursing into test_1, recursion_misses 1
  [  424.982201]
  [  424.982207] ================================
  [  424.982216] WARNING: inconsistent lock state
  [  424.982219] 6.18.0-rc1-gbb1b9387787c-dirty #1 Tainted: G        W  OE
  [  424.982221] --------------------------------
  [  424.982223] inconsistent {INITIAL USE} -> {IN-NMI} usage.
  [  424.982225] new_name/11207 [HC1[1]:SC0[0]:HE0:SE1] takes:
  [  424.982229] ffffe8ffffd9c000 (&loc_l->lock){....}-{2:2}, at:
bpf_lru_pop_free+0x2c6/0x1a50
  [  424.982244] {INITIAL USE} state was registered at:
  [  424.982246]   lock_acquire+0x154/0x2d0
  [  424.982252]   _raw_spin_lock_irqsave+0x39/0x60
  [  424.982259]   bpf_lru_pop_free+0x2c6/0x1a50
  [  424.982262]   htab_lru_map_update_elem+0x17e/0xa90
  [  424.982266]   bpf_map_update_value+0x5aa/0x1230
  [  424.982272]   __sys_bpf+0x33b4/0x4ef0
  [  424.982275]   __x64_sys_bpf+0x78/0xe0
  [  424.982278]   do_syscall_64+0x6a/0x2f0
  [  424.982282]   entry_SYSCALL_64_after_hwframe+0x76/0x7e
  [  424.982287] irq event stamp: 236
  [  424.982288] hardirqs last  enabled at (235): [<ffffffff959e4e70>]
do_syscall_64+0x30/0x2f0
  [  424.982292] hardirqs last disabled at (236): [<ffffffff959e65df>]
exc_nmi+0x7f/0x110
  [  424.982296] softirqs last  enabled at (0): [<ffffffff933fe7cf>]
copy_process+0x1c3f/0x6ab0
  [  424.982302] softirqs last disabled at (0): [<0000000000000000>] 0x0
  [  424.982305]
  [  424.982305] other info that might help us debug this:
  [  424.982306]  Possible unsafe locking scenario:
  [  424.982306]
  [  424.982307]        CPU0
  [  424.982308]        ----
  [  424.982309]   lock(&loc_l->lock);
  [  424.982311]   <Interrupt>
  [  424.982312]     lock(&loc_l->lock);
  [  424.982314]
  [  424.982314]  *** DEADLOCK ***
  [  424.982314]
  [  424.982315] no locks held by new_name/11207.
  [  424.982317]
  [  424.982317] stack backtrace:
  [  424.982326] CPU: 1 UID: 0 PID: 11207 Comm: new_name Tainted: G
    W  OE       6.18.0-rc1-gbb1b9387787c-dirty #1 PREEMPT(full)
  [  424.982332] Tainted: [W]=WARN, [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
  [  424.982334] Hardware name: QEMU Ubuntu 25.04 PC (i440FX + PIIX,
1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
  [  424.982337] Call Trace:
  [  424.982340]  <NMI>

  [  424.982342]  dump_stack_lvl+0x5d/0x80
  [  424.982356]  print_usage_bug.part.0+0x22b/0x2c0
  [  424.982360]  lock_acquire+0x278/0x2d0
  [  424.982364]  ? __irq_work_queue_local+0x133/0x360
  [  424.982371]  ? bpf_lru_pop_free+0x2c6/0x1a50
  [  424.982375]  _raw_spin_lock_irqsave+0x39/0x60
  [  424.982379]  ? bpf_lru_pop_free+0x2c6/0x1a50
  [  424.982382]  bpf_lru_pop_free+0x2c6/0x1a50

  [  424.982387]  ? arch_irq_work_raise+0x3f/0x60
  [  424.982394]  ? __pfx___irq_work_queue_local+0x10/0x10
  [  424.982399]  htab_lru_map_update_elem+0x17e/0xa90
  [  424.982405]  ? __pfx_htab_lru_map_update_elem+0x10/0x10
  [  424.982408]  ? __kasan_check_byte+0x16/0x60
  [  424.982414]  ? __htab_map_lookup_elem+0x95/0x220
  [  424.982420]  bpf_prog_2c77131b3c031599_oncpu_lru_map+0xe4/0x168
  [  424.982423]  __perf_event_overflow+0x8e8/0xea0
  [  424.982430]  ? __pfx___perf_event_overflow+0x10/0x10
  [  424.982436]  handle_pmi_common+0x3fe/0x810
  [  424.982441]  ? __pfx_handle_pmi_common+0x10/0x10
  [  424.982452]  ? __pfx_intel_bts_interrupt+0x10/0x10
  [  424.982458]  intel_pmu_handle_irq+0x1c5/0x5d0
  [  424.982461]  ? lock_acquire+0x1ef/0x2d0
  [  424.982465]  ? nmi_handle.part.0+0x2f/0x380
  [  424.982469]  perf_event_nmi_handler+0x3e/0x70
  [  424.982476]  nmi_handle.part.0+0x13f/0x380
  [  424.982480]  ? trace_rcu_watching+0x105/0x170
  [  424.982486]  default_do_nmi+0x3b/0x110
  [  424.982490]  ? irqentry_nmi_enter+0x6f/0x80
  [  424.982493]  exc_nmi+0xe3/0x110
  [  424.982497]  end_repeat_nmi+0xf/0x53
  [  424.982502] RIP: 0010:fput_close_sync+0x56/0x1a0
  [  424.982509] Code: 48 89 e5 48 c7 04 24 b3 8a b5 41 48 c7 44 24 08
5c a2 3e 96 48 c1 ed 03 48 c7 44 24 10 10 a7 e0 93 42 c7 44 2d 00 f1
f1 f1 f1 <42> c7 44 2d 04 00 f3 f3 f3 65 48 8b 05 91 98 56 04 48 89 44
24 58
  [  424.982513] RSP: 0018:ffffc900099d7e88 EFLAGS: 00000a06
  [  424.982517] RAX: 0000000000000000 RBX: ffff888109fb48c0 RCX:
0000000000000000
  [  424.982520] RDX: 1ffff110099572bb RSI: 0000000000000008 RDI:
ffff888109fb4a20
  [  424.982522] RBP: 1ffff9200133afd1 R08: ffff888109fb48c0 R09:
ffff888109278b40
  [  424.982524] R10: ffff888109fb4920 R11: 0000000000000000 R12:
0000000000000003
  [  424.982526] R13: dffffc0000000000 R14: 0000000000000003 R15:
0000000000000000
  [  424.982532]  ? fput_close_sync+0x56/0x1a0
  [  424.982537]  ? fput_close_sync+0x56/0x1a0
  [  424.982541]  </NMI>
  [  424.982542]  <TASK>
  [  424.982544]  ? __pfx_fput_close_sync+0x10/0x10
  [  424.982548]  ? do_raw_spin_unlock+0x59/0x250
  [  424.982553]  __x64_sys_close+0x7d/0xd0
  [  424.982559]  do_syscall_64+0x6a/0x2f0
  [  424.982563]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
  [  424.982566] RIP: 0033:0x7faae0f88fe2
  [  424.982569] Code: 08 0f 85 71 3a ff ff 49 89 fb 48 89 f0 48 89 d7
48 89 ce 4c 89 c2 4d 89 ca 4c 8b 44 24 08 4c 8b 4c 24 10 4c 89 5c 24
08 0f 05 <c3> 66 2e 0f 1f 84 00 00 00 00 00 66 2e 0f 1f 84 00 00 00 00
00 66
  [  424.982571] RSP: 002b:00007ffe58ee5b08 EFLAGS: 00000246 ORIG_RAX:
0000000000000003
  [  424.982574] RAX: ffffffffffffffda RBX: 00007faae0a6cb00 RCX:
00007faae0f88fe2
  [  424.982577] RDX: 0000000000000000 RSI: 0000000000000000 RDI:
0000000000000072
  [  424.982579] RBP: 00007ffe58ee5b30 R08: 0000000000000000 R09:
0000000000000000
  [  424.982581] R10: 0000000000000000 R11: 0000000000000246 R12:
0000000000000008
  [  424.982583] R13: 0000000000000000 R14: 0000556f5e250c90 R15:
00007faae11e9000
  [  424.982588]  </TASK>
  [  424.982606] perf: interrupt took too long (14417 > 12551),

lowering kernel.perf_event_max_sample_rate to 13000

We'll need to figure this out first.

[0] https://github.com/kernel-patches/bpf/actions/runs/18884827710/job/53898669092

Quoted 6 lines

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

Quoted 1 line · Andrii Nakryiko

On Tue, Oct 28, 2025 at 11:10 AM <[email protected]> wrote:

Quoted 2 lines

  [  424.982379]  ? bpf_lru_pop_free+0x2c6/0x1a50
  [  424.982382]  bpf_lru_pop_free+0x2c6/0x1a50

Right, this is the classic NMI vs spinlock deadlock:

Process Context (CPU 0) NMI Context (CPU 0)
======================= ===================

syscall()
|
+-> htab_lru_map_update_elem()
|
+-> bpf_lru_pop_free()
|
+-> spin_lock_irqsave(&lock)
| +-------------------+
| | LOCK ACQUIRED [Y] |
| | IRQs DISABLED |
| +-------------------+
|
+-> [Critical Section]
| |
| | Working with LRU...
| |
| | +-----------------------+
| |<---------------------+ ! NMI FIRES! |
| | +-----------------------+
| | | (IRQs disabled but |
| | | NMI ignores that!) |
| | +-----------------------+
| | |
| | [INTERRUPTED] |
| | [Context saved] |
| | v
| | perf_event_nmi_handler()
| | |
| | +-> BPF program
| | |
| | +-> htab_lru_map_
| | | update_elem()
| | |
| | +-> bpf_lru_pop_
| | | free()
| | |
| | +-> spin_lock_
| | | irqsave(&lock)
| | | +------------+
| | | | TRIES TO |
| | | | ACQUIRE |
| | | | SAME LOCK! |
| | | +------------+
| | | |
| | | v
| | | +------------+
| |<--------------------------------+---+ ! DEADLOCK |
| | | +------------+
| | | | Lock held |
| | Still holding lock... | | by process |
| | Waiting for NMI to finish ---+ | | context |
| | | | | |
| | | | | NMI waits |
| | | | | for same |
| | | | | lock |
| | | | +------------+
| | | | |
| | | | v
| | | | [Spin forever]
| | | | |
| | | +--------+
| | | (Circular wait)
| | |
| | +-> SYSTEM HUNG
| |
| +-> [Never reached]
|
+-> spin_unlock_irqrestore(&lock)
[Never reached]

+---------------------------------------------------------------------+
|                       DEADLOCK SUMMARY                              |
+---------------------------------------------------------------------+
|                                                                     |
| Process Context: Holds &loc_l->lock, waiting for NMI to finish      |
|                                                                     |
| NMI Context:     Trying to acquire &loc_l->lock                     |
|                  (same lock, same CPU)                              |
|                                                                     |
| Result:          Both contexts wait for each other = DEADLOCK       |
|                                                                     |
+---------------------------------------------------------------------+

We can fix this by converting the raw_spinlock_t to trylock-based approach, similar to the fix for ringbuf in
commit a650d38915c1 ("bpf: Convert ringbuf map to rqspinlock").

In bpf_common_lru_pop_free(), we could use:

    if (!raw_res_spin_lock_irqsave(&loc_l->lock, flags))
        return NULL;

However, this deadlock is pre-existing and not introduced by this series. It's better to send a separate fix for this deadlock.

Hi Menglong, could you follow up on the deadlock fix?

Thanks,
Leon

Right, this is the classic NMI vs spinlock deadlock:

Quoted 3 lines · Leon Hwang

Right, this is the classic NMI vs spinlock deadlock:

Leon,

please stop copy pasting what AI told you.
I'd rather see a human with typos and grammar mistakes.