]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
bpf: net_sched: Add basic bpf qdisc kfuncs
authorAmery Hung <amery.hung@bytedance.com>
Wed, 9 Apr 2025 21:45:59 +0000 (14:45 -0700)
committerMartin KaFai Lau <martin.lau@kernel.org>
Thu, 17 Apr 2025 17:54:40 +0000 (10:54 -0700)
Add basic kfuncs for working on skb in qdisc.

Both bpf_qdisc_skb_drop() and bpf_kfree_skb() can be used to release
a reference to an skb. However, bpf_qdisc_skb_drop() can only be called
in .enqueue where a to_free skb list is available from kernel to defer
the release. bpf_kfree_skb() should be used elsewhere. It is also used
in bpf_obj_free_fields() when cleaning up skb in maps and collections.

bpf_skb_get_hash() returns the flow hash of an skb, which can be used
to build flow-based queueing algorithms.

Finally, allow users to create read-only dynptr via bpf_dynptr_from_skb().

Signed-off-by: Amery Hung <amery.hung@bytedance.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://patch.msgid.link/20250409214606.2000194-4-ameryhung@gmail.com
net/sched/bpf_qdisc.c

index 7e5cb7294f406fb341acc7996352a4cdf7dd57e6..5ef13cc6356024f1a6148e5e9651eabbef5e3415 100644 (file)
@@ -8,6 +8,9 @@
 #include <net/pkt_sched.h>
 #include <net/pkt_cls.h>
 
+#define QDISC_OP_IDX(op)       (offsetof(struct Qdisc_ops, op) / sizeof(void (*)(void)))
+#define QDISC_MOFF_IDX(moff)   (moff / sizeof(void (*)(void)))
+
 static struct bpf_struct_ops bpf_Qdisc_ops;
 
 struct bpf_sk_buff_ptr {
@@ -123,6 +126,95 @@ static int bpf_qdisc_btf_struct_access(struct bpf_verifier_log *log,
        return 0;
 }
 
+__bpf_kfunc_start_defs();
+
+/* bpf_skb_get_hash - Get the flow hash of an skb.
+ * @skb: The skb to get the flow hash from.
+ */
+__bpf_kfunc u32 bpf_skb_get_hash(struct sk_buff *skb)
+{
+       return skb_get_hash(skb);
+}
+
+/* bpf_kfree_skb - Release an skb's reference and drop it immediately.
+ * @skb: The skb whose reference to be released and dropped.
+ */
+__bpf_kfunc void bpf_kfree_skb(struct sk_buff *skb)
+{
+       kfree_skb(skb);
+}
+
+/* bpf_qdisc_skb_drop - Drop an skb by adding it to a deferred free list.
+ * @skb: The skb whose reference to be released and dropped.
+ * @to_free_list: The list of skbs to be dropped.
+ */
+__bpf_kfunc void bpf_qdisc_skb_drop(struct sk_buff *skb,
+                                   struct bpf_sk_buff_ptr *to_free_list)
+{
+       __qdisc_drop(skb, (struct sk_buff **)to_free_list);
+}
+
+__bpf_kfunc_end_defs();
+
+BTF_KFUNCS_START(qdisc_kfunc_ids)
+BTF_ID_FLAGS(func, bpf_skb_get_hash, KF_TRUSTED_ARGS)
+BTF_ID_FLAGS(func, bpf_kfree_skb, KF_RELEASE)
+BTF_ID_FLAGS(func, bpf_qdisc_skb_drop, KF_RELEASE)
+BTF_ID_FLAGS(func, bpf_dynptr_from_skb, KF_TRUSTED_ARGS)
+BTF_KFUNCS_END(qdisc_kfunc_ids)
+
+BTF_SET_START(qdisc_common_kfunc_set)
+BTF_ID(func, bpf_skb_get_hash)
+BTF_ID(func, bpf_kfree_skb)
+BTF_ID(func, bpf_dynptr_from_skb)
+BTF_SET_END(qdisc_common_kfunc_set)
+
+BTF_SET_START(qdisc_enqueue_kfunc_set)
+BTF_ID(func, bpf_qdisc_skb_drop)
+BTF_SET_END(qdisc_enqueue_kfunc_set)
+
+enum qdisc_ops_kf_flags {
+       QDISC_OPS_KF_COMMON             = 0,
+       QDISC_OPS_KF_ENQUEUE            = 1 << 0,
+};
+
+static const u32 qdisc_ops_context_flags[] = {
+       [QDISC_OP_IDX(enqueue)]         = QDISC_OPS_KF_ENQUEUE,
+       [QDISC_OP_IDX(dequeue)]         = QDISC_OPS_KF_COMMON,
+       [QDISC_OP_IDX(init)]            = QDISC_OPS_KF_COMMON,
+       [QDISC_OP_IDX(reset)]           = QDISC_OPS_KF_COMMON,
+       [QDISC_OP_IDX(destroy)]         = QDISC_OPS_KF_COMMON,
+};
+
+static int bpf_qdisc_kfunc_filter(const struct bpf_prog *prog, u32 kfunc_id)
+{
+       u32 moff, flags;
+
+       if (!btf_id_set8_contains(&qdisc_kfunc_ids, kfunc_id))
+               return 0;
+
+       if (prog->aux->st_ops != &bpf_Qdisc_ops)
+               return -EACCES;
+
+       moff = prog->aux->attach_st_ops_member_off;
+       flags = qdisc_ops_context_flags[QDISC_MOFF_IDX(moff)];
+
+       if ((flags & QDISC_OPS_KF_ENQUEUE) &&
+           btf_id_set_contains(&qdisc_enqueue_kfunc_set, kfunc_id))
+               return 0;
+
+       if (btf_id_set_contains(&qdisc_common_kfunc_set, kfunc_id))
+               return 0;
+
+       return -EACCES;
+}
+
+static const struct btf_kfunc_id_set bpf_qdisc_kfunc_set = {
+       .owner = THIS_MODULE,
+       .set   = &qdisc_kfunc_ids,
+       .filter = bpf_qdisc_kfunc_filter,
+};
+
 static const struct bpf_verifier_ops bpf_qdisc_verifier_ops = {
        .get_func_proto         = bpf_base_func_proto,
        .is_valid_access        = bpf_qdisc_is_valid_access,
@@ -209,8 +301,25 @@ static struct bpf_struct_ops bpf_Qdisc_ops = {
        .owner = THIS_MODULE,
 };
 
+BTF_ID_LIST(bpf_sk_buff_dtor_ids)
+BTF_ID(func, bpf_kfree_skb)
+
 static int __init bpf_qdisc_kfunc_init(void)
 {
-       return register_bpf_struct_ops(&bpf_Qdisc_ops, Qdisc_ops);
+       int ret;
+       const struct btf_id_dtor_kfunc skb_kfunc_dtors[] = {
+               {
+                       .btf_id       = bpf_sk_buff_ids[0],
+                       .kfunc_btf_id = bpf_sk_buff_dtor_ids[0]
+               },
+       };
+
+       ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &bpf_qdisc_kfunc_set);
+       ret = ret ?: register_btf_id_dtor_kfuncs(skb_kfunc_dtors,
+                                                ARRAY_SIZE(skb_kfunc_dtors),
+                                                THIS_MODULE);
+       ret = ret ?: register_bpf_struct_ops(&bpf_Qdisc_ops, Qdisc_ops);
+
+       return ret;
 }
 late_initcall(bpf_qdisc_kfunc_init);