Re: [bpf-next PATCH v2] bpf: Add rx_queue_mapping to bpf_sock

[Nambiar, Amritha]

On 5/22/2020 4:34 PM, Daniel Borkmann wrote:

On 5/22/20 8:28 PM, Amritha Nambiar wrote:

Add "rx_queue_mapping" to bpf_sock. This gives read access for the
existing field (sk_rx_queue_mapping) of struct sock from bpf_sock.
Semantics for the bpf_sock rx_queue_mapping access are similar to
sk_rx_queue_get(), i.e the value NO_QUEUE_MAPPING is not allowed
and -1 is returned in that case.

This adds the "what this patch does" but could you also add a description for

the use-case in here?

Will fix in v3.

v2: fixed build error for CONFIG_XPS wrapping, reported by
     kbuild test robot <l...@intel.com>

Signed-off-by: Amritha Nambiar <amritha.namb...@intel.com>
---
  include/uapi/linux/bpf.h |    1 +
  net/core/filter.c        |   18 ++++++++++++++++++
  2 files changed, 19 insertions(+)

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index 97e1fd19ff58..d2acd5aeae8d 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -3530,6 +3530,7 @@ struct bpf_sock {
      __u32 dst_ip4;
      __u32 dst_ip6[4];
      __u32 state;
+    __u32 rx_queue_mapping;
  };
  struct bpf_tcp_sock {
diff --git a/net/core/filter.c b/net/core/filter.c
index bd2853d23b50..c4ba92204b73 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c

@@ -6829,6 +6829,7 @@ bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,

      case offsetof(struct bpf_sock, protocol):
      case offsetof(struct bpf_sock, dst_port):
      case offsetof(struct bpf_sock, src_port):
+    case offsetof(struct bpf_sock, rx_queue_mapping):
      case bpf_ctx_range(struct bpf_sock, src_ip4):
      case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
      case bpf_ctx_range(struct bpf_sock, dst_ip4):

@@ -7872,6 +7873,23 @@ u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,

                              skc_state),
                         target_size));
          break;
+    case offsetof(struct bpf_sock, rx_queue_mapping):
+#ifdef CONFIG_XPS
+        *insn++ = BPF_LDX_MEM(
+            BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
+            si->dst_reg, si->src_reg,
+            bpf_target_off(struct sock, sk_rx_queue_mapping,
+                       sizeof_field(struct sock,
+                            sk_rx_queue_mapping),
+                       target_size));
+        *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
+                      1);
+        *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
+#else
+        *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);

This should be -1 as queue mapping as well if XPS is not configured, no?
Otherwise, how do you tell it apart from an actual mapping to 0 if XPS is
built-in?

Agree. My bad. Will fix in v3.

+        *target_size = 2;
+#endif
+        break;
      }
      return insn - insn_buf;