bpf: add support for generic xdp

Follow-up to commit c7272ca72009 ("bpf: add initial support for
attaching xdp progs") to also support generic XDP. This adds an
indicator for loaded generic XDP programs when programs are loaded
as shown in c7272ca72009, but the driver still lacks native XDP
support.

  # ip link
  [...]
  3: eno1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 xdpgeneric qdisc [...]
      link/ether 0c:c4:7a:03:f9:25 brd ff:ff:ff:ff:ff:ff
  [...]

In case the driver does support native XDP, but the user wants
to load the program as generic XDP (e.g. for testing purposes),
then this can be done with the same semantics as in c7272ca72009,
but with 'xdpgeneric' instead of 'xdp' command for loading:

  # ip -force link set dev eno1 xdpgeneric obj xdp.o

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: David S. Miller <davem@davemloft.net>

diff --git a/ip/iplink.c b/ip/iplink.c
index da3f9a779351c935f27a4e9c978ab0c8f5b06545..ae1c70ebcc814cb8bf11985879d773796fa084d8 100644 (file)
--- a/ip/iplink.c
+++ b/ip/iplink.c
@@ -612,9 +612,12 @@ int iplink_parse(int argc, char **argv, struct iplink_req *req,
                        if (get_integer(&mtu, *argv, 0))
                                invarg("Invalid \"mtu\" value\n", *argv);
                        addattr_l(&req->n, sizeof(*req), IFLA_MTU, &mtu, 4);
-               } else if (strcmp(*argv, "xdp") == 0) {
+               } else if (strcmp(*argv, "xdpgeneric") == 0 ||
+                          strcmp(*argv, "xdp") == 0) {
+                       bool generic = strcmp(*argv, "xdpgeneric") == 0;
+
                        NEXT_ARG();
-                       if (xdp_parse(&argc, &argv, req))
+                       if (xdp_parse(&argc, &argv, req, generic))
                                exit(-1);
                } else if (strcmp(*argv, "netns") == 0) {
                        NEXT_ARG();

diff --git a/ip/iplink_xdp.c b/ip/iplink_xdp.c
index a81ed971896c8c410fde94dca6476fc2a8670159..a1380eec7db7962b916b3ef6a3d23a6386a0f2c2 100644 (file)
--- a/ip/iplink_xdp.c
+++ b/ip/iplink_xdp.c
@@ -19,41 +19,56 @@
 
 extern int force;
 
+struct xdp_req {
+       struct iplink_req *req;
+       __u32 flags;
+};
+
 static void xdp_ebpf_cb(void *raw, int fd, const char *annotation)
 {
-       __u32 flags = !force ? XDP_FLAGS_UPDATE_IF_NOEXIST : 0;
-       struct iplink_req *req = raw;
-       struct rtattr *xdp;
+       struct xdp_req *xdp = raw;
+       struct iplink_req *req = xdp->req;
+       struct rtattr *xdp_attr;
 
-       xdp = addattr_nest(&req->n, sizeof(*req), IFLA_XDP);
+       xdp_attr = addattr_nest(&req->n, sizeof(*req), IFLA_XDP);
        addattr32(&req->n, sizeof(*req), IFLA_XDP_FD, fd);
-       addattr32(&req->n, sizeof(*req), IFLA_XDP_FLAGS, flags);
-       addattr_nest_end(&req->n, xdp);
+       if (xdp->flags)
+               addattr32(&req->n, sizeof(*req), IFLA_XDP_FLAGS, xdp->flags);
+       addattr_nest_end(&req->n, xdp_attr);
 }
 
 static const struct bpf_cfg_ops bpf_cb_ops = {
        .ebpf_cb = xdp_ebpf_cb,
 };
 
-static int xdp_delete(struct iplink_req *req)
+static int xdp_delete(struct xdp_req *xdp)
 {
-       xdp_ebpf_cb(req, -1, NULL);
+       xdp_ebpf_cb(xdp, -1, NULL);
        return 0;
 }
 
-int xdp_parse(int *argc, char ***argv, struct iplink_req *req)
+int xdp_parse(int *argc, char ***argv, struct iplink_req *req, bool generic)
 {
        struct bpf_cfg_in cfg = {
                .argc = *argc,
                .argv = *argv,
        };
+       struct xdp_req xdp = {
+               .req = req,
+       };
+
+       if (!force)
+               xdp.flags |= XDP_FLAGS_UPDATE_IF_NOEXIST;
+       if (generic)
+               xdp.flags |= XDP_FLAGS_SKB_MODE;
 
        if (*argc == 1) {
                if (strcmp(**argv, "none") == 0 ||
                    strcmp(**argv, "off") == 0)
-                       return xdp_delete(req);
+                       return xdp_delete(&xdp);
        }
-       if (bpf_parse_common(BPF_PROG_TYPE_XDP, &cfg, &bpf_cb_ops, req))
+
+       if (bpf_parse_common(BPF_PROG_TYPE_XDP, &cfg, &bpf_cb_ops, &xdp))
                return -1;
 
        *argc = cfg.argc;
@@ -64,12 +79,17 @@ int xdp_parse(int *argc, char ***argv, struct iplink_req *req)
 void xdp_dump(FILE *fp, struct rtattr *xdp)
 {
        struct rtattr *tb[IFLA_XDP_MAX + 1];
+       __u32 flags = 0;
 
        parse_rtattr_nested(tb, IFLA_XDP_MAX, xdp);
+
        if (!tb[IFLA_XDP_ATTACHED] ||
            !rta_getattr_u8(tb[IFLA_XDP_ATTACHED]))
                return;
 
-       fprintf(fp, "xdp ");
-       /* More to come here in future for 'ip -d link' (digest, etc) ... */
+       if (tb[IFLA_XDP_FLAGS])
+               flags = rta_getattr_u32(tb[IFLA_XDP_FLAGS]);
+
+       fprintf(fp, "xdp%s ",
+               flags & XDP_FLAGS_SKB_MODE ? "generic" : "");
 }

diff --git a/ip/xdp.h b/ip/xdp.h
index bc6964588cf14ac22b3323f5732fa5e5df51cbf3..1b95e0f63a99d23e3923ea2b036999c351759f7c 100644 (file)
--- a/ip/xdp.h
+++ b/ip/xdp.h
@@ -3,7 +3,7 @@
 
 #include "utils.h"
 
-int xdp_parse(int *argc, char ***argv, struct iplink_req *req);
+int xdp_parse(int *argc, char ***argv, struct iplink_req *req, bool generic);
 void xdp_dump(FILE *fp, struct rtattr *tb);
 
 #endif /* __XDP__ */

diff --git a/man/man8/ip-link.8.in b/man/man8/ip-link.8.in
index a5ddfe7a106e72f96c8413a3589ab5a60243fe49..52571b724af46f5c36449f8aebd174ca15e6259e 100644 (file)
--- a/man/man8/ip-link.8.in
+++ b/man/man8/ip-link.8.in
@@ -126,7 +126,7 @@ ip-link \- network device configuration
 .RB "[ " port_guid " eui64 ] ]"
 .br
 .in -9
-.RB "[ " xdp  " { " off " | "
+.RB "[ { " xdp " | " xdpgeneric  " } { " off " | "
 .br
 .in +8
 .BR object
@@ -1572,8 +1572,23 @@ which may impact security and/or performance. (e.g. VF multicast promiscuous mod
 
 .TP
 .B xdp object "|" pinned "|" off
-set (or unset) a XDP ("express data path") BPF program to run on every
+set (or unset) a XDP ("eXpress Data Path") BPF program to run on every
 packet at driver level.
+.B ip link
+output will indicate a
+.B xdp
+flag for the networking device. If the driver does not have native XDP
+support, the kernel will fall back to a slower, driver-independent "generic"
+XDP variant. The
+.B ip link
+output will in that case indicate
+.B xdpgeneric
+instead of
+.B xdp
+only. If the driver does have native XDP support, but the program is
+loaded under
+.B xdpgeneric object "|" pinned
+then the kernel will use the generic XDP variant instead of the native one.
 
 .B off
 (or