[PATCH iproute2-next] bpf: don't offload perf array maps

[PATCH iproute2-next] bpf: don't offload perf array maps

[Jakub Kicinski]

Perf arrays are handled specially by the kernel, don't request
offload even when used by an offloaded program.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 lib/bpf.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/lib/bpf.c b/lib/bpf.c
index d9a406bf55f2..4e26c0df76c5 100644
--- a/lib/bpf.c
+++ b/lib/bpf.c
@@ -97,6 +97,11 @@ static const struct bpf_prog_meta __bpf_prog_meta[] = {
 	},
 };
 
+static bool bpf_map_offload_neutral(enum bpf_map_type type)
+{
+	return type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
+}
+
 static const char *bpf_prog_to_subdir(enum bpf_prog_type type)
 {
 	assert(type < ARRAY_SIZE(__bpf_prog_meta) &&
@@ -1594,7 +1599,7 @@ static int bpf_map_attach(const char *name, struct bpf_elf_ctx *ctx,
 			  const struct bpf_elf_map *map, struct bpf_map_ext *ext,
 			  int *have_map_in_map)
 {
-	int fd, ret, map_inner_fd = 0;
+	int fd, ifindex, ret, map_inner_fd = 0;
 
 	fd = bpf_probe_pinned(name, ctx, map->pinning);
 	if (fd > 0) {
@@ -1631,10 +1636,10 @@ static int bpf_map_attach(const char *name, struct bpf_elf_ctx *ctx,
 		}
 	}
 
+	ifindex = bpf_map_offload_neutral(map->type) ? 0 : ctx->ifindex;
 	errno = 0;
 	fd = bpf_map_create(map->type, map->size_key, map->size_value,
-			    map->max_elem, map->flags, map_inner_fd,
-			    ctx->ifindex);
+			    map->max_elem, map->flags, map_inner_fd, ifindex);
 
 	if (fd < 0 || ctx->verbose) {
 		bpf_map_report(fd, name, map, ctx, map_inner_fd);
-- 
2.17.0

Re: [PATCH iproute2-next] bpf: don't offload perf array maps

[Daniel Borkmann]

On 05/05/2018 02:37 AM, Jakub Kicinski wrote:
> Perf arrays are handled specially by the kernel, don't request
> offload even when used by an offloaded program.
> 
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>

Acked-by: Daniel Borkmann <daniel@iogearbox.net>

Re: [PATCH iproute2-next] bpf: don't offload perf array maps

[David Ahern]

On 5/4/18 6:37 PM, Jakub Kicinski wrote:
> Perf arrays are handled specially by the kernel, don't request
> offload even when used by an offloaded program.
> 
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
> ---
>  lib/bpf.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)

applied to iproute2-next.