From: Magnus Karlsson <magnus.karlsson@xxxxxxxxx> Document in the XDP_REDIRECT manual section that drivers must call xdp_do_flush() before napi_complete_done(). The two reasons behind this can be found following the links below. Signed-off-by: Magnus Karlsson <magnus.karlsson@xxxxxxxxx> Link: https://lore.kernel.org/r/20221220185903.1105011-1-sbohrer@xxxxxxxxxxxxxx Link: https://lore.kernel.org/all/20210624160609.292325-1-toke@xxxxxxxxxx/ --- net/core/filter.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) diff --git a/net/core/filter.c b/net/core/filter.c index ab811293ae5d..7a2b67893afd 100644 --- a/net/core/filter.c +++ b/net/core/filter.c @@ -4128,9 +4128,13 @@ static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { * bpf_redirect_info to actually enqueue the frame into a map type-specific * bulk queue structure. * - * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(), - * which will flush all the different bulk queues, thus completing the - * redirect. + * 3. Before exiting its NAPI poll loop, the driver will call + * xdp_do_flush(), which will flush all the different bulk queues, + * thus completing the redirect. Note that xdp_do_flush() must be + * called before napi_complete_done() in the driver, as the + * XDP_REDIRECT logic relies on being inside a single NAPI instance + * through to the xdp_do_flush() call for RCU protection of all + * in-kernel data structures. */ /* * Pointers to the map entries will be kept around for this whole sequence of base-commit: 501543b4fff0ff70bde28a829eb8835081ccef2f -- 2.34.1
