From patchwork Fri Jun 9 14:01:34 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Pattrick X-Patchwork-Id: 1792979 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@legolas.ozlabs.org Authentication-Results: legolas.ozlabs.org; spf=pass (sender SPF authorized) smtp.mailfrom=openvswitch.org (client-ip=140.211.166.136; helo=smtp3.osuosl.org; envelope-from=ovs-dev-bounces@openvswitch.org; receiver=) Authentication-Results: legolas.ozlabs.org; dkim=fail reason="signature verification failed" (1024-bit key; unprotected) header.d=redhat.com header.i=@redhat.com header.a=rsa-sha256 header.s=mimecast20190719 header.b=CNR88EJO; dkim-atps=neutral Received: from smtp3.osuosl.org (smtp3.osuosl.org [140.211.166.136]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature ECDSA (P-384) server-digest SHA384) (No client certificate requested) by legolas.ozlabs.org (Postfix) with ESMTPS id 4Qd2nY2sXWz20QH for ; Sat, 10 Jun 2023 00:01:53 +1000 (AEST) Received: from localhost (localhost [127.0.0.1]) by smtp3.osuosl.org (Postfix) with ESMTP id 3A2DD615C3; Fri, 9 Jun 2023 14:01:51 +0000 (UTC) DKIM-Filter: OpenDKIM Filter v2.11.0 smtp3.osuosl.org 3A2DD615C3 Authentication-Results: smtp3.osuosl.org; dkim=fail reason="signature verification failed" (1024-bit key) header.d=redhat.com header.i=@redhat.com header.a=rsa-sha256 header.s=mimecast20190719 header.b=CNR88EJO X-Virus-Scanned: amavisd-new at osuosl.org Received: from smtp3.osuosl.org ([127.0.0.1]) by localhost (smtp3.osuosl.org [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 6gYASpxf81qp; Fri, 9 Jun 2023 14:01:49 +0000 (UTC) Received: from lists.linuxfoundation.org (lf-lists.osuosl.org [IPv6:2605:bc80:3010:104::8cd3:938]) by smtp3.osuosl.org (Postfix) with ESMTPS id C06AB60C33; Fri, 9 Jun 2023 14:01:48 +0000 (UTC) DKIM-Filter: OpenDKIM Filter v2.11.0 smtp3.osuosl.org C06AB60C33 Received: from lf-lists.osuosl.org (localhost [127.0.0.1]) by lists.linuxfoundation.org (Postfix) with ESMTP id 9AE08C0037; Fri, 9 Jun 2023 14:01:48 +0000 (UTC) X-Original-To: dev@openvswitch.org Delivered-To: ovs-dev@lists.linuxfoundation.org Received: from smtp4.osuosl.org (smtp4.osuosl.org [IPv6:2605:bc80:3010::137]) by lists.linuxfoundation.org (Postfix) with ESMTP id 97A3DC0029 for ; Fri, 9 Jun 2023 14:01:47 +0000 (UTC) Received: from localhost (localhost [127.0.0.1]) by smtp4.osuosl.org (Postfix) with ESMTP id 701F942472 for ; Fri, 9 Jun 2023 14:01:47 +0000 (UTC) DKIM-Filter: OpenDKIM Filter v2.11.0 smtp4.osuosl.org 701F942472 Authentication-Results: smtp4.osuosl.org; dkim=pass (1024-bit key) header.d=redhat.com header.i=@redhat.com header.a=rsa-sha256 header.s=mimecast20190719 header.b=CNR88EJO X-Virus-Scanned: amavisd-new at osuosl.org Received: from smtp4.osuosl.org ([127.0.0.1]) by localhost (smtp4.osuosl.org [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id IYRE34aX_b7g for ; Fri, 9 Jun 2023 14:01:45 +0000 (UTC) X-Greylist: domain auto-whitelisted by SQLgrey-1.8.0 DKIM-Filter: OpenDKIM Filter v2.11.0 smtp4.osuosl.org 59FCE4246F Received: from us-smtp-delivery-124.mimecast.com (us-smtp-delivery-124.mimecast.com [170.10.133.124]) by smtp4.osuosl.org (Postfix) with ESMTPS id 59FCE4246F for ; Fri, 9 Jun 2023 14:01:45 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1686319304; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=hcUajaMbHj6T2gj9v7rEY0Q4SAe40UxANEPeiQJ1Xlo=; b=CNR88EJOVSvfAM/SQ+/XtR774bFJDJFyyA1vaCBkSDwvvV4oHNmDsKAi97Xd82sPpoRGNS uozvnsD+5qTTa82r0SbdWShF+RLLVZZJaFfdIOkGf4XHJAA71yGsJY0Gi0dZKX+uIEg42m RRqbJg7dQd9hLneME0xDN5QD5VrMctk= Received: from mimecast-mx02.redhat.com (mx3-rdu2.redhat.com [66.187.233.73]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id us-mta-662-9oiXFxn6OAKyARfFLvwaCg-1; Fri, 09 Jun 2023 10:01:40 -0400 X-MC-Unique: 9oiXFxn6OAKyARfFLvwaCg-1 Received: from smtp.corp.redhat.com (int-mx08.intmail.prod.int.rdu2.redhat.com [10.11.54.8]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mimecast-mx02.redhat.com (Postfix) with ESMTPS id 50540296A616; Fri, 9 Jun 2023 14:01:40 +0000 (UTC) Received: from mpattric.remote.csb (unknown [10.22.8.151]) by smtp.corp.redhat.com (Postfix) with ESMTP id C7F7FC16044; Fri, 9 Jun 2023 14:01:39 +0000 (UTC) From: Mike Pattrick To: dev@openvswitch.org Date: Fri, 9 Jun 2023 10:01:34 -0400 Message-Id: <20230609140134.640641-1-mkp@redhat.com> In-Reply-To: <20230525203454.2579117-1-mkp@redhat.com> References: <20230525203454.2579117-1-mkp@redhat.com> MIME-Version: 1.0 X-Scanned-By: MIMEDefang 3.1 on 10.11.54.8 X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Cc: Simon Horman Subject: [ovs-dev] [PATCH v4] ofproto-dpif-xlate: Reduce stack usage in recursive xlate functions X-BeenThere: ovs-dev@openvswitch.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: ovs-dev-bounces@openvswitch.org Sender: "dev" Several xlate actions used in recursive translation currently store a large amount of information on the stack. This can result in handler threads quickly running out of stack space despite before xlate_resubmit_resource_check() is able to terminate translation. This patch reduces stack usage by over 3kb from several translation actions. This patch also moves some trace function from do_xlate_actions into its own function to reduce stack usage. Reported-at: https://bugzilla.redhat.com/show_bug.cgi?id=2104779 Signed-off-by: Mike Pattrick Reviewed-by: Simon Horman --- Since v1: - Refactored code into specialized functions and renamed variables for improved readability. Since v2: - Removed inline keywords Since v3: - Improved formatting. --- ofproto/ofproto-dpif-xlate.c | 214 ++++++++++++++++++++++------------- 1 file changed, 138 insertions(+), 76 deletions(-) diff --git a/ofproto/ofproto-dpif-xlate.c b/ofproto/ofproto-dpif-xlate.c index 29f4daa63..2c45831d1 100644 --- a/ofproto/ofproto-dpif-xlate.c +++ b/ofproto/ofproto-dpif-xlate.c @@ -501,6 +501,85 @@ ctx_cancel_freeze(struct xlate_ctx *ctx) static void finish_freezing(struct xlate_ctx *ctx); +/* These functions and structure are used to save stack space in actions that + * need to retain a large amount of xlate_ctx state. */ +struct xretained_state { + union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)]; + uint64_t actset_stub[1024 / 8]; + struct ofpbuf old_stack; + struct ofpbuf old_action_set; + struct flow old_flow; + struct flow old_base; + struct flow_tnl flow_tnl_mask; +}; + +/* The return of this function must be freed by xretain_state_restore(). */ +static struct xretained_state * +xretain_state_save(struct xlate_ctx *ctx) +{ + struct xretained_state *retained = xmalloc(sizeof *retained); + + retained->old_flow = ctx->xin->flow; + retained->old_stack = ctx->stack; + retained->old_action_set = ctx->action_set; + ofpbuf_use_stub(&ctx->stack, retained->new_stack, + sizeof retained->new_stack); + ofpbuf_use_stub(&ctx->action_set, retained->actset_stub, + sizeof retained->actset_stub); + + return retained; +} + +static void +xretain_tunnel_mask_save(struct xlate_ctx *ctx, + struct xretained_state *retained, + const struct xport *out_dev) +{ + struct flow *flow = &ctx->xin->flow; + + retained->flow_tnl_mask = ctx->wc->masks.tunnel; +} + +static void +xretain_base_flow_save(const struct xlate_ctx *ctx, + struct xretained_state *retained) +{ + retained->old_base = ctx->base_flow; +} + +static void +xretain_base_flow_restore(struct xlate_ctx *ctx, + const struct xretained_state *retained) +{ + ctx->base_flow = retained->old_base; +} + +static void +xretain_flow_restore(struct xlate_ctx *ctx, + const struct xretained_state *retained) +{ + ctx->xin->flow = retained->old_flow; +} + +static void +xretain_tunnel_mask_restore(struct xlate_ctx *ctx, + const struct xretained_state *retained) +{ + ctx->wc->masks.tunnel = retained->flow_tnl_mask; +} + +static void +xretain_state_restore(struct xlate_ctx *ctx, struct xretained_state *retained) +{ + ctx->xin->flow = retained->old_flow; + ofpbuf_uninit(&ctx->action_set); + ctx->action_set = retained->old_action_set; + ofpbuf_uninit(&ctx->stack); + ctx->stack = retained->old_stack; + + free(retained); +} + /* A controller may use OFPP_NONE as the ingress port to indicate that * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for * when an input bundle is needed for validation (e.g., mirroring or @@ -3915,20 +3994,17 @@ static void patch_port_output(struct xlate_ctx *ctx, const struct xport *in_dev, struct xport *out_dev, bool is_last_action) { - struct flow *flow = &ctx->xin->flow; - struct flow old_flow = ctx->xin->flow; - struct flow_tnl old_flow_tnl_wc = ctx->wc->masks.tunnel; + struct ovs_list *old_trace = ctx->xin->trace; bool old_conntrack = ctx->conntracked; + struct flow *flow = &ctx->xin->flow; bool old_was_mpls = ctx->was_mpls; + struct xretained_state *retained_state; ovs_version_t old_version = ctx->xin->tables_version; - struct ofpbuf old_stack = ctx->stack; - uint8_t new_stack[1024]; - struct ofpbuf old_action_set = ctx->action_set; - struct ovs_list *old_trace = ctx->xin->trace; - uint64_t actset_stub[1024 / 8]; - ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack); - ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub); + retained_state = xretain_state_save(ctx); + + xretain_tunnel_mask_save(ctx, retained_state, out_dev); + flow->in_port.ofp_port = out_dev->ofp_port; flow->metadata = htonll(0); memset(&flow->tunnel, 0, sizeof flow->tunnel); @@ -3967,14 +4043,15 @@ patch_port_output(struct xlate_ctx *ctx, const struct xport *in_dev, } else { /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and * the learning action look at the packet, then drop it. */ - struct flow old_base_flow = ctx->base_flow; size_t old_size = ctx->odp_actions->size; + + xretain_base_flow_save(ctx, retained_state); mirror_mask_t old_mirrors2 = ctx->mirrors; xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true, false, is_last_action, clone_xlate_actions); ctx->mirrors = old_mirrors2; - ctx->base_flow = old_base_flow; + xretain_base_flow_restore(ctx, retained_state); ctx->odp_actions->size = old_size; /* Undo changes that may have been done for freezing. */ @@ -3986,18 +4063,15 @@ patch_port_output(struct xlate_ctx *ctx, const struct xport *in_dev, if (independent_mirrors) { ctx->mirrors = old_mirrors; } - ctx->xin->flow = old_flow; ctx->xbridge = in_dev->xbridge; - ofpbuf_uninit(&ctx->action_set); - ctx->action_set = old_action_set; - ofpbuf_uninit(&ctx->stack); - ctx->stack = old_stack; /* Restore calling bridge's lookup version. */ ctx->xin->tables_version = old_version; - /* Restore to calling bridge tunneling information */ - ctx->wc->masks.tunnel = old_flow_tnl_wc; + /* Restore to calling bridge tunneling information; the ctx flow, actions, + * and stack. And free the retained state. */ + xretain_tunnel_mask_restore(ctx, retained_state); + xretain_state_restore(ctx, retained_state); /* The out bridge popping MPLS should have no effect on the original * bridge. */ @@ -4247,7 +4321,7 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port, const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port); struct flow_wildcards *wc = ctx->wc; struct flow *flow = &ctx->xin->flow; - struct flow_tnl flow_tnl; + struct flow_tnl *flow_tnl = NULL; union flow_vlan_hdr flow_vlans[FLOW_MAX_VLAN_HEADERS]; uint8_t flow_nw_tos; odp_port_t out_port, odp_port, odp_tnl_port; @@ -4261,7 +4335,6 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port, /* If 'struct flow' gets additional metadata, we'll need to zero it out * before traversing a patch port. */ BUILD_ASSERT_DECL(FLOW_WC_SEQ == 42); - memset(&flow_tnl, 0, sizeof flow_tnl); if (!check_output_prerequisites(ctx, xport, flow, check_stp)) { return; @@ -4305,7 +4378,8 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port, * the Logical (tunnel) Port are not visible for any further * matches, while explicit set actions on tunnel metadata are. */ - flow_tnl = flow->tunnel; + flow_tnl = xmalloc(sizeof *flow_tnl); + *flow_tnl = flow->tunnel; odp_port = tnl_port_send(xport->ofport, flow, ctx->wc); if (odp_port == ODPP_NONE) { xlate_report(ctx, OFT_WARN, "Tunneling decided against output"); @@ -4336,7 +4410,7 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port, tnl_type = tnl_port_get_type(xport->ofport); commit_odp_tunnel_action(flow, &ctx->base_flow, ctx->odp_actions, tnl_type); - flow->tunnel = flow_tnl; /* Restore tunnel metadata */ + flow->tunnel = *flow_tnl; /* Restore tunnel metadata. */ } } else { odp_port = xport->odp_port; @@ -4380,7 +4454,11 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port, /* Output to native tunnel port. */ native_tunnel_output(ctx, xport, flow, odp_port, truncate, is_last_action); - flow->tunnel = flow_tnl; /* Restore tunnel metadata */ + if (flow_tnl) { + flow->tunnel = *flow_tnl; /* Restore tunnel metadata. */ + } else { + memset(&flow->tunnel, 0, sizeof flow->tunnel); + } } else if (terminate_native_tunnel(ctx, xport, flow, wc, &odp_tnl_port)) { @@ -4423,7 +4501,7 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port, xport->xbundle)); } - out: +out: /* Restore flow */ memcpy(flow->vlans, flow_vlans, sizeof flow->vlans); flow->nw_tos = flow_nw_tos; @@ -4431,6 +4509,7 @@ compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port, flow->dl_src = flow_dl_src; flow->packet_type = flow_packet_type; flow->dl_type = flow_dl_type; + free(flow_tnl); } static void @@ -5874,21 +5953,12 @@ clone_xlate_actions(const struct ofpact *actions, size_t actions_len, struct xlate_ctx *ctx, bool is_last_action, bool group_bucket_action OVS_UNUSED) { - struct ofpbuf old_stack = ctx->stack; - union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)]; - ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack); - ofpbuf_put(&ctx->stack, old_stack.data, old_stack.size); - - struct ofpbuf old_action_set = ctx->action_set; - uint64_t actset_stub[1024 / 8]; - ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub); - ofpbuf_put(&ctx->action_set, old_action_set.data, old_action_set.size); - + struct xretained_state *retained_state; size_t offset, ac_offset; - struct flow old_flow = ctx->xin->flow; + + retained_state = xretain_state_save(ctx); if (reversible_actions(actions, actions_len) || is_last_action) { - old_flow = ctx->xin->flow; do_xlate_actions(actions, actions_len, ctx, is_last_action, false); if (!ctx->freezing) { xlate_action_set(ctx); @@ -5903,7 +5973,8 @@ clone_xlate_actions(const struct ofpact *actions, size_t actions_len, * avoid emitting those actions twice. Once inside * the clone, another time for the action after clone. */ xlate_commit_actions(ctx); - struct flow old_base = ctx->base_flow; + xretain_base_flow_save(ctx, retained_state); + bool old_was_mpls = ctx->was_mpls; bool old_conntracked = ctx->conntracked; @@ -5960,14 +6031,10 @@ dp_clone_done: ctx->was_mpls = old_was_mpls; /* Restore the 'base_flow' for the next action. */ - ctx->base_flow = old_base; + xretain_base_flow_restore(ctx, retained_state); xlate_done: - ofpbuf_uninit(&ctx->action_set); - ctx->action_set = old_action_set; - ofpbuf_uninit(&ctx->stack); - ctx->stack = old_stack; - ctx->xin->flow = old_flow; + xretain_state_restore(ctx, retained_state); } static void @@ -6471,19 +6538,13 @@ xlate_check_pkt_larger(struct xlate_ctx *ctx, return; } - struct ofpbuf old_stack = ctx->stack; - union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)]; - ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack); - ofpbuf_put(&ctx->stack, old_stack.data, old_stack.size); + struct xretained_state *retained_state; - struct ofpbuf old_action_set = ctx->action_set; - uint64_t actset_stub[1024 / 8]; - ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub); - ofpbuf_put(&ctx->action_set, old_action_set.data, old_action_set.size); + retained_state = xretain_state_save(ctx); - struct flow old_flow = ctx->xin->flow; xlate_commit_actions(ctx); - struct flow old_base = ctx->base_flow; + xretain_base_flow_save(ctx, retained_state); + bool old_was_mpls = ctx->was_mpls; bool old_conntracked = ctx->conntracked; @@ -6504,10 +6565,10 @@ xlate_check_pkt_larger(struct xlate_ctx *ctx, } nl_msg_end_nested(ctx->odp_actions, offset_attr); - ctx->base_flow = old_base; + xretain_base_flow_restore(ctx, retained_state); + xretain_flow_restore(ctx, retained_state); ctx->was_mpls = old_was_mpls; ctx->conntracked = old_conntracked; - ctx->xin->flow = old_flow; /* If the flow translation for the IF_GREATER case requires freezing, * then ctx->exit would be true. Reset to false so that we can @@ -6530,15 +6591,11 @@ xlate_check_pkt_larger(struct xlate_ctx *ctx, nl_msg_end_nested(ctx->odp_actions, offset_attr); nl_msg_end_nested(ctx->odp_actions, offset); - ofpbuf_uninit(&ctx->action_set); - ctx->action_set = old_action_set; - ofpbuf_uninit(&ctx->stack); - ctx->stack = old_stack; - ctx->base_flow = old_base; ctx->was_mpls = old_was_mpls; ctx->conntracked = old_conntracked; - ctx->xin->flow = old_flow; ctx->exit = old_exit; + xretain_base_flow_restore(ctx, retained_state); + xretain_state_restore(ctx, retained_state); } static void @@ -6989,6 +7046,24 @@ xlate_ofpact_unroll_xlate(struct xlate_ctx *ctx, "cookie=%#"PRIx64, a->rule_table_id, a->rule_cookie); } +static void +xlate_trace(struct xlate_ctx *ctx, const struct ofpact *a) { + struct ofputil_port_map map = OFPUTIL_PORT_MAP_INITIALIZER(&map); + + if (ctx->xin->names) { + struct ofproto_dpif *ofprotop; + ofprotop = ofproto_dpif_lookup_by_name(ctx->xbridge->name); + ofproto_append_ports_to_map(&map, ofprotop->up.ports); + } + + struct ds s = DS_EMPTY_INITIALIZER; + struct ofpact_format_params fp = { .s = &s, .port_map = &map }; + ofpacts_format(a, OFPACT_ALIGN(a->len), &fp); + xlate_report(ctx, OFT_ACTION, "%s", ds_cstr(&s)); + ds_destroy(&s); + ofputil_port_map_destroy(&map); +} + static void do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, struct xlate_ctx *ctx, bool is_last_action, @@ -7031,20 +7106,7 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, } if (OVS_UNLIKELY(ctx->xin->trace)) { - struct ofputil_port_map map = OFPUTIL_PORT_MAP_INITIALIZER(&map); - - if (ctx->xin->names) { - struct ofproto_dpif *ofprotop; - ofprotop = ofproto_dpif_lookup_by_name(ctx->xbridge->name); - ofproto_append_ports_to_map(&map, ofprotop->up.ports); - } - - struct ds s = DS_EMPTY_INITIALIZER; - struct ofpact_format_params fp = { .s = &s, .port_map = &map }; - ofpacts_format(a, OFPACT_ALIGN(a->len), &fp); - xlate_report(ctx, OFT_ACTION, "%s", ds_cstr(&s)); - ds_destroy(&s); - ofputil_port_map_destroy(&map); + xlate_trace(ctx, a); } switch (a->type) {