diff mbox series

[ovs-dev,v5,12/12] netdev-offload-tc: Add offload support for sFlow

Message ID 20201029112340.14167-13-cmi@nvidia.com
State New
Headers show
Series Add offload support for sFlow | expand

Commit Message

Chris Mi Oct. 29, 2020, 11:23 a.m. UTC
Create a unique group ID to map the sFlow info when offloading sFlow
action to TC. When showing the offloaded datapath flows, translate the
group ID from TC sample action to sFlow info using the mapping.

Signed-off-by: Chris Mi <cmi@nvidia.com>
Reviewed-by: Eli Britstein <elibr@nvidia.com>
---
 lib/netdev-offload-tc.c | 149 +++++++++++++++++++++++++++++++++++++---
 lib/tc.c                |  61 +++++++++++++++-
 lib/tc.h                |   9 ++-
 3 files changed, 207 insertions(+), 12 deletions(-)
diff mbox series

Patch

diff --git a/lib/netdev-offload-tc.c b/lib/netdev-offload-tc.c
index aef5312a9..36798acc6 100644
--- a/lib/netdev-offload-tc.c
+++ b/lib/netdev-offload-tc.c
@@ -989,6 +989,18 @@  parse_tc_flower_to_match(struct tc_flower *flower,
         action = flower->actions;
         for (i = 0; i < flower->action_count; i++, action++) {
             switch (action->type) {
+            case TC_ACT_SAMPLE: {
+                const struct gid_node *node;
+
+                node = gid_find(action->sample.action_group_id);
+                if (!node) {
+                    VLOG_ERR_RL(&error_rl, "gid node is NULL, gid: %d",
+                                action->sample.action_group_id);
+                    return ENOENT;
+                }
+                nl_msg_put(buf, node->sflow.sflow, node->sflow.sflow_len);
+            }
+            break;
             case TC_ACT_VLAN_POP: {
                 nl_msg_put_flag(buf, OVS_ACTION_ATTR_POP_VLAN);
             }
@@ -1668,6 +1680,78 @@  flower_match_to_tun_opt(struct tc_flower *flower, const struct flow_tnl *tnl,
     flower->mask.tunnel.metadata.present.len = tnl->metadata.present.len;
 }
 
+static int
+parse_userspace_userdata(const struct nlattr *actions,
+                         struct dpif_sflow_attr *sflow_attr)
+{
+    const struct nlattr *nla;
+    unsigned int left;
+
+    NL_NESTED_FOR_EACH_UNSAFE (nla, left, actions) {
+        if (nl_attr_type(nla) == OVS_USERSPACE_ATTR_USERDATA) {
+            struct user_action_cookie *cookie;
+
+            cookie = CONST_CAST(struct user_action_cookie *, nl_attr_get(nla));
+            if (cookie->type == USER_ACTION_COOKIE_SFLOW) {
+                sflow_attr->userdata = CONST_CAST(void *, nl_attr_get(nla));
+                sflow_attr->userdata_len = nl_attr_get_size(nla);
+                return 0;
+            }
+        }
+    }
+
+    VLOG_ERR_RL(&error_rl, "%s: no sFlow cookie", __func__);
+    return EINVAL;
+}
+
+static int
+parse_action_userspace(const struct nlattr *actions,
+                       struct dpif_sflow_attr *sflow_attr)
+{
+    const struct nlattr *nla;
+    unsigned int left;
+
+    NL_NESTED_FOR_EACH_UNSAFE (nla, left, actions) {
+        if (nl_attr_type(nla) == OVS_ACTION_ATTR_USERSPACE) {
+            return parse_userspace_userdata(nla, sflow_attr);
+        }
+    }
+
+    VLOG_ERR_RL(&error_rl, "%s: no OVS_ACTION_ATTR_USERSPACE attribute",
+                __func__);
+    return EINVAL;
+}
+
+static int
+parse_sample_action(const struct nlattr *actions,
+                    struct dpif_sflow_attr *sflow_attr,
+                    struct tc_action *tc_action)
+{
+    const struct nlattr *nla;
+    unsigned int left;
+    int ret = EINVAL;
+
+    sflow_attr->sflow = actions;
+    sflow_attr->sflow_len = actions->nla_len;
+
+    NL_NESTED_FOR_EACH_UNSAFE (nla, left, actions) {
+        if (nl_attr_type(nla) == OVS_SAMPLE_ATTR_ACTIONS) {
+            ret = parse_action_userspace(nla, sflow_attr);
+        } else if (nl_attr_type(nla) == OVS_SAMPLE_ATTR_PROBABILITY) {
+            tc_action->type = TC_ACT_SAMPLE;
+            tc_action->sample.action_rate = UINT32_MAX / nl_attr_get_u32(nla);
+        } else {
+            return EINVAL;
+        }
+    }
+
+    if (tc_action->sample.action_rate) {
+        return ret;
+    }
+
+    return EINVAL;
+}
+
 static int
 netdev_tc_flow_put(struct netdev *netdev, struct match *match,
                    struct nlattr *actions, size_t actions_len,
@@ -1684,6 +1768,7 @@  netdev_tc_flow_put(struct netdev *netdev, struct match *match,
     struct tc_action *action;
     bool recirc_act = false;
     uint32_t block_id = 0;
+    uint32_t group_id = 0;
     struct nlattr *nla;
     struct tcf_id id;
     uint32_t chain;
@@ -1973,7 +2058,8 @@  netdev_tc_flow_put(struct netdev *netdev, struct match *match,
     NL_ATTR_FOR_EACH(nla, left, actions, actions_len) {
         if (flower.action_count >= TCA_ACT_MAX_NUM) {
             VLOG_DBG_RL(&rl, "Can only support %d actions", TCA_ACT_MAX_NUM);
-            return EOPNOTSUPP;
+            err = EOPNOTSUPP;
+            goto out;
         }
         action = &flower.actions[flower.action_count];
         if (nl_attr_type(nla) == OVS_ACTION_ATTR_OUTPUT) {
@@ -1983,7 +2069,8 @@  netdev_tc_flow_put(struct netdev *netdev, struct match *match,
 
             if (!outdev) {
                 VLOG_DBG_RL(&rl, "Can't find netdev for output port %d", port);
-                return ENODEV;
+                err = ENODEV;
+                goto out;
             }
             action->out.ifindex_out = netdev_get_ifindex(outdev);
             action->out.ingress = is_internal_port(netdev_get_type(outdev));
@@ -2021,7 +2108,7 @@  netdev_tc_flow_put(struct netdev *netdev, struct match *match,
 
             err = parse_put_flow_set_action(&flower, action, set, set_len);
             if (err) {
-                return err;
+                goto out;
             }
             if (action->type == TC_ACT_ENCAP) {
                 action->encap.tp_dst = info->tp_dst_port;
@@ -2034,7 +2121,7 @@  netdev_tc_flow_put(struct netdev *netdev, struct match *match,
             err = parse_put_flow_set_masked_action(&flower, action, set,
                                                    set_len, true);
             if (err) {
-                return err;
+                goto out;
             }
         } else if (nl_attr_type(nla) == OVS_ACTION_ATTR_CT) {
             const struct nlattr *ct = nl_attr_get(nla);
@@ -2042,7 +2129,7 @@  netdev_tc_flow_put(struct netdev *netdev, struct match *match,
 
             err = parse_put_flow_ct_action(&flower, action, ct, ct_len);
             if (err) {
-                return err;
+                goto out;
             }
         } else if (nl_attr_type(nla) == OVS_ACTION_ATTR_CT_CLEAR) {
             action->type = TC_ACT_CT;
@@ -2061,17 +2148,49 @@  netdev_tc_flow_put(struct netdev *netdev, struct match *match,
             struct dpif_sflow_attr sflow_attr;
 
             memset(&sflow_attr, 0, sizeof sflow_attr);
-            gid_alloc_ctx(&sflow_attr);
+            if (flower.tunnel) {
+                sflow_attr.tunnel = CONST_CAST(struct flow_tnl *, tnl);
+            }
+            err = parse_sample_action(nla, &sflow_attr, action);
+            if (err) {
+                goto out;
+            }
+            group_id = gid_alloc_ctx(&sflow_attr);
+            action->sample.action_group_id = group_id;
+            flower.action_count++;
+        } else if (nl_attr_type(nla) == OVS_ACTION_ATTR_USERSPACE) {
+            struct dpif_sflow_attr sflow_attr;
+
+            /* If there is a sFlow cookie inside of a userspace attribute,
+             * but no sample attribute, that means sampling rate is 1.
+             */
+            memset(&sflow_attr, 0, sizeof sflow_attr);
+            if (flower.tunnel) {
+                sflow_attr.tunnel = CONST_CAST(struct flow_tnl *, tnl);
+            }
+            err = parse_userspace_userdata(nla, &sflow_attr);
+            if (err) {
+                goto out;
+            }
+            sflow_attr.sflow = nla;
+            sflow_attr.sflow_len = nla->nla_len;
+            group_id = gid_alloc_ctx(&sflow_attr);
+            action->type = TC_ACT_SAMPLE;
+            action->sample.action_group_id = group_id;
+            action->sample.action_rate = 1;
+            flower.action_count++;
         } else {
             VLOG_DBG_RL(&rl, "unsupported put action type: %d",
                         nl_attr_type(nla));
-            return EOPNOTSUPP;
+            err = EOPNOTSUPP;
+            goto out;
         }
     }
 
     if ((chain || recirc_act) && !info->recirc_id_shared_with_tc) {
         VLOG_ERR_RL(&error_rl, "flow_put: recirc_id sharing not supported");
-        return EOPNOTSUPP;
+        err = EOPNOTSUPP;
+        goto out;
     }
 
     if (get_ufid_tc_mapping(ufid, &id) == 0) {
@@ -2083,20 +2202,30 @@  netdev_tc_flow_put(struct netdev *netdev, struct match *match,
     prio = get_prio_for_tc_flower(&flower);
     if (prio == 0) {
         VLOG_ERR_RL(&rl, "couldn't get tc prio: %s", ovs_strerror(ENOSPC));
-        return ENOSPC;
+        err = ENOSPC;
+        goto out;
     }
 
     flower.act_cookie.data = ufid;
     flower.act_cookie.len = sizeof *ufid;
 
     block_id = get_block_id_from_netdev(netdev);
-    id = tc_make_tcf_id_chain(ifindex, block_id, chain, prio, hook);
+    id = tc_make_tcf_id_chain(ifindex, block_id, chain, prio, hook, group_id);
     err = tc_replace_flower(&id, &flower);
     if (!err) {
         if (stats) {
             memset(stats, 0, sizeof *stats);
         }
         add_ufid_tc_mapping(netdev, ufid, &id);
+    } else {
+        goto out;
+    }
+
+    return 0;
+
+out:
+    if (group_id) {
+        gid_free(group_id);
     }
 
     return err;
diff --git a/lib/tc.c b/lib/tc.c
index 8761304c9..2e31648fc 100644
--- a/lib/tc.c
+++ b/lib/tc.c
@@ -23,14 +23,15 @@ 
 #include <linux/if_packet.h>
 #include <linux/rtnetlink.h>
 #include <linux/tc_act/tc_csum.h>
+#include <linux/tc_act/tc_ct.h>
 #include <linux/tc_act/tc_gact.h>
 #include <linux/tc_act/tc_mirred.h>
 #include <linux/tc_act/tc_mpls.h>
 #include <linux/tc_act/tc_pedit.h>
+#include <linux/tc_act/tc_sample.h>
 #include <linux/tc_act/tc_skbedit.h>
 #include <linux/tc_act/tc_tunnel_key.h>
 #include <linux/tc_act/tc_vlan.h>
-#include <linux/tc_act/tc_ct.h>
 #include <linux/gen_stats.h>
 #include <net/if.h>
 #include <unistd.h>
@@ -1289,6 +1290,38 @@  nl_parse_act_gact(struct nlattr *options, struct tc_flower *flower)
     return 0;
 }
 
+static const struct nl_policy sample_policy[] = {
+    [TCA_SAMPLE_PARMS] = { .type = NL_A_UNSPEC,
+                           .min_len = sizeof(struct tc_sample),
+                           .optional = false, },
+    [TCA_SAMPLE_PSAMPLE_GROUP] = { .type = NL_A_U32,
+                                   .optional = false, },
+    [TCA_SAMPLE_RATE] = { .type = NL_A_U32,
+                          .optional = false, },
+};
+
+static int
+nl_parse_act_sample(struct nlattr *options, struct tc_flower *flower)
+{
+    struct nlattr *sample_attrs[ARRAY_SIZE(sample_policy)];
+    struct tc_action *action;
+
+    if (!nl_parse_nested(options, sample_policy, sample_attrs,
+                         ARRAY_SIZE(sample_policy))) {
+        VLOG_ERR_RL(&error_rl, "failed to parse sample action options");
+        return EPROTO;
+    }
+
+    action = &flower->actions[flower->action_count++];
+    action->type = TC_ACT_SAMPLE;
+    action->sample.action_group_id =
+        nl_attr_get_u32(sample_attrs[TCA_SAMPLE_PSAMPLE_GROUP]);
+    action->sample.action_rate =
+        nl_attr_get_u32(sample_attrs[TCA_SAMPLE_RATE]);
+
+    return 0;
+}
+
 static const struct nl_policy mirred_policy[] = {
     [TCA_MIRRED_PARMS] = { .type = NL_A_UNSPEC,
                            .min_len = sizeof(struct tc_mirred),
@@ -1697,6 +1730,8 @@  nl_parse_single_action(struct nlattr *action, struct tc_flower *flower,
         /* Added for TC rule only (not in OvS rule) so ignore. */
     } else if (!strcmp(act_kind, "ct")) {
         nl_parse_act_ct(act_options, flower);
+    } else if (!strcmp(act_kind, "sample")) {
+        nl_parse_act_sample(act_options, flower);
     } else {
         VLOG_ERR_RL(&error_rl, "unknown tc action kind: %s", act_kind);
         err = EINVAL;
@@ -2292,6 +2327,23 @@  nl_msg_put_act_mirred(struct ofpbuf *request, int ifindex, int action,
     nl_msg_end_nested(request, offset);
 }
 
+static void
+nl_msg_put_act_sample(struct ofpbuf *request, uint32_t rate, uint32_t group_id)
+{
+    size_t offset;
+
+    nl_msg_put_string(request, TCA_ACT_KIND, "sample");
+    offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS | NLA_F_NESTED);
+    {
+        struct tc_sample parm = { .action = TC_ACT_PIPE };
+
+        nl_msg_put_unspec(request, TCA_SAMPLE_PARMS, &parm, sizeof parm);
+        nl_msg_put_u32(request, TCA_SAMPLE_RATE, rate);
+        nl_msg_put_u32(request, TCA_SAMPLE_PSAMPLE_GROUP, group_id);
+    }
+    nl_msg_end_nested(request, offset);
+}
+
 static inline void
 nl_msg_put_act_cookie(struct ofpbuf *request, struct tc_cookie *ck) {
     if (ck->len) {
@@ -2551,6 +2603,13 @@  nl_msg_put_flower_acts(struct ofpbuf *request, struct tc_flower *flower)
                 nl_msg_end_nested(request, act_offset);
             }
             break;
+            case TC_ACT_SAMPLE: {
+                act_offset = nl_msg_start_nested(request, act_index++);
+                nl_msg_put_act_sample(request, action->sample.action_rate,
+                                      action->sample.action_group_id);
+                nl_msg_end_nested(request, act_offset);
+            }
+            break;
             case TC_ACT_OUTPUT: {
                 if (!released && flower->tunnel) {
                     act_offset = nl_msg_start_nested(request, act_index++);
diff --git a/lib/tc.h b/lib/tc.h
index cc2ad025d..143e225d1 100644
--- a/lib/tc.h
+++ b/lib/tc.h
@@ -171,6 +171,7 @@  enum tc_action_type {
     TC_ACT_MPLS_SET,
     TC_ACT_GOTO,
     TC_ACT_CT,
+    TC_ACT_SAMPLE,
 };
 
 enum nat_type {
@@ -253,6 +254,11 @@  struct tc_action {
             bool force;
             bool commit;
         } ct;
+
+        struct {
+            uint32_t action_rate;
+            uint32_t action_group_id;
+        } sample;
      };
 
      enum tc_action_type type;
@@ -292,11 +298,12 @@  tc_make_tcf_id(int ifindex, uint32_t block_id, uint16_t prio,
 
 static inline struct tcf_id
 tc_make_tcf_id_chain(int ifindex, uint32_t block_id, uint32_t chain,
-                     uint16_t prio, enum tc_qdisc_hook hook)
+                     uint16_t prio, enum tc_qdisc_hook hook, uint32_t group_id)
 {
     struct tcf_id id = tc_make_tcf_id(ifindex, block_id, prio, hook);
 
     id.chain = chain;
+    id.sflow_group_id = group_id;
 
     return id;
 }