add port-based ingress policing based packet-per-second rate-limiting

OVS has support for using policing to enforce a rate limit in kilobits per second. This is configured using OVSDB. f.e. $ ovs-vsctl set interface tap0 ingress_policing_rate=1000 $ ovs-vsctl set interface tap0 ingress_policing_burst=100 This patch adds a related feature, allowing policing to enforce a rate limit in kilo-packets per second. This is also configured using OVSDB. $ ovs-vsctl set interface tap0 ingress_policing_kpkts_rate=1000 $ ovs-vsctl set interface tap0 ingress_policing_kpkts_burst=100 The kilo-bit and kilo-packet rate limits may be used separately or in combination. Add separate action for BPS and PPS in netlink message. Revise code and change action result to pipe to allow traffic pipe into second action. This patch implements the feature for: * OVSDB (northbound API) * TC policer when used both with and without TC offload (kernel API) Signed-off-by: Yong Xu <yong.xu@corigine.com> Signed-off-by: Simon Horman <simon.horman@netronome.com>
2025-08-31 06:15:47 +00:00 · 2021-06-09 11:52:08 +02:00
parent b6c5f30cfa
commit c2567e533f
11 changed files with 179 additions and 54 deletions
--- a/acinclude.m4
+++ b/acinclude.m4
@@ -209,10 +209,10 @@ dnl Configure Linux tc compat.
 AC_DEFUN([OVS_CHECK_LINUX_TC], [
  AC_COMPILE_IFELSE([
    AC_LANG_PROGRAM([#include <linux/pkt_cls.h>], [
-        int x = TCA_FLOWER_KEY_CT_FLAGS_REPLY;
+        int x = TCA_POLICE_PKTRATE64;
    ])],
-    [AC_DEFINE([HAVE_TCA_FLOWER_KEY_CT_FLAGS_REPLY], [1],
-               [Define to 1 if TCA_FLOWER_KEY_CT_FLAGS_REPLY is available.])])
+    [AC_DEFINE([HAVE_TCA_POLICE_PKTRATE64], [1],
+               [Define to 1 if TCA_POLICE_PKTRATE64 is available.])])

  AC_CHECK_MEMBERS([struct tcf_t.firstuse], [], [], [#include <linux/pkt_cls.h>])

--- a/include/linux/pkt_cls.h
+++ b/include/linux/pkt_cls.h
@@ -1,7 +1,7 @@
 #ifndef __LINUX_PKT_CLS_WRAPPER_H
 #define __LINUX_PKT_CLS_WRAPPER_H 1

-#if defined(__KERNEL__) || defined(HAVE_TCA_FLOWER_KEY_CT_FLAGS_REPLY)
+#if defined(__KERNEL__) || defined(HAVE_TCA_POLICE_PKTRATE64)
 #include_next <linux/pkt_cls.h>
 #else

@@ -104,6 +104,10 @@ enum {
 	TCA_POLICE_RESULT,
 	TCA_POLICE_TM,
 	TCA_POLICE_PAD,
+	TCA_POLICE_RATE64,
+	TCA_POLICE_PEAKRATE64,
+	TCA_POLICE_PKTRATE64,
+	TCA_POLICE_PKTBURST64,
 	__TCA_POLICE_MAX
 #define TCA_POLICE_RESULT TCA_POLICE_RESULT
 };
--- a/lib/netdev-dpdk.c
+++ b/lib/netdev-dpdk.c
@@ -3389,7 +3389,9 @@ netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst)

 static int
 netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate,
-                         uint32_t policer_burst)
+                         uint32_t policer_burst,
+                         uint32_t policer_kpkts_rate OVS_UNUSED,
+                         uint32_t policer_kpkts_burst OVS_UNUSED)
 {
    struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
    struct ingress_policer *policer;
--- a/lib/netdev-linux-private.h
+++ b/lib/netdev-linux-private.h
@@ -75,8 +75,10 @@ struct netdev_linux {
    int mtu;
    unsigned int ifi_flags;
    long long int carrier_resets;
-    uint32_t kbits_rate;        /* Policing data. */
+    uint32_t kbits_rate;        /* Policing data - kbps */
    uint32_t kbits_burst;
+    uint32_t kpkts_rate;        /* Policing data - kpps */
+    uint32_t kpkts_burst;
    int vport_stats_error;      /* Cached error code from vport_get_stats().
                                   0 or an errno value. */
    int netdev_mtu_error;       /* Cached error code from SIOCGIFMTU
--- a/lib/netdev-linux.c
+++ b/lib/netdev-linux.c
@@ -485,8 +485,10 @@ static struct tcmsg *netdev_linux_tc_make_request(const struct netdev *,
                                                  int type,
                                                  unsigned int flags,
                                                  struct ofpbuf *);
-static int tc_add_policer(struct netdev *,
-                          uint32_t kbits_rate, uint32_t kbits_burst);
+
+static int tc_add_policer(struct netdev *, uint32_t kbits_rate,
+                          uint32_t kbits_burst, uint32_t kpkts_rate,
+                          uint32_t kpkts_burst);

 static int tc_parse_qdisc(const struct ofpbuf *, const char **kind,
                          struct nlattr **options);
@@ -2595,24 +2597,62 @@ tc_matchall_fill_police(uint32_t kbits_rate, uint32_t kbits_burst)
 }

 static void
-nl_msg_put_act_police(struct ofpbuf *request, struct tc_police police)
+nl_msg_act_police_start_nest(struct ofpbuf *request, uint32_t prio,
+                             size_t *offset, size_t *act_offset)
 {
-    size_t offset;
-
+    *act_offset = nl_msg_start_nested(request, prio);
    nl_msg_put_string(request, TCA_ACT_KIND, "police");
-    offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
-    nl_msg_put_unspec(request, TCA_POLICE_TBF, &police, sizeof police);
-    tc_put_rtab(request, TCA_POLICE_RATE, &police.rate);
-    nl_msg_put_u32(request, TCA_POLICE_RESULT, TC_ACT_UNSPEC);
+    *offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
+}
+
+static void
+nl_msg_act_police_end_nest(struct ofpbuf *request, size_t offset,
+                           size_t act_offset)
+{
+    nl_msg_put_u32(request, TCA_POLICE_RESULT, TC_ACT_PIPE);
    nl_msg_end_nested(request, offset);
+    nl_msg_end_nested(request, act_offset);
+}
+
+static void
+nl_msg_put_act_police(struct ofpbuf *request, struct tc_police police,
+                      uint32_t kpkts_rate, uint32_t kpkts_burst)
+{
+    size_t offset, act_offset;
+    uint32_t prio = 0;
+    /* used for PPS, set rate as 0 to act as a single action */
+    struct tc_police null_police = {0};
+
+    if (police.rate.rate) {
+        nl_msg_act_police_start_nest(request, ++prio, &offset, &act_offset);
+        tc_put_rtab(request, TCA_POLICE_RATE, &police.rate);
+        nl_msg_put_unspec(request, TCA_POLICE_TBF, &police, sizeof police);
+        nl_msg_act_police_end_nest(request, offset, act_offset);
+    }
+    if (kpkts_rate) {
+        unsigned int pkt_burst_ticks, pps_rate, size;
+        nl_msg_act_police_start_nest(request, ++prio, &offset, &act_offset);
+        pps_rate = kpkts_rate * 1000;
+        size = MIN(UINT32_MAX / 1000, kpkts_burst) * 1000;
+        /* Here tc_bytes_to_ticks is used to convert packets rather than bytes
+           to ticks. */
+        pkt_burst_ticks = tc_bytes_to_ticks(pps_rate, size);
+        nl_msg_put_u64(request, TCA_POLICE_PKTRATE64, (uint64_t) pps_rate);
+        nl_msg_put_u64(request, TCA_POLICE_PKTBURST64,
+                       (uint64_t) pkt_burst_ticks);
+        nl_msg_put_unspec(request, TCA_POLICE_TBF, &null_police,
+                          sizeof null_police);
+        nl_msg_act_police_end_nest(request, offset, act_offset);
+    }
 }

 static int
 tc_add_matchall_policer(struct netdev *netdev, uint32_t kbits_rate,
-                        uint32_t kbits_burst)
+                        uint32_t kbits_burst, uint32_t kpkts_rate,
+                        uint32_t kpkts_burst)
 {
    uint16_t eth_type = (OVS_FORCE uint16_t) htons(ETH_P_ALL);
-    size_t basic_offset, action_offset, inner_offset;
+    size_t basic_offset, action_offset;
    uint16_t prio = TC_RESERVED_PRIORITY_POLICE;
    int ifindex, err = 0;
    struct tc_police pol_act;
@@ -2636,9 +2676,7 @@ tc_add_matchall_policer(struct netdev *netdev, uint32_t kbits_rate,
    nl_msg_put_string(&request, TCA_KIND, "matchall");
    basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
    action_offset = nl_msg_start_nested(&request, TCA_MATCHALL_ACT);
-    inner_offset = nl_msg_start_nested(&request, 1);
-    nl_msg_put_act_police(&request, pol_act);
-    nl_msg_end_nested(&request, inner_offset);
+    nl_msg_put_act_police(&request, pol_act, kpkts_rate, kpkts_burst);
    nl_msg_end_nested(&request, action_offset);
    nl_msg_end_nested(&request, basic_offset);

@@ -2678,8 +2716,9 @@ tc_del_matchall_policer(struct netdev *netdev)
 /* Attempts to set input rate limiting (policing) policy.  Returns 0 if
 * successful, otherwise a positive errno value. */
 static int
-netdev_linux_set_policing(struct netdev *netdev_,
-                          uint32_t kbits_rate, uint32_t kbits_burst)
+netdev_linux_set_policing(struct netdev *netdev_, uint32_t kbits_rate,
+                          uint32_t kbits_burst, uint32_t kpkts_rate,
+                          uint32_t kpkts_burst)
 {
    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
    const char *netdev_name = netdev_get_name(netdev_);
@@ -2690,6 +2729,10 @@ netdev_linux_set_policing(struct netdev *netdev_,
                   : !kbits_burst ? 8000 /* Default to 8000 kbits if 0. */
                   : kbits_burst);       /* Stick with user-specified value. */

+    kpkts_burst = (!kpkts_rate ? 0       /* Force to 0 if no rate specified. */
+                   : !kpkts_burst ? 16   /* Default to 16 kpkts if 0. */
+                   : kpkts_burst);       /* Stick with user-specified value. */
+
    ovs_mutex_lock(&netdev->mutex);
    if (netdev_linux_netnsid_is_remote(netdev)) {
        error = EOPNOTSUPP;
@@ -2699,7 +2742,9 @@ netdev_linux_set_policing(struct netdev *netdev_,
    if (netdev->cache_valid & VALID_POLICING) {
        error = netdev->netdev_policing_error;
        if (error || (netdev->kbits_rate == kbits_rate &&
-                      netdev->kbits_burst == kbits_burst)) {
+                      netdev->kpkts_rate == kpkts_rate &&
+                      netdev->kbits_burst == kbits_burst &&
+                      netdev->kpkts_burst == kpkts_burst)) {
            /* Assume that settings haven't changed since we last set them. */
            goto out;
        }
@@ -2711,8 +2756,9 @@ netdev_linux_set_policing(struct netdev *netdev_,
    /* Use matchall for policing when offloadling ovs with tc-flower. */
    if (netdev_is_flow_api_enabled()) {
        error = tc_del_matchall_policer(netdev_);
-        if (kbits_rate) {
-            error = tc_add_matchall_policer(netdev_, kbits_rate, kbits_burst);
+        if (kbits_rate || kpkts_rate) {
+            error = tc_add_matchall_policer(netdev_, kbits_rate, kbits_burst,
+                                            kpkts_rate, kpkts_burst);
        }
        ovs_mutex_unlock(&netdev->mutex);
        return error;
@@ -2731,7 +2777,7 @@ netdev_linux_set_policing(struct netdev *netdev_,
        goto out;
    }

-    if (kbits_rate) {
+    if (kbits_rate || kpkts_rate) {
        error = tc_add_del_qdisc(ifindex, true, 0, TC_INGRESS);
        if (error) {
            VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s",
@@ -2739,7 +2785,8 @@ netdev_linux_set_policing(struct netdev *netdev_,
            goto out;
        }

-        error = tc_add_policer(netdev_, kbits_rate, kbits_burst);
+        error = tc_add_policer(netdev_, kbits_rate, kbits_burst,
+                               kpkts_rate, kpkts_burst);
        if (error){
            VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s",
                    netdev_name, ovs_strerror(error));
@@ -2749,6 +2796,8 @@ netdev_linux_set_policing(struct netdev *netdev_,

    netdev->kbits_rate = kbits_rate;
    netdev->kbits_burst = kbits_burst;
+    netdev->kpkts_rate = kpkts_rate;
+    netdev->kpkts_burst = kpkts_burst;

 out:
    if (!error || error == ENODEV) {
@@ -5525,7 +5574,8 @@ netdev_linux_tc_make_request(const struct netdev *netdev, int type,
 }

 /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
- * of 'kbits_burst'.
+ * of 'kbits_burst', with a rate of 'kpkts_rate' and a burst size of
+ * 'kpkts_burst'.
 *
 * This function is equivalent to running:
 *     /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
@@ -5538,14 +5588,14 @@ netdev_linux_tc_make_request(const struct netdev *netdev, int type,
 * Returns 0 if successful, otherwise a positive errno value.
 */
 static int
-tc_add_policer(struct netdev *netdev,
-               uint32_t kbits_rate, uint32_t kbits_burst)
+tc_add_policer(struct netdev *netdev, uint32_t kbits_rate,
+               uint32_t kbits_burst, uint32_t kpkts_rate,
+               uint32_t kpkts_burst)
 {
+    size_t basic_offset, police_offset;
    struct tc_police tc_police;
    struct ofpbuf request;
    struct tcmsg *tcmsg;
-    size_t basic_offset;
-    size_t police_offset;
    int error;
    int mtu = 65535;

@@ -5563,7 +5613,6 @@ tc_add_policer(struct netdev *netdev,
     * tc's point of view.  Whatever. */
    tc_police.burst = tc_bytes_to_ticks(
        tc_police.rate.rate, MIN(UINT32_MAX / 1024, kbits_burst) * 1024 / 8);
-
    tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTFILTER,
                                         NLM_F_EXCL | NLM_F_CREATE, &request);
    if (!tcmsg) {
@@ -5572,12 +5621,10 @@ tc_add_policer(struct netdev *netdev,
    tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
    tcmsg->tcm_info = tc_make_handle(49,
                                     (OVS_FORCE uint16_t) htons(ETH_P_ALL));
-
    nl_msg_put_string(&request, TCA_KIND, "basic");
    basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
-    police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
-    nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
-    tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
+    police_offset = nl_msg_start_nested(&request, TCA_BASIC_ACT);
+    nl_msg_put_act_police(&request, tc_police, kpkts_rate, kpkts_burst);
    nl_msg_end_nested(&request, police_offset);
    nl_msg_end_nested(&request, basic_offset);

--- a/lib/netdev-provider.h
+++ b/lib/netdev-provider.h
@@ -514,13 +514,16 @@ struct netdev_class {
     * NETDEV_PT_LEGACY_L2. */
    enum netdev_pt_mode (*get_pt_mode)(const struct netdev *netdev);

-    /* Attempts to set input rate limiting (policing) policy, such that up to
-     * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative
-     * burst size of 'kbits' kb.
+    /* Attempts to set input rate limiting (policing) policy, such that:
+     * - up to 'kbits_rate' kbps of traffic is accepted, with a maximum
+     *   accumulative burst size of 'kbits' kb; and
+     * - up to 'kpkts' kpps of traffic is accepted, with a maximum
+     *   accumulative burst size of 'kpkts' kilo packets.
     *
     * This function may be set to null if policing is not supported. */
    int (*set_policing)(struct netdev *netdev, unsigned int kbits_rate,
-                        unsigned int kbits_burst);
+                        unsigned int kbits_burst, unsigned int kpkts_rate,
+                        unsigned int kpkts_burst);

    /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves
     * it empty if 'netdev' does not support QoS.  Any names added to 'types'
--- a/lib/netdev.c
+++ b/lib/netdev.c
@@ -1619,17 +1619,20 @@ netdev_get_custom_stats(const struct netdev *netdev,
    return error;
 }

-
-/* Attempts to set input rate limiting (policing) policy, such that up to
- * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
- * size of 'kbits' kb. */
+/* Attempts to set input rate limiting (policing) policy, such that:
+ * - up to 'kbits_rate' kbps of traffic is accepted, with a maximum
+ *   accumulative burst size of 'kbits' kb; and
+ * - up to 'kpkts' kpps of traffic is accepted, with a maximum
+ *   accumulative burst size of 'kpkts' kilo packets.
+ */
 int
 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
-                    uint32_t kbits_burst)
+                    uint32_t kbits_burst, uint32_t kpkts_rate,
+                    uint32_t kpkts_burst)
 {
    return (netdev->netdev_class->set_policing
            ? netdev->netdev_class->set_policing(netdev,
-                    kbits_rate, kbits_burst)
+                    kbits_rate, kbits_burst, kpkts_rate, kpkts_burst)
            : EOPNOTSUPP);
 }

--- a/lib/netdev.h
+++ b/lib/netdev.h
@@ -289,7 +289,8 @@ struct netdev_queue_stats {
 };

 int netdev_set_policing(struct netdev *, uint32_t kbits_rate,
-                        uint32_t kbits_burst);
+                        uint32_t kbits_burst, uint32_t kpkts_rate,
+                        uint32_t kpkts_burst);

 int netdev_get_qos_types(const struct netdev *, struct sset *types);
 int netdev_get_qos_capabilities(const struct netdev *,
--- a/vswitchd/bridge.c
+++ b/vswitchd/bridge.c
@@ -4895,8 +4895,10 @@ iface_configure_qos(struct iface *iface, const struct ovsrec_qos *qos)
    }

    netdev_set_policing(iface->netdev,
-                        MIN(UINT32_MAX, iface->cfg->ingress_policing_rate),
-                        MIN(UINT32_MAX, iface->cfg->ingress_policing_burst));
+                MIN(UINT32_MAX, iface->cfg->ingress_policing_rate),
+                MIN(UINT32_MAX, iface->cfg->ingress_policing_burst),
+                MIN(UINT32_MAX, iface->cfg->ingress_policing_kpkts_rate),
+                MIN(UINT32_MAX, iface->cfg->ingress_policing_kpkts_burst));

    ofpbuf_uninit(&queues_buf);
 }
--- a/vswitchd/vswitch.ovsschema
+++ b/vswitchd/vswitch.ovsschema
@@ -1,6 +1,6 @@
 {"name": "Open_vSwitch",
- "version": "8.2.0",
- "cksum": "1076640191 26427",
+ "version": "8.3.0",
+ "cksum": "3781850481 26690",
 "tables": {
   "Open_vSwitch": {
     "columns": {
@@ -242,6 +242,12 @@
       "ingress_policing_burst": {
         "type": {"key": {"type": "integer",
                          "minInteger": 0}}},
+       "ingress_policing_kpkts_rate": {
+         "type": {"key": {"type": "integer",
+                          "minInteger": 0}}},
+       "ingress_policing_kpkts_burst": {
+         "type": {"key": {"type": "integer",
+                          "minInteger": 0}}},
       "mac_in_use": {
         "type": {"key": {"type": "string"},
                  "min": 0, "max": 1},
--- a/vswitchd/vswitch.xml
+++ b/vswitchd/vswitch.xml
@@ -3654,8 +3654,18 @@ ovs-vsctl add-port br0 p0 -- set Interface p0 type=patch options:peer=p1 \
        table="Queue"/> tables).
      </p>
      <p>
-        Policing is currently implemented on Linux and OVS with DPDK. Both
-        implementations use a simple ``token bucket'' approach:
+        Policing settings can be set with byte rate or packet rate, and they
+        can be configured together, in which case they take effect together,
+        that means the smaller speed limit of them is in effect.
+      </p>
+      <p>
+        Currently, byte rate policing is implemented on Linux and OVS with
+        DPDK, while packet rate policing is only implemented on Linux.  Both
+        Linux and OVS DPDK implementations use a simple ``token bucket''
+        approach.
+      </p>
+      <p>
+        Byte rate policing:
      </p>
      <ul>
        <li>
@@ -3673,6 +3683,26 @@ ovs-vsctl add-port br0 p0 -- set Interface p0 type=patch options:peer=p1 \
          rate specified by <ref column="ingress_policing_rate"/>.
        </li>
      </ul>
+      <p>
+        Packet rate policing:
+      </p>
+      <ul>
+        <li>
+          The size of the bucket corresponds to <ref
+          column="ingress_policing_kpkts_burst"/>.  Initially the bucket is
+          full.
+        </li>
+        <li>
+          Whenever a packet is received, it will consume one token from the
+          current bucket.  If the token is available in the bucket, it's
+          removed and the packet is forwarded.  Otherwise, the packet is
+          dropped.
+        </li>
+        <li>
+          Whenever it is not full, the bucket is refilled with tokens at the
+          rate specified by <ref column="ingress_policing_kpkts_rate"/>.
+        </li>
+      </ul>
      <p>
        Policing interacts badly with some network protocols, and especially
        with fragmented IP packets.  Suppose that there is enough network
@@ -3698,6 +3728,14 @@ ovs-vsctl add-port br0 p0 -- set Interface p0 type=patch options:peer=p1 \
        </p>
      </column>

+      <column name="ingress_policing_kpkts_rate">
+        <p>
+          Maximum rate for data received on this interface, in kpps (1 kpps is
+          1000 pps).  Data received faster than this rate is dropped.  Set to
+          <code>0</code> (the default) to disable policing.
+        </p>
+      </column>
+
      <column name="ingress_policing_burst">
        <p>Maximum burst size for data received on this interface, in kb.  The
        default burst size if set to <code>0</code> is 8000 kbit.  This value
@@ -3712,6 +3750,23 @@ ovs-vsctl add-port br0 p0 -- set Interface p0 type=patch options:peer=p1 \
          closer to achieving the full rate.
        </p>
      </column>
+
+      <column name="ingress_policing_kpkts_burst">
+        <p>
+          Maximum burst size for data received on this interface, in kpkts (1
+          kpkts is 1000 packets).  The default burst size if set to
+          <code>0</code> is 16 kpkts.  This value has no effect if
+          <ref column="ingress_policing_kpkts_rate"/> is <code>0</code>.
+        </p>
+        <p>
+          Specifying a larger burst size lets the algorithm be more forgiving,
+          which is important for protocols like TCP that react severely to
+          dropped packets.  Specifying a value that is numerically at least as
+          large as 80% of <ref column="ingress_policing_kpkts_rate"/> helps TCP
+          come closer to achieving the full rate.
+        </p>
+      </column>
+
    </group>

    <group title="Bidirectional Forwarding Detection (BFD)">