Support for nexthop group statistics

PATCH Superseded Petr Machata 4 participants 23 messages Feb 29, 2024 – Mar 4, 2024

Subsystems (3)

Networking Network drivers IPv4/IPv6

lore

Share & tools

lore

Petr Machata

author

2024-02-29 18:20 UTC #0 lore raw

v2 cover Support for nexthop group statistics

ECMP is a fundamental component in L3 designs. However, it's fragile. Many factors influence whether an ECMP group will operate as intended: hash policy (i.e. the set of fields that contribute to ECMP hash calculation), neighbor validity, hash seed (which might lead to polarization) or the type of ECMP group used (hash-threshold or resilient).

At the same time, collection of statistics that would help an operator determine that the group performs as desired, is difficult.

A solution that we present in this patchset is to add counters to next hop group entries. For SW-datapath deployments, this will on its own allow collection and evaluation of relevant statistics. For HW-datapath deployments, we further add a way to request that HW counters be installed for a given group, in-kernel interfaces to collect the HW statistics, and netlink interfaces to query them.

For example:

    # ip nexthop replace id 4000 group 4001/4002 hw_stats on

    # ip -s -d nexthop show id 4000
    id 4000 group 4001/4002 scope global proto unspec offload hw_stats on used on
      stats:
        id 4001 packets 5002 packets_hw 5000
        id 4002 packets 4999 packets_hw 4999

The point of the patchset is visibility of ECMP balance, and that is influenced by packet headers, not their payload. Correspondingly, we only include packet counters in the statistics, not byte counters.

We also decided to model HW statistics as a nexthop group attribute, not an arbitrary nexthop one. The latter would count any traffic going through a given nexthop, regardless of which ECMP group it is in, or any at all. The reason is again hat the point of the patchset is ECMP balance visibility, not arbitrary inspection of how busy a particular nexthop is. Implementation of individual-nexthop statistics is certainly possible, and could well follow the general approach we are taking in this patchset. For resilient groups, per-bucket statistics could be done in a similar manner as well.

This patchset contains the core code. mlxsw support will be sent in a follow-up patch set.

This patchset progresses as follows:

- Patches #1 and #2 add support for a new next-hop object attribute, NHA_OP_FLAGS. That is meant to carry various op-specific signaling, in particular whether SW- and HW-collected nexthop stats should be part of the get or dump response. The idea is to avoid wasting message space, and time for collection of HW statistics, when the values are not needed.

- Patches #3 and #4 add SW-datapath stats and corresponding UAPI.

- Patches #5, #6 and #7 add support fro HW-datapath stats and UAPI. Individual drivers still need to contribute the appropriate HW-specific support code.

v2:
- Patch #2:
- Change OP_FLAGS to u32, enforce through NLA_POLICY_MASK
- Patch #3:
- Set err on nexthop_create_group() error path
- Patch #4:
- Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS
- Rename jump target in nla_put_nh_group_stats() to avoid having to rename further in the patchset.
- Patch #7:
- Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS_HW
- Do not cancel outside of nesting in nla_put_nh_group_stats()

Ido Schimmel (5):
net: nexthop: Add nexthop group entry stats
net: nexthop: Expose nexthop group stats to user space
net: nexthop: Add hardware statistics notifications
net: nexthop: Add ability to enable / disable hardware statistics
net: nexthop: Expose nexthop group HW stats to user space

Petr Machata (2):
net: nexthop: Adjust netlink policy parsing for a new attribute
net: nexthop: Add NHA_OP_FLAGS

 include/net/nexthop.h        |  29 ++++
 include/uapi/linux/nexthop.h |  45 +++++
 net/ipv4/nexthop.c           | 317 ++++++++++++++++++++++++++++++-----
 3 files changed, 351 insertions(+), 40 deletions(-)

Signature · 1 line

2.43.0

Petr Machata

author

2024-02-29 18:20 UTC #1 lore raw

PATCH v2 1/7 net: nexthop: Adjust netlink policy parsing for a new attribute

A following patch will introduce a new attribute, op-specific flags to adjust the behavior of an operation. Different operations will recognize different flags.

- To make the differentiation possible, stop sharing the policies for get and del operations.

- To allow querying for presence of the attribute, have all the attribute arrays sized to NHA_MAX, regardless of what is permitted by policy, and pass the corresponding value to nlmsg_parse() as well.

Signed-off-by Petr Machata <[email protected]>

---
 net/ipv4/nexthop.c | 58 ++++++++++++++++++++++------------------------
 1 file changed, 28 insertions(+), 30 deletions(-)

net/ipv4/nexthop.c +28 ~~-30~~

index 70509da4f080..bcd4df2f1cad 100644
@@ -43,6 +43,10 @@ static const struct nla_policy rtm_nh_policy_get[] = {
 	[NHA_ID]		= { .type = NLA_U32 },
 };
 
+static const struct nla_policy rtm_nh_policy_del[] = {
+	[NHA_ID]		= { .type = NLA_U32 },
+};
+
 static const struct nla_policy rtm_nh_policy_dump[] = {
 	[NHA_OIF]		= { .type = NLA_U32 },
 	[NHA_GROUPS]		= { .type = NLA_FLAG },

@@ -2966,9 +2970,9 @@ static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
 	return err;
 }
 
-static int __nh_valid_get_del_req(const struct nlmsghdr *nlh,
-				  struct nlattr **tb, u32 *id,
-				  struct netlink_ext_ack *extack)
+static int nh_valid_get_del_req(const struct nlmsghdr *nlh,
+				struct nlattr **tb, u32 *id,
+				struct netlink_ext_ack *extack)
 {
 	struct nhmsg *nhm = nlmsg_data(nlh);

@@ -2991,26 +2995,12 @@ static int __nh_valid_get_del_req(const struct nlmsghdr *nlh,
 	return 0;
 }
 
-static int nh_valid_get_del_req(const struct nlmsghdr *nlh, u32 *id,
-				struct netlink_ext_ack *extack)
-{
-	struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get)];
-	int err;
-
-	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
-			  ARRAY_SIZE(rtm_nh_policy_get) - 1,
-			  rtm_nh_policy_get, extack);
-	if (err < 0)
-		return err;
-
-	return __nh_valid_get_del_req(nlh, tb, id, extack);
-}
-
 /* rtnl */
 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
 			   struct netlink_ext_ack *extack)
 {
 	struct net *net = sock_net(skb->sk);
+	struct nlattr *tb[NHA_MAX + 1];
 	struct nl_info nlinfo = {
 		.nlh = nlh,
 		.nl_net = net,

@@ -3020,7 +3010,12 @@ static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
 	int err;
 	u32 id;
 
-	err = nh_valid_get_del_req(nlh, &id, extack);
+	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, NHA_MAX,
+			  rtm_nh_policy_del, extack);
+	if (err < 0)
+		return err;
+
+	err = nh_valid_get_del_req(nlh, tb, &id, extack);
 	if (err)
 		return err;

@@ -3038,12 +3033,18 @@ static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
 			   struct netlink_ext_ack *extack)
 {
 	struct net *net = sock_net(in_skb->sk);
+	struct nlattr *tb[NHA_MAX + 1];
 	struct sk_buff *skb = NULL;
 	struct nexthop *nh;
 	int err;
 	u32 id;
 
-	err = nh_valid_get_del_req(nlh, &id, extack);
+	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, NHA_MAX,
+			  rtm_nh_policy_get, extack);
+	if (err < 0)
+		return err;
+
+	err = nh_valid_get_del_req(nlh, tb, &id, extack);
 	if (err)
 		return err;

@@ -3157,11 +3158,10 @@ static int nh_valid_dump_req(const struct nlmsghdr *nlh,
 			     struct nh_dump_filter *filter,
 			     struct netlink_callback *cb)
 {
-	struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump)];
+	struct nlattr *tb[NHA_MAX + 1];
 	int err;
 
-	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
-			  ARRAY_SIZE(rtm_nh_policy_dump) - 1,
+	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, NHA_MAX,
 			  rtm_nh_policy_dump, cb->extack);
 	if (err < 0)
 		return err;

@@ -3300,11 +3300,10 @@ static int nh_valid_dump_bucket_req(const struct nlmsghdr *nlh,
 				    struct netlink_callback *cb)
 {
 	struct nlattr *res_tb[ARRAY_SIZE(rtm_nh_res_bucket_policy_dump)];
-	struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump_bucket)];
+	struct nlattr *tb[NHA_MAX + 1];
 	int err;
 
-	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
-			  ARRAY_SIZE(rtm_nh_policy_dump_bucket) - 1,
+	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, NHA_MAX,
 			  rtm_nh_policy_dump_bucket, NULL);
 	if (err < 0)
 		return err;

@@ -3474,16 +3473,15 @@ static int nh_valid_get_bucket_req(const struct nlmsghdr *nlh,
 				   u32 *id, u16 *bucket_index,
 				   struct netlink_ext_ack *extack)
 {
-	struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get_bucket)];
+	struct nlattr *tb[NHA_MAX + 1];
 	int err;
 
-	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
-			  ARRAY_SIZE(rtm_nh_policy_get_bucket) - 1,
+	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, NHA_MAX,
 			  rtm_nh_policy_get_bucket, extack);
 	if (err < 0)
 		return err;
 
-	err = __nh_valid_get_del_req(nlh, tb, id, extack);
+	err = nh_valid_get_del_req(nlh, tb, id, extack);
 	if (err)
 		return err;

Signature · 1 line

2.43.0

David Ahern

maintainer Reviewed-by on v2 1/7

2024-03-01 15:31 UTC #2 lore raw

Re: [PATCH net-next v2 1/7] net: nexthop: Adjust netlink policy parsing for a new attribute

Quoted 7 lines · Petr Machata

A following patch will introduce a new attribute, op-specific flags to adjust the behavior of an operation. Different operations will recognize different flags.
- To make the differentiation possible, stop sharing the policies for get and del operations.

Quoted 9 lines

- To allow querying for presence of the attribute, have all the attribute arrays sized to NHA_MAX, regardless of what is permitted by policy, and pass the corresponding value to nlmsg_parse() as well.
Signed-off-by Petr Machata <[email protected]>
---
 net/ipv4/nexthop.c | 58 ++++++++++++++++++++++------------------------
 1 file changed, 28 insertions(+), 30 deletions(-)

Reviewed-by David Ahern <[email protected]>

Petr Machata

author

2024-02-29 18:20 UTC #3 lore raw

PATCH v2 2/7 net: nexthop: Add NHA_OP_FLAGS

In reply to Petr Machata

In order to add per-nexthop statistics, but still not increase netlink message size for consumers that do not care about them, there needs to be a toggle through which the user indicates their desire to get the statistics. To that end, add a new attribute, NHA_OP_FLAGS. The idea is to be able to use the attribute for carrying of arbitrary operation-specific flags, i.e. not make it specific for get / dump.

Add the new attribute to get and dump policies, but do not actually allow any flags yet -- those will come later as the flags themselves are defined. Add the necessary parsing code.

Signed-off-by Petr Machata <[email protected]>

---

Notes:
v2:
- Change OP_FLAGS to u32, enforce through NLA_POLICY_MASK

 include/uapi/linux/nexthop.h |  3 +++
 net/ipv4/nexthop.c           | 24 ++++++++++++++++++++----
 2 files changed, 23 insertions(+), 4 deletions(-)

include/uapi/linux/nexthop.h +3 -0

index d8ffa8c9ca78..086444e2946c 100644
@@ -60,6 +60,9 @@ enum {
 	/* nested; nexthop bucket attributes */
 	NHA_RES_BUCKET,
 
+	/* u32; operation-specific flags */
+	NHA_OP_FLAGS,
+
 	__NHA_MAX,
 };

net/ipv4/nexthop.c +20 -4

index bcd4df2f1cad..816ae8ee3e06 100644
@@ -41,6 +41,7 @@ static const struct nla_policy rtm_nh_policy_new[] = {
 
 static const struct nla_policy rtm_nh_policy_get[] = {
 	[NHA_ID]		= { .type = NLA_U32 },
+	[NHA_OP_FLAGS]		= NLA_POLICY_MASK(NLA_U32, 0),
 };
 
 static const struct nla_policy rtm_nh_policy_del[] = {

@@ -52,6 +53,7 @@ static const struct nla_policy rtm_nh_policy_dump[] = {
 	[NHA_GROUPS]		= { .type = NLA_FLAG },
 	[NHA_MASTER]		= { .type = NLA_U32 },
 	[NHA_FDB]		= { .type = NLA_FLAG },
+	[NHA_OP_FLAGS]		= NLA_POLICY_MASK(NLA_U32, 0),
 };
 
 static const struct nla_policy rtm_nh_res_policy_new[] = {

@@ -2971,7 +2973,7 @@ static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
 }
 
 static int nh_valid_get_del_req(const struct nlmsghdr *nlh,
-				struct nlattr **tb, u32 *id,
+				struct nlattr **tb, u32 *id, u32 *op_flags,
 				struct netlink_ext_ack *extack)
 {
 	struct nhmsg *nhm = nlmsg_data(nlh);

@@ -2992,6 +2994,11 @@ static int nh_valid_get_del_req(const struct nlmsghdr *nlh,
 		return -EINVAL;
 	}
 
+	if (tb[NHA_OP_FLAGS])
+		*op_flags = nla_get_u32(tb[NHA_OP_FLAGS]);
+	else
+		*op_flags = 0;
+
 	return 0;
 }

@@ -3007,6 +3014,7 @@ static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
 		.portid = NETLINK_CB(skb).portid,
 	};
 	struct nexthop *nh;
+	u32 op_flags;
 	int err;
 	u32 id;

@@ -3015,7 +3023,7 @@ static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
 	if (err < 0)
 		return err;
 
-	err = nh_valid_get_del_req(nlh, tb, &id, extack);
+	err = nh_valid_get_del_req(nlh, tb, &id, &op_flags, extack);
 	if (err)
 		return err;

@@ -3036,6 +3044,7 @@ static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
 	struct nlattr *tb[NHA_MAX + 1];
 	struct sk_buff *skb = NULL;
 	struct nexthop *nh;
+	u32 op_flags;
 	int err;
 	u32 id;

@@ -3044,7 +3053,7 @@ static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
 	if (err < 0)
 		return err;
 
-	err = nh_valid_get_del_req(nlh, tb, &id, extack);
+	err = nh_valid_get_del_req(nlh, tb, &id, &op_flags, extack);
 	if (err)
 		return err;

@@ -3080,6 +3089,7 @@ struct nh_dump_filter {
 	bool group_filter;
 	bool fdb_filter;
 	u32 res_bucket_nh_id;
+	u32 op_flags;
 };
 
 static bool nh_dump_filtered(struct nexthop *nh,

@@ -3151,6 +3161,11 @@ static int __nh_valid_dump_req(const struct nlmsghdr *nlh, struct nlattr **tb,
 		return -EINVAL;
 	}
 
+	if (tb[NHA_OP_FLAGS])
+		filter->op_flags = nla_get_u32(tb[NHA_OP_FLAGS]);
+	else
+		filter->op_flags = 0;
+
 	return 0;
 }

@@ -3474,6 +3489,7 @@ static int nh_valid_get_bucket_req(const struct nlmsghdr *nlh,
 				   struct netlink_ext_ack *extack)
 {
 	struct nlattr *tb[NHA_MAX + 1];
+	u32 op_flags;
 	int err;
 
 	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, NHA_MAX,

@@ -3481,7 +3497,7 @@ static int nh_valid_get_bucket_req(const struct nlmsghdr *nlh,
 	if (err < 0)
 		return err;
 
-	err = nh_valid_get_del_req(nlh, tb, id, extack);
+	err = nh_valid_get_del_req(nlh, tb, id, &op_flags, extack);
 	if (err)
 		return err;

Signature · 1 line

2.43.0

David Ahern

maintainer Reviewed-by on v2 2/7

2024-03-01 15:36 UTC #4 lore raw

Re: [PATCH net-next v2 2/7] net: nexthop: Add NHA_OP_FLAGS

Quoted 14 lines · Petr Machata

In order to add per-nexthop statistics, but still not increase netlink message size for consumers that do not care about them, there needs to be a toggle through which the user indicates their desire to get the statistics. To that end, add a new attribute, NHA_OP_FLAGS. The idea is to be able to use the attribute for carrying of arbitrary operation-specific flags, i.e. not make it specific for get / dump.
Add the new attribute to get and dump policies, but do not actually allow any flags yet -- those will come later as the flags themselves are defined. Add the necessary parsing code.
Signed-off-by Petr Machata <[email protected]>
---

Quoted 8 lines

Notes:
v2:
- Change OP_FLAGS to u32, enforce through NLA_POLICY_MASK

 include/uapi/linux/nexthop.h |  3 +++
 net/ipv4/nexthop.c           | 24 ++++++++++++++++++++----
 2 files changed, 23 insertions(+), 4 deletions(-)

Reviewed-by David Ahern <[email protected]>

Petr Machata

author

2024-02-29 18:20 UTC #5 lore raw

PATCH v2 3/7 net: nexthop: Add nexthop group entry stats

In reply to Petr Machata

From: Ido Schimmel <[email protected]>

Add nexthop group entry stats to count the number of packets forwarded via each nexthop in the group. The stats will be exposed to user space for better data path observability in the next patch.

The per-CPU stats pointer is placed at the beginning of 'struct nh_grp_entry', so that all the fields accessed for the data path reside on the same cache line:

struct nh_grp_entry {
        struct nexthop *           nh;                   /*     0     8 */
        struct nh_grp_entry_stats * stats;               /*     8     8 */
        u8                         weight;               /*    16     1 */

        /* XXX 7 bytes hole, try to pack */

        union {
                struct {
                        atomic_t   upper_bound;          /*    24     4 */
                } hthr;                                  /*    24     4 */
                struct {
                        struct list_head uw_nh_entry;    /*    24    16 */
                        u16        count_buckets;        /*    40     2 */
                        u16        wants_buckets;        /*    42     2 */
                } res;                                   /*    24    24 */
        };                                               /*    24    24 */
        struct list_head           nh_list;              /*    48    16 */
        /* --- cacheline 1 boundary (64 bytes) --- */
        struct nexthop *           nh_parent;            /*    64     8 */

        /* size: 72, cachelines: 2, members: 6 */
        /* sum members: 65, holes: 1, sum holes: 7 */
        /* last cacheline: 8 bytes */
};

Signed-off-by Ido Schimmel <[email protected]>

Signed-off-by Petr Machata <[email protected]>

---

Notes:
v2:
- Set err on nexthop_create_group() error path

 include/net/nexthop.h |  6 ++++++
 net/ipv4/nexthop.c    | 25 +++++++++++++++++++++----
 2 files changed, 27 insertions(+), 4 deletions(-)

include/net/nexthop.h +6 -0

index 77e99cba60ad..4bf1875445d8 100644
@@ -95,8 +95,14 @@ struct nh_res_table {
 	struct nh_res_bucket	nh_buckets[] __counted_by(num_nh_buckets);
 };
 
+struct nh_grp_entry_stats {
+	u64 packets;
+	struct u64_stats_sync syncp;
+};
+
 struct nh_grp_entry {
 	struct nexthop	*nh;
+	struct nh_grp_entry_stats __percpu	*stats;
 	u8		weight;
 
 	union {

net/ipv4/nexthop.c +21 -4

index 816ae8ee3e06..4be66622e24f 100644
@@ -480,6 +480,7 @@ static void nexthop_free_group(struct nexthop *nh)
 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 
 		WARN_ON(!list_empty(&nhge->nh_list));
+		free_percpu(nhge->stats);
 		nexthop_put(nhge->nh);
 	}

@@ -1182,6 +1183,7 @@ static struct nexthop *nexthop_select_path_fdb(struct nh_group *nhg, int hash)
 		if (hash > atomic_read(&nhge->hthr.upper_bound))
 			continue;
 
+		this_cpu_inc(nhge->stats->packets);
 		return nhge->nh;
 	}

@@ -1191,7 +1193,7 @@ static struct nexthop *nexthop_select_path_fdb(struct nh_group *nhg, int hash)
 
 static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash)
 {
-	struct nexthop *rc = NULL;
+	struct nh_grp_entry *nhge0 = NULL;
 	int i;
 
 	if (nhg->fdb_nh)

@@ -1206,16 +1208,20 @@ static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash)
 		if (!nexthop_is_good_nh(nhge->nh))
 			continue;
 
-		if (!rc)
-			rc = nhge->nh;
+		if (!nhge0)
+			nhge0 = nhge;
 
 		if (hash > atomic_read(&nhge->hthr.upper_bound))
 			continue;
 
+		this_cpu_inc(nhge->stats->packets);
 		return nhge->nh;
 	}
 
-	return rc ? : nhg->nh_entries[0].nh;
+	if (!nhge0)
+		nhge0 = &nhg->nh_entries[0];
+	this_cpu_inc(nhge0->stats->packets);
+	return nhge0->nh;
 }
 
 static struct nexthop *nexthop_select_path_res(struct nh_group *nhg, int hash)

@@ -1231,6 +1237,7 @@ static struct nexthop *nexthop_select_path_res(struct nh_group *nhg, int hash)
 	bucket = &res_table->nh_buckets[bucket_index];
 	nh_res_bucket_set_busy(bucket);
 	nhge = rcu_dereference(bucket->nh_entry);
+	this_cpu_inc(nhge->stats->packets);
 	return nhge->nh;
 }

@@ -1804,6 +1811,7 @@ static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge,
 			newg->has_v4 = true;
 
 		list_del(&nhges[i].nh_list);
+		new_nhges[j].stats = nhges[i].stats;
 		new_nhges[j].nh_parent = nhges[i].nh_parent;
 		new_nhges[j].nh = nhges[i].nh;
 		new_nhges[j].weight = nhges[i].weight;

@@ -1819,6 +1827,7 @@ static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge,
 	rcu_assign_pointer(nhp->nh_grp, newg);
 
 	list_del(&nhge->nh_list);
+	free_percpu(nhge->stats);
 	nexthop_put(nhge->nh);
 
 	/* Removal of a NH from a resilient group is notified through

@@ -2483,6 +2492,13 @@ static struct nexthop *nexthop_create_group(struct net *net,
 		if (nhi->family == AF_INET)
 			nhg->has_v4 = true;
 
+		nhg->nh_entries[i].stats =
+			netdev_alloc_pcpu_stats(struct nh_grp_entry_stats);
+		if (!nhg->nh_entries[i].stats) {
+			err = -ENOMEM;
+			nexthop_put(nhe);
+			goto out_no_nh;
+		}
 		nhg->nh_entries[i].nh = nhe;
 		nhg->nh_entries[i].weight = entry[i].weight + 1;
 		list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);

@@ -2522,6 +2538,7 @@ static struct nexthop *nexthop_create_group(struct net *net,
 out_no_nh:
 	for (i--; i >= 0; --i) {
 		list_del(&nhg->nh_entries[i].nh_list);
+		free_percpu(nhg->nh_entries[i].stats);
 		nexthop_put(nhg->nh_entries[i].nh);
 	}

Signature · 1 line

2.43.0

David Ahern

maintainer Reviewed-by on v2 3/7

2024-03-01 15:36 UTC #6 lore raw

Re: [PATCH net-next v2 3/7] net: nexthop: Add nexthop group entry stats

Quoted 40 lines · Petr Machata

From: Ido Schimmel <[email protected]>

Add nexthop group entry stats to count the number of packets forwarded via each nexthop in the group. The stats will be exposed to user space for better data path observability in the next patch.

The per-CPU stats pointer is placed at the beginning of 'struct nh_grp_entry', so that all the fields accessed for the data path reside on the same cache line:

struct nh_grp_entry {
        struct nexthop *           nh;                   /*     0     8 */
        struct nh_grp_entry_stats * stats;               /*     8     8 */
        u8                         weight;               /*    16     1 */

        /* XXX 7 bytes hole, try to pack */

        union {
                struct {
                        atomic_t   upper_bound;          /*    24     4 */
                } hthr;                                  /*    24     4 */
                struct {
                        struct list_head uw_nh_entry;    /*    24    16 */
                        u16        count_buckets;        /*    40     2 */
                        u16        wants_buckets;        /*    42     2 */
                } res;                                   /*    24    24 */
        };                                               /*    24    24 */
        struct list_head           nh_list;              /*    48    16 */
        /* --- cacheline 1 boundary (64 bytes) --- */
        struct nexthop *           nh_parent;            /*    64     8 */

        /* size: 72, cachelines: 2, members: 6 */
        /* sum members: 65, holes: 1, sum holes: 7 */
        /* last cacheline: 8 bytes */
};

Signed-off-by Ido Schimmel <[email protected]>

Signed-off-by Petr Machata <[email protected]>

---

Quoted 8 lines

Notes:
v2:
- Set err on nexthop_create_group() error path

 include/net/nexthop.h |  6 ++++++
 net/ipv4/nexthop.c    | 25 +++++++++++++++++++++----
 2 files changed, 27 insertions(+), 4 deletions(-)

Reviewed-by David Ahern <[email protected]>

Petr Machata

author

2024-02-29 18:20 UTC #7 lore raw

PATCH v2 4/7 net: nexthop: Expose nexthop group stats to user space

In reply to Petr Machata

From: Ido Schimmel <[email protected]>

Add netlink support for reading NH group stats.

This data is only for statistics of the traffic in the SW datapath. HW nexthop group statistics will be added in the following patches.

Emission of the stats is keyed to a new op_stats flag to avoid cluttering the netlink message with stats if the user doesn't need them:
NHA_OP_FLAG_DUMP_STATS.

Co-developed-by Petr Machata <[email protected]>

Signed-off-by Petr Machata <[email protected]>

Signed-off-by Ido Schimmel <[email protected]>

---

Notes:
v2:
- Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS
- Rename jump target in nla_put_nh_group_stats() to avoid having to rename further in the patchset.

 include/uapi/linux/nexthop.h | 30 ++++++++++++
 net/ipv4/nexthop.c           | 92 ++++++++++++++++++++++++++++++++----
 2 files changed, 114 insertions(+), 8 deletions(-)

include/uapi/linux/nexthop.h +30 -0

index 086444e2946c..f4db63c17085 100644
@@ -30,6 +30,8 @@ enum {
 
 #define NEXTHOP_GRP_TYPE_MAX (__NEXTHOP_GRP_TYPE_MAX - 1)
 
+#define NHA_OP_FLAG_DUMP_STATS		BIT(0)
+
 enum {
 	NHA_UNSPEC,
 	NHA_ID,		/* u32; id for nexthop. id == 0 means auto-assign */

@@ -63,6 +65,9 @@ enum {
 	/* u32; operation-specific flags */
 	NHA_OP_FLAGS,
 
+	/* nested; nexthop group stats */
+	NHA_GROUP_STATS,
+
 	__NHA_MAX,
 };

@@ -104,4 +109,29 @@ enum {
 
 #define NHA_RES_BUCKET_MAX	(__NHA_RES_BUCKET_MAX - 1)
 
+enum {
+	NHA_GROUP_STATS_UNSPEC,
+
+	/* nested; nexthop group entry stats */
+	NHA_GROUP_STATS_ENTRY,
+
+	__NHA_GROUP_STATS_MAX,
+};
+
+#define NHA_GROUP_STATS_MAX	(__NHA_GROUP_STATS_MAX - 1)
+
+enum {
+	NHA_GROUP_STATS_ENTRY_UNSPEC,
+
+	/* u32; nexthop id of the nexthop group entry */
+	NHA_GROUP_STATS_ENTRY_ID,
+
+	/* uint; number of packets forwarded via the nexthop group entry */
+	NHA_GROUP_STATS_ENTRY_PACKETS,
+
+	__NHA_GROUP_STATS_ENTRY_MAX,
+};
+
+#define NHA_GROUP_STATS_ENTRY_MAX	(__NHA_GROUP_STATS_ENTRY_MAX - 1)
+
 #endif

net/ipv4/nexthop.c +84 -8

147 lines not on this page · raw message

Signature · 1 line

2.43.0

David Ahern

maintainer Reviewed-by on v2 4/7

2024-03-01 15:45 UTC #8 lore raw

Re: [PATCH net-next v2 4/7] net: nexthop: Expose nexthop group stats to user space

Quoted 40 lines · Petr Machata

From: Ido Schimmel <[email protected]>
Add netlink support for reading NH group stats.
This data is only for statistics of the traffic in the SW datapath. HW nexthop group statistics will be added in the following patches.
Emission of the stats is keyed to a new op_stats flag to avoid cluttering the netlink message with stats if the user doesn't need them:
NHA_OP_FLAG_DUMP_STATS.
Co-developed-by Petr Machata <[email protected]>
Signed-off-by Petr Machata <[email protected]>
Signed-off-by Ido Schimmel <[email protected]>
---
Notes:
v2:
- Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS
- Rename jump target in nla_put_nh_group_stats() to avoid having to rename further in the patchset.
 include/uapi/linux/nexthop.h | 30 ++++++++++++
 net/ipv4/nexthop.c           | 92 ++++++++++++++++++++++++++++++++----
 2 files changed, 114 insertions(+), 8 deletions(-)
include/uapi/linux/nexthop.h +8 -0
@@ -104,4 +109,29 @@ enum {
 
 #define NHA_RES_BUCKET_MAX	(__NHA_RES_BUCKET_MAX - 1)
 
+enum {
+	NHA_GROUP_STATS_UNSPEC,
+
+	/* nested; nexthop group entry stats */
+	NHA_GROUP_STATS_ENTRY,
+
+	__NHA_GROUP_STATS_MAX,
+};

include/uapi/linux/nexthop.h enum { · line 129

Quoted 10 lines

include/uapi/linux/nexthop.h +10 -0

+
+#define NHA_GROUP_STATS_MAX	(__NHA_GROUP_STATS_MAX - 1)
+
+enum {
+	NHA_GROUP_STATS_ENTRY_UNSPEC,
+
+	/* u32; nexthop id of the nexthop group entry */
+	NHA_GROUP_STATS_ENTRY_ID,
+
+	/* uint; number of packets forwarded via the nexthop group entry */

why not make it a u64?

include/uapi/linux/nexthop.h enum { · line 137

Quoted 8 lines

include/uapi/linux/nexthop.h +7 -0

+	NHA_GROUP_STATS_ENTRY_PACKETS,
+
+	__NHA_GROUP_STATS_ENTRY_MAX,
+};
+
+#define NHA_GROUP_STATS_ENTRY_MAX	(__NHA_GROUP_STATS_ENTRY_MAX - 1)
+
 #endif

Reviewed-by David Ahern <[email protected]>

Petr Machata

author

2024-03-01 17:18 UTC #9 lore raw

Re: [PATCH net-next v2 4/7] net: nexthop: Expose nexthop group stats to user space

Quoted 10 lines · David Ahern

include/uapi/linux/nexthop.h enum { · line 129
Quoted 7 lines · Petr Machata
include/uapi/linux/nexthop.h +7 -0
+enum {
+	NHA_GROUP_STATS_ENTRY_UNSPEC,
+
+	/* u32; nexthop id of the nexthop group entry */
+	NHA_GROUP_STATS_ENTRY_ID,
+
+	/* uint; number of packets forwarded via the nexthop group entry */
why not make it a u64?

uint will store it as 32-bit if possible and as 64-bit if necessary.

Jakub Kicinski

maintainer

2024-03-01 17:28 UTC #10 lore raw

Re: [PATCH net-next v2 4/7] net: nexthop: Expose nexthop group stats to user space

In reply to David Ahern

Quoted 3 lines · On Fri, 1 Mar 2024 08:45:52 -0700 David Ahern wrote:

include/uapi/linux/nexthop.h enum { · line 129
Quoted 1 line
include/uapi/linux/nexthop.h +1 -0
+	/* uint; number of packets forwarded via the nexthop group entry */  
why not make it a u64?

I think it should be our default type. Explicit length should only be picked if there's a clear reason. Too many review cycles wasted steering people onto correct types...

Now, I'm not sure Pablo merged the sint/uint support into libmnl, I can share the patches to handle them in iproute2, if that helps. Or just copy/paste from YNL.

I need to find the courage to ask you how we can start using YNL in iproute2, I really worry that we're not getting CLI tools written for new families any more :(

David Ahern

maintainer

2024-03-02 02:45 UTC #11 lore raw

Re: [PATCH net-next v2 4/7] net: nexthop: Expose nexthop group stats to user space

Quoted 9 lines · Jakub Kicinski

Quoted 3 lines · On Fri, 1 Mar 2024 08:45:52 -0700 David Ahern wrote:
include/uapi/linux/nexthop.h enum { · line 129
Quoted 1 line
include/uapi/linux/nexthop.h +1 -0
+	/* uint; number of packets forwarded via the nexthop group entry */  
why not make it a u64?
I think it should be our default type. Explicit length should only be picked if there's a clear reason. Too many review cycles wasted steering people onto correct types...

Quoted 7 lines

Now, I'm not sure Pablo merged the sint/uint support into libmnl, I can share the patches to handle them in iproute2, if that helps. Or just copy/paste from YNL.
I need to find the courage to ask you how we can start using YNL in iproute2, I really worry that we're not getting CLI tools written for new families any more :(

I need to find time to play around with the cli tool more; my attempts last weekend were a bit rough hunting for the right syntax.

Jakub Kicinski

maintainer

2024-03-03 03:36 UTC #12 lore raw

Re: [PATCH net-next v2 4/7] net: nexthop: Expose nexthop group stats to user space

Quoted 6 lines · On Fri, 1 Mar 2024 19:45:27 -0700 David Ahern wrote:

Quoted 3 lines
I need to find the courage to ask you how we can start using YNL in iproute2, I really worry that we're not getting CLI tools written for new families any more :(
I need to find time to play around with the cli tool more; my attempts last weekend were a bit rough hunting for the right syntax.

Do you mean playing with the Python CLI? :(
The thought of packaging that did cross my mind, but I always viewed it as a quick PoC / development tool. Maybe you're right tho, I should stop fighting it.

Eric Dumazet

2024-03-01 17:32 UTC #13 lore raw

Re: [PATCH net-next v2 4/7] net: nexthop: Expose nexthop group stats to user space

In reply to Petr Machata

Quoted 114 lines · Petr Machata

From: Ido Schimmel <[email protected]>

Add netlink support for reading NH group stats.

This data is only for statistics of the traffic in the SW datapath. HW nexthop group statistics will be added in the following patches.

Emission of the stats is keyed to a new op_stats flag to avoid cluttering the netlink message with stats if the user doesn't need them:
NHA_OP_FLAG_DUMP_STATS.

Co-developed-by Petr Machata <[email protected]>

Signed-off-by Petr Machata <[email protected]>

Signed-off-by Ido Schimmel <[email protected]>

---

Notes:
v2:
- Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS
- Rename jump target in nla_put_nh_group_stats() to avoid having to rename further in the patchset.

 include/uapi/linux/nexthop.h | 30 ++++++++++++
 net/ipv4/nexthop.c           | 92 ++++++++++++++++++++++++++++++++----
 2 files changed, 114 insertions(+), 8 deletions(-)

include/uapi/linux/nexthop.h +30 -0

index 086444e2946c..f4db63c17085 100644
@@ -30,6 +30,8 @@ enum {
 
 #define NEXTHOP_GRP_TYPE_MAX (__NEXTHOP_GRP_TYPE_MAX - 1)
 
+#define NHA_OP_FLAG_DUMP_STATS         BIT(0)
+
 enum {
        NHA_UNSPEC,
        NHA_ID,         /* u32; id for nexthop. id == 0 means auto-assign */

@@ -63,6 +65,9 @@ enum {
        /* u32; operation-specific flags */
        NHA_OP_FLAGS,
 
+       /* nested; nexthop group stats */
+       NHA_GROUP_STATS,
+
        __NHA_MAX,
 };

@@ -104,4 +109,29 @@ enum {
 
 #define NHA_RES_BUCKET_MAX     (__NHA_RES_BUCKET_MAX - 1)
 
+enum {
+       NHA_GROUP_STATS_UNSPEC,
+
+       /* nested; nexthop group entry stats */
+       NHA_GROUP_STATS_ENTRY,
+
+       __NHA_GROUP_STATS_MAX,
+};
+
+#define NHA_GROUP_STATS_MAX    (__NHA_GROUP_STATS_MAX - 1)
+
+enum {
+       NHA_GROUP_STATS_ENTRY_UNSPEC,
+
+       /* u32; nexthop id of the nexthop group entry */
+       NHA_GROUP_STATS_ENTRY_ID,
+
+       /* uint; number of packets forwarded via the nexthop group entry */
+       NHA_GROUP_STATS_ENTRY_PACKETS,
+
+       __NHA_GROUP_STATS_ENTRY_MAX,
+};
+
+#define NHA_GROUP_STATS_ENTRY_MAX      (__NHA_GROUP_STATS_ENTRY_MAX - 1)
+
 #endif

net/ipv4/nexthop.c +8 -3

index 4be66622e24f..0ede8777bd66 100644
@@ -41,7 +41,8 @@ static const struct nla_policy rtm_nh_policy_new[] = {
 
 static const struct nla_policy rtm_nh_policy_get[] = {
        [NHA_ID]                = { .type = NLA_U32 },
-       [NHA_OP_FLAGS]          = NLA_POLICY_MASK(NLA_U32, 0),
+       [NHA_OP_FLAGS]          = NLA_POLICY_MASK(NLA_U32,
+                                                 NHA_OP_FLAG_DUMP_STATS),
 };
 
 static const struct nla_policy rtm_nh_policy_del[] = {

@@ -53,7 +54,8 @@ static const struct nla_policy rtm_nh_policy_dump[] = {
        [NHA_GROUPS]            = { .type = NLA_FLAG },
        [NHA_MASTER]            = { .type = NLA_U32 },
        [NHA_FDB]               = { .type = NLA_FLAG },
-       [NHA_OP_FLAGS]          = NLA_POLICY_MASK(NLA_U32, 0),
+       [NHA_OP_FLAGS]          = NLA_POLICY_MASK(NLA_U32,
+                                                 NHA_OP_FLAG_DUMP_STATS),
 };
 
 static const struct nla_policy rtm_nh_res_policy_new[] = {

@@ -661,8 +663,77 @@ static int nla_put_nh_group_res(struct sk_buff *skb, struct nh_group *nhg)
        return -EMSGSIZE;
 }
 
-static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
+static void nh_grp_entry_stats_read(struct nh_grp_entry *nhge,
+                                   struct nh_grp_entry_stats *stats)
 {
+       int i;
+

net/ipv4/nexthop.c static int nla_put_nh_group_res(struct sk_buff *skb, struct nh_group *nhg) · line 680

Quoted 10 lines

net/ipv4/nexthop.c +10 -0

+       memset(stats, 0, sizeof(*stats));
+       for_each_possible_cpu(i) {
+               struct nh_grp_entry_stats *cpu_stats;
+               unsigned int start;
+               u64 packets;
+
+               cpu_stats = per_cpu_ptr(nhge->stats, i);
+               do {
+                       start = u64_stats_fetch_begin(&cpu_stats->syncp);
+                       packets = cpu_stats->packets;

This is not safe, even on 64bit arches.

You should use u64_stats_t, u64_stats_read(), u64_stats_add() ...

Quoted 5 lines

net/ipv4/nexthop.c +5 -0

+               } while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
+
+               stats->packets += packets;
+       }
+}

Petr Machata

author

2024-03-04 12:45 UTC #14 lore raw

Re: [PATCH net-next v2 4/7] net: nexthop: Expose nexthop group stats to user space

Quoted 24 lines · Eric Dumazet

Quoted 10 lines · Petr Machata

net/ipv4/nexthop.c +4 -1

@@ -661,8 +663,77 @@ static int nla_put_nh_group_res(struct sk_buff *skb, struct nh_group *nhg)
        return -EMSGSIZE;
 }
 
-static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
+static void nh_grp_entry_stats_read(struct nh_grp_entry *nhge,
+                                   struct nh_grp_entry_stats *stats)
 {
+       int i;
+

net/ipv4/nexthop.c static int nla_put_nh_group_res(struct sk_buff *skb, struct nh_group *nhg) · line 680

Quoted 10 lines

net/ipv4/nexthop.c +10 -0

+       memset(stats, 0, sizeof(*stats));
+       for_each_possible_cpu(i) {
+               struct nh_grp_entry_stats *cpu_stats;
+               unsigned int start;
+               u64 packets;
+
+               cpu_stats = per_cpu_ptr(nhge->stats, i);
+               do {
+                       start = u64_stats_fetch_begin(&cpu_stats->syncp);
+                       packets = cpu_stats->packets;

This is not safe, even on 64bit arches.

Quoted 1 line

You should use u64_stats_t, u64_stats_read(), u64_stats_add() ...

OK.

Petr Machata

author

2024-02-29 18:20 UTC #15 lore raw

PATCH v2 5/7 net: nexthop: Add hardware statistics notifications

In reply to Petr Machata

From: Ido Schimmel <[email protected]>

Add hw_stats field to several notifier structures to communicate to the drivers that HW statistics should be configured for nexthops within a given group.

Signed-off-by Ido Schimmel <[email protected]>

Signed-off-by Petr Machata <[email protected]>

---
 include/net/nexthop.h | 3 +++
 net/ipv4/nexthop.c    | 2 ++
 2 files changed, 5 insertions(+)

include/net/nexthop.h +3 -0

index 4bf1875445d8..a8dad8f48ca8 100644
@@ -130,6 +130,7 @@ struct nh_group {
 	bool			resilient;
 	bool			fdb_nh;
 	bool			has_v4;
+	bool			hw_stats;
 
 	struct nh_res_table __rcu *res_table;
 	struct nh_grp_entry	nh_entries[] __counted_by(num_nh);

@@ -193,6 +194,7 @@ struct nh_notifier_grp_entry_info {
 struct nh_notifier_grp_info {
 	u16 num_nh;
 	bool is_fdb;
+	bool hw_stats;
 	struct nh_notifier_grp_entry_info nh_entries[] __counted_by(num_nh);
 };

@@ -206,6 +208,7 @@ struct nh_notifier_res_bucket_info {
 
 struct nh_notifier_res_table_info {
 	u16 num_nh_buckets;
+	bool hw_stats;
 	struct nh_notifier_single_info nhs[] __counted_by(num_nh_buckets);
 };

net/ipv4/nexthop.c +2 -0

index 0ede8777bd66..02629ba7a75d 100644
@@ -139,6 +139,7 @@ static int nh_notifier_mpath_info_init(struct nh_notifier_info *info,
 
 	info->nh_grp->num_nh = num_nh;
 	info->nh_grp->is_fdb = nhg->fdb_nh;
+	info->nh_grp->hw_stats = nhg->hw_stats;
 
 	for (i = 0; i < num_nh; i++) {
 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];

@@ -170,6 +171,7 @@ static int nh_notifier_res_table_info_init(struct nh_notifier_info *info,
 		return -ENOMEM;
 
 	info->nh_res_table->num_nh_buckets = num_nh_buckets;
+	info->nh_res_table->hw_stats = nhg->hw_stats;
 
 	for (i = 0; i < num_nh_buckets; i++) {
 		struct nh_res_bucket *bucket = &res_table->nh_buckets[i];

Signature · 1 line

2.43.0

David Ahern

maintainer Reviewed-by on v2 5/7

2024-03-01 15:47 UTC #16 lore raw

Re: [PATCH net-next v2 5/7] net: nexthop: Add hardware statistics notifications

Quoted 6 lines · Petr Machata

From: Ido Schimmel <[email protected]>
Add hw_stats field to several notifier structures to communicate to the drivers that HW statistics should be configured for nexthops within a given group.

Quoted 7 lines

Signed-off-by Ido Schimmel <[email protected]>
Signed-off-by Petr Machata <[email protected]>
---
 include/net/nexthop.h | 3 +++
 net/ipv4/nexthop.c    | 2 ++
 2 files changed, 5 insertions(+)

Reviewed-by David Ahern <[email protected]>

Petr Machata

author

2024-02-29 18:20 UTC #17 lore raw

PATCH v2 6/7 net: nexthop: Add ability to enable / disable hardware statistics

In reply to Petr Machata

From: Ido Schimmel <[email protected]>

Add netlink support for enabling collection of HW statistics on nexthop groups.

Signed-off-by Ido Schimmel <[email protected]>

Signed-off-by Petr Machata <[email protected]>

---
 include/net/nexthop.h        |  2 ++
 include/uapi/linux/nexthop.h |  3 +++
 net/ipv4/nexthop.c           | 15 ++++++++++++++-
 3 files changed, 19 insertions(+), 1 deletion(-)

include/net/nexthop.h +2 -0

index a8dad8f48ca8..20cd337b4a9c 100644
@@ -47,6 +47,8 @@ struct nh_config {
 	bool		nh_grp_res_has_idle_timer;
 	bool		nh_grp_res_has_unbalanced_timer;
 
+	bool		nh_hw_stats;
+
 	struct nlattr	*nh_encap;
 	u16		nh_encap_type;

include/uapi/linux/nexthop.h +3 -0

index f4db63c17085..b86af80d4e09 100644
@@ -68,6 +68,9 @@ enum {
 	/* nested; nexthop group stats */
 	NHA_GROUP_STATS,
 
+	/* u32; nexthop hardware stats enable */
+	NHA_HW_STATS_ENABLE,
+
 	__NHA_MAX,
 };

net/ipv4/nexthop.c +14 -1

index 02629ba7a75d..15f108c440ae 100644
@@ -37,6 +37,7 @@ static const struct nla_policy rtm_nh_policy_new[] = {
 	[NHA_ENCAP]		= { .type = NLA_NESTED },
 	[NHA_FDB]		= { .type = NLA_FLAG },
 	[NHA_RES_GROUP]		= { .type = NLA_NESTED },
+	[NHA_HW_STATS_ENABLE]	= NLA_POLICY_MAX(NLA_U32, 1),
 };
 
 static const struct nla_policy rtm_nh_policy_get[] = {

@@ -765,7 +766,8 @@ static int nla_put_nh_group(struct sk_buff *skb, struct nexthop *nh,
 		goto nla_put_failure;
 
 	if (op_flags & NHA_OP_FLAG_DUMP_STATS &&
-	    nla_put_nh_group_stats(skb, nh))
+	    (nla_put_u32(skb, NHA_HW_STATS_ENABLE, nhg->hw_stats) ||
+	     nla_put_nh_group_stats(skb, nh)))
 		goto nla_put_failure;
 
 	return 0;

@@ -1189,6 +1191,7 @@ static int nh_check_attr_group(struct net *net,
 		if (!tb[i])
 			continue;
 		switch (i) {
+		case NHA_HW_STATS_ENABLE:
 		case NHA_FDB:
 			continue;
 		case NHA_RES_GROUP:

@@ -2609,6 +2612,9 @@ static struct nexthop *nexthop_create_group(struct net *net,
 	if (cfg->nh_fdb)
 		nhg->fdb_nh = 1;
 
+	if (cfg->nh_hw_stats)
+		nhg->hw_stats = true;
+
 	rcu_assign_pointer(nh->nh_grp, nhg);
 
 	return nh;

@@ -2951,6 +2957,9 @@ static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
 			err = rtm_to_nh_config_grp_res(tb[NHA_RES_GROUP],
 						       cfg, extack);
 
+		if (tb[NHA_HW_STATS_ENABLE])
+			cfg->nh_hw_stats = nla_get_u32(tb[NHA_HW_STATS_ENABLE]);
+
 		/* no other attributes should be set */
 		goto out;
 	}

@@ -3042,6 +3051,10 @@ static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
 		goto out;
 	}
 
+	if (tb[NHA_HW_STATS_ENABLE]) {
+		NL_SET_ERR_MSG(extack, "Cannot enable nexthop hardware statistics for non-group nexthops");
+		goto out;
+	}
 
 	err = 0;
 out:

Signature · 1 line

2.43.0

David Ahern

maintainer Reviewed-by on v2 6/7

2024-03-01 15:49 UTC #18 lore raw

Re: [PATCH net-next v2 6/7] net: nexthop: Add ability to enable / disable hardware statistics

Quoted 20 lines · Petr Machata

From: Ido Schimmel <[email protected]>
Add netlink support for enabling collection of HW statistics on nexthop groups.
Signed-off-by Ido Schimmel <[email protected]>
Signed-off-by Petr Machata <[email protected]>
---
 include/net/nexthop.h        |  2 ++
 include/uapi/linux/nexthop.h |  3 +++
 net/ipv4/nexthop.c           | 15 ++++++++++++++-
 3 files changed, 19 insertions(+), 1 deletion(-)
net/ipv4/nexthop.c +0 -0
index 02629ba7a75d..15f108c440ae 100644

net/ipv4/nexthop.c static const struct nla_policy rtm_nh_policy_new[] = { · line 40

Quoted 4 lines

net/ipv4/nexthop.c +1 -0

@@ -37,6 +37,7 @@ static const struct nla_policy rtm_nh_policy_new[] = {
 	[NHA_ENCAP]		= { .type = NLA_NESTED },
 	[NHA_FDB]		= { .type = NLA_FLAG },
 	[NHA_RES_GROUP]		= { .type = NLA_NESTED },
+	[NHA_HW_STATS_ENABLE]	= NLA_POLICY_MAX(NLA_U32, 1),

numbers typically need a name or comment.

Reviewed-by David Ahern <[email protected]>

Petr Machata

author

2024-03-01 17:23 UTC #19 lore raw

Re: [PATCH net-next v2 6/7] net: nexthop: Add ability to enable / disable hardware statistics

Quoted 12 lines · David Ahern

net/ipv4/nexthop.c static const struct nla_policy rtm_nh_policy_new[] = { · line 40
Quoted 9 lines · Petr Machata
net/ipv4/nexthop.c +1 -0
index 02629ba7a75d..15f108c440ae 100644
@@ -37,6 +37,7 @@ static const struct nla_policy rtm_nh_policy_new[] = {
 	[NHA_ENCAP]		= { .type = NLA_NESTED },
 	[NHA_FDB]		= { .type = NLA_FLAG },
 	[NHA_RES_GROUP]		= { .type = NLA_NESTED },
+	[NHA_HW_STATS_ENABLE]	= NLA_POLICY_MAX(NLA_U32, 1),
numbers typically need a name or comment.

How about this?

	[NHA_HW_STATS_ENABLE]	= NLA_POLICY_MAX(NLA_U32, true),

Too weird maybe?

Petr Machata

author

2024-02-29 18:20 UTC #20 lore raw

PATCH v2 7/7 net: nexthop: Expose nexthop group HW stats to user space

In reply to Petr Machata

From: Ido Schimmel <[email protected]>

Add netlink support for reading NH group hardware stats.

Stats collection is done through a new notifier, NEXTHOP_EVENT_HW_STATS_REPORT_DELTA. Drivers that implement HW counters for a given NH group are thereby asked to collect the stats and report back to core by calling nh_grp_hw_stats_report_delta(). This is similar to what netdevice L3 stats do.

Besides exposing number of packets that passed in the HW datapath, also include information on whether any driver actually realizes the counters. The core can tell based on whether it got any _report_delta() reports from the drivers. This allows enabling the statistics at the group at any time, with drivers opting into supporting them. This is also in line with what netdevice L3 stats are doing.

So as not to waste time and space, tie the collection and reporting of HW stats with a new op flag, NHA_OP_FLAG_DUMP_HW_STATS.

Co-developed-by Petr Machata <[email protected]>

Signed-off-by Petr Machata <[email protected]>

Signed-off-by Ido Schimmel <[email protected]>

Reviewed-by Kees Cook <[email protected]> # For the __counted_by bits

---

Notes:
v2:
- Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS_HW
- Do not cancel outside of nesting in nla_put_nh_group_stats()

 include/net/nexthop.h        |  18 +++++
 include/uapi/linux/nexthop.h |   9 +++
 net/ipv4/nexthop.c           | 133 ++++++++++++++++++++++++++++++++---
 3 files changed, 151 insertions(+), 9 deletions(-)

include/net/nexthop.h +18 -0

index 20cd337b4a9c..235f94ab16a8 100644
@@ -122,6 +122,7 @@ struct nh_grp_entry {
 
 	struct list_head nh_list;
 	struct nexthop	*nh_parent;  /* nexthop of group with this entry */
+	u64		packets_hw;
 };
 
 struct nh_group {

@@ -166,6 +167,7 @@ enum nexthop_event_type {
 	NEXTHOP_EVENT_REPLACE,
 	NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE,
 	NEXTHOP_EVENT_BUCKET_REPLACE,
+	NEXTHOP_EVENT_HW_STATS_REPORT_DELTA,
 };
 
 enum nh_notifier_info_type {

@@ -173,6 +175,7 @@ enum nh_notifier_info_type {
 	NH_NOTIFIER_INFO_TYPE_GRP,
 	NH_NOTIFIER_INFO_TYPE_RES_TABLE,
 	NH_NOTIFIER_INFO_TYPE_RES_BUCKET,
+	NH_NOTIFIER_INFO_TYPE_GRP_HW_STATS,
 };
 
 struct nh_notifier_single_info {

@@ -214,6 +217,17 @@ struct nh_notifier_res_table_info {
 	struct nh_notifier_single_info nhs[] __counted_by(num_nh_buckets);
 };
 
+struct nh_notifier_grp_hw_stats_entry_info {
+	u32 id;
+	u64 packets;
+};
+
+struct nh_notifier_grp_hw_stats_info {
+	u16 num_nh;
+	bool hw_stats_used;
+	struct nh_notifier_grp_hw_stats_entry_info stats[] __counted_by(num_nh);
+};
+
 struct nh_notifier_info {
 	struct net *net;
 	struct netlink_ext_ack *extack;

@@ -224,6 +238,7 @@ struct nh_notifier_info {
 		struct nh_notifier_grp_info *nh_grp;
 		struct nh_notifier_res_table_info *nh_res_table;
 		struct nh_notifier_res_bucket_info *nh_res_bucket;
+		struct nh_notifier_grp_hw_stats_info *nh_grp_hw_stats;
 	};
 };

@@ -236,6 +251,9 @@ void nexthop_bucket_set_hw_flags(struct net *net, u32 id, u16 bucket_index,
 				 bool offload, bool trap);
 void nexthop_res_grp_activity_update(struct net *net, u32 id, u16 num_buckets,
 				     unsigned long *activity);
+void nh_grp_hw_stats_report_delta(struct nh_notifier_grp_hw_stats_info *info,
+				  unsigned int nh_idx,
+				  u64 delta_packets);
 
 /* caller is holding rcu or rtnl; no reference taken to nexthop */
 struct nexthop *nexthop_find_by_id(struct net *net, u32 id);

include/uapi/linux/nexthop.h +9 -0

index b86af80d4e09..dd8787f9cf39 100644
@@ -31,6 +31,7 @@ enum {
 #define NEXTHOP_GRP_TYPE_MAX (__NEXTHOP_GRP_TYPE_MAX - 1)
 
 #define NHA_OP_FLAG_DUMP_STATS		BIT(0)
+#define NHA_OP_FLAG_DUMP_HW_STATS	BIT(1)
 
 enum {
 	NHA_UNSPEC,

@@ -71,6 +72,9 @@ enum {
 	/* u32; nexthop hardware stats enable */
 	NHA_HW_STATS_ENABLE,
 
+	/* u32; read-only; whether any driver collects HW stats */
+	NHA_HW_STATS_USED,
+
 	__NHA_MAX,
 };

@@ -132,6 +136,11 @@ enum {
 	/* uint; number of packets forwarded via the nexthop group entry */
 	NHA_GROUP_STATS_ENTRY_PACKETS,
 
+	/* uint; number of packets forwarded via the nexthop group entry in
+	 * hardware
+	 */
+	NHA_GROUP_STATS_ENTRY_PACKETS_HW,
+
 	__NHA_GROUP_STATS_ENTRY_MAX,
 };

net/ipv4/nexthop.c +124 -9

185 lines not on this page · raw message

Signature · 1 line

2.43.0

David Ahern

maintainer Reviewed-by on v2 7/7

2024-03-01 15:55 UTC #21 lore raw

Re: [PATCH net-next v2 7/7] net: nexthop: Expose nexthop group HW stats to user space

Quoted 53 lines · Petr Machata

From: Ido Schimmel <[email protected]>
Add netlink support for reading NH group hardware stats.
Stats collection is done through a new notifier, NEXTHOP_EVENT_HW_STATS_REPORT_DELTA. Drivers that implement HW counters for a given NH group are thereby asked to collect the stats and report back to core by calling nh_grp_hw_stats_report_delta(). This is similar to what netdevice L3 stats do.
Besides exposing number of packets that passed in the HW datapath, also include information on whether any driver actually realizes the counters. The core can tell based on whether it got any _report_delta() reports from the drivers. This allows enabling the statistics at the group at any time, with drivers opting into supporting them. This is also in line with what netdevice L3 stats are doing.
So as not to waste time and space, tie the collection and reporting of HW stats with a new op flag, NHA_OP_FLAG_DUMP_HW_STATS.
Co-developed-by Petr Machata <[email protected]>
Signed-off-by Petr Machata <[email protected]>
Signed-off-by Ido Schimmel <[email protected]>
Reviewed-by Kees Cook <[email protected]> # For the __counted_by bits
---
Notes:
v2:
- Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS_HW
- Do not cancel outside of nesting in nla_put_nh_group_stats()
 include/net/nexthop.h        |  18 +++++
 include/uapi/linux/nexthop.h |   9 +++
 net/ipv4/nexthop.c           | 133 ++++++++++++++++++++++++++++++++---
 3 files changed, 151 insertions(+), 9 deletions(-)
net/ipv4/nexthop.c +2 -1
index 15f108c440ae..169a003cc855 100644
@@ -43,7 +43,8 @@ static const struct nla_policy rtm_nh_policy_new[] = {
 static const struct nla_policy rtm_nh_policy_get[] = {
 	[NHA_ID]		= { .type = NLA_U32 },
 	[NHA_OP_FLAGS]		= NLA_POLICY_MASK(NLA_U32,
-						  NHA_OP_FLAG_DUMP_STATS),
+						  NHA_OP_FLAG_DUMP_STATS |
+						  NHA_OP_FLAG_DUMP_HW_STATS),
 };
 
 static const struct nla_policy rtm_nh_policy_del[] = {

net/ipv4/nexthop.c static const struct nla_policy rtm_nh_policy_dump[] = { · line 61

Quoted 6 lines

net/ipv4/nexthop.c +2 -1

@@ -56,7 +57,8 @@ static const struct nla_policy rtm_nh_policy_dump[] = {
 	[NHA_MASTER]		= { .type = NLA_U32 },
 	[NHA_FDB]		= { .type = NLA_FLAG },
 	[NHA_OP_FLAGS]		= NLA_POLICY_MASK(NLA_U32,
-						  NHA_OP_FLAG_DUMP_STATS),
+						  NHA_OP_FLAG_DUMP_STATS |
+						  NHA_OP_FLAG_DUMP_HW_STATS),

2 instances of the same mask; worth giving it a name.

Reviewed-by David Ahern <[email protected]>

Petr Machata

author

2024-03-01 17:25 UTC #22 lore raw

Re: [PATCH net-next v2 7/7] net: nexthop: Expose nexthop group HW stats to user space

Quoted 14 lines · David Ahern

Quoted 13 lines · Petr Machata

net/ipv4/nexthop.c +2 -1

index 15f108c440ae..169a003cc855 100644
@@ -43,7 +43,8 @@ static const struct nla_policy rtm_nh_policy_new[] = {
 static const struct nla_policy rtm_nh_policy_get[] = {
 	[NHA_ID]		= { .type = NLA_U32 },
 	[NHA_OP_FLAGS]		= NLA_POLICY_MASK(NLA_U32,
-						  NHA_OP_FLAG_DUMP_STATS),
+						  NHA_OP_FLAG_DUMP_STATS |
+						  NHA_OP_FLAG_DUMP_HW_STATS),
 };

Quoted 10 lines

net/ipv4/nexthop.c static const struct nla_policy rtm_nh_policy_dump[] = { · line 61

Quoted 8 lines

net/ipv4/nexthop.c +2 -1

 static const struct nla_policy rtm_nh_policy_del[] = {

@@ -56,7 +57,8 @@ static const struct nla_policy rtm_nh_policy_dump[] = {
 	[NHA_MASTER]		= { .type = NLA_U32 },
 	[NHA_FDB]		= { .type = NLA_FLAG },
 	[NHA_OP_FLAGS]		= NLA_POLICY_MASK(NLA_U32,
-						  NHA_OP_FLAG_DUMP_STATS),
+						  NHA_OP_FLAG_DUMP_STATS |
+						  NHA_OP_FLAG_DUMP_HW_STATS),

2 instances of the same mask; worth giving it a name.

OK. I'll stick this somewhere above the policies:

#define NHA_OP_FLAGS_DUMP_STATS_ALL (NHA_OP_FLAG_DUMP_STATS | \
NHA_OP_FLAG_DUMP_HW_STATS)