Skip to contentSkip to conversation

Follow-up BPF helper improvements

PATCH Applied Daniel Borkmann 5 participants 13 messages Oct 10, 2020 – Oct 12, 2020

Subsystems (8)
lore
Share & tools
lore

Daniel Borkmann

author maintainer
#0 lore raw

v6 cover Follow-up BPF helper improvements

This series addresses most of the feedback [0] that was to be followed up from the last series, that is, UAPI helper comment improvements and getting rid of the ifindex obj file hacks in the selftest by using a BPF map instead. The __sk_buff data/data_end pointer work, I'm planning to do in a later round as well as the mem*() BPF improvements we have in Cilium for libbpf. Next, the series adds two features, i) a helper called redirect_peer() to improve latency on netns switch, and ii) to allow map in map with dynamic inner array map sizes. Selftests for each are added as well. For details, please check individual patches, thanks!

[0] https://lore.kernel.org/bpf/[email protected]/

v5 -> v6:
- Going with Andrii's suggestion to make the misconfigured verifier test more robust, and only probe on -EOPNOTSUPP (Andrii)
v4 -> v5:
- Replace cnt == -EOPNOTSUPP check with cnt < 0; I've used < 0 here as I think it's useful to keep the existing cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) for error detection (Andrii)
v3 -> v4:
- Rename new array map flag to BPF_F_INNER_MAP (Alexei)
v2 -> v3:
- Remove tab that slipped into uapi helper desc (Jakub)
- Rework map in map for array to error from map_gen_lookup (Andrii)
v1 -> v2:
- Fixed selftest comment wrt inner1/inner2 value (Yonghong)

Daniel Borkmann (6):
bpf: improve bpf_redirect_neigh helper description
bpf: add redirect_peer helper
bpf: allow for map-in-map with dynamic inner array map entries bpf, selftests: add test for different array inner map size bpf, selftests: make redirect_neigh test more extensible bpf, selftests: add redirect_peer selftest

 drivers/net/veth.c                            |   9 +
 include/linux/bpf.h                           |   2 +-
 include/linux/netdevice.h                     |   4 +
 include/uapi/linux/bpf.h                      |  30 ++-
 kernel/bpf/arraymap.c                         |  17 +-
 kernel/bpf/hashtab.c                          |   6 +-
 kernel/bpf/verifier.c                         |   6 +-
 net/core/dev.c                                |  15 +-
 net/core/filter.c                             |  54 ++++-
 net/xdp/xskmap.c                              |   2 +-
 tools/include/uapi/linux/bpf.h                |  30 ++-
 .../selftests/bpf/prog_tests/btf_map_in_map.c |  39 +++-
 .../selftests/bpf/progs/test_btf_map_in_map.c |  43 ++++
 .../selftests/bpf/progs/test_tc_neigh.c       |  40 ++--
 .../selftests/bpf/progs/test_tc_peer.c        |  45 ++++
 tools/testing/selftests/bpf/test_tc_neigh.sh  | 168 ---------------
 .../testing/selftests/bpf/test_tc_redirect.sh | 204 ++++++++++++++++++
 17 files changed, 489 insertions(+), 225 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/progs/test_tc_peer.c
 delete mode 100755 tools/testing/selftests/bpf/test_tc_neigh.sh
 create mode 100755 tools/testing/selftests/bpf/test_tc_redirect.sh
Signature · 1 line
2.17.1
author maintainer
#1 lore raw

PATCH v6 1/6 bpf: improve bpf_redirect_neigh helper description

Follow-up to address David's feedback that we should better describe internals of the bpf_redirect_neigh() helper.

---
 include/uapi/linux/bpf.h       | 10 +++++++---
 tools/include/uapi/linux/bpf.h | 10 +++++++---
 2 files changed, 14 insertions(+), 6 deletions(-)
include/uapi/linux/bpf.h +7 -3
index 42d2df799397..4272cc53d478 100644
@@ -3679,10 +3679,14 @@ union bpf_attr {
  * 		Redirect the packet to another net device of index *ifindex*
  * 		and fill in L2 addresses from neighboring subsystem. This helper
  * 		is somewhat similar to **bpf_redirect**\ (), except that it
- * 		fills in e.g. MAC addresses based on the L3 information from
- * 		the packet. This helper is supported for IPv4 and IPv6 protocols.
+ * 		populates L2 addresses as well, meaning, internally, the helper
+ * 		performs a FIB lookup based on the skb's networking header to
+ * 		get the address of the next hop and then relies on the neighbor
+ * 		lookup for the L2 address of the nexthop.
+ *
  * 		The *flags* argument is reserved and must be 0. The helper is
- * 		currently only supported for tc BPF program types.
+ * 		currently only supported for tc BPF program types, and enabled
+ * 		for IPv4 and IPv6 protocols.
  * 	Return
  * 		The helper returns **TC_ACT_REDIRECT** on success or
  * 		**TC_ACT_SHOT** on error.
tools/include/uapi/linux/bpf.h +7 -3
index 42d2df799397..4272cc53d478 100644
@@ -3679,10 +3679,14 @@ union bpf_attr {
  * 		Redirect the packet to another net device of index *ifindex*
  * 		and fill in L2 addresses from neighboring subsystem. This helper
  * 		is somewhat similar to **bpf_redirect**\ (), except that it
- * 		fills in e.g. MAC addresses based on the L3 information from
- * 		the packet. This helper is supported for IPv4 and IPv6 protocols.
+ * 		populates L2 addresses as well, meaning, internally, the helper
+ * 		performs a FIB lookup based on the skb's networking header to
+ * 		get the address of the next hop and then relies on the neighbor
+ * 		lookup for the L2 address of the nexthop.
+ *
  * 		The *flags* argument is reserved and must be 0. The helper is
- * 		currently only supported for tc BPF program types.
+ * 		currently only supported for tc BPF program types, and enabled
+ * 		for IPv4 and IPv6 protocols.
  * 	Return
  * 		The helper returns **TC_ACT_REDIRECT** on success or
  * 		**TC_ACT_SHOT** on error.
Signature · 1 line
2.17.1
author maintainer
#2 lore raw

PATCH v6 2/6 bpf: add redirect_peer helper

In reply to Daniel Borkmann

Add an efficient ingress to ingress netns switch that can be used out of tc BPF programs in order to redirect traffic from host ns ingress into a container veth device ingress without having to go via CPU backlog queue [0]. For local containers this can also be utilized and path via CPU backlog queue only needs to be taken once, not twice. On a high level this borrows from ipvlan which does similar switch in __netif_receive_skb_core() and then iterates via another_round. This helps to reduce latency for mentioned use cases.

Pod to remote pod with redirect(), TCP_RR [1]:

  # percpu_netperf 10.217.1.33
          RT_LATENCY:         122.450         (per CPU:         122.666         122.401         122.333         122.401 )
        MEAN_LATENCY:         121.210         (per CPU:         121.100         121.260         121.320         121.160 )
      STDDEV_LATENCY:         120.040         (per CPU:         119.420         119.910         125.460         115.370 )
         MIN_LATENCY:          46.500         (per CPU:          47.000          47.000          47.000          45.000 )
         P50_LATENCY:         118.500         (per CPU:         118.000         119.000         118.000         119.000 )
         P90_LATENCY:         127.500         (per CPU:         127.000         128.000         127.000         128.000 )
         P99_LATENCY:         130.750         (per CPU:         131.000         131.000         129.000         132.000 )
    TRANSACTION_RATE:       32666.400         (per CPU:        8152.200        8169.842        8174.439        8169.897 )

Pod to remote pod with redirect_peer(), TCP_RR:

  # percpu_netperf 10.217.1.33
          RT_LATENCY:          44.449         (per CPU:          43.767          43.127          45.279          45.622 )
        MEAN_LATENCY:          45.065         (per CPU:          44.030          45.530          45.190          45.510 )
      STDDEV_LATENCY:          84.823         (per CPU:          66.770          97.290          84.380          90.850 )
         MIN_LATENCY:          33.500         (per CPU:          33.000          33.000          34.000          34.000 )
         P50_LATENCY:          43.250         (per CPU:          43.000          43.000          43.000          44.000 )
         P90_LATENCY:          46.750         (per CPU:          46.000          47.000          47.000          47.000 )
         P99_LATENCY:          52.750         (per CPU:          51.000          54.000          53.000          53.000 )
    TRANSACTION_RATE:       90039.500         (per CPU:       22848.186       23187.089       22085.077       21919.130 )

[0] https://linuxplumbersconf.org/event/7/contributions/674/attachments/568/1002/plumbers_2020_cilium_load_balancer.pdf
[1] https://github.com/borkmann/netperf_scripts/blob/master/percpu_netperf

---
 drivers/net/veth.c             |  9 ++++++
 include/linux/netdevice.h      |  4 +++
 include/uapi/linux/bpf.h       | 17 +++++++++++
 net/core/dev.c                 | 15 ++++++++--
 net/core/filter.c              | 54 +++++++++++++++++++++++++++++-----
 tools/include/uapi/linux/bpf.h | 17 +++++++++++
 6 files changed, 106 insertions(+), 10 deletions(-)
drivers/net/veth.c +9 -0
index 091e5b4ba042..8c737668008a 100644
@@ -420,6 +420,14 @@ static int veth_select_rxq(struct net_device *dev)
 	return smp_processor_id() % dev->real_num_rx_queues;
 }
 
+static struct net_device *veth_peer_dev(struct net_device *dev)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+
+	/* Callers must be under RCU read side. */
+	return rcu_dereference(priv->peer);
+}
+
 static int veth_xdp_xmit(struct net_device *dev, int n,
 			 struct xdp_frame **frames,
 			 u32 flags, bool ndo_xmit)
@@ -1224,6 +1232,7 @@ static const struct net_device_ops veth_netdev_ops = {
 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
 	.ndo_bpf		= veth_xdp,
 	.ndo_xdp_xmit		= veth_ndo_xdp_xmit,
+	.ndo_get_peer_dev	= veth_peer_dev,
 };
 
 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
include/linux/netdevice.h +4 -0
index 28cfa53daf72..0533f86018dd 100644
@@ -1277,6 +1277,9 @@ struct netdev_net_notifier {
  * int (*ndo_tunnel_ctl)(struct net_device *dev, struct ip_tunnel_parm *p,
  *			 int cmd);
  *	Add, change, delete or get information on an IPv4 tunnel.
+ * struct net_device *(*ndo_get_peer_dev)(struct net_device *dev);
+ *	If a device is paired with a peer device, return the peer instance.
+ *	The caller must be under RCU read context.
  */
 struct net_device_ops {
 	int			(*ndo_init)(struct net_device *dev);
@@ -1484,6 +1487,7 @@ struct net_device_ops {
 	struct devlink_port *	(*ndo_get_devlink_port)(struct net_device *dev);
 	int			(*ndo_tunnel_ctl)(struct net_device *dev,
 						  struct ip_tunnel_parm *p, int cmd);
+	struct net_device *	(*ndo_get_peer_dev)(struct net_device *dev);
 };
 
 /**
include/uapi/linux/bpf.h +17 -0
index 4272cc53d478..b97bc5abb3b8 100644
@@ -3719,6 +3719,22 @@ union bpf_attr {
  *		never return NULL.
  *	Return
  *		A pointer pointing to the kernel percpu variable on this cpu.
+ *
+ * long bpf_redirect_peer(u32 ifindex, u64 flags)
+ * 	Description
+ * 		Redirect the packet to another net device of index *ifindex*.
+ * 		This helper is somewhat similar to **bpf_redirect**\ (), except
+ * 		that the redirection happens to the *ifindex*' peer device and
+ * 		the netns switch takes place from ingress to ingress without
+ * 		going through the CPU's backlog queue.
+ *
+ * 		The *flags* argument is reserved and must be 0. The helper is
+ * 		currently only supported for tc BPF program types at the ingress
+ * 		hook and for veth device types. The peer device must reside in a
+ * 		different network namespace.
+ * 	Return
+ * 		The helper returns **TC_ACT_REDIRECT** on success or
+ * 		**TC_ACT_SHOT** on error.
  */
 #define __BPF_FUNC_MAPPER(FN)		\
 	FN(unspec),			\
@@ -3876,6 +3892,7 @@ union bpf_attr {
 	FN(redirect_neigh),		\
 	FN(bpf_per_cpu_ptr),            \
 	FN(bpf_this_cpu_ptr),		\
+	FN(redirect_peer),		\
 	/* */
 
 /* integer value in 'imm' field of BPF_CALL instruction selects which helper
net/core/dev.c +12 -3
index 9d55bf5d1a65..7dd015823593 100644
@@ -4930,7 +4930,7 @@ EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
 
 static inline struct sk_buff *
 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
-		   struct net_device *orig_dev)
+		   struct net_device *orig_dev, bool *another)
 {
 #ifdef CONFIG_NET_CLS_ACT
 	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
@@ -4974,7 +4974,11 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
 		 * redirecting to another netdev
 		 */
 		__skb_push(skb, skb->mac_len);
-		skb_do_redirect(skb);
+		if (skb_do_redirect(skb) == -EAGAIN) {
+			__skb_pull(skb, skb->mac_len);
+			*another = true;
+			break;
+		}
 		return NULL;
 	case TC_ACT_CONSUMED:
 		return NULL;
@@ -5163,7 +5167,12 @@ static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
 skip_taps:
 #ifdef CONFIG_NET_INGRESS
 	if (static_branch_unlikely(&ingress_needed_key)) {
-		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
+		bool another = false;
+
+		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
+					 &another);
+		if (another)
+			goto another_round;
 		if (!skb)
 			goto out;
 
net/core/filter.c +47 -7

86 lines not on this page · raw message

tools/include/uapi/linux/bpf.h +17 -0
index 4272cc53d478..b97bc5abb3b8 100644
@@ -3719,6 +3719,22 @@ union bpf_attr {
  *		never return NULL.
  *	Return
  *		A pointer pointing to the kernel percpu variable on this cpu.
+ *
+ * long bpf_redirect_peer(u32 ifindex, u64 flags)
+ * 	Description
+ * 		Redirect the packet to another net device of index *ifindex*.
+ * 		This helper is somewhat similar to **bpf_redirect**\ (), except
+ * 		that the redirection happens to the *ifindex*' peer device and
+ * 		the netns switch takes place from ingress to ingress without
+ * 		going through the CPU's backlog queue.
+ *
+ * 		The *flags* argument is reserved and must be 0. The helper is
+ * 		currently only supported for tc BPF program types at the ingress
+ * 		hook and for veth device types. The peer device must reside in a
+ * 		different network namespace.
+ * 	Return
+ * 		The helper returns **TC_ACT_REDIRECT** on success or
+ * 		**TC_ACT_SHOT** on error.
  */
 #define __BPF_FUNC_MAPPER(FN)		\
 	FN(unspec),			\
@@ -3876,6 +3892,7 @@ union bpf_attr {
 	FN(redirect_neigh),		\
 	FN(bpf_per_cpu_ptr),            \
 	FN(bpf_this_cpu_ptr),		\
+	FN(redirect_peer),		\
 	/* */
 
 /* integer value in 'imm' field of BPF_CALL instruction selects which helper
Signature · 1 line
2.17.1
#3 lore raw

Re: [PATCH bpf-next v6 2/6] bpf: add redirect_peer helper

Quoted 32 lines · On Sun, 11 Oct 2020 01:40:02 +0200 Daniel Borkmann <[email protected]> wrote:

Add an efficient ingress to ingress netns switch that can be used out of tc BPF programs in order to redirect traffic from host ns ingress into a container veth device ingress without having to go via CPU backlog queue [0]. For local containers this can also be utilized and path via CPU backlog queue only needs to be taken once, not twice. On a high level this borrows from ipvlan which does similar switch in __netif_receive_skb_core() and then iterates via another_round. This helps to reduce latency for mentioned use cases.

Pod to remote pod with redirect(), TCP_RR [1]:

  # percpu_netperf 10.217.1.33
          RT_LATENCY:         122.450         (per CPU:         122.666         122.401         122.333         122.401 )
        MEAN_LATENCY:         121.210         (per CPU:         121.100         121.260         121.320         121.160 )
      STDDEV_LATENCY:         120.040         (per CPU:         119.420         119.910         125.460         115.370 )
         MIN_LATENCY:          46.500         (per CPU:          47.000          47.000          47.000          45.000 )
         P50_LATENCY:         118.500         (per CPU:         118.000         119.000         118.000         119.000 )
         P90_LATENCY:         127.500         (per CPU:         127.000         128.000         127.000         128.000 )
         P99_LATENCY:         130.750         (per CPU:         131.000         131.000         129.000         132.000 )
    TRANSACTION_RATE:       32666.400         (per CPU:        8152.200        8169.842        8174.439        8169.897 )

Pod to remote pod with redirect_peer(), TCP_RR:

  # percpu_netperf 10.217.1.33
          RT_LATENCY:          44.449         (per CPU:          43.767          43.127          45.279          45.622 )
        MEAN_LATENCY:          45.065         (per CPU:          44.030          45.530          45.190          45.510 )
      STDDEV_LATENCY:          84.823         (per CPU:          66.770          97.290          84.380          90.850 )
         MIN_LATENCY:          33.500         (per CPU:          33.000          33.000          34.000          34.000 )
         P50_LATENCY:          43.250         (per CPU:          43.000          43.000          43.000          44.000 )
         P90_LATENCY:          46.750         (per CPU:          46.000          47.000          47.000          47.000 )
         P99_LATENCY:          52.750         (per CPU:          51.000          54.000          53.000          53.000 )
Quoted 1 line
    TRANSACTION_RATE:       90039.500         (per CPU:       22848.186       23187.089       22085.077       21919.130 )

This is awesome results and great work Daniel! :-)

I wonder if we can also support this from XDP, which can also native redirect into veth. Originally I though we could add the peer netdev in the devmap, but AFAIK Toke showed me that this was not possible.

Quoted 13 lines

[0] https://linuxplumbersconf.org/event/7/contributions/674/attachments/568/1002/plumbers_2020_cilium_load_balancer.pdf
[1] https://github.com/borkmann/netperf_scripts/blob/master/percpu_netperf

---
 drivers/net/veth.c             |  9 ++++++
 include/linux/netdevice.h      |  4 +++
 include/uapi/linux/bpf.h       | 17 +++++++++++
 net/core/dev.c                 | 15 ++++++++--
 net/core/filter.c              | 54 +++++++++++++++++++++++++++++-----
 tools/include/uapi/linux/bpf.h | 17 +++++++++++
 6 files changed, 106 insertions(+), 10 deletions(-)

[...]

Quoted 72 lines
net/core/dev.c +12 -3
index 9d55bf5d1a65..7dd015823593 100644
@@ -4930,7 +4930,7 @@ EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
 
 static inline struct sk_buff *
 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
-		   struct net_device *orig_dev)
+		   struct net_device *orig_dev, bool *another)
 {
 #ifdef CONFIG_NET_CLS_ACT
 	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
@@ -4974,7 +4974,11 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
 		 * redirecting to another netdev
 		 */
 		__skb_push(skb, skb->mac_len);
-		skb_do_redirect(skb);
+		if (skb_do_redirect(skb) == -EAGAIN) {
+			__skb_pull(skb, skb->mac_len);
+			*another = true;
+			break;
+		}
 		return NULL;
 	case TC_ACT_CONSUMED:
 		return NULL;
@@ -5163,7 +5167,12 @@ static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
 skip_taps:
 #ifdef CONFIG_NET_INGRESS
 	if (static_branch_unlikely(&ingress_needed_key)) {
-		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
+		bool another = false;
+
+		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
+					 &another);
+		if (another)
+			goto another_round;
 		if (!skb)
 			goto out;
 
net/core/filter.c +7 -6
index 5da44b11e1ec..fab951c6be57 100644
@@ -2380,8 +2380,9 @@ static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev)
 
 /* Internal, non-exposed redirect flags. */
 enum {
-	BPF_F_NEIGH = (1ULL << 1),
-#define BPF_F_REDIRECT_INTERNAL	(BPF_F_NEIGH)
+	BPF_F_NEIGH	= (1ULL << 1),
+	BPF_F_PEER	= (1ULL << 2),
+#define BPF_F_REDIRECT_INTERNAL	(BPF_F_NEIGH | BPF_F_PEER)
 };
 
 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
@@ -2430,19 +2431,35 @@ EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
 int skb_do_redirect(struct sk_buff *skb)
 {
 	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct net *net = dev_net(skb->dev);
 	struct net_device *dev;
 	u32 flags = ri->flags;
 
-	dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
+	dev = dev_get_by_index_rcu(net, ri->tgt_index);
 	ri->tgt_index = 0;
-	if (unlikely(!dev)) {
-		kfree_skb(skb);
-		return -EINVAL;
+	ri->flags = 0;
+	if (unlikely(!dev))

net/core/filter.c EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); · line 2451

Quoted 10 lines
net/core/filter.c +10 -0
+		goto out_drop;
+	if (flags & BPF_F_PEER) {
+		const struct net_device_ops *ops = dev->netdev_ops;
+
+		if (unlikely(!ops->ndo_get_peer_dev ||
+			     !skb_at_tc_ingress(skb)))
+			goto out_drop;
+		dev = ops->ndo_get_peer_dev(dev);
+		if (unlikely(!dev ||
+			     !is_skb_forwardable(dev, skb) ||

Again a MTU "transmissing" check on ingress "receive" path, but we can take that discussion after this is merged, as this keeps the current behavior.

net/core/filter.c EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); · line 2454

Quoted 3 lines
net/core/filter.c +3 -0
+			     net_eq(net, dev_net(dev))))
+			goto out_drop;
+		skb->dev = dev;

Don't we need to clean some more state when this packet gets redirected into another namespace?

Like skb_scrub_packet(), or is that not needed? (p.s. I would like to avoid it, as it e.g. clears the skb->mark.)

Quoted 10 lines
net/core/filter.c +4 -1
+		return -EAGAIN;
 	}
-
 	return flags & BPF_F_NEIGH ?
 	       __bpf_redirect_neigh(skb, dev) :
 	       __bpf_redirect(skb, dev, flags);
+out_drop:
+	kfree_skb(skb);
+	return -EINVAL;
 }
Signature · 4 lines
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer
author maintainer
#4 lore raw

Re: [PATCH bpf-next v6 2/6] bpf: add redirect_peer helper

Quoted 35 lines · Jesper Dangaard Brouer
Quoted 32 lines · On Sun, 11 Oct 2020 01:40:02 +0200 Daniel Borkmann <[email protected]> wrote:

Add an efficient ingress to ingress netns switch that can be used out of tc BPF programs in order to redirect traffic from host ns ingress into a container veth device ingress without having to go via CPU backlog queue [0]. For local containers this can also be utilized and path via CPU backlog queue only needs to be taken once, not twice. On a high level this borrows from ipvlan which does similar switch in __netif_receive_skb_core() and then iterates via another_round. This helps to reduce latency for mentioned use cases.

Pod to remote pod with redirect(), TCP_RR [1]:

   # percpu_netperf 10.217.1.33
           RT_LATENCY:         122.450         (per CPU:         122.666         122.401         122.333         122.401 )
         MEAN_LATENCY:         121.210         (per CPU:         121.100         121.260         121.320         121.160 )
       STDDEV_LATENCY:         120.040         (per CPU:         119.420         119.910         125.460         115.370 )
          MIN_LATENCY:          46.500         (per CPU:          47.000          47.000          47.000          45.000 )
          P50_LATENCY:         118.500         (per CPU:         118.000         119.000         118.000         119.000 )
          P90_LATENCY:         127.500         (per CPU:         127.000         128.000         127.000         128.000 )
          P99_LATENCY:         130.750         (per CPU:         131.000         131.000         129.000         132.000 )
     TRANSACTION_RATE:       32666.400         (per CPU:        8152.200        8169.842        8174.439        8169.897 )

Pod to remote pod with redirect_peer(), TCP_RR:

   # percpu_netperf 10.217.1.33
           RT_LATENCY:          44.449         (per CPU:          43.767          43.127          45.279          45.622 )
         MEAN_LATENCY:          45.065         (per CPU:          44.030          45.530          45.190          45.510 )
       STDDEV_LATENCY:          84.823         (per CPU:          66.770          97.290          84.380          90.850 )
          MIN_LATENCY:          33.500         (per CPU:          33.000          33.000          34.000          34.000 )
          P50_LATENCY:          43.250         (per CPU:          43.000          43.000          43.000          44.000 )
          P90_LATENCY:          46.750         (per CPU:          46.000          47.000          47.000          47.000 )
          P99_LATENCY:          52.750         (per CPU:          51.000          54.000          53.000          53.000 )
Quoted 7 lines
Quoted 1 line
     TRANSACTION_RATE:       90039.500         (per CPU:       22848.186       23187.089       22085.077       21919.130 )

This is awesome results and great work Daniel! :-)

I wonder if we can also support this from XDP, which can also native redirect into veth. Originally I though we could add the peer netdev in the devmap, but AFAIK Toke showed me that this was not possible.

I think it should be possible with similar principle. What was the limitation that you ran into with devmap for XDP?

Quoted 92 lines
Quoted 13 lines
   [0] https://linuxplumbersconf.org/event/7/contributions/674/attachments/568/1002/plumbers_2020_cilium_load_balancer.pdf
   [1] https://github.com/borkmann/netperf_scripts/blob/master/percpu_netperf
---
  drivers/net/veth.c             |  9 ++++++
  include/linux/netdevice.h      |  4 +++
  include/uapi/linux/bpf.h       | 17 +++++++++++
  net/core/dev.c                 | 15 ++++++++--
  net/core/filter.c              | 54 +++++++++++++++++++++++++++++-----
  tools/include/uapi/linux/bpf.h | 17 +++++++++++
  6 files changed, 106 insertions(+), 10 deletions(-)

[...]

Quoted 78 lines
net/core/dev.c +12 -3
index 9d55bf5d1a65..7dd015823593 100644
@@ -4930,7 +4930,7 @@ EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
 
  static inline struct sk_buff *
  sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
-		   struct net_device *orig_dev)
+		   struct net_device *orig_dev, bool *another)
  {
  #ifdef CONFIG_NET_CLS_ACT
  	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
@@ -4974,7 +4974,11 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
  		 * redirecting to another netdev
  		 */
  		__skb_push(skb, skb->mac_len);
-		skb_do_redirect(skb);
+		if (skb_do_redirect(skb) == -EAGAIN) {
+			__skb_pull(skb, skb->mac_len);
+			*another = true;
+			break;
+		}
  		return NULL;
  	case TC_ACT_CONSUMED:
  		return NULL;
@@ -5163,7 +5167,12 @@ static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
  skip_taps:
  #ifdef CONFIG_NET_INGRESS
  	if (static_branch_unlikely(&ingress_needed_key)) {
-		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
+		bool another = false;
+
+		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
+					 &another);
+		if (another)
+			goto another_round;
  		if (!skb)
  			goto out;
 
net/core/filter.c +13 -6
index 5da44b11e1ec..fab951c6be57 100644
@@ -2380,8 +2380,9 @@ static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev)
 
  /* Internal, non-exposed redirect flags. */
  enum {
-	BPF_F_NEIGH = (1ULL << 1),
-#define BPF_F_REDIRECT_INTERNAL	(BPF_F_NEIGH)
+	BPF_F_NEIGH	= (1ULL << 1),
+	BPF_F_PEER	= (1ULL << 2),
+#define BPF_F_REDIRECT_INTERNAL	(BPF_F_NEIGH | BPF_F_PEER)
  };
 
  BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
@@ -2430,19 +2431,35 @@ EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
  int skb_do_redirect(struct sk_buff *skb)
  {
  	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct net *net = dev_net(skb->dev);
  	struct net_device *dev;
  	u32 flags = ri->flags;
 
-	dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
+	dev = dev_get_by_index_rcu(net, ri->tgt_index);
  	ri->tgt_index = 0;
-	if (unlikely(!dev)) {
-		kfree_skb(skb);
-		return -EINVAL;
+	ri->flags = 0;
+	if (unlikely(!dev))
+		goto out_drop;
+	if (flags & BPF_F_PEER) {
+		const struct net_device_ops *ops = dev->netdev_ops;
+
+		if (unlikely(!ops->ndo_get_peer_dev ||
+			     !skb_at_tc_ingress(skb)))
Quoted 8 lines

net/core/filter.c EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); · line 2451

Quoted 4 lines
net/core/filter.c +4 -0
+			goto out_drop;
+		dev = ops->ndo_get_peer_dev(dev);
+		if (unlikely(!dev ||
+			     !is_skb_forwardable(dev, skb) ||

Again a MTU "transmissing" check on ingress "receive" path, but we can take that discussion after this is merged, as this keeps the current behavior.

Yep, agree; also it checks whether dev is up which we need here too.

Quoted 9 lines

net/core/filter.c EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); · line 2454

Quoted 3 lines
net/core/filter.c +3 -0
+			     net_eq(net, dev_net(dev))))
+			goto out_drop;
+		skb->dev = dev;

Don't we need to clean some more state when this packet gets redirected into another namespace?

Like skb_scrub_packet(), or is that not needed? (p.s. I would like to avoid it, as it e.g. clears the skb->mark.)

Not needed, the traffic egress path from a netns is scrubbing already, and ingress traffic in hostns the BPF prog can do if needed given it has full control, this is similar to how ipvlan does it for traffic into container.

Quoted 13 lines
Quoted 10 lines
net/core/filter.c +4 -1
+		return -EAGAIN;
  	}
-
  	return flags & BPF_F_NEIGH ?
  	       __bpf_redirect_neigh(skb, dev) :
  	       __bpf_redirect(skb, dev, flags);
+out_drop:
+	kfree_skb(skb);
+	return -EINVAL;
  }
#5 lore raw

Re: [PATCH bpf-next v6 2/6] bpf: add redirect_peer helper

Quoted 2 lines · Daniel Borkmann
Quoted 2 lines

This is awesome results and great work Daniel! :-)

+1

Quoted 8 lines
Quoted 4 lines

I wonder if we can also support this from XDP, which can also native redirect into veth.  Originally I though we could add the peer netdev in the devmap, but AFAIK Toke showed me that this was not possible.

I think it should be possible with similar principle. What was the limitation
that you ran into with devmap for XDP?

Should just need an API to set the namespace of the redirect device - something that devmap can be extended to include now.

#6 lore raw

Re: [PATCH bpf-next v6 2/6] bpf: add redirect_peer helper

Quoted 13 lines · On Sun, 11 Oct 2020 20:50:12 -0600 David Ahern <[email protected]> wrote:
Quoted 2 lines · Daniel Borkmann
Quoted 2 lines

This is awesome results and great work Daniel! :-)

+1

Quoted 7 lines
Quoted 4 lines

I wonder if we can also support this from XDP, which can also native redirect into veth.  Originally I though we could add the peer netdev in the devmap, but AFAIK Toke showed me that this was not possible.

I think it should be possible with similar principle. What was the limitation that you ran into with devmap for XDP?

If you add a device to devmap and afterwards move this device into a namespace, then the device is removed from the devmap. This is because devmap detect/react on NETDEV_UNREGISTER and remove the net_device.

Quoted 2 lines

Should just need an API to set the namespace of the redirect device - something that devmap can be extended to include now.

Perhaps for other devices being moved into a namespace.

Specifically for veth the XDP redirect (veth_ndo_xdp_xmit) already pickup the peer net_device, and *queue* the xdp_frame, thus it's not directly relevant for the XDP redirect (except we also have an intermediate queue which is likely bad for the TCP_RR test).

I just tried to test native-XDP redirect into a veth with samples/bpf/ xdp_redirect_map, which doesn't work. Packets are actually getting silently dropped. After digging into the kernel code, I realized this is because the *peer*-veth device didn't have a XDP-prog loaded. The xdp_redirect_map loads a dummy-XDP prog on the veth-device (it can see), as a way to enable the ndo_xdp_xmit (which we have discussed before it a broken way to do this, but it have become a defacto way).

Signature · 4 lines
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer
author maintainer
#7 lore raw

PATCH v6 3/6 bpf: allow for map-in-map with dynamic inner array map entries

In reply to Daniel Borkmann

Recent work in f4d05259213f ("bpf: Add map_meta_equal map ops") and 134fede4eecf ("bpf: Relax max_entries check for most of the inner map types") added support for dynamic inner max elements for most map-in-map types. Exceptions were maps like array or prog array where the map_gen_lookup() callback uses the maps' max_entries field as a constant when emitting instructions.

We recently implemented Maglev consistent hashing into Cilium's load balancer which uses map-in-map with an outer map being hash and inner being array holding the Maglev backend table for each service. This has been designed this way in order to reduce overall memory consumption given the outer hash map allows to avoid preallocating a large, flat memory area for all services. Also, the number of service mappings is not always known a-priori.

The use case for dynamic inner array map entries is to further reduce memory overhead, for example, some services might just have a small number of back ends while others could have a large number. Right now the Maglev backend table for small and large number of backends would need to have the same inner array map entries which adds a lot of unneeded overhead.

Dynamic inner array map entries can be realized by avoiding the inlined code generation for their lookup. The lookup will still be efficient since it will be calling into array_map_lookup_elem() directly and thus avoiding retpoline. The patch adds a BPF_F_INNER_MAP flag to map creation which therefore skips inline code generation and relaxes array_map_meta_equal() check to ignore both maps' max_entries. This also still allows to have faster lookups for map-in-map when BPF_F_INNER_MAP is not specified and hence dynamic max_entries not needed.

Example code generation where inner map is dynamic sized array:

  # bpftool p d x i 125
  int handle__sys_enter(void * ctx):
  ; int handle__sys_enter(void *ctx)
     0: (b4) w1 = 0
  ; int key = 0;
     1: (63) *(u32 *)(r10 -4) = r1
     2: (bf) r2 = r10
  ;
     3: (07) r2 += -4
  ; inner_map = bpf_map_lookup_elem(&outer_arr_dyn, &key);
     4: (18) r1 = map[id:468]
     6: (07) r1 += 272
     7: (61) r0 = *(u32 *)(r2 +0)
     8: (35) if r0 >= 0x3 goto pc+5
     9: (67) r0 <<= 3
    10: (0f) r0 += r1
    11: (79) r0 = *(u64 *)(r0 +0)
    12: (15) if r0 == 0x0 goto pc+1
    13: (05) goto pc+1
    14: (b7) r0 = 0
    15: (b4) w6 = -1
  ; if (!inner_map)
    16: (15) if r0 == 0x0 goto pc+6
    17: (bf) r2 = r10
  ;
    18: (07) r2 += -4
  ; val = bpf_map_lookup_elem(inner_map, &key);
    19: (bf) r1 = r0                               | No inlining but instead
    20: (85) call array_map_lookup_elem#149280     | call to array_map_lookup_elem()
  ; return val ? *val : -1;                        | for inner array lookup.
    21: (15) if r0 == 0x0 goto pc+1
  ; return val ? *val : -1;
    22: (61) r6 = *(u32 *)(r0 +0)
  ; }
    23: (bc) w0 = w6
    24: (95) exit
---
 include/linux/bpf.h            |  2 +-
 include/uapi/linux/bpf.h       |  3 +++
 kernel/bpf/arraymap.c          | 17 +++++++++++------
 kernel/bpf/hashtab.c           |  6 +++---
 kernel/bpf/verifier.c          |  6 ++++--
 net/xdp/xskmap.c               |  2 +-
 tools/include/uapi/linux/bpf.h |  3 +++
 7 files changed, 26 insertions(+), 13 deletions(-)
include/linux/bpf.h +1 -1
index dc63eeed4fd9..2b16bf48aab6 100644
@@ -82,7 +82,7 @@ struct bpf_map_ops {
 	void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file,
 				int fd);
 	void (*map_fd_put_ptr)(void *ptr);
-	u32 (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf);
+	int (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf);
 	u32 (*map_fd_sys_lookup_elem)(void *ptr);
 	void (*map_seq_show_elem)(struct bpf_map *map, void *key,
 				  struct seq_file *m);
include/uapi/linux/bpf.h +3 -0
index b97bc5abb3b8..bf5a99d803e4 100644
@@ -435,6 +435,9 @@ enum {
 
 /* Share perf_event among processes */
 	BPF_F_PRESERVE_ELEMS	= (1U << 11),
+
+/* Create a map that is suitable to be an inner map with dynamic max entries */
+	BPF_F_INNER_MAP		= (1U << 12),
 };
 
 /* Flags for BPF_PROG_QUERY. */
kernel/bpf/arraymap.c +11 -6
index bd777dd6f967..c6c81eceb68f 100644
@@ -16,7 +16,7 @@
 
 #define ARRAY_CREATE_FLAG_MASK \
 	(BPF_F_NUMA_NODE | BPF_F_MMAPABLE | BPF_F_ACCESS_MASK | \
-	 BPF_F_PRESERVE_ELEMS)
+	 BPF_F_PRESERVE_ELEMS | BPF_F_INNER_MAP)
 
 static void bpf_array_free_percpu(struct bpf_array *array)
 {
@@ -62,7 +62,7 @@ int array_map_alloc_check(union bpf_attr *attr)
 		return -EINVAL;
 
 	if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
-	    attr->map_flags & BPF_F_MMAPABLE)
+	    attr->map_flags & (BPF_F_MMAPABLE | BPF_F_INNER_MAP))
 		return -EINVAL;
 
 	if (attr->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
@@ -214,7 +214,7 @@ static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
 }
 
 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
-static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
+static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	struct bpf_insn *insn = insn_buf;
@@ -223,6 +223,9 @@ static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 	const int map_ptr = BPF_REG_1;
 	const int index = BPF_REG_2;
 
+	if (map->map_flags & BPF_F_INNER_MAP)
+		return -EOPNOTSUPP;
+
 	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
 	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
 	if (!map->bypass_spec_v1) {
@@ -496,8 +499,10 @@ static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
 static bool array_map_meta_equal(const struct bpf_map *meta0,
 				 const struct bpf_map *meta1)
 {
-	return meta0->max_entries == meta1->max_entries &&
-		bpf_map_meta_equal(meta0, meta1);
+	if (!bpf_map_meta_equal(meta0, meta1))
+		return false;
+	return meta0->map_flags & BPF_F_INNER_MAP ? true :
+	       meta0->max_entries == meta1->max_entries;
 }
 
 struct bpf_iter_seq_array_map_info {
@@ -1251,7 +1256,7 @@ static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
 	return READ_ONCE(*inner_map);
 }
 
-static u32 array_of_map_gen_lookup(struct bpf_map *map,
+static int array_of_map_gen_lookup(struct bpf_map *map,
 				   struct bpf_insn *insn_buf)
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
kernel/bpf/hashtab.c +3 -3
index 3395cf140d22..1815e97d4c9c 100644
@@ -612,7 +612,7 @@ static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
  * bpf_prog
  *   __htab_map_lookup_elem
  */
-static u32 htab_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
+static int htab_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 {
 	struct bpf_insn *insn = insn_buf;
 	const int ret = BPF_REG_0;
@@ -651,7 +651,7 @@ static void *htab_lru_map_lookup_elem_sys(struct bpf_map *map, void *key)
 	return __htab_lru_map_lookup_elem(map, key, false);
 }
 
-static u32 htab_lru_map_gen_lookup(struct bpf_map *map,
+static int htab_lru_map_gen_lookup(struct bpf_map *map,
 				   struct bpf_insn *insn_buf)
 {
 	struct bpf_insn *insn = insn_buf;
@@ -2070,7 +2070,7 @@ static void *htab_of_map_lookup_elem(struct bpf_map *map, void *key)
 	return READ_ONCE(*inner_map);
 }
 
-static u32 htab_of_map_gen_lookup(struct bpf_map *map,
+static int htab_of_map_gen_lookup(struct bpf_map *map,
 				  struct bpf_insn *insn_buf)
 {
 	struct bpf_insn *insn = insn_buf;
kernel/bpf/verifier.c +4 -2
index f3e36eade3d4..fa5badc9279a 100644
@@ -11049,7 +11049,9 @@ static int fixup_bpf_calls(struct bpf_verifier_env *env)
 			if (insn->imm == BPF_FUNC_map_lookup_elem &&
 			    ops->map_gen_lookup) {
 				cnt = ops->map_gen_lookup(map_ptr, insn_buf);
-				if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
+				if (cnt == -EOPNOTSUPP)
+					goto patch_map_ops_generic;
+				if (cnt <= 0 || cnt >= ARRAY_SIZE(insn_buf)) {
 					verbose(env, "bpf verifier is misconfigured\n");
 					return -EINVAL;
 				}
@@ -11079,7 +11081,7 @@ static int fixup_bpf_calls(struct bpf_verifier_env *env)
 				     (int (*)(struct bpf_map *map, void *value))NULL));
 			BUILD_BUG_ON(!__same_type(ops->map_peek_elem,
 				     (int (*)(struct bpf_map *map, void *value))NULL));
-
+patch_map_ops_generic:
 			switch (insn->imm) {
 			case BPF_FUNC_map_lookup_elem:
 				insn->imm = BPF_CAST_CALL(ops->map_lookup_elem) -
net/xdp/xskmap.c +1 -1
index 0c5df593bc56..49da2b8ace8b 100644
@@ -132,7 +132,7 @@ static int xsk_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 	return 0;
 }
 
-static u32 xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
+static int xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 {
 	const int ret = BPF_REG_0, mp = BPF_REG_1, index = BPF_REG_2;
 	struct bpf_insn *insn = insn_buf;
tools/include/uapi/linux/bpf.h +3 -0
index b97bc5abb3b8..bf5a99d803e4 100644
@@ -435,6 +435,9 @@ enum {
 
 /* Share perf_event among processes */
 	BPF_F_PRESERVE_ELEMS	= (1U << 11),
+
+/* Create a map that is suitable to be an inner map with dynamic max entries */
+	BPF_F_INNER_MAP		= (1U << 12),
 };
 
 /* Flags for BPF_PROG_QUERY. */
Signature · 1 line
2.17.1
Acked-by on v6 3/6
#8 lore raw

Re: [PATCH bpf-next v6 3/6] bpf: allow for map-in-map with dynamic inner array map entries

Quoted 67 lines · Daniel Borkmann

Recent work in f4d05259213f ("bpf: Add map_meta_equal map ops") and 134fede4eecf ("bpf: Relax max_entries check for most of the inner map types") added support for dynamic inner max elements for most map-in-map types. Exceptions were maps like array or prog array where the map_gen_lookup() callback uses the maps' max_entries field as a constant when emitting instructions.

We recently implemented Maglev consistent hashing into Cilium's load balancer which uses map-in-map with an outer map being hash and inner being array holding the Maglev backend table for each service. This has been designed this way in order to reduce overall memory consumption given the outer hash map allows to avoid preallocating a large, flat memory area for all services. Also, the number of service mappings is not always known a-priori.

The use case for dynamic inner array map entries is to further reduce memory overhead, for example, some services might just have a small number of back ends while others could have a large number. Right now the Maglev backend table for small and large number of backends would need to have the same inner array map entries which adds a lot of unneeded overhead.

Dynamic inner array map entries can be realized by avoiding the inlined code generation for their lookup. The lookup will still be efficient since it will be calling into array_map_lookup_elem() directly and thus avoiding retpoline. The patch adds a BPF_F_INNER_MAP flag to map creation which therefore skips inline code generation and relaxes array_map_meta_equal() check to ignore both maps' max_entries. This also still allows to have faster lookups for map-in-map when BPF_F_INNER_MAP is not specified and hence dynamic max_entries not needed.

Example code generation where inner map is dynamic sized array:

  # bpftool p d x i 125
  int handle__sys_enter(void * ctx):
  ; int handle__sys_enter(void *ctx)
     0: (b4) w1 = 0
  ; int key = 0;
     1: (63) *(u32 *)(r10 -4) = r1
     2: (bf) r2 = r10
  ;
     3: (07) r2 += -4
  ; inner_map = bpf_map_lookup_elem(&outer_arr_dyn, &key);
     4: (18) r1 = map[id:468]
     6: (07) r1 += 272
     7: (61) r0 = *(u32 *)(r2 +0)
     8: (35) if r0 >= 0x3 goto pc+5
     9: (67) r0 <<= 3
    10: (0f) r0 += r1
    11: (79) r0 = *(u64 *)(r0 +0)
    12: (15) if r0 == 0x0 goto pc+1
    13: (05) goto pc+1
    14: (b7) r0 = 0
    15: (b4) w6 = -1
  ; if (!inner_map)
    16: (15) if r0 == 0x0 goto pc+6
    17: (bf) r2 = r10
  ;
    18: (07) r2 += -4
  ; val = bpf_map_lookup_elem(inner_map, &key);
    19: (bf) r1 = r0                               | No inlining but instead
    20: (85) call array_map_lookup_elem#149280     | call to array_map_lookup_elem()
  ; return val ? *val : -1;                        | for inner array lookup.
    21: (15) if r0 == 0x0 goto pc+1
  ; return val ? *val : -1;
    22: (61) r6 = *(u32 *)(r0 +0)
  ; }
    23: (bc) w0 = w6
    24: (95) exit
Quoted 3 lines

---

Looks good, thanks!

[...]

author maintainer
#9 lore raw

PATCH v6 4/6 bpf, selftests: add test for different array inner map size

In reply to Daniel Borkmann

Extend the "diff_size" subtest to also include a non-inlined array map variant where dynamic inner #elems are possible.

---
 .../selftests/bpf/prog_tests/btf_map_in_map.c | 39 ++++++++++++-----
 .../selftests/bpf/progs/test_btf_map_in_map.c | 43 +++++++++++++++++++
 2 files changed, 72 insertions(+), 10 deletions(-)
tools/testing/selftests/bpf/prog_tests/btf_map_in_map.c +29 -10
index 540fea4c91a5..76ebe4c250f1 100644
@@ -55,10 +55,10 @@ static int kern_sync_rcu(void)
 
 static void test_lookup_update(void)
 {
-	int err, key = 0, val, i;
+	int map1_fd, map2_fd, map3_fd, map4_fd, map5_fd, map1_id, map2_id;
+	int outer_arr_fd, outer_hash_fd, outer_arr_dyn_fd;
 	struct test_btf_map_in_map *skel;
-	int outer_arr_fd, outer_hash_fd;
-	int fd, map1_fd, map2_fd, map1_id, map2_id;
+	int err, key = 0, val, i, fd;
 
 	skel = test_btf_map_in_map__open_and_load();
 	if (CHECK(!skel, "skel_open", "failed to open&load skeleton\n"))
@@ -70,32 +70,45 @@ static void test_lookup_update(void)
 
 	map1_fd = bpf_map__fd(skel->maps.inner_map1);
 	map2_fd = bpf_map__fd(skel->maps.inner_map2);
+	map3_fd = bpf_map__fd(skel->maps.inner_map3);
+	map4_fd = bpf_map__fd(skel->maps.inner_map4);
+	map5_fd = bpf_map__fd(skel->maps.inner_map5);
+	outer_arr_dyn_fd = bpf_map__fd(skel->maps.outer_arr_dyn);
 	outer_arr_fd = bpf_map__fd(skel->maps.outer_arr);
 	outer_hash_fd = bpf_map__fd(skel->maps.outer_hash);
 
-	/* inner1 = input, inner2 = input + 1 */
-	map1_fd = bpf_map__fd(skel->maps.inner_map1);
+	/* inner1 = input, inner2 = input + 1, inner3 = input + 2 */
 	bpf_map_update_elem(outer_arr_fd, &key, &map1_fd, 0);
-	map2_fd = bpf_map__fd(skel->maps.inner_map2);
 	bpf_map_update_elem(outer_hash_fd, &key, &map2_fd, 0);
+	bpf_map_update_elem(outer_arr_dyn_fd, &key, &map3_fd, 0);
 	skel->bss->input = 1;
 	usleep(1);
-
 	bpf_map_lookup_elem(map1_fd, &key, &val);
 	CHECK(val != 1, "inner1", "got %d != exp %d\n", val, 1);
 	bpf_map_lookup_elem(map2_fd, &key, &val);
 	CHECK(val != 2, "inner2", "got %d != exp %d\n", val, 2);
+	bpf_map_lookup_elem(map3_fd, &key, &val);
+	CHECK(val != 3, "inner3", "got %d != exp %d\n", val, 3);
 
-	/* inner1 = input + 1, inner2 = input */
+	/* inner2 = input, inner1 = input + 1, inner4 = input + 2 */
 	bpf_map_update_elem(outer_arr_fd, &key, &map2_fd, 0);
 	bpf_map_update_elem(outer_hash_fd, &key, &map1_fd, 0);
+	bpf_map_update_elem(outer_arr_dyn_fd, &key, &map4_fd, 0);
 	skel->bss->input = 3;
 	usleep(1);
-
 	bpf_map_lookup_elem(map1_fd, &key, &val);
 	CHECK(val != 4, "inner1", "got %d != exp %d\n", val, 4);
 	bpf_map_lookup_elem(map2_fd, &key, &val);
 	CHECK(val != 3, "inner2", "got %d != exp %d\n", val, 3);
+	bpf_map_lookup_elem(map4_fd, &key, &val);
+	CHECK(val != 5, "inner4", "got %d != exp %d\n", val, 5);
+
+	/* inner5 = input + 2 */
+	bpf_map_update_elem(outer_arr_dyn_fd, &key, &map5_fd, 0);
+	skel->bss->input = 5;
+	usleep(1);
+	bpf_map_lookup_elem(map5_fd, &key, &val);
+	CHECK(val != 7, "inner5", "got %d != exp %d\n", val, 7);
 
 	for (i = 0; i < 5; i++) {
 		val = i % 2 ? map1_fd : map2_fd;
@@ -106,7 +119,13 @@ static void test_lookup_update(void)
 		}
 		err = bpf_map_update_elem(outer_arr_fd, &key, &val, 0);
 		if (CHECK_FAIL(err)) {
-			printf("failed to update hash_of_maps on iter #%d\n", i);
+			printf("failed to update array_of_maps on iter #%d\n", i);
+			goto cleanup;
+		}
+		val = i % 2 ? map4_fd : map5_fd;
+		err = bpf_map_update_elem(outer_arr_dyn_fd, &key, &val, 0);
+		if (CHECK_FAIL(err)) {
+			printf("failed to update array_of_maps (dyn) on iter #%d\n", i);
 			goto cleanup;
 		}
 	}
tools/testing/selftests/bpf/progs/test_btf_map_in_map.c +43 -0

55 lines not on this page · raw message

Signature · 1 line
2.17.1
author maintainer
#10 lore raw

PATCH v6 5/6 bpf, selftests: make redirect_neigh test more extensible

In reply to Daniel Borkmann

Rename into test_tc_redirect.sh and move setup and test code into separate functions so they can be reused for newly added tests in here. Also remove the crude hack to override ifindex inside the object file via xxd and sed and just use a simple map instead. Map given iproute2 does not support BTF fully and therefore neither global data at this point.

---
 .../selftests/bpf/progs/test_tc_neigh.c       |  40 ++--
 tools/testing/selftests/bpf/test_tc_neigh.sh  | 168 ---------------
 .../testing/selftests/bpf/test_tc_redirect.sh | 197 ++++++++++++++++++
 3 files changed, 219 insertions(+), 186 deletions(-)
 delete mode 100755 tools/testing/selftests/bpf/test_tc_neigh.sh
 create mode 100755 tools/testing/selftests/bpf/test_tc_redirect.sh
tools/testing/selftests/bpf/progs/test_tc_neigh.c +22 -18
index 889a72c3024f..fe182616b112 100644
@@ -13,17 +13,10 @@
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_endian.h>
 
-#ifndef barrier_data
-# define barrier_data(ptr)	asm volatile("": :"r"(ptr) :"memory")
-#endif
-
 #ifndef ctx_ptr
 # define ctx_ptr(field)		(void *)(long)(field)
 #endif
 
-#define dst_to_src_tmp		0xeeddddeeU
-#define src_to_dst_tmp		0xeeffffeeU
-
 #define ip4_src			0xac100164 /* 172.16.1.100 */
 #define ip4_dst			0xac100264 /* 172.16.2.100 */
 
@@ -39,6 +32,18 @@
 				 a.s6_addr32[3] == b.s6_addr32[3])
 #endif
 
+enum {
+	dev_src,
+	dev_dst,
+};
+
+struct bpf_map_def SEC("maps") ifindex_map = {
+	.type		= BPF_MAP_TYPE_ARRAY,
+	.key_size	= sizeof(int),
+	.value_size	= sizeof(int),
+	.max_entries	= 2,
+};
+
 static __always_inline bool is_remote_ep_v4(struct __sk_buff *skb,
 					    __be32 addr)
 {
@@ -73,7 +78,14 @@ static __always_inline bool is_remote_ep_v6(struct __sk_buff *skb,
 	return v6_equal(ip6h->daddr, addr);
 }
 
-SEC("chk_neigh") int tc_chk(struct __sk_buff *skb)
+static __always_inline int get_dev_ifindex(int which)
+{
+	int *ifindex = bpf_map_lookup_elem(&ifindex_map, &which);
+
+	return ifindex ? *ifindex : 0;
+}
+
+SEC("chk_egress") int tc_chk(struct __sk_buff *skb)
 {
 	void *data_end = ctx_ptr(skb->data_end);
 	void *data = ctx_ptr(skb->data);
@@ -87,7 +99,6 @@ SEC("chk_neigh") int tc_chk(struct __sk_buff *skb)
 
 SEC("dst_ingress") int tc_dst(struct __sk_buff *skb)
 {
-	int idx = dst_to_src_tmp;
 	__u8 zero[ETH_ALEN * 2];
 	bool redirect = false;
 
@@ -103,19 +114,15 @@ SEC("dst_ingress") int tc_dst(struct __sk_buff *skb)
 	if (!redirect)
 		return TC_ACT_OK;
 
-	barrier_data(&idx);
-	idx = bpf_ntohl(idx);
-
 	__builtin_memset(&zero, 0, sizeof(zero));
 	if (bpf_skb_store_bytes(skb, 0, &zero, sizeof(zero), 0) < 0)
 		return TC_ACT_SHOT;
 
-	return bpf_redirect_neigh(idx, 0);
+	return bpf_redirect_neigh(get_dev_ifindex(dev_src), 0);
 }
 
 SEC("src_ingress") int tc_src(struct __sk_buff *skb)
 {
-	int idx = src_to_dst_tmp;
 	__u8 zero[ETH_ALEN * 2];
 	bool redirect = false;
 
@@ -131,14 +138,11 @@ SEC("src_ingress") int tc_src(struct __sk_buff *skb)
 	if (!redirect)
 		return TC_ACT_OK;
 
-	barrier_data(&idx);
-	idx = bpf_ntohl(idx);
-
 	__builtin_memset(&zero, 0, sizeof(zero));
 	if (bpf_skb_store_bytes(skb, 0, &zero, sizeof(zero), 0) < 0)
 		return TC_ACT_SHOT;
 
-	return bpf_redirect_neigh(idx, 0);
+	return bpf_redirect_neigh(get_dev_ifindex(dev_dst), 0);
 }
 
 char __license[] SEC("license") = "GPL";
tools/testing/selftests/bpf/test_tc_neigh.sh deleted +0 -168

168 lines not on this page · raw message

tools/testing/selftests/bpf/test_tc_redirect.sh new file +197 -0

197 lines not on this page · raw message

Signature · 1 line
2.17.1
author maintainer
#11 lore raw

PATCH v6 6/6 bpf, selftests: add redirect_peer selftest

In reply to Daniel Borkmann

Extend the test_tc_redirect test and add a small test that exercises the new redirect_peer() helper for the IPv4 and IPv6 case.

---
 .../selftests/bpf/progs/test_tc_peer.c        | 45 +++++++++++++++++++
 .../testing/selftests/bpf/test_tc_redirect.sh | 25 +++++++----
 2 files changed, 61 insertions(+), 9 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/progs/test_tc_peer.c
tools/testing/selftests/bpf/progs/test_tc_peer.c new file +45 -0
new file mode 100644
index 000000000000..fc84a7685aa2
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <stdint.h>
+#include <stdbool.h>
+
+#include <linux/bpf.h>
+#include <linux/stddef.h>
+#include <linux/pkt_cls.h>
+
+#include <bpf/bpf_helpers.h>
+
+enum {
+	dev_src,
+	dev_dst,
+};
+
+struct bpf_map_def SEC("maps") ifindex_map = {
+	.type		= BPF_MAP_TYPE_ARRAY,
+	.key_size	= sizeof(int),
+	.value_size	= sizeof(int),
+	.max_entries	= 2,
+};
+
+static __always_inline int get_dev_ifindex(int which)
+{
+	int *ifindex = bpf_map_lookup_elem(&ifindex_map, &which);
+
+	return ifindex ? *ifindex : 0;
+}
+
+SEC("chk_egress") int tc_chk(struct __sk_buff *skb)
+{
+	return TC_ACT_SHOT;
+}
+
+SEC("dst_ingress") int tc_dst(struct __sk_buff *skb)
+{
+	return bpf_redirect_peer(get_dev_ifindex(dev_src), 0);
+}
+
+SEC("src_ingress") int tc_src(struct __sk_buff *skb)
+{
+	return bpf_redirect_peer(get_dev_ifindex(dev_dst), 0);
+}
+
+char __license[] SEC("license") = "GPL";
tools/testing/selftests/bpf/test_tc_redirect.sh +16 -9
index 6ad441405132..6d7482562140 100755
@@ -4,9 +4,10 @@
 # This test sets up 3 netns (src <-> fwd <-> dst). There is no direct veth link
 # between src and dst. The netns fwd has veth links to each src and dst. The
 # client is in src and server in dst. The test installs a TC BPF program to each
-# host facing veth in fwd which calls into bpf_redirect_peer() to perform the
-# neigh addr population and redirect; it also installs a dropper prog on the
-# egress side to drop skbs if neigh addrs were not populated.
+# host facing veth in fwd which calls into i) bpf_redirect_neigh() to perform the
+# neigh addr population and redirect or ii) bpf_redirect_peer() for namespace
+# switch from ingress side; it also installs a checker prog on the egress side
+# to drop unexpected traffic.
 
 if [[ $EUID -ne 0 ]]; then
 	echo "This script must be run as root"
@@ -166,15 +167,17 @@ hex_mem_str()
 	perl -e 'print join(" ", unpack("(H2)8", pack("L", @ARGV)))' $1
 }
 
-netns_setup_neigh()
+netns_setup_bpf()
 {
+	local obj=$1
+
 	ip netns exec ${NS_FWD} tc qdisc add dev veth_src_fwd clsact
-	ip netns exec ${NS_FWD} tc filter add dev veth_src_fwd ingress bpf da obj test_tc_neigh.o sec src_ingress
-	ip netns exec ${NS_FWD} tc filter add dev veth_src_fwd egress  bpf da obj test_tc_neigh.o sec chk_egress
+	ip netns exec ${NS_FWD} tc filter add dev veth_src_fwd ingress bpf da obj $obj sec src_ingress
+	ip netns exec ${NS_FWD} tc filter add dev veth_src_fwd egress  bpf da obj $obj sec chk_egress
 
 	ip netns exec ${NS_FWD} tc qdisc add dev veth_dst_fwd clsact
-	ip netns exec ${NS_FWD} tc filter add dev veth_dst_fwd ingress bpf da obj test_tc_neigh.o sec dst_ingress
-	ip netns exec ${NS_FWD} tc filter add dev veth_dst_fwd egress  bpf da obj test_tc_neigh.o sec chk_egress
+	ip netns exec ${NS_FWD} tc filter add dev veth_dst_fwd ingress bpf da obj $obj sec dst_ingress
+	ip netns exec ${NS_FWD} tc filter add dev veth_dst_fwd egress  bpf da obj $obj sec chk_egress
 
 	veth_src=$(ip netns exec ${NS_FWD} cat /sys/class/net/veth_src_fwd/ifindex)
 	veth_dst=$(ip netns exec ${NS_FWD} cat /sys/class/net/veth_dst_fwd/ifindex)
@@ -193,5 +196,9 @@ trap netns_cleanup EXIT
 set -e
 
 netns_setup
-netns_setup_neigh
+netns_setup_bpf test_tc_neigh.o
+netns_test_connectivity
+netns_cleanup
+netns_setup
+netns_setup_bpf test_tc_peer.o
 netns_test_connectivity
Signature · 1 line
2.17.1
#12 lore raw
In reply to Daniel Borkmann
This series was applied to bpf/bpf-next.git (refs/heads/master):

Hello:

This series was applied to bpf/bpf-next.git (refs/heads/master):

Quoted 11 lines · On Sun, 11 Oct 2020 01:40:00 +0200 you wrote:

This series addresses most of the feedback [0] that was to be followed up from the last series, that is, UAPI helper comment improvements and getting rid of the ifindex obj file hacks in the selftest by using a BPF map instead. The __sk_buff data/data_end pointer work, I'm planning to do in a later round as well as the mem*() BPF improvements we have in Cilium for libbpf. Next, the series adds two features, i) a helper called redirect_peer() to improve latency on netns switch, and ii) to allow map in map with dynamic inner array map sizes. Selftests for each are added as well. For details, please check individual patches, thanks!

[...]

Here is the summary with links:
- [bpf-next,v6,1/6] bpf: improve bpf_redirect_neigh helper description

    https://git.kernel.org/bpf/bpf-next/c/dd2ce6a5373c
  - [bpf-next,v6,2/6] bpf: add redirect_peer helper
    https://git.kernel.org/bpf/bpf-next/c/9aa1206e8f48
  - [bpf-next,v6,3/6] bpf: allow for map-in-map with dynamic inner array map entries
    https://git.kernel.org/bpf/bpf-next/c/4a8f87e60f6d
  - [bpf-next,v6,4/6] bpf, selftests: add test for different array inner map size
    https://git.kernel.org/bpf/bpf-next/c/6775dab73bdc
  - [bpf-next,v6,5/6] bpf, selftests: make redirect_neigh test more extensible
    https://git.kernel.org/bpf/bpf-next/c/57a73fe7c198
  - [bpf-next,v6,6/6] bpf, selftests: add redirect_peer selftest
    https://git.kernel.org/bpf/bpf-next/c/9f4c53ca23a2

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html