[RFC PATCH bpf-next v6 2/3] net: Add additional bit to support clockid_t timestamp type

From: Abhishek Chauhan
Date: Fri May 03 2024 - 23:14:15 EST


tstamp_type is now set based on actual clockid_t compressed
into 2 bits.

To make the design scalable for future needs this commit bring in
the change to extend the tstamp_type:1 to tstamp_type:2 to support
other clockid_t timestamp.

We now support CLOCK_TAI as part of tstamp_type as part of this
commit with exisiting support CLOCK_MONOTONIC and CLOCK_REALTIME.

Link: https://lore.kernel.org/netdev/bc037db4-58bb-4861-ac31-a361a93841d3@xxxxxxxxx/
Signed-off-by: Abhishek Chauhan <quic_abchauha@xxxxxxxxxxx>
---
Changes since v5
- Took care of documentation comments of tstamp_type
in skbuff.h as mentioned by Willem.
- Use of complete words instead of abbrevation in
macro definitions as mentioned by Willem.
- Fixed indentation problems
- Removed BPF_SKB_TSTAMP_UNSPEC and marked it
Deprecated as documentation, and introduced
BPF_SKB_CLOCK_REALTIME instead.
- BUILD_BUG_ON for additional enums introduced.
- __ip_make_skb and ip6_make_skb now has
tcp checks to mark tcp packet as mono tstamp base.
- separated the selftests/bpf changes into another patch.
- Made changes as per Martin in selftest bpf code and
tool/include/uapi/linux/bpf.h

Changes since v4
- Made changes to BPF code in filter.c as per
Martin's comments
- Minor fixes on comments given on documentation
from Willem in skbuff.h (removed obvious ones)
- Made changes to ctx_rewrite.c and test_tc_dtime.c
- test_tc_dtime.c i am not really sure if i took care
of all the changes as i am not too familiar with
the framework.
- Introduce common mask SKB_TSTAMP_TYPE_MASK instead
of multiple SKB mask.
- Optimisation on BPF code as suggested by Martin.
- Set default case to SKB_CLOCK_REALTME.

Changes since v3
- Carefully reviewed BPF APIs and made changes in
BPF code as well.
- Re-used actual clockid_t values since skbuff.h
indirectly includes uapi/linux/time.h
- Added CLOCK_TAI as part of the skb_set_delivery_time
handling instead of CLOCK_USER
- Added default in switch for unsupported and invalid
timestamp with an WARN_ONCE
- All of the above comments were given by Willem
- Made changes in filter.c as per Martin's comments
to handle invalid cases in bpf code with addition of
SKB_TAI_DELIVERY_TIME_MASK

Changes since v2
- Minor changes to commit subject

Changes since v1
- identified additional changes in BPF framework.
- Bit shift in SKB_MONO_DELIVERY_TIME_MASK and TC_AT_INGRESS_MASK.
- Made changes in skb_set_delivery_time to keep changes similar to
previous code for mono_delivery_time and just setting tstamp_type
bit 1 for userspace timestamp.


include/linux/skbuff.h | 21 +++++++++++--------
include/uapi/linux/bpf.h | 15 +++++++++-----
net/core/filter.c | 44 +++++++++++++++++++++++-----------------
net/ipv4/ip_output.c | 5 ++++-
net/ipv4/raw.c | 2 +-
net/ipv6/ip6_output.c | 5 ++++-
net/ipv6/raw.c | 2 +-
net/packet/af_packet.c | 7 +++----
8 files changed, 61 insertions(+), 40 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index de3915e2bfdb..fe7d8dbef77e 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -709,6 +709,8 @@ typedef unsigned char *sk_buff_data_t;
enum skb_tstamp_type {
SKB_CLOCK_REALTIME,
SKB_CLOCK_MONOTONIC,
+ SKB_CLOCK_TAI,
+ __SKB_CLOCK_MAX = SKB_CLOCK_TAI,
};

/**
@@ -829,8 +831,7 @@ enum skb_tstamp_type {
* @decrypted: Decrypted SKB
* @slow_gro: state present at GRO time, slower prepare step required
* @tstamp_type: When set, skb->tstamp has the
- * delivery_time in mono clock base Otherwise, the
- * timestamp is considered real clock base.
+ * delivery_time clock base of skb->tstamp.
* @napi_id: id of the NAPI struct this skb came from
* @sender_cpu: (aka @napi_id) source CPU in XPS
* @alloc_cpu: CPU which did the skb allocation.
@@ -958,7 +959,7 @@ struct sk_buff {
/* private: */
__u8 __mono_tc_offset[0];
/* public: */
- __u8 tstamp_type:1; /* See skb_tstamp_type */
+ __u8 tstamp_type:2; /* See skb_tstamp_type */
#ifdef CONFIG_NET_XGRESS
__u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
__u8 tc_skip_classify:1;
@@ -1088,15 +1089,16 @@ struct sk_buff {
#endif
#define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)

-/* if you move tc_at_ingress or mono_delivery_time
+/* if you move tc_at_ingress or tstamp_type
* around, you also must adapt these constants.
*/
#ifdef __BIG_ENDIAN_BITFIELD
-#define SKB_MONO_DELIVERY_TIME_MASK (1 << 7)
-#define TC_AT_INGRESS_MASK (1 << 6)
+#define SKB_TSTAMP_TYPE_MASK (3 << 6)
+#define SKB_TSTAMP_TYPE_RSHIFT (6)
+#define TC_AT_INGRESS_MASK (1 << 5)
#else
-#define SKB_MONO_DELIVERY_TIME_MASK (1 << 0)
-#define TC_AT_INGRESS_MASK (1 << 1)
+#define SKB_TSTAMP_TYPE_MASK (3)
+#define TC_AT_INGRESS_MASK (1 << 2)
#endif
#define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset)

@@ -4213,6 +4215,9 @@ static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb,
case CLOCK_MONOTONIC:
tstamp_type = SKB_CLOCK_MONOTONIC;
break;
+ case CLOCK_TAI:
+ tstamp_type = SKB_CLOCK_TAI;
+ break;
default:
WARN_ON_ONCE(1);
kt = 0;
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index 90706a47f6ff..25ea393cf084 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -6207,12 +6207,17 @@ union { \
__u64 :64; \
} __attribute__((aligned(8)))

+/* The enum used in skb->tstamp_type. It specifies the clock type
+ * of the time stored in the skb->tstamp.
+ */
enum {
- BPF_SKB_TSTAMP_UNSPEC,
- BPF_SKB_TSTAMP_DELIVERY_MONO, /* tstamp has mono delivery time */
- /* For any BPF_SKB_TSTAMP_* that the bpf prog cannot handle,
- * the bpf prog should handle it like BPF_SKB_TSTAMP_UNSPEC
- * and try to deduce it by ingress, egress or skb->sk->sk_clockid.
+ BPF_SKB_TSTAMP_UNSPEC = 0, /* DEPRECATED */
+ BPF_SKB_TSTAMP_DELIVERY_MONO = 1, /* DEPRECATED */
+ BPF_SKB_CLOCK_REALTIME = 0,
+ BPF_SKB_CLOCK_MONOTONIC = 1,
+ BPF_SKB_CLOCK_TAI = 2,
+ /* For any future BPF_SKB_CLOCK_* that the bpf prog cannot handle,
+ * the bpf prog can try to deduce it by ingress/egress/skb->sk->sk_clockid.
*/
};

diff --git a/net/core/filter.c b/net/core/filter.c
index a3781a796da4..9f3df4a0d1ee 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -7726,16 +7726,20 @@ BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
return -EOPNOTSUPP;

switch (tstamp_type) {
- case BPF_SKB_TSTAMP_DELIVERY_MONO:
+ case BPF_SKB_CLOCK_MONOTONIC:
if (!tstamp)
return -EINVAL;
skb->tstamp = tstamp;
skb->tstamp_type = SKB_CLOCK_MONOTONIC;
break;
- case BPF_SKB_TSTAMP_UNSPEC:
- if (tstamp)
+ case BPF_SKB_CLOCK_TAI:
+ if (!tstamp)
return -EINVAL;
- skb->tstamp = 0;
+ skb->tstamp = tstamp;
+ skb->tstamp_type = SKB_CLOCK_TAI;
+ break;
+ case BPF_SKB_CLOCK_REALTIME:
+ skb->tstamp = tstamp;
skb->tstamp_type = SKB_CLOCK_REALTIME;
break;
default:
@@ -9387,16 +9391,17 @@ static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
{
__u8 value_reg = si->dst_reg;
__u8 skb_reg = si->src_reg;
- /* AX is needed because src_reg and dst_reg could be the same */
- __u8 tmp_reg = BPF_REG_AX;
-
- *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
- SKB_BF_MONO_TC_OFFSET);
- *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
- SKB_MONO_DELIVERY_TIME_MASK, 2);
- *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
- *insn++ = BPF_JMP_A(1);
- *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
+ BUILD_BUG_ON(__SKB_CLOCK_MAX != (int)BPF_SKB_CLOCK_TAI);
+ BUILD_BUG_ON(SKB_CLOCK_REALTIME != (int)BPF_SKB_CLOCK_REALTIME);
+ BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != (int)BPF_SKB_CLOCK_MONOTONIC);
+ BUILD_BUG_ON(SKB_CLOCK_TAI != (int)BPF_SKB_CLOCK_TAI);
+ *insn++ = BPF_LDX_MEM(BPF_B, value_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
+ *insn++ = BPF_ALU32_IMM(BPF_AND, value_reg, SKB_TSTAMP_TYPE_MASK);
+#ifdef __BIG_ENDIAN_BITFIELD
+ *insn++ = BPF_ALU32_IMM(BPF_RSH, value_reg, SKB_TSTAMP_TYPE_RSHIFT);
+#else
+ BUILD_BUG_ON(!(SKB_TSTAMP_TYPE_MASK & 0x1));
+#endif

return insn;
}
@@ -9439,10 +9444,11 @@ static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
__u8 tmp_reg = BPF_REG_AX;

*insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
- *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
- TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
- *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
- TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
+ /* check if ingress mask bits is set */
+ *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
+ *insn++ = BPF_JMP_A(4);
+ *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, SKB_TSTAMP_TYPE_MASK, 1);
+ *insn++ = BPF_JMP_A(2);
/* skb->tc_at_ingress && skb->tstamp_type,
* read 0 as the (rcv) timestamp.
*/
@@ -9479,7 +9485,7 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
/* goto <store> */
*insn++ = BPF_JMP_A(2);
/* <clear>: skb->tstamp_type */
- *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
+ *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_TSTAMP_TYPE_MASK);
*insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
}
#endif
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index fe86cadfa85b..c3d852eecb01 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1457,7 +1457,10 @@ struct sk_buff *__ip_make_skb(struct sock *sk,

skb->priority = (cork->tos != -1) ? cork->priority: READ_ONCE(sk->sk_priority);
skb->mark = cork->mark;
- skb->tstamp = cork->transmit_time;
+ if (sk_is_tcp(sk))
+ skb_set_delivery_type_by_clockid(skb, cork->transmit_time, CLOCK_MONOTONIC);
+ else
+ skb_set_delivery_type_by_clockid(skb, cork->transmit_time, sk->sk_clockid);
/*
* Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
* on dst refcount
diff --git a/net/ipv4/raw.c b/net/ipv4/raw.c
index 4cb43401e0e0..1a0953650356 100644
--- a/net/ipv4/raw.c
+++ b/net/ipv4/raw.c
@@ -360,7 +360,7 @@ static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
skb->protocol = htons(ETH_P_IP);
skb->priority = READ_ONCE(sk->sk_priority);
skb->mark = sockc->mark;
- skb->tstamp = sockc->transmit_time;
+ skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, sk->sk_clockid);
skb_dst_set(skb, &rt->dst);
*rtp = NULL;

diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index 05067bd44775..797a9764e8fe 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -1924,7 +1924,10 @@ struct sk_buff *__ip6_make_skb(struct sock *sk,

skb->priority = READ_ONCE(sk->sk_priority);
skb->mark = cork->base.mark;
- skb->tstamp = cork->base.transmit_time;
+ if (sk_is_tcp(sk))
+ skb_set_delivery_type_by_clockid(skb, cork->base.transmit_time, CLOCK_MONOTONIC);
+ else
+ skb_set_delivery_type_by_clockid(skb, cork->base.transmit_time, sk->sk_clockid);

ip6_cork_steal_dst(skb, cork);
IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
diff --git a/net/ipv6/raw.c b/net/ipv6/raw.c
index 2eedf255600b..f838366e8256 100644
--- a/net/ipv6/raw.c
+++ b/net/ipv6/raw.c
@@ -621,7 +621,7 @@ static int rawv6_send_hdrinc(struct sock *sk, struct msghdr *msg, int length,
skb->protocol = htons(ETH_P_IPV6);
skb->priority = READ_ONCE(sk->sk_priority);
skb->mark = sockc->mark;
- skb->tstamp = sockc->transmit_time;
+ skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, sk->sk_clockid);

skb_put(skb, length);
skb_reset_network_header(skb);
diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index 8c6d3fbb4ed8..89b54021d196 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -2056,8 +2056,7 @@ static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
skb->dev = dev;
skb->priority = READ_ONCE(sk->sk_priority);
skb->mark = READ_ONCE(sk->sk_mark);
- skb->tstamp = sockc.transmit_time;
-
+ skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
skb_setup_tx_timestamp(skb, sockc.tsflags);

if (unlikely(extra_len == 4))
@@ -2585,7 +2584,7 @@ static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
skb->dev = dev;
skb->priority = READ_ONCE(po->sk.sk_priority);
skb->mark = READ_ONCE(po->sk.sk_mark);
- skb->tstamp = sockc->transmit_time;
+ skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid);
skb_setup_tx_timestamp(skb, sockc->tsflags);
skb_zcopy_set_nouarg(skb, ph.raw);

@@ -3063,7 +3062,7 @@ static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
skb->dev = dev;
skb->priority = READ_ONCE(sk->sk_priority);
skb->mark = sockc.mark;
- skb->tstamp = sockc.transmit_time;
+ skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);

if (unlikely(extra_len == 4))
skb->no_fcs = 1;
--
2.25.1