[PATCH bpf-next 0/3] net: skbuff: skb bitfield compaction - bpf

[PATCH bpf-next 0/3] net: skbuff: skb bitfield compaction - bpf

[Jakub Kicinski]

I'm trying to make more of the sk_buff bits optional.
Move the BPF-accessed bits a little - because they must
be at coding-time-constant offsets they must precede any
optional bit. While at it clean up the naming a bit.

Jakub Kicinski (3):
  net: skbuff: rename __pkt_vlan_present_offset to __mono_tc_offset
  net: skbuff: reorder bytes 2 and 3 of the bitfield
  net: skbuff: move the fields BPF cares about directly next to the
    offset marker

 include/linux/skbuff.h                        | 36 +++++++++----------
 net/core/filter.c                             |  8 ++---
 .../selftests/bpf/prog_tests/ctx_rewrite.c    |  6 ++--
 3 files changed, 25 insertions(+), 25 deletions(-)

-- 
2.39.2

[PATCH bpf-next 1/3] net: skbuff: rename __pkt_vlan_present_offset to __mono_tc_offset

[Jakub Kicinski]

vlan_present is gone since
commit 354259fa73e2 ("net: remove skb->vlan_present")
rename the offset field to what BPF is currently looking
for in this byte - mono_delivery_time and tc_at_ingress.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 include/linux/skbuff.h                               | 4 ++--
 net/core/filter.c                                    | 8 ++++----
 tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c | 6 +++---
 3 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index ff7ad331fb82..004009b3930f 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -956,7 +956,7 @@ struct sk_buff {
 	__u8			csum_valid:1;
 
 	/* private: */
-	__u8			__pkt_vlan_present_offset[0];
+	__u8			__mono_tc_offset[0];
 	/* public: */
 	__u8			remcsum_offload:1;
 	__u8			csum_complete_sw:1;
@@ -1080,7 +1080,7 @@ struct sk_buff {
 #define TC_AT_INGRESS_MASK		(1 << 7)
 #define SKB_MONO_DELIVERY_TIME_MASK	(1 << 5)
 #endif
-#define PKT_VLAN_PRESENT_OFFSET	offsetof(struct sk_buff, __pkt_vlan_present_offset)
+#define SKB_BF_MONO_TC_OFFSET		offsetof(struct sk_buff, __mono_tc_offset)
 
 #ifdef __KERNEL__
 /*
diff --git a/net/core/filter.c b/net/core/filter.c
index 50f649f1b4a9..3370efad1dda 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -9185,7 +9185,7 @@ static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
 	__u8 tmp_reg = BPF_REG_AX;
 
 	*insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
-			      PKT_VLAN_PRESENT_OFFSET);
+			      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);
@@ -9232,7 +9232,7 @@ static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
 		/* 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, PKT_VLAN_PRESENT_OFFSET);
+		*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,
@@ -9267,14 +9267,14 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
 	if (!prog->tstamp_type_access) {
 		__u8 tmp_reg = BPF_REG_AX;
 
-		*insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
+		*insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
 		/* Writing __sk_buff->tstamp as ingress, goto <clear> */
 		*insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
 		/* goto <store> */
 		*insn++ = BPF_JMP_A(2);
 		/* <clear>: mono_delivery_time */
 		*insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
-		*insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
+		*insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
 	}
 #endif
 
diff --git a/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c b/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c
index d5fe3d4b936c..ae7b6e50e405 100644
--- a/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c
+++ b/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c
@@ -68,17 +68,17 @@ static struct test_case test_cases[] = {
 #if defined(__x86_64__) || defined(__aarch64__)
 	{
 		N(SCHED_CLS, struct __sk_buff, tstamp),
-		.read  = "r11 = *(u8 *)($ctx + sk_buff::__pkt_vlan_present_offset);"
+		.read  = "r11 = *(u8 *)($ctx + sk_buff::__mono_tc_offset);"
 			 "w11 &= 160;"
 			 "if w11 != 0xa0 goto pc+2;"
 			 "$dst = 0;"
 			 "goto pc+1;"
 			 "$dst = *(u64 *)($ctx + sk_buff::tstamp);",
-		.write = "r11 = *(u8 *)($ctx + sk_buff::__pkt_vlan_present_offset);"
+		.write = "r11 = *(u8 *)($ctx + sk_buff::__mono_tc_offset);"
 			 "if w11 & 0x80 goto pc+1;"
 			 "goto pc+2;"
 			 "w11 &= -33;"
-			 "*(u8 *)($ctx + sk_buff::__pkt_vlan_present_offset) = r11;"
+			 "*(u8 *)($ctx + sk_buff::__mono_tc_offset) = r11;"
 			 "*(u64 *)($ctx + sk_buff::tstamp) = $src;",
 	},
 #endif
-- 
2.39.2

[PATCH bpf-next 2/3] net: skbuff: reorder bytes 2 and 3 of the bitfield

[Jakub Kicinski]

BPF needs to know the offsets of fields it tries to access.
Zero-length fields are added to make offsetof() work.
This unfortunately partitions the bitfield (fields across
the zero-length members can't be coalesced).

Reorder bytes 2 and 3, BPF needs to know the offset of fields
previously in byte 3 and some fields in byte 2 should really
be optional.

The two bytes are always in the same cacheline so it should
not matter.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 include/linux/skbuff.h | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 004009b3930f..c4122797d465 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -945,16 +945,6 @@ struct sk_buff {
 	__u8			ip_summed:2;
 	__u8			ooo_okay:1;
 
-	__u8			l4_hash:1;
-	__u8			sw_hash:1;
-	__u8			wifi_acked_valid:1;
-	__u8			wifi_acked:1;
-	__u8			no_fcs:1;
-	/* Indicates the inner headers are valid in the skbuff. */
-	__u8			encapsulation:1;
-	__u8			encap_hdr_csum:1;
-	__u8			csum_valid:1;
-
 	/* private: */
 	__u8			__mono_tc_offset[0];
 	/* public: */
@@ -967,6 +957,16 @@ struct sk_buff {
 	__u8			tc_skip_classify:1;
 	__u8			tc_at_ingress:1;	/* See TC_AT_INGRESS_MASK */
 #endif
+
+	__u8			l4_hash:1;
+	__u8			sw_hash:1;
+	__u8			wifi_acked_valid:1;
+	__u8			wifi_acked:1;
+	__u8			no_fcs:1;
+	/* Indicates the inner headers are valid in the skbuff. */
+	__u8			encapsulation:1;
+	__u8			encap_hdr_csum:1;
+	__u8			csum_valid:1;
 #ifdef CONFIG_IPV6_NDISC_NODETYPE
 	__u8			ndisc_nodetype:2;
 #endif
-- 
2.39.2

[PATCH bpf-next 3/3] net: skbuff: move the fields BPF cares about directly next to the offset marker

[Jakub Kicinski]

To avoid more possible BPF dependencies with moving bitfields
around keep the fields BPF cares about right next to the offset
marker.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 include/linux/skbuff.h | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index c4122797d465..3716818b804d 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -948,15 +948,15 @@ struct sk_buff {
 	/* private: */
 	__u8			__mono_tc_offset[0];
 	/* public: */
-	__u8			remcsum_offload:1;
-	__u8			csum_complete_sw:1;
-	__u8			csum_level:2;
-	__u8			dst_pending_confirm:1;
 	__u8			mono_delivery_time:1;	/* See SKB_MONO_DELIVERY_TIME_MASK */
 #ifdef CONFIG_NET_CLS_ACT
-	__u8			tc_skip_classify:1;
 	__u8			tc_at_ingress:1;	/* See TC_AT_INGRESS_MASK */
+	__u8			tc_skip_classify:1;
 #endif
+	__u8			remcsum_offload:1;
+	__u8			csum_complete_sw:1;
+	__u8			csum_level:2;
+	__u8			dst_pending_confirm:1;
 
 	__u8			l4_hash:1;
 	__u8			sw_hash:1;
@@ -1074,11 +1074,11 @@ struct sk_buff {
  * around, you also must adapt these constants.
  */
 #ifdef __BIG_ENDIAN_BITFIELD
-#define TC_AT_INGRESS_MASK		(1 << 0)
-#define SKB_MONO_DELIVERY_TIME_MASK	(1 << 2)
+#define SKB_MONO_DELIVERY_TIME_MASK	(1 << 7)
+#define TC_AT_INGRESS_MASK		(1 << 6)
 #else
-#define TC_AT_INGRESS_MASK		(1 << 7)
-#define SKB_MONO_DELIVERY_TIME_MASK	(1 << 5)
+#define SKB_MONO_DELIVERY_TIME_MASK	(1 << 0)
+#define TC_AT_INGRESS_MASK		(1 << 1)
 #endif
 #define SKB_BF_MONO_TC_OFFSET		offsetof(struct sk_buff, __mono_tc_offset)
 
-- 
2.39.2

Re: [PATCH bpf-next 1/3] net: skbuff: rename __pkt_vlan_present_offset to __mono_tc_offset

[Martin KaFai Lau]

On 3/7/23 4:31 PM, Jakub Kicinski wrote:
> vlan_present is gone since
> commit 354259fa73e2 ("net: remove skb->vlan_present")
> rename the offset field to what BPF is currently looking
> for in this byte - mono_delivery_time and tc_at_ingress.
> 
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> ---
>   include/linux/skbuff.h                               | 4 ++--
>   net/core/filter.c                                    | 8 ++++----
>   tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c | 6 +++---
>   3 files changed, 9 insertions(+), 9 deletions(-)
> 
> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
> index ff7ad331fb82..004009b3930f 100644
> --- a/include/linux/skbuff.h
> +++ b/include/linux/skbuff.h
> @@ -956,7 +956,7 @@ struct sk_buff {
>   	__u8			csum_valid:1;
>   
>   	/* private: */
> -	__u8			__pkt_vlan_present_offset[0];
> +	__u8			__mono_tc_offset[0];
>   	/* public: */
>   	__u8			remcsum_offload:1;
>   	__u8			csum_complete_sw:1;
> @@ -1080,7 +1080,7 @@ struct sk_buff {
>   #define TC_AT_INGRESS_MASK		(1 << 7)
>   #define SKB_MONO_DELIVERY_TIME_MASK	(1 << 5)
>   #endif
> -#define PKT_VLAN_PRESENT_OFFSET	offsetof(struct sk_buff, __pkt_vlan_present_offset)
> +#define SKB_BF_MONO_TC_OFFSET		offsetof(struct sk_buff, __mono_tc_offset)
>   
>   #ifdef __KERNEL__
>   /*
> diff --git a/net/core/filter.c b/net/core/filter.c
> index 50f649f1b4a9..3370efad1dda 100644
> --- a/net/core/filter.c
> +++ b/net/core/filter.c
> @@ -9185,7 +9185,7 @@ static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
>   	__u8 tmp_reg = BPF_REG_AX;
>   
>   	*insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
> -			      PKT_VLAN_PRESENT_OFFSET);
> +			      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);
> @@ -9232,7 +9232,7 @@ static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
>   		/* 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, PKT_VLAN_PRESENT_OFFSET);
> +		*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,
> @@ -9267,14 +9267,14 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
>   	if (!prog->tstamp_type_access) {
>   		__u8 tmp_reg = BPF_REG_AX;
>   
> -		*insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
> +		*insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
>   		/* Writing __sk_buff->tstamp as ingress, goto <clear> */
>   		*insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
>   		/* goto <store> */
>   		*insn++ = BPF_JMP_A(2);
>   		/* <clear>: mono_delivery_time */
>   		*insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
> -		*insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
> +		*insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
>   	}
>   #endif
>   
> diff --git a/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c b/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c
> index d5fe3d4b936c..ae7b6e50e405 100644
> --- a/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c
> +++ b/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c
> @@ -68,17 +68,17 @@ static struct test_case test_cases[] = {
>   #if defined(__x86_64__) || defined(__aarch64__)
>   	{
>   		N(SCHED_CLS, struct __sk_buff, tstamp),
> -		.read  = "r11 = *(u8 *)($ctx + sk_buff::__pkt_vlan_present_offset);"
> +		.read  = "r11 = *(u8 *)($ctx + sk_buff::__mono_tc_offset);"
>   			 "w11 &= 160;"

The mask needs to be adjusted also after patch 3. This selftest is failing: 
https://github.com/kernel-patches/bpf/actions/runs/4369844423/jobs/7644281598

>   			 "if w11 != 0xa0 goto pc+2;"
>   			 "$dst = 0;"
>   			 "goto pc+1;"
>   			 "$dst = *(u64 *)($ctx + sk_buff::tstamp);",
> -		.write = "r11 = *(u8 *)($ctx + sk_buff::__pkt_vlan_present_offset);"
> +		.write = "r11 = *(u8 *)($ctx + sk_buff::__mono_tc_offset);"
>   			 "if w11 & 0x80 goto pc+1;"

It probably needs to adjust here also

>   			 "goto pc+2;"
>   			 "w11 &= -33;"

and here.

> -			 "*(u8 *)($ctx + sk_buff::__pkt_vlan_present_offset) = r11;"
> +			 "*(u8 *)($ctx + sk_buff::__mono_tc_offset) = r11;"
>   			 "*(u64 *)($ctx + sk_buff::tstamp) = $src;",
>   	},
>   #endif