| Message ID | 20240424222028.1080134-3-quic_abchauha@quicinc.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | BPF |
| Headers | show |
| Series | Replace mono_delivery_time with tstamp_type | expand |
Abhishek Chauhan wrote: > 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@linux.dev/ > Signed-off-by: Abhishek Chauhan <quic_abchauha@quicinc.com> > --- > diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h > index e464d0ebc9c1..3ad0de07d261 100644 > --- a/include/linux/skbuff.h > +++ b/include/linux/skbuff.h > @@ -711,6 +711,8 @@ 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, > }; > > /** > @@ -831,8 +833,8 @@ 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 in mono clock base or clock base of skb->tstamp. drop "in mono clock base or " > + * Otherwise, the timestamp is considered real clock base drop this: whenever in realtime clock base, tstamp_type is zero, so the above shorter statement always holds. > * @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. > @@ -960,7 +962,7 @@ 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; > @@ -1090,15 +1092,17 @@ 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:2 > * 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_RSH (6) > +#define TC_AT_INGRESS_RSH (5) I had to find BPF_RSH to understand this abbreviation. use SHIFT instead of RSH, as that is so domain specific? > +#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) > > - if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO) { > + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO || > + skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI) { Peculiar indentation? Just FYI that I'm not the best person to review the BPF part. Thankfully Martin is helping you with that.
On 4/25/2024 7:42 AM, Willem de Bruijn wrote: > Abhishek Chauhan wrote: >> 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@linux.dev/ >> Signed-off-by: Abhishek Chauhan <quic_abchauha@quicinc.com> >> --- > >> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h >> index e464d0ebc9c1..3ad0de07d261 100644 >> --- a/include/linux/skbuff.h >> +++ b/include/linux/skbuff.h >> @@ -711,6 +711,8 @@ 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, >> }; >> >> /** >> @@ -831,8 +833,8 @@ 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 in mono clock base or clock base of skb->tstamp. > > drop "in mono clock base or " > Noted >> + * Otherwise, the timestamp is considered real clock base > Noted > drop this: whenever in realtime clock base, tstamp_type is zero, so > the above shorter statement always holds. > >> * @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. >> @@ -960,7 +962,7 @@ 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; >> @@ -1090,15 +1092,17 @@ 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:2 >> * 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_RSH (6) >> +#define TC_AT_INGRESS_RSH (5) > > I had to find BPF_RSH to understand this abbreviation. > > use SHIFT instead of RSH, as that is so domain specific? > Noted! I will use complete words instead of abbreviations >> +#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) >> > >> - if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO) { >> + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO || >> + skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI) { > > Peculiar indentation? > Let me check why the indentation here is messed up. Ideally i run checkpatch(shows 0 errors or warnings) and also check before raising a patch. Internally it looks good but on the patch it shows differently. > Just FYI that I'm not the best person to review the BPF part. > Thankfully Martin is helping you with that. > I will wait for comments from Martin as well. > >
On 4/24/24 3:20 PM, Abhishek Chauhan wrote: > diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h > index e464d0ebc9c1..3ad0de07d261 100644 > --- a/include/linux/skbuff.h > +++ b/include/linux/skbuff.h > @@ -711,6 +711,8 @@ 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, > }; > > /** > @@ -831,8 +833,8 @@ 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 in mono clock base or clock base of skb->tstamp. > + * Otherwise, the timestamp is considered real clock base > * @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. > @@ -960,7 +962,7 @@ 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; > @@ -1090,15 +1092,17 @@ 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:2 > * 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_RSH (6) > +#define TC_AT_INGRESS_RSH (5) TC_AT_INGRESS_RSH is not used. > +#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) > > @@ -4204,6 +4208,12 @@ static inline void skb_set_tstamp_type_frm_clkid(struct sk_buff *skb, > case CLOCK_MONOTONIC: > skb->tstamp_type = SKB_CLOCK_MONOTONIC; > break; > + case CLOCK_TAI: > + skb->tstamp_type = SKB_CLOCK_TAI; > + break; > + default: > + WARN_ONCE(true, "clockid %d not supported", clockid); > + skb->tstamp_type = SKB_CLOCK_REALTIME; > } > } > > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h > index cee0a7915c08..1376ed5ece10 100644 > --- a/include/uapi/linux/bpf.h > +++ b/include/uapi/linux/bpf.h The bpf.h needs to be sync to tools/include/uapi/linux/bpf.h. Otherwise, the bpf CI cannot compile the tests: https://patchwork.kernel.org/project/netdevbpf/patch/20240424222028.1080134-2-quic_abchauha@quicinc.com/ Please monitor the bpf CI test result after submitting the patches. > @@ -6209,6 +6209,7 @@ union { \ > enum { > BPF_SKB_TSTAMP_UNSPEC, > BPF_SKB_TSTAMP_DELIVERY_MONO, /* tstamp has mono delivery time */ > + BPF_SKB_TSTAMP_DELIVERY_TAI, /* tstamp has tai 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. SKB_CLOCK_TAI is properly defined as an enum now and there is a WARN for clock other than REAL, MONO, and TAI. I think it is time to remove UNSPEC and give it back the proper name REALTIME. I want to take this chance to do some renaming: /* 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 = 0, /* DEPRECATED */ BPF_SKB_TSTAMP_DELIVERY_MONO = 1, /* DEPRECATED */ BPF_SKB_CLOCK_REALTIME = 0, /* Realtime clock */ BPF_SKB_CLOCK_MONOTONIC = 1, /* Monotonic clock */ BPF_SKB_CLOCK_TAI = 2, /* TAI clock */ /* 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 957c2fc724eb..c67622f4fe98 100644 > --- a/net/core/filter.c > +++ b/net/core/filter.c > @@ -7733,6 +7733,12 @@ BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, > skb->tstamp = tstamp; > skb->tstamp_type = SKB_CLOCK_MONOTONIC; > break; > + case BPF_SKB_TSTAMP_DELIVERY_TAI: > + if (!tstamp) > + return -EINVAL; > + skb->tstamp = tstamp; > + skb->tstamp_type = SKB_CLOCK_TAI; > + break; > case BPF_SKB_TSTAMP_UNSPEC: > if (tstamp) Allow to store any realtime tstamp here since BPF_SKB_TSTAMP_UNSPEC becomes BPF_SKB_CLOCK_REALTIME. Like: BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, u64, tstamp, u32, tstamp_type) { /* ... */ case BPF_SKB_CLOCK_TAI: if (!tstamp) return -EINVAL; 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; /* ... */ } > return -EINVAL; > @@ -9388,17 +9394,17 @@ 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 != BPF_SKB_TSTAMP_DELIVERY_TAI); Add these also: BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != BPF_SKB_CLOCK_MONOTONIC); BUILD_BUG_ON(SKB_CLOCK_TAI != 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_RSH); > +#else > + BUILD_BUG_ON(!(SKB_TSTAMP_TYPE_MASK & 0x1)); > +#endif > + *insn++ = BPF_JMP32_IMM(BPF_JNE, value_reg, SKB_TSTAMP_TYPE_MASK, 1); > + /* Both the bits set then mark it BPF_SKB_TSTAMP_UNSPEC */ > + *insn++ = BPF_MOV64_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC); The kernel should not have both bits set in skb->tstamp_type. No need to add two extra bpf insns to check this. If there is a bug in the kernel, it is better to be uncovered instead of hiding it under BPF_SKB_TSTAMP_UNSPEC (which is renamed to BPF_SKB_CLOCK_REALTIME anyway). Hence, the last two bpf insns should be removed. > return insn; > } > > @@ -9430,6 +9436,7 @@ static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, > __u8 value_reg = si->dst_reg; > __u8 skb_reg = si->src_reg; > > +BUILD_BUG_ON(__SKB_CLOCK_MAX != BPF_SKB_TSTAMP_DELIVERY_TAI); It is a dup of the one in bpf_convert_tstamp_type_read and can be removed. > #ifdef CONFIG_NET_XGRESS > /* If the tstamp_type is read, > * the bpf prog is aware the tstamp could have delivery time. > @@ -9440,11 +9447,12 @@ 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); > - /* skb->tc_at_ingress && skb->tstamp_type:1, > + /* 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:2, > * read 0 as the (rcv) timestamp. > */ > *insn++ = BPF_MOV64_IMM(value_reg, 0); > @@ -9469,7 +9477,7 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, > * the bpf prog is aware the tstamp could have delivery time. > * Thus, write skb->tstamp as is if tstamp_type_access is true. > * Otherwise, writing at ingress will have to clear the > - * mono_delivery_time (skb->tstamp_type:1)bit also. > + * mono_delivery_time (skb->tstamp_type:2)bit also. > */ > if (!prog->tstamp_type_access) { > __u8 tmp_reg = BPF_REG_AX; > @@ -9479,8 +9487,8 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, > *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); > /* goto <store> */ > *insn++ = BPF_JMP_A(2); > - /* <clear>: mono_delivery_time or (skb->tstamp_type:1) */ > - *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK); > + /* <clear>: skb->tstamp_type:2 */ > + *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 591226dcde26..f195b31d6e75 100644 > --- a/net/ipv4/ip_output.c > +++ b/net/ipv4/ip_output.c > @@ -1457,7 +1457,7 @@ 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; > + skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); hmm... I think this will break for tcp. This sequence in particular: tcp_v4_timewait_ack() tcp_v4_send_ack() ip_send_unicast_reply() ip_push_pending_frames() ip_finish_skb() __ip_make_skb() /* sk_clockid is REAL but cork->transmit_time should be in mono */ skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);; I think I hit it from time to time when running the test in this patch set. [ ... ] > diff --git a/tools/testing/selftests/bpf/progs/test_tc_dtime.c b/tools/testing/selftests/bpf/progs/test_tc_dtime.c > index 74ec09f040b7..19dba6d88265 100644 > --- a/tools/testing/selftests/bpf/progs/test_tc_dtime.c > +++ b/tools/testing/selftests/bpf/progs/test_tc_dtime.c Please separate the selftests/bpf changes into another patch. > @@ -227,6 +227,12 @@ int egress_host(struct __sk_buff *skb) > inc_dtimes(EGRESS_ENDHOST); > else > inc_errs(EGRESS_ENDHOST); > + } else if (skb_proto(skb_type) == IPPROTO_UDP) { > + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI && > + skb->tstamp) > + inc_dtimes(EGRESS_ENDHOST); > + else > + inc_errs(EGRESS_ENDHOST); > } else { > if (skb->tstamp_type == BPF_SKB_TSTAMP_UNSPEC && > skb->tstamp) > @@ -255,6 +261,9 @@ int ingress_host(struct __sk_buff *skb) > if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO && > skb->tstamp == EGRESS_FWDNS_MAGIC) > inc_dtimes(INGRESS_ENDHOST); > + else if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI && > + skb->tstamp == EGRESS_FWDNS_MAGIC) > + inc_dtimes(INGRESS_ENDHOST); > else > inc_errs(INGRESS_ENDHOST); > > @@ -323,12 +332,14 @@ int ingress_fwdns_prio101(struct __sk_buff *skb) > /* Should have handled in prio100 */ > return TC_ACT_SHOT; > > - if (skb_proto(skb_type) == IPPROTO_UDP) > + if (skb_proto(skb_type) == IPPROTO_UDP && > + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) > expected_dtime = 0; The IPPROTO_UDP check and expected_dtime can be removed. The UDP test can expect the same EGRESS_ENDHOST_MAGIC in the skb->tstamp since the TAI tstamp is also forwarded from egress to ingress now. > > if (skb->tstamp_type) { > if (fwdns_clear_dtime() || > - skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO || > + (skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO && > + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) || > skb->tstamp != expected_dtime) > inc_errs(INGRESS_FWDNS_P101); > else > @@ -338,7 +349,8 @@ int ingress_fwdns_prio101(struct __sk_buff *skb) > inc_errs(INGRESS_FWDNS_P101); > } > > - if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO) { > + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO || > + skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI) { No need to check BPF_SKB_TSTAMP_DELIVERY_TAI such that the bpf_skb_set_tstamp() helper can still be tested. There are some other minor changes needed for the test_tc_dtime.c and the tc_redirect.c. I quickly made the changes and put them here (first patch): https://git.kernel.org/pub/scm/linux/kernel/git/martin.lau/bpf-next.git/log/?h=skb.tstamp_type > skb->tstamp = INGRESS_FWDNS_MAGIC; > } else { > if (bpf_skb_set_tstamp(skb, INGRESS_FWDNS_MAGIC, > @@ -370,7 +382,8 @@ int egress_fwdns_prio101(struct __sk_buff *skb) > > if (skb->tstamp_type) { > if (fwdns_clear_dtime() || > - skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO || > + (skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO && > + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) || > skb->tstamp != INGRESS_FWDNS_MAGIC) > inc_errs(EGRESS_FWDNS_P101); > else > @@ -380,7 +393,8 @@ int egress_fwdns_prio101(struct __sk_buff *skb) > inc_errs(EGRESS_FWDNS_P101); > } > > - if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO) { > + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO || > + skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI) { > skb->tstamp = EGRESS_FWDNS_MAGIC; > } else { > if (bpf_skb_set_tstamp(skb, EGRESS_FWDNS_MAGIC,
On 4/25/2024 9:37 PM, Martin KaFai Lau wrote: > On 4/24/24 3:20 PM, Abhishek Chauhan wrote: >> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h >> index e464d0ebc9c1..3ad0de07d261 100644 >> --- a/include/linux/skbuff.h >> +++ b/include/linux/skbuff.h >> @@ -711,6 +711,8 @@ 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, >> }; >> /** >> @@ -831,8 +833,8 @@ 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 in mono clock base or clock base of skb->tstamp. >> + * Otherwise, the timestamp is considered real clock base >> * @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. >> @@ -960,7 +962,7 @@ 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; >> @@ -1090,15 +1092,17 @@ 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:2 >> * 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_RSH (6) >> +#define TC_AT_INGRESS_RSH (5) > > TC_AT_INGRESS_RSH is not used. > Noted. I will remove it. >> +#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) >> @@ -4204,6 +4208,12 @@ static inline void skb_set_tstamp_type_frm_clkid(struct sk_buff *skb, >> case CLOCK_MONOTONIC: >> skb->tstamp_type = SKB_CLOCK_MONOTONIC; >> break; >> + case CLOCK_TAI: >> + skb->tstamp_type = SKB_CLOCK_TAI; >> + break; >> + default: >> + WARN_ONCE(true, "clockid %d not supported", clockid); >> + skb->tstamp_type = SKB_CLOCK_REALTIME; >> } >> } >> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h >> index cee0a7915c08..1376ed5ece10 100644 >> --- a/include/uapi/linux/bpf.h >> +++ b/include/uapi/linux/bpf.h > > The bpf.h needs to be sync to tools/include/uapi/linux/bpf.h. Oh i see. There is a bpf.h in tools as well . I was not aware of this part. I will further check this. > Otherwise, the bpf CI cannot compile the tests: > > https://patchwork.kernel.org/project/netdevbpf/patch/20240424222028.1080134-2-quic_abchauha@quicinc.com/ > > Please monitor the bpf CI test result after submitting the patches. > >> @@ -6209,6 +6209,7 @@ union { \ >> enum { >> BPF_SKB_TSTAMP_UNSPEC, >> BPF_SKB_TSTAMP_DELIVERY_MONO, /* tstamp has mono delivery time */ >> + BPF_SKB_TSTAMP_DELIVERY_TAI, /* tstamp has tai 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. > > SKB_CLOCK_TAI is properly defined as an enum now and there is a > WARN for clock other than REAL, MONO, and TAI. I think it is > time to remove UNSPEC and give it back the proper name REALTIME. > > I want to take this chance to do some renaming: > > /* 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 = 0, /* DEPRECATED */ > BPF_SKB_TSTAMP_DELIVERY_MONO = 1, /* DEPRECATED */ > BPF_SKB_CLOCK_REALTIME = 0, /* Realtime clock */ > BPF_SKB_CLOCK_MONOTONIC = 1, /* Monotonic clock */ > BPF_SKB_CLOCK_TAI = 2, /* TAI clock */ > /* 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. > */ > }; > Okay let me evalute this and get back > >> diff --git a/net/core/filter.c b/net/core/filter.c >> index 957c2fc724eb..c67622f4fe98 100644 >> --- a/net/core/filter.c >> +++ b/net/core/filter.c >> @@ -7733,6 +7733,12 @@ BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, >> skb->tstamp = tstamp; >> skb->tstamp_type = SKB_CLOCK_MONOTONIC; >> break; >> + case BPF_SKB_TSTAMP_DELIVERY_TAI: >> + if (!tstamp) >> + return -EINVAL; >> + skb->tstamp = tstamp; >> + skb->tstamp_type = SKB_CLOCK_TAI; >> + break; >> case BPF_SKB_TSTAMP_UNSPEC: >> if (tstamp) > > Allow to store any realtime tstamp here since BPF_SKB_TSTAMP_UNSPEC > becomes BPF_SKB_CLOCK_REALTIME. > > Like: > > BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, > u64, tstamp, u32, tstamp_type) > { > /* ... */ > case BPF_SKB_CLOCK_TAI: > if (!tstamp) > return -EINVAL; > 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; > > /* ... */ > } Noted! > >> return -EINVAL; > >> @@ -9388,17 +9394,17 @@ 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 != BPF_SKB_TSTAMP_DELIVERY_TAI); > > Add these also: > Noted! > BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); > BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != BPF_SKB_CLOCK_MONOTONIC); > BUILD_BUG_ON(SKB_CLOCK_TAI != 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_RSH); >> +#else >> + BUILD_BUG_ON(!(SKB_TSTAMP_TYPE_MASK & 0x1)); >> +#endif >> + *insn++ = BPF_JMP32_IMM(BPF_JNE, value_reg, SKB_TSTAMP_TYPE_MASK, 1); >> + /* Both the bits set then mark it BPF_SKB_TSTAMP_UNSPEC */ >> + *insn++ = BPF_MOV64_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC); > > The kernel should not have both bits set in skb->tstamp_type. No need to > add two extra bpf insns to check this. If there is a bug in the kernel, > it is better to be uncovered instead of hiding it under BPF_SKB_TSTAMP_UNSPEC (which > is renamed to BPF_SKB_CLOCK_REALTIME anyway). > Hence, the last two bpf insns should be removed. > Got it !. >> return insn; >> } >> @@ -9430,6 +9436,7 @@ static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, >> __u8 value_reg = si->dst_reg; >> __u8 skb_reg = si->src_reg; >> +BUILD_BUG_ON(__SKB_CLOCK_MAX != BPF_SKB_TSTAMP_DELIVERY_TAI); > > It is a dup of the one in bpf_convert_tstamp_type_read and can be removed. > Noted! >> #ifdef CONFIG_NET_XGRESS >> /* If the tstamp_type is read, >> * the bpf prog is aware the tstamp could have delivery time. >> @@ -9440,11 +9447,12 @@ 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); >> - /* skb->tc_at_ingress && skb->tstamp_type:1, >> + /* 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:2, >> * read 0 as the (rcv) timestamp. >> */ >> *insn++ = BPF_MOV64_IMM(value_reg, 0); >> @@ -9469,7 +9477,7 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, >> * the bpf prog is aware the tstamp could have delivery time. >> * Thus, write skb->tstamp as is if tstamp_type_access is true. >> * Otherwise, writing at ingress will have to clear the >> - * mono_delivery_time (skb->tstamp_type:1)bit also. >> + * mono_delivery_time (skb->tstamp_type:2)bit also. >> */ >> if (!prog->tstamp_type_access) { >> __u8 tmp_reg = BPF_REG_AX; >> @@ -9479,8 +9487,8 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, >> *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); >> /* goto <store> */ >> *insn++ = BPF_JMP_A(2); >> - /* <clear>: mono_delivery_time or (skb->tstamp_type:1) */ >> - *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK); >> + /* <clear>: skb->tstamp_type:2 */ >> + *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 591226dcde26..f195b31d6e75 100644 >> --- a/net/ipv4/ip_output.c >> +++ b/net/ipv4/ip_output.c >> @@ -1457,7 +1457,7 @@ 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; >> + skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); > > hmm... I think this will break for tcp. This sequence in particular: > > tcp_v4_timewait_ack() > tcp_v4_send_ack() > ip_send_unicast_reply() > ip_push_pending_frames() > ip_finish_skb() > __ip_make_skb() > /* sk_clockid is REAL but cork->transmit_time should be in mono */ > skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);; > > I think I hit it from time to time when running the test in this patch set. > do you think i need to check for protocol type here . since tcp uses Mono and the rest according to the new design is based on sk->sk_clockid if (iph->protocol == IPPROTO_TCP) skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, CLOCK_MONOTONIC); else skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); > [ ... ] > >> diff --git a/tools/testing/selftests/bpf/progs/test_tc_dtime.c b/tools/testing/selftests/bpf/progs/test_tc_dtime.c >> index 74ec09f040b7..19dba6d88265 100644 >> --- a/tools/testing/selftests/bpf/progs/test_tc_dtime.c >> +++ b/tools/testing/selftests/bpf/progs/test_tc_dtime.c > > Please separate the selftests/bpf changes into another patch. > I will do that. >> @@ -227,6 +227,12 @@ int egress_host(struct __sk_buff *skb) >> inc_dtimes(EGRESS_ENDHOST); >> else >> inc_errs(EGRESS_ENDHOST); >> + } else if (skb_proto(skb_type) == IPPROTO_UDP) { >> + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI && >> + skb->tstamp) >> + inc_dtimes(EGRESS_ENDHOST); >> + else >> + inc_errs(EGRESS_ENDHOST); >> } else { >> if (skb->tstamp_type == BPF_SKB_TSTAMP_UNSPEC && >> skb->tstamp) >> @@ -255,6 +261,9 @@ int ingress_host(struct __sk_buff *skb) >> if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO && >> skb->tstamp == EGRESS_FWDNS_MAGIC) >> inc_dtimes(INGRESS_ENDHOST); >> + else if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI && >> + skb->tstamp == EGRESS_FWDNS_MAGIC) >> + inc_dtimes(INGRESS_ENDHOST); >> else >> inc_errs(INGRESS_ENDHOST); >> @@ -323,12 +332,14 @@ int ingress_fwdns_prio101(struct __sk_buff *skb) >> /* Should have handled in prio100 */ >> return TC_ACT_SHOT; >> - if (skb_proto(skb_type) == IPPROTO_UDP) >> + if (skb_proto(skb_type) == IPPROTO_UDP && >> + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) >> expected_dtime = 0; > > The IPPROTO_UDP check and expected_dtime can be removed. The UDP test > can expect the same EGRESS_ENDHOST_MAGIC in the skb->tstamp since > the TAI tstamp is also forwarded from egress to ingress now. > >> if (skb->tstamp_type) { >> if (fwdns_clear_dtime() || >> - skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO || >> + (skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO && >> + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) || >> skb->tstamp != expected_dtime) >> inc_errs(INGRESS_FWDNS_P101); >> else >> @@ -338,7 +349,8 @@ int ingress_fwdns_prio101(struct __sk_buff *skb) >> inc_errs(INGRESS_FWDNS_P101); >> } >> - if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO) { >> + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO || >> + skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI) { > > No need to check BPF_SKB_TSTAMP_DELIVERY_TAI such that the > bpf_skb_set_tstamp() helper can still be tested. > > There are some other minor changes needed for the test_tc_dtime.c and > the tc_redirect.c. I quickly made the changes and put them here (first patch): > > https://git.kernel.org/pub/scm/linux/kernel/git/martin.lau/bpf-next.git/log/?h=skb.tstamp_type > Thanks for your help Martin. I will check the changes you have made in BPF test framework and see what i have missed. > > >> skb->tstamp = INGRESS_FWDNS_MAGIC; >> } else { >> if (bpf_skb_set_tstamp(skb, INGRESS_FWDNS_MAGIC, >> @@ -370,7 +382,8 @@ int egress_fwdns_prio101(struct __sk_buff *skb) >> if (skb->tstamp_type) { >> if (fwdns_clear_dtime() || >> - skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO || >> + (skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO && >> + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) || >> skb->tstamp != INGRESS_FWDNS_MAGIC) >> inc_errs(EGRESS_FWDNS_P101); >> else >> @@ -380,7 +393,8 @@ int egress_fwdns_prio101(struct __sk_buff *skb) >> inc_errs(EGRESS_FWDNS_P101); >> } >> - if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO) { >> + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO || >> + skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI) { >> skb->tstamp = EGRESS_FWDNS_MAGIC; >> } else { >> if (bpf_skb_set_tstamp(skb, EGRESS_FWDNS_MAGIC, >
On 4/26/24 11:46 AM, Abhishek Chauhan (ABC) wrote: >>> diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c >>> index 591226dcde26..f195b31d6e75 100644 >>> --- a/net/ipv4/ip_output.c >>> +++ b/net/ipv4/ip_output.c >>> @@ -1457,7 +1457,7 @@ 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; >>> + skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); >> hmm... I think this will break for tcp. This sequence in particular: >> >> tcp_v4_timewait_ack() >> tcp_v4_send_ack() >> ip_send_unicast_reply() >> ip_push_pending_frames() >> ip_finish_skb() >> __ip_make_skb() >> /* sk_clockid is REAL but cork->transmit_time should be in mono */ >> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);; >> >> I think I hit it from time to time when running the test in this patch set. >> > do you think i need to check for protocol type here . since tcp uses Mono and the rest according to the new design is based on > sk->sk_clockid > if (iph->protocol == IPPROTO_TCP) > skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, CLOCK_MONOTONIC); > else > skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); Looks ok. iph->protocol is from sk->sk_protocol. I would defer to Willem input here. There is at least one more place that needs this protocol check, __ip6_make_skb().
On 4/24/24 3:20 PM, Abhishek Chauhan wrote: > diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c > index a9e819115622..63e4cc30d18d 100644 > --- a/net/ipv6/ip6_output.c > +++ b/net/ipv6/ip6_output.c > @@ -955,7 +955,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, > if (iter.frag) > ip6_fraglist_prepare(skb, &iter); > > - skb_set_delivery_time(skb, tstamp, tstamp_type); > + skb_set_tstamp_type_frm_clkid(skb, tstamp, tstamp_type); > err = output(net, sk, skb); > if (!err) > IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), > @@ -1016,7 +1016,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, > /* > * Put this fragment into the sending queue. > */ > - skb_set_delivery_time(frag, tstamp, tstamp_type); > + skb_set_tstamp_type_frm_clkid(frag, tstamp, tstamp_type); > err = output(net, sk, frag); > if (err) > goto fail; When replying another thread and looking closer at the ip6 changes, these two line changes should not be needed.
On 4/26/2024 4:50 PM, Martin KaFai Lau wrote: > On 4/26/24 11:46 AM, Abhishek Chauhan (ABC) wrote: >>>> diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c >>>> index 591226dcde26..f195b31d6e75 100644 >>>> --- a/net/ipv4/ip_output.c >>>> +++ b/net/ipv4/ip_output.c >>>> @@ -1457,7 +1457,7 @@ 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; >>>> + skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); >>> hmm... I think this will break for tcp. This sequence in particular: >>> >>> tcp_v4_timewait_ack() >>> tcp_v4_send_ack() >>> ip_send_unicast_reply() >>> ip_push_pending_frames() >>> ip_finish_skb() >>> __ip_make_skb() >>> /* sk_clockid is REAL but cork->transmit_time should be in mono */ >>> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);; >>> >>> I think I hit it from time to time when running the test in this patch set. >>> >> do you think i need to check for protocol type here . since tcp uses Mono and the rest according to the new design is based on >> sk->sk_clockid >> if (iph->protocol == IPPROTO_TCP) >> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, CLOCK_MONOTONIC); >> else >> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); > > Looks ok. iph->protocol is from sk->sk_protocol. I would defer to Willem input here. > > There is at least one more place that needs this protocol check, __ip6_make_skb(). Sounds good. I will wait for Willem to comment here.
On 4/26/2024 4:55 PM, Martin KaFai Lau wrote: > On 4/24/24 3:20 PM, Abhishek Chauhan wrote: >> diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c >> index a9e819115622..63e4cc30d18d 100644 >> --- a/net/ipv6/ip6_output.c >> +++ b/net/ipv6/ip6_output.c >> @@ -955,7 +955,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, >> if (iter.frag) >> ip6_fraglist_prepare(skb, &iter); >> - skb_set_delivery_time(skb, tstamp, tstamp_type); >> + skb_set_tstamp_type_frm_clkid(skb, tstamp, tstamp_type); >> err = output(net, sk, skb); >> if (!err) >> IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), >> @@ -1016,7 +1016,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, >> /* >> * Put this fragment into the sending queue. >> */ >> - skb_set_delivery_time(frag, tstamp, tstamp_type); >> + skb_set_tstamp_type_frm_clkid(frag, tstamp, tstamp_type); >> err = output(net, sk, frag); >> if (err) >> goto fail; > > When replying another thread and looking closer at the ip6 changes, these two line changes should not be needed. Similar code exists in ip_output.c for ipv4 packets => ip_do_fragment => I was thinking do we need require that code or not. Since in both functionality are the same, only difference is protocol. From what i see is the for Frag cases in both ipv4 and ipv6, previously skb->tstamp_type was being set for each fragments. If we are planning to not do it for ip6 the same should follow for ip4 too.
Abhishek Chauhan (ABC) wrote: > > > On 4/26/2024 4:50 PM, Martin KaFai Lau wrote: > > On 4/26/24 11:46 AM, Abhishek Chauhan (ABC) wrote: > >>>> diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c > >>>> index 591226dcde26..f195b31d6e75 100644 > >>>> --- a/net/ipv4/ip_output.c > >>>> +++ b/net/ipv4/ip_output.c > >>>> @@ -1457,7 +1457,7 @@ 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; > >>>> + skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); > >>> hmm... I think this will break for tcp. This sequence in particular: Good catch, thanks! > >>> > >>> tcp_v4_timewait_ack() > >>> tcp_v4_send_ack() > >>> ip_send_unicast_reply() > >>> ip_push_pending_frames() > >>> ip_finish_skb() > >>> __ip_make_skb() > >>> /* sk_clockid is REAL but cork->transmit_time should be in mono */ > >>> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);; > >>> > >>> I think I hit it from time to time when running the test in this patch set. > >>> > >> do you think i need to check for protocol type here . since tcp uses Mono and the rest according to the new design is based on > >> sk->sk_clockid > >> if (iph->protocol == IPPROTO_TCP) > >> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, CLOCK_MONOTONIC); > >> else > >> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); > > > > Looks ok. iph->protocol is from sk->sk_protocol. I would defer to Willem input here. > > > > There is at least one more place that needs this protocol check, __ip6_make_skb(). > > Sounds good. I will wait for Willem to comment here. This would be sk_is_tcp(sk). I think we want to avoid special casing if we can. Note the if. If TCP always uses monotonic, we could consider initializing sk_clockid to CLOCK_MONONOTIC in tcp_init_sock. I guess TCP logic currently entirely ignores sk_clockid. If we are to start using this, then setsocktop SO_TXTIME must explicitly fail or ignore for TCP sockets, or silently skip the write. All of that is more complexity. Than is maybe warranted for this one case. So no objections from me to special casing using sk_is_tcp(sk) either.
On 4/30/2024 1:26 PM, Willem de Bruijn wrote: > Abhishek Chauhan (ABC) wrote: >> >> >> On 4/26/2024 4:50 PM, Martin KaFai Lau wrote: >>> On 4/26/24 11:46 AM, Abhishek Chauhan (ABC) wrote: >>>>>> diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c >>>>>> index 591226dcde26..f195b31d6e75 100644 >>>>>> --- a/net/ipv4/ip_output.c >>>>>> +++ b/net/ipv4/ip_output.c >>>>>> @@ -1457,7 +1457,7 @@ 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; >>>>>> + skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); >>>>> hmm... I think this will break for tcp. This sequence in particular: > > Good catch, thanks! > >>>>> >>>>> tcp_v4_timewait_ack() >>>>> tcp_v4_send_ack() >>>>> ip_send_unicast_reply() >>>>> ip_push_pending_frames() >>>>> ip_finish_skb() >>>>> __ip_make_skb() >>>>> /* sk_clockid is REAL but cork->transmit_time should be in mono */ >>>>> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);; >>>>> >>>>> I think I hit it from time to time when running the test in this patch set. >>>>> >>>> do you think i need to check for protocol type here . since tcp uses Mono and the rest according to the new design is based on >>>> sk->sk_clockid >>>> if (iph->protocol == IPPROTO_TCP) >>>> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, CLOCK_MONOTONIC); >>>> else >>>> skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid); >>> >>> Looks ok. iph->protocol is from sk->sk_protocol. I would defer to Willem input here. >>> >>> There is at least one more place that needs this protocol check, __ip6_make_skb(). >> >> Sounds good. I will wait for Willem to comment here. > > This would be sk_is_tcp(sk). > > I think we want to avoid special casing if we can. Note the if. > > If TCP always uses monotonic, we could consider initializing > sk_clockid to CLOCK_MONONOTIC in tcp_init_sock. > > I guess TCP logic currently entirely ignores sk_clockid. If we are to > start using this, then setsocktop SO_TXTIME must explicitly fail or > ignore for TCP sockets, or silently skip the write. > > All of that is more complexity. Than is maybe warranted for this one > case. So no objections from me to special casing using sk_is_tcp(sk) > either. > > Thanks Willem and Martin for all the support and reviews. I will take care of this in the next RFC patch - adding the sk_is_tcp check and setting tstamp_type to mono for tcp > >
> BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, > u64, tstamp, u32, tstamp_type) > { > /* ... */ > case BPF_SKB_CLOCK_TAI: > if (!tstamp) > return -EINVAL; > 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; > > /* ... */ > } > >> return -EINVAL; > >> @@ -9388,17 +9394,17 @@ 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 != BPF_SKB_TSTAMP_DELIVERY_TAI); > > Add these also: > > BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); > BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != BPF_SKB_CLOCK_MONOTONIC); > BUILD_BUG_ON(SKB_CLOCK_TAI != BPF_SKB_CLOCK_TAI); > Martin, The above suggestion of adding BUILD_BUG_ON always gives me a warning stating the following. Some systems considers warning as error if compiler flags are enabled. I believe this requires your suggestion before i raise RFC v6 patchset to either keep the BUILD_BUG_ON or remove it completely. /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9395:34: warning: comparison between ‘enum skb_tstamp_type’ and ‘enum <anonymous>’ [-Wenum-compare] 9395 | BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); | ^~ /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:451:9: note: in definition of macro ‘__compiletime_assert’ 451 | if (!(condition)) \ | ^~~~~~~~~ /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:471:2: note: in expansion of macro ‘_compiletime_assert’ 471 | _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__) | ^~~~~~~~~~~~~~~~~~~ /local/mnt/workspace/kernel_master/linux-next/include/linux/build_bug.h:39:37: note: in expansion of macro ‘compiletime_assert’ 39 | #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg) | ^~~~~~~~~~~~~~~~~~ /local/mnt/workspace/kernel_master/linux-next/include/linux/build_bug.h:50:2: note: in expansion of macro ‘BUILD_BUG_ON_MSG’ 50 | BUILD_BUG_ON_MSG(condition, "BUILD_BUG_ON failed: " #condition) | ^~~~~~~~~~~~~~~~ /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9395:2: note: in expansion of macro ‘BUILD_BUG_ON’ 9395 | BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); | ^~~~~~~~~~~~ /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9396:35: warning: comparison between ‘enum skb_tstamp_type’ and ‘enum <anonymous>’ [-Wenum-compare] 9396 | BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != BPF_SKB_CLOCK_MONOTONIC); | ^~ /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:451:9: note: in definition of macro ‘__compiletime_assert’ 451 | if (!(condition)) \ | ^~~~~~~~~ | ^~
On 5/3/24 2:33 PM, Abhishek Chauhan (ABC) wrote: > >> BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, >> u64, tstamp, u32, tstamp_type) >> { >> /* ... */ >> case BPF_SKB_CLOCK_TAI: >> if (!tstamp) >> return -EINVAL; >> 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; >> >> /* ... */ >> } >> >>> return -EINVAL; >> >>> @@ -9388,17 +9394,17 @@ 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 != BPF_SKB_TSTAMP_DELIVERY_TAI); >> >> Add these also: >> >> BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); >> BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != BPF_SKB_CLOCK_MONOTONIC); >> BUILD_BUG_ON(SKB_CLOCK_TAI != BPF_SKB_CLOCK_TAI); >> > > Martin, The above suggestion of adding BUILD_BUG_ON always gives me a warning stating the following. > > Some systems considers warning as error if compiler flags are enabled. I believe this requires your suggestion before i raise RFC v6 patchset to either keep the > BUILD_BUG_ON or remove it completely. cast it? > > /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9395:34: warning: comparison between ‘enum skb_tstamp_type’ and ‘enum <anonymous>’ [-Wenum-compare] > 9395 | BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); > | ^~ > /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:451:9: note: in definition of macro ‘__compiletime_assert’ > 451 | if (!(condition)) \ > | ^~~~~~~~~ > /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:471:2: note: in expansion of macro ‘_compiletime_assert’ > 471 | _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__) > | ^~~~~~~~~~~~~~~~~~~ > /local/mnt/workspace/kernel_master/linux-next/include/linux/build_bug.h:39:37: note: in expansion of macro ‘compiletime_assert’ > 39 | #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg) > | ^~~~~~~~~~~~~~~~~~ > /local/mnt/workspace/kernel_master/linux-next/include/linux/build_bug.h:50:2: note: in expansion of macro ‘BUILD_BUG_ON_MSG’ > 50 | BUILD_BUG_ON_MSG(condition, "BUILD_BUG_ON failed: " #condition) > | ^~~~~~~~~~~~~~~~ > /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9395:2: note: in expansion of macro ‘BUILD_BUG_ON’ > 9395 | BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); > | ^~~~~~~~~~~~ > /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9396:35: warning: comparison between ‘enum skb_tstamp_type’ and ‘enum <anonymous>’ [-Wenum-compare] > 9396 | BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != BPF_SKB_CLOCK_MONOTONIC); > | ^~ > /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:451:9: note: in definition of macro ‘__compiletime_assert’ > 451 | if (!(condition)) \ > | ^~~~~~~~~ > > | ^~ > > >
On 5/3/2024 2:41 PM, Martin KaFai Lau wrote: > On 5/3/24 2:33 PM, Abhishek Chauhan (ABC) wrote: >> >>> BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, >>> u64, tstamp, u32, tstamp_type) >>> { >>> /* ... */ >>> case BPF_SKB_CLOCK_TAI: >>> if (!tstamp) >>> return -EINVAL; >>> 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; >>> >>> /* ... */ >>> } >>> >>>> return -EINVAL; >>> >>>> @@ -9388,17 +9394,17 @@ 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 != BPF_SKB_TSTAMP_DELIVERY_TAI); >>> >>> Add these also: >>> >>> BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); >>> BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != BPF_SKB_CLOCK_MONOTONIC); >>> BUILD_BUG_ON(SKB_CLOCK_TAI != BPF_SKB_CLOCK_TAI); >>> >> >> Martin, The above suggestion of adding BUILD_BUG_ON always gives me a warning stating the following. >> >> Some systems considers warning as error if compiler flags are enabled. I believe this requires your suggestion before i raise RFC v6 patchset to either keep the >> BUILD_BUG_ON or remove it completely. > > cast it? > Thanks Martin. Will do the same. Casting worked for me!. >> >> /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9395:34: warning: comparison between ‘enum skb_tstamp_type’ and ‘enum <anonymous>’ [-Wenum-compare] >> 9395 | BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); >> | ^~ >> /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:451:9: note: in definition of macro ‘__compiletime_assert’ >> 451 | if (!(condition)) \ >> | ^~~~~~~~~ >> /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:471:2: note: in expansion of macro ‘_compiletime_assert’ >> 471 | _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__) >> | ^~~~~~~~~~~~~~~~~~~ >> /local/mnt/workspace/kernel_master/linux-next/include/linux/build_bug.h:39:37: note: in expansion of macro ‘compiletime_assert’ >> 39 | #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg) >> | ^~~~~~~~~~~~~~~~~~ >> /local/mnt/workspace/kernel_master/linux-next/include/linux/build_bug.h:50:2: note: in expansion of macro ‘BUILD_BUG_ON_MSG’ >> 50 | BUILD_BUG_ON_MSG(condition, "BUILD_BUG_ON failed: " #condition) >> | ^~~~~~~~~~~~~~~~ >> /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9395:2: note: in expansion of macro ‘BUILD_BUG_ON’ >> 9395 | BUILD_BUG_ON(SKB_CLOCK_REALTIME != BPF_SKB_CLOCK_REALTIME); >> | ^~~~~~~~~~~~ >> /local/mnt/workspace/kernel_master/linux-next/net/core/filter.c:9396:35: warning: comparison between ‘enum skb_tstamp_type’ and ‘enum <anonymous>’ [-Wenum-compare] >> 9396 | BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != BPF_SKB_CLOCK_MONOTONIC); >> | ^~ >> /local/mnt/workspace/kernel_master/linux-next/include/linux/compiler_types.h:451:9: note: in definition of macro ‘__compiletime_assert’ >> 451 | if (!(condition)) \ >> | ^~~~~~~~~ >> >> | ^~ >> >> >> >
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h index e464d0ebc9c1..3ad0de07d261 100644 --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -711,6 +711,8 @@ 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, }; /** @@ -831,8 +833,8 @@ 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 in mono clock base or clock base of skb->tstamp. + * Otherwise, the timestamp is considered real clock base * @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. @@ -960,7 +962,7 @@ 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; @@ -1090,15 +1092,17 @@ 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:2 * 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_RSH (6) +#define TC_AT_INGRESS_RSH (5) +#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) @@ -4204,6 +4208,12 @@ static inline void skb_set_tstamp_type_frm_clkid(struct sk_buff *skb, case CLOCK_MONOTONIC: skb->tstamp_type = SKB_CLOCK_MONOTONIC; break; + case CLOCK_TAI: + skb->tstamp_type = SKB_CLOCK_TAI; + break; + default: + WARN_ONCE(true, "clockid %d not supported", clockid); + skb->tstamp_type = SKB_CLOCK_REALTIME; } } diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index cee0a7915c08..1376ed5ece10 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -6209,6 +6209,7 @@ union { \ enum { BPF_SKB_TSTAMP_UNSPEC, BPF_SKB_TSTAMP_DELIVERY_MONO, /* tstamp has mono delivery time */ + BPF_SKB_TSTAMP_DELIVERY_TAI, /* tstamp has tai 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. diff --git a/net/core/filter.c b/net/core/filter.c index 957c2fc724eb..c67622f4fe98 100644 --- a/net/core/filter.c +++ b/net/core/filter.c @@ -7733,6 +7733,12 @@ BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, skb->tstamp = tstamp; skb->tstamp_type = SKB_CLOCK_MONOTONIC; break; + case BPF_SKB_TSTAMP_DELIVERY_TAI: + if (!tstamp) + return -EINVAL; + skb->tstamp = tstamp; + skb->tstamp_type = SKB_CLOCK_TAI; + break; case BPF_SKB_TSTAMP_UNSPEC: if (tstamp) return -EINVAL; @@ -9388,17 +9394,17 @@ 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 != BPF_SKB_TSTAMP_DELIVERY_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_RSH); +#else + BUILD_BUG_ON(!(SKB_TSTAMP_TYPE_MASK & 0x1)); +#endif + *insn++ = BPF_JMP32_IMM(BPF_JNE, value_reg, SKB_TSTAMP_TYPE_MASK, 1); + /* Both the bits set then mark it BPF_SKB_TSTAMP_UNSPEC */ + *insn++ = BPF_MOV64_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC); return insn; } @@ -9430,6 +9436,7 @@ static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, __u8 value_reg = si->dst_reg; __u8 skb_reg = si->src_reg; +BUILD_BUG_ON(__SKB_CLOCK_MAX != BPF_SKB_TSTAMP_DELIVERY_TAI); #ifdef CONFIG_NET_XGRESS /* If the tstamp_type is read, * the bpf prog is aware the tstamp could have delivery time. @@ -9440,11 +9447,12 @@ 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); - /* skb->tc_at_ingress && skb->tstamp_type:1, + /* 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:2, * read 0 as the (rcv) timestamp. */ *insn++ = BPF_MOV64_IMM(value_reg, 0); @@ -9469,7 +9477,7 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, * the bpf prog is aware the tstamp could have delivery time. * Thus, write skb->tstamp as is if tstamp_type_access is true. * Otherwise, writing at ingress will have to clear the - * mono_delivery_time (skb->tstamp_type:1)bit also. + * mono_delivery_time (skb->tstamp_type:2)bit also. */ if (!prog->tstamp_type_access) { __u8 tmp_reg = BPF_REG_AX; @@ -9479,8 +9487,8 @@ static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); /* goto <store> */ *insn++ = BPF_JMP_A(2); - /* <clear>: mono_delivery_time or (skb->tstamp_type:1) */ - *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK); + /* <clear>: skb->tstamp_type:2 */ + *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 591226dcde26..f195b31d6e75 100644 --- a/net/ipv4/ip_output.c +++ b/net/ipv4/ip_output.c @@ -1457,7 +1457,7 @@ 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; + skb_set_tstamp_type_frm_clkid(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 dcb11f22cbf2..8b370369cdd8 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_tstamp_type_frm_clkid(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 a9e819115622..63e4cc30d18d 100644 --- a/net/ipv6/ip6_output.c +++ b/net/ipv6/ip6_output.c @@ -955,7 +955,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, if (iter.frag) ip6_fraglist_prepare(skb, &iter); - skb_set_delivery_time(skb, tstamp, tstamp_type); + skb_set_tstamp_type_frm_clkid(skb, tstamp, tstamp_type); err = output(net, sk, skb); if (!err) IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), @@ -1016,7 +1016,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, /* * Put this fragment into the sending queue. */ - skb_set_delivery_time(frag, tstamp, tstamp_type); + skb_set_tstamp_type_frm_clkid(frag, tstamp, tstamp_type); err = output(net, sk, frag); if (err) goto fail; @@ -1924,7 +1924,7 @@ 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; + skb_set_tstamp_type_frm_clkid(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 0d896ca7b589..5649362577ab 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_tstamp_type_frm_clkid(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..6a4a86c26d2a 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_tstamp_type_frm_clkid(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_tstamp_type_frm_clkid(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_tstamp_type_frm_clkid(skb, sockc.transmit_time, sk->sk_clockid); if (unlikely(extra_len == 4)) skb->no_fcs = 1; diff --git a/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c b/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c index 3b7c57fe55a5..71940f4ef0fb 100644 --- a/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c +++ b/tools/testing/selftests/bpf/prog_tests/ctx_rewrite.c @@ -69,15 +69,17 @@ static struct test_case test_cases[] = { { N(SCHED_CLS, struct __sk_buff, tstamp), .read = "r11 = *(u8 *)($ctx + sk_buff::__mono_tc_offset);" - "w11 &= 3;" - "if w11 != 0x3 goto pc+2;" + "if w11 == 0x4 goto pc+1;" + "goto pc+4;" + "if w11 == 0x3 goto pc+1;" + "goto pc+2;" "$dst = 0;" "goto pc+1;" "$dst = *(u64 *)($ctx + sk_buff::tstamp);", .write = "r11 = *(u8 *)($ctx + sk_buff::__mono_tc_offset);" - "if w11 & 0x2 goto pc+1;" + "if w11 & 0x4 goto pc+1;" "goto pc+2;" - "w11 &= -2;" + "w11 &= -3;" "*(u8 *)($ctx + sk_buff::__mono_tc_offset) = r11;" "*(u64 *)($ctx + sk_buff::tstamp) = $src;", }, diff --git a/tools/testing/selftests/bpf/progs/test_tc_dtime.c b/tools/testing/selftests/bpf/progs/test_tc_dtime.c index 74ec09f040b7..19dba6d88265 100644 --- a/tools/testing/selftests/bpf/progs/test_tc_dtime.c +++ b/tools/testing/selftests/bpf/progs/test_tc_dtime.c @@ -227,6 +227,12 @@ int egress_host(struct __sk_buff *skb) inc_dtimes(EGRESS_ENDHOST); else inc_errs(EGRESS_ENDHOST); + } else if (skb_proto(skb_type) == IPPROTO_UDP) { + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI && + skb->tstamp) + inc_dtimes(EGRESS_ENDHOST); + else + inc_errs(EGRESS_ENDHOST); } else { if (skb->tstamp_type == BPF_SKB_TSTAMP_UNSPEC && skb->tstamp) @@ -255,6 +261,9 @@ int ingress_host(struct __sk_buff *skb) if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO && skb->tstamp == EGRESS_FWDNS_MAGIC) inc_dtimes(INGRESS_ENDHOST); + else if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI && + skb->tstamp == EGRESS_FWDNS_MAGIC) + inc_dtimes(INGRESS_ENDHOST); else inc_errs(INGRESS_ENDHOST); @@ -323,12 +332,14 @@ int ingress_fwdns_prio101(struct __sk_buff *skb) /* Should have handled in prio100 */ return TC_ACT_SHOT; - if (skb_proto(skb_type) == IPPROTO_UDP) + if (skb_proto(skb_type) == IPPROTO_UDP && + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) expected_dtime = 0; if (skb->tstamp_type) { if (fwdns_clear_dtime() || - skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO || + (skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO && + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) || skb->tstamp != expected_dtime) inc_errs(INGRESS_FWDNS_P101); else @@ -338,7 +349,8 @@ int ingress_fwdns_prio101(struct __sk_buff *skb) inc_errs(INGRESS_FWDNS_P101); } - if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO) { + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO || + skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI) { skb->tstamp = INGRESS_FWDNS_MAGIC; } else { if (bpf_skb_set_tstamp(skb, INGRESS_FWDNS_MAGIC, @@ -370,7 +382,8 @@ int egress_fwdns_prio101(struct __sk_buff *skb) if (skb->tstamp_type) { if (fwdns_clear_dtime() || - skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO || + (skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_MONO && + skb->tstamp_type != BPF_SKB_TSTAMP_DELIVERY_TAI) || skb->tstamp != INGRESS_FWDNS_MAGIC) inc_errs(EGRESS_FWDNS_P101); else @@ -380,7 +393,8 @@ int egress_fwdns_prio101(struct __sk_buff *skb) inc_errs(EGRESS_FWDNS_P101); } - if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO) { + if (skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_MONO || + skb->tstamp_type == BPF_SKB_TSTAMP_DELIVERY_TAI) { skb->tstamp = EGRESS_FWDNS_MAGIC; } else { if (bpf_skb_set_tstamp(skb, EGRESS_FWDNS_MAGIC,
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@linux.dev/ Signed-off-by: Abhishek Chauhan <quic_abchauha@quicinc.com> --- 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 | 26 +++++++---- include/uapi/linux/bpf.h | 1 + net/core/filter.c | 46 +++++++++++-------- net/ipv4/ip_output.c | 2 +- net/ipv4/raw.c | 2 +- net/ipv6/ip6_output.c | 6 +-- net/ipv6/raw.c | 2 +- net/packet/af_packet.c | 7 ++- .../selftests/bpf/prog_tests/ctx_rewrite.c | 10 ++-- .../selftests/bpf/progs/test_tc_dtime.c | 24 ++++++++-- 10 files changed, 80 insertions(+), 46 deletions(-)