| Message ID | 20240809123915.27812-3-justin.iurman@uliege.be (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: ipv6: ioam6: introduce tunsrc | expand |
On 8/9/24 14:39, Justin Iurman wrote: > This patch provides a new feature (i.e., "tunsrc") for the tunnel (i.e., > "encap") mode of ioam6. Just like seg6 already does, except it is > attached to a route. The "tunsrc" is optional: when not provided (by > default), the automatic resolution is applied. Using "tunsrc" when > possible has a benefit: performance. It's customary to include performances figures in performance related changeset ;) > > Signed-off-by: Justin Iurman <justin.iurman@uliege.be> > --- > include/uapi/linux/ioam6_iptunnel.h | 7 +++++ > net/ipv6/ioam6_iptunnel.c | 48 ++++++++++++++++++++++++++--- > 2 files changed, 51 insertions(+), 4 deletions(-) > > diff --git a/include/uapi/linux/ioam6_iptunnel.h b/include/uapi/linux/ioam6_iptunnel.h > index 38f6a8fdfd34..6cdbd0da7ad8 100644 > --- a/include/uapi/linux/ioam6_iptunnel.h > +++ b/include/uapi/linux/ioam6_iptunnel.h > @@ -50,6 +50,13 @@ enum { > IOAM6_IPTUNNEL_FREQ_K, /* u32 */ > IOAM6_IPTUNNEL_FREQ_N, /* u32 */ > > + /* Tunnel src address. > + * For encap,auto modes. > + * Optional (automatic if > + * not provided). > + */ > + IOAM6_IPTUNNEL_SRC, /* struct in6_addr */ > + > __IOAM6_IPTUNNEL_MAX, > }; > > diff --git a/net/ipv6/ioam6_iptunnel.c b/net/ipv6/ioam6_iptunnel.c > index cd2522f04edf..e0e73faf9969 100644 > --- a/net/ipv6/ioam6_iptunnel.c > +++ b/net/ipv6/ioam6_iptunnel.c > @@ -42,6 +42,8 @@ struct ioam6_lwt { > struct ioam6_lwt_freq freq; > atomic_t pkt_cnt; > u8 mode; > + bool has_tunsrc; > + struct in6_addr tunsrc; > struct in6_addr tundst; > struct ioam6_lwt_encap tuninfo; > }; > @@ -72,6 +74,7 @@ static const struct nla_policy ioam6_iptunnel_policy[IOAM6_IPTUNNEL_MAX + 1] = { > [IOAM6_IPTUNNEL_MODE] = NLA_POLICY_RANGE(NLA_U8, > IOAM6_IPTUNNEL_MODE_MIN, > IOAM6_IPTUNNEL_MODE_MAX), > + [IOAM6_IPTUNNEL_SRC] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), > [IOAM6_IPTUNNEL_DST] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), > [IOAM6_IPTUNNEL_TRACE] = NLA_POLICY_EXACT_LEN( > sizeof(struct ioam6_trace_hdr)), > @@ -144,6 +147,11 @@ static int ioam6_build_state(struct net *net, struct nlattr *nla, > else > mode = nla_get_u8(tb[IOAM6_IPTUNNEL_MODE]); > > + if (tb[IOAM6_IPTUNNEL_SRC] && mode == IOAM6_IPTUNNEL_MODE_INLINE) { > + NL_SET_ERR_MSG(extack, "no tunnel source expected in this mode"); > + return -EINVAL; > + } when mode is IOAM6_IPTUNNEL_MODE_AUTO, the data path could still add the encapsulation for forwarded packets, why explicitly preventing this optimization in such scenario? > + > if (!tb[IOAM6_IPTUNNEL_DST] && mode != IOAM6_IPTUNNEL_MODE_INLINE) { > NL_SET_ERR_MSG(extack, "this mode needs a tunnel destination"); > return -EINVAL; > @@ -178,6 +186,14 @@ static int ioam6_build_state(struct net *net, struct nlattr *nla, > ilwt->freq.n = freq_n; > > ilwt->mode = mode; > + > + if (!tb[IOAM6_IPTUNNEL_SRC]) { > + ilwt->has_tunsrc = false; > + } else { > + ilwt->has_tunsrc = true; > + ilwt->tunsrc = nla_get_in6_addr(tb[IOAM6_IPTUNNEL_SRC]); Since you are going to use the source address only if != ANY, I think it would be cleaner to refuse such addresses here. That will avoid an additional check in the datapath. Cheers, Paolo
On 8/13/24 13:06, Paolo Abeni wrote: > On 8/9/24 14:39, Justin Iurman wrote: >> This patch provides a new feature (i.e., "tunsrc") for the tunnel (i.e., >> "encap") mode of ioam6. Just like seg6 already does, except it is >> attached to a route. The "tunsrc" is optional: when not provided (by >> default), the automatic resolution is applied. Using "tunsrc" when >> possible has a benefit: performance. > > It's customary to include performances figures in performance related > changeset ;) Indeed, I realized it too late... thx for the reminder! Before (= "encap" mode): https://ibb.co/bNCzvf7 After (= "encap" mode with "tunsrc"): https://ibb.co/PT8L6yq I'll add these again to -v2. >> Signed-off-by: Justin Iurman <justin.iurman@uliege.be> >> --- >> include/uapi/linux/ioam6_iptunnel.h | 7 +++++ >> net/ipv6/ioam6_iptunnel.c | 48 ++++++++++++++++++++++++++--- >> 2 files changed, 51 insertions(+), 4 deletions(-) >> >> diff --git a/include/uapi/linux/ioam6_iptunnel.h >> b/include/uapi/linux/ioam6_iptunnel.h >> index 38f6a8fdfd34..6cdbd0da7ad8 100644 >> --- a/include/uapi/linux/ioam6_iptunnel.h >> +++ b/include/uapi/linux/ioam6_iptunnel.h >> @@ -50,6 +50,13 @@ enum { >> IOAM6_IPTUNNEL_FREQ_K, /* u32 */ >> IOAM6_IPTUNNEL_FREQ_N, /* u32 */ >> + /* Tunnel src address. >> + * For encap,auto modes. >> + * Optional (automatic if >> + * not provided). >> + */ >> + IOAM6_IPTUNNEL_SRC, /* struct in6_addr */ >> + >> __IOAM6_IPTUNNEL_MAX, >> }; >> diff --git a/net/ipv6/ioam6_iptunnel.c b/net/ipv6/ioam6_iptunnel.c >> index cd2522f04edf..e0e73faf9969 100644 >> --- a/net/ipv6/ioam6_iptunnel.c >> +++ b/net/ipv6/ioam6_iptunnel.c >> @@ -42,6 +42,8 @@ struct ioam6_lwt { >> struct ioam6_lwt_freq freq; >> atomic_t pkt_cnt; >> u8 mode; >> + bool has_tunsrc; >> + struct in6_addr tunsrc; >> struct in6_addr tundst; >> struct ioam6_lwt_encap tuninfo; >> }; >> @@ -72,6 +74,7 @@ static const struct nla_policy >> ioam6_iptunnel_policy[IOAM6_IPTUNNEL_MAX + 1] = { >> [IOAM6_IPTUNNEL_MODE] = NLA_POLICY_RANGE(NLA_U8, >> IOAM6_IPTUNNEL_MODE_MIN, >> IOAM6_IPTUNNEL_MODE_MAX), >> + [IOAM6_IPTUNNEL_SRC] = NLA_POLICY_EXACT_LEN(sizeof(struct >> in6_addr)), >> [IOAM6_IPTUNNEL_DST] = NLA_POLICY_EXACT_LEN(sizeof(struct >> in6_addr)), >> [IOAM6_IPTUNNEL_TRACE] = NLA_POLICY_EXACT_LEN( >> sizeof(struct ioam6_trace_hdr)), >> @@ -144,6 +147,11 @@ static int ioam6_build_state(struct net *net, >> struct nlattr *nla, >> else >> mode = nla_get_u8(tb[IOAM6_IPTUNNEL_MODE]); >> + if (tb[IOAM6_IPTUNNEL_SRC] && mode == IOAM6_IPTUNNEL_MODE_INLINE) { >> + NL_SET_ERR_MSG(extack, "no tunnel source expected in this >> mode"); >> + return -EINVAL; >> + } > > when mode is IOAM6_IPTUNNEL_MODE_AUTO, the data path could still add the > encapsulation for forwarded packets, why explicitly preventing this > optimization in such scenario? Good catch! I guess we can just ignore "tunsrc" if provided with inline mode, instead of returning an error. >> + >> if (!tb[IOAM6_IPTUNNEL_DST] && mode != >> IOAM6_IPTUNNEL_MODE_INLINE) { >> NL_SET_ERR_MSG(extack, "this mode needs a tunnel destination"); >> return -EINVAL; >> @@ -178,6 +186,14 @@ static int ioam6_build_state(struct net *net, >> struct nlattr *nla, >> ilwt->freq.n = freq_n; >> ilwt->mode = mode; >> + >> + if (!tb[IOAM6_IPTUNNEL_SRC]) { >> + ilwt->has_tunsrc = false; >> + } else { >> + ilwt->has_tunsrc = true; >> + ilwt->tunsrc = nla_get_in6_addr(tb[IOAM6_IPTUNNEL_SRC]); > > Since you are going to use the source address only if != ANY, I think it > would be cleaner to refuse such addresses here. That will avoid an > additional check in the datapath. +1. Thanks, Justin > Cheers, > > Paolo >
On 8/13/24 13:06, Paolo Abeni wrote: > On 8/9/24 14:39, Justin Iurman wrote: >> This patch provides a new feature (i.e., "tunsrc") for the tunnel (i.e., >> "encap") mode of ioam6. Just like seg6 already does, except it is >> attached to a route. The "tunsrc" is optional: when not provided (by >> default), the automatic resolution is applied. Using "tunsrc" when >> possible has a benefit: performance. > > It's customary to include performances figures in performance related > changeset ;) > >> >> Signed-off-by: Justin Iurman <justin.iurman@uliege.be> >> --- >> include/uapi/linux/ioam6_iptunnel.h | 7 +++++ >> net/ipv6/ioam6_iptunnel.c | 48 ++++++++++++++++++++++++++--- >> 2 files changed, 51 insertions(+), 4 deletions(-) >> >> diff --git a/include/uapi/linux/ioam6_iptunnel.h >> b/include/uapi/linux/ioam6_iptunnel.h >> index 38f6a8fdfd34..6cdbd0da7ad8 100644 >> --- a/include/uapi/linux/ioam6_iptunnel.h >> +++ b/include/uapi/linux/ioam6_iptunnel.h >> @@ -50,6 +50,13 @@ enum { >> IOAM6_IPTUNNEL_FREQ_K, /* u32 */ >> IOAM6_IPTUNNEL_FREQ_N, /* u32 */ >> + /* Tunnel src address. >> + * For encap,auto modes. >> + * Optional (automatic if >> + * not provided). >> + */ >> + IOAM6_IPTUNNEL_SRC, /* struct in6_addr */ >> + >> __IOAM6_IPTUNNEL_MAX, >> }; >> diff --git a/net/ipv6/ioam6_iptunnel.c b/net/ipv6/ioam6_iptunnel.c >> index cd2522f04edf..e0e73faf9969 100644 >> --- a/net/ipv6/ioam6_iptunnel.c >> +++ b/net/ipv6/ioam6_iptunnel.c >> @@ -42,6 +42,8 @@ struct ioam6_lwt { >> struct ioam6_lwt_freq freq; >> atomic_t pkt_cnt; >> u8 mode; >> + bool has_tunsrc; >> + struct in6_addr tunsrc; >> struct in6_addr tundst; >> struct ioam6_lwt_encap tuninfo; >> }; >> @@ -72,6 +74,7 @@ static const struct nla_policy >> ioam6_iptunnel_policy[IOAM6_IPTUNNEL_MAX + 1] = { >> [IOAM6_IPTUNNEL_MODE] = NLA_POLICY_RANGE(NLA_U8, >> IOAM6_IPTUNNEL_MODE_MIN, >> IOAM6_IPTUNNEL_MODE_MAX), >> + [IOAM6_IPTUNNEL_SRC] = NLA_POLICY_EXACT_LEN(sizeof(struct >> in6_addr)), >> [IOAM6_IPTUNNEL_DST] = NLA_POLICY_EXACT_LEN(sizeof(struct >> in6_addr)), >> [IOAM6_IPTUNNEL_TRACE] = NLA_POLICY_EXACT_LEN( >> sizeof(struct ioam6_trace_hdr)), >> @@ -144,6 +147,11 @@ static int ioam6_build_state(struct net *net, >> struct nlattr *nla, >> else >> mode = nla_get_u8(tb[IOAM6_IPTUNNEL_MODE]); >> + if (tb[IOAM6_IPTUNNEL_SRC] && mode == IOAM6_IPTUNNEL_MODE_INLINE) { >> + NL_SET_ERR_MSG(extack, "no tunnel source expected in this >> mode"); >> + return -EINVAL; >> + } > > when mode is IOAM6_IPTUNNEL_MODE_AUTO, the data path could still add the > encapsulation for forwarded packets, why explicitly preventing this > optimization in such scenario? Actually, this check is correct. We don't want the "tunsrc" with "inline" mode since it's useless. If the "auto" mode is chosen, then it's fine (same for the "encap" mode). Preventing "tunsrc" for the "inline" mode does *not* impact the auto mode. Basically: tb[IOAM6_IPTUNNEL_SRC] && mode == IOAM6_IPTUNNEL_MODE_INLINE -> error tb[IOAM6_IPTUNNEL_SRC] && mode == IOAM6_IPTUNNEL_MODE_ENCAP -> OK tb[IOAM6_IPTUNNEL_SRC] && mode == IOAM6_IPTUNNEL_MODE_AUTO -> OK It aligns better with the semantics of "tundst", which is a MUST for "encap"/"auto" modes (and forbidden for the "inline" mode). In the case of "tunsrc", it is a MAY for "encap"/"auto" modes (and forbidden for the "inline" mode). >> + >> if (!tb[IOAM6_IPTUNNEL_DST] && mode != >> IOAM6_IPTUNNEL_MODE_INLINE) { >> NL_SET_ERR_MSG(extack, "this mode needs a tunnel destination"); >> return -EINVAL; >> @@ -178,6 +186,14 @@ static int ioam6_build_state(struct net *net, >> struct nlattr *nla, >> ilwt->freq.n = freq_n; >> ilwt->mode = mode; >> + >> + if (!tb[IOAM6_IPTUNNEL_SRC]) { >> + ilwt->has_tunsrc = false; >> + } else { >> + ilwt->has_tunsrc = true; >> + ilwt->tunsrc = nla_get_in6_addr(tb[IOAM6_IPTUNNEL_SRC]); > > Since you are going to use the source address only if != ANY, I think it > would be cleaner to refuse such addresses here. That will avoid an > additional check in the datapath. > > Cheers, > > Paolo >
diff --git a/include/uapi/linux/ioam6_iptunnel.h b/include/uapi/linux/ioam6_iptunnel.h index 38f6a8fdfd34..6cdbd0da7ad8 100644 --- a/include/uapi/linux/ioam6_iptunnel.h +++ b/include/uapi/linux/ioam6_iptunnel.h @@ -50,6 +50,13 @@ enum { IOAM6_IPTUNNEL_FREQ_K, /* u32 */ IOAM6_IPTUNNEL_FREQ_N, /* u32 */ + /* Tunnel src address. + * For encap,auto modes. + * Optional (automatic if + * not provided). + */ + IOAM6_IPTUNNEL_SRC, /* struct in6_addr */ + __IOAM6_IPTUNNEL_MAX, }; diff --git a/net/ipv6/ioam6_iptunnel.c b/net/ipv6/ioam6_iptunnel.c index cd2522f04edf..e0e73faf9969 100644 --- a/net/ipv6/ioam6_iptunnel.c +++ b/net/ipv6/ioam6_iptunnel.c @@ -42,6 +42,8 @@ struct ioam6_lwt { struct ioam6_lwt_freq freq; atomic_t pkt_cnt; u8 mode; + bool has_tunsrc; + struct in6_addr tunsrc; struct in6_addr tundst; struct ioam6_lwt_encap tuninfo; }; @@ -72,6 +74,7 @@ static const struct nla_policy ioam6_iptunnel_policy[IOAM6_IPTUNNEL_MAX + 1] = { [IOAM6_IPTUNNEL_MODE] = NLA_POLICY_RANGE(NLA_U8, IOAM6_IPTUNNEL_MODE_MIN, IOAM6_IPTUNNEL_MODE_MAX), + [IOAM6_IPTUNNEL_SRC] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), [IOAM6_IPTUNNEL_DST] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), [IOAM6_IPTUNNEL_TRACE] = NLA_POLICY_EXACT_LEN( sizeof(struct ioam6_trace_hdr)), @@ -144,6 +147,11 @@ static int ioam6_build_state(struct net *net, struct nlattr *nla, else mode = nla_get_u8(tb[IOAM6_IPTUNNEL_MODE]); + if (tb[IOAM6_IPTUNNEL_SRC] && mode == IOAM6_IPTUNNEL_MODE_INLINE) { + NL_SET_ERR_MSG(extack, "no tunnel source expected in this mode"); + return -EINVAL; + } + if (!tb[IOAM6_IPTUNNEL_DST] && mode != IOAM6_IPTUNNEL_MODE_INLINE) { NL_SET_ERR_MSG(extack, "this mode needs a tunnel destination"); return -EINVAL; @@ -178,6 +186,14 @@ static int ioam6_build_state(struct net *net, struct nlattr *nla, ilwt->freq.n = freq_n; ilwt->mode = mode; + + if (!tb[IOAM6_IPTUNNEL_SRC]) { + ilwt->has_tunsrc = false; + } else { + ilwt->has_tunsrc = true; + ilwt->tunsrc = nla_get_in6_addr(tb[IOAM6_IPTUNNEL_SRC]); + } + if (tb[IOAM6_IPTUNNEL_DST]) ilwt->tundst = nla_get_in6_addr(tb[IOAM6_IPTUNNEL_DST]); @@ -257,6 +273,8 @@ static int ioam6_do_inline(struct net *net, struct sk_buff *skb, static int ioam6_do_encap(struct net *net, struct sk_buff *skb, struct ioam6_lwt_encap *tuninfo, + bool has_tunsrc, + struct in6_addr *tunsrc, struct in6_addr *tundst) { struct dst_entry *dst = skb_dst(skb); @@ -286,8 +304,13 @@ static int ioam6_do_encap(struct net *net, struct sk_buff *skb, hdr->nexthdr = NEXTHDR_HOP; hdr->payload_len = cpu_to_be16(skb->len - sizeof(*hdr)); hdr->daddr = *tundst; - ipv6_dev_get_saddr(net, dst->dev, &hdr->daddr, - IPV6_PREFER_SRC_PUBLIC, &hdr->saddr); + + if (has_tunsrc && !ipv6_addr_any(tunsrc)) { + memcpy(&hdr->saddr, tunsrc, sizeof(*tunsrc)); + } else { + ipv6_dev_get_saddr(net, dst->dev, &hdr->daddr, + IPV6_PREFER_SRC_PUBLIC, &hdr->saddr); + } skb_postpush_rcsum(skb, hdr, len); @@ -329,7 +352,9 @@ static int ioam6_output(struct net *net, struct sock *sk, struct sk_buff *skb) case IOAM6_IPTUNNEL_MODE_ENCAP: do_encap: /* Encapsulation (ip6ip6) */ - err = ioam6_do_encap(net, skb, &ilwt->tuninfo, &ilwt->tundst); + err = ioam6_do_encap(net, skb, &ilwt->tuninfo, + ilwt->has_tunsrc, &ilwt->tunsrc, + &ilwt->tundst); if (unlikely(err)) goto drop; @@ -415,6 +440,13 @@ static int ioam6_fill_encap_info(struct sk_buff *skb, goto ret; if (ilwt->mode != IOAM6_IPTUNNEL_MODE_INLINE) { + if (ilwt->has_tunsrc) { + err = nla_put_in6_addr(skb, IOAM6_IPTUNNEL_SRC, + &ilwt->tunsrc); + if (err) + goto ret; + } + err = nla_put_in6_addr(skb, IOAM6_IPTUNNEL_DST, &ilwt->tundst); if (err) goto ret; @@ -436,8 +468,12 @@ static int ioam6_encap_nlsize(struct lwtunnel_state *lwtstate) nla_total_size(sizeof(ilwt->mode)) + nla_total_size(sizeof(ilwt->tuninfo.traceh)); - if (ilwt->mode != IOAM6_IPTUNNEL_MODE_INLINE) + if (ilwt->mode != IOAM6_IPTUNNEL_MODE_INLINE) { + if (ilwt->has_tunsrc) + nlsize += nla_total_size(sizeof(ilwt->tunsrc)); + nlsize += nla_total_size(sizeof(ilwt->tundst)); + } return nlsize; } @@ -452,8 +488,12 @@ static int ioam6_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b) return (ilwt_a->freq.k != ilwt_b->freq.k || ilwt_a->freq.n != ilwt_b->freq.n || ilwt_a->mode != ilwt_b->mode || + ilwt_a->has_tunsrc != ilwt_b->has_tunsrc || (ilwt_a->mode != IOAM6_IPTUNNEL_MODE_INLINE && !ipv6_addr_equal(&ilwt_a->tundst, &ilwt_b->tundst)) || + (ilwt_a->mode != IOAM6_IPTUNNEL_MODE_INLINE && + ilwt_a->has_tunsrc && + !ipv6_addr_equal(&ilwt_a->tunsrc, &ilwt_b->tunsrc)) || trace_a->namespace_id != trace_b->namespace_id); }
This patch provides a new feature (i.e., "tunsrc") for the tunnel (i.e., "encap") mode of ioam6. Just like seg6 already does, except it is attached to a route. The "tunsrc" is optional: when not provided (by default), the automatic resolution is applied. Using "tunsrc" when possible has a benefit: performance. Signed-off-by: Justin Iurman <justin.iurman@uliege.be> --- include/uapi/linux/ioam6_iptunnel.h | 7 +++++ net/ipv6/ioam6_iptunnel.c | 48 ++++++++++++++++++++++++++--- 2 files changed, 51 insertions(+), 4 deletions(-)