Message ID | c87caa37757cdf6e323c89748fd0a0408fd47da2.1715807303.git.lorenzo@kernel.org |
---|---|
State | Changes Requested |
Headers | show |
Series | netfilter: Add the capability to offload flowtable in XDP layer | expand |
On Wed, 15 May 2024 at 23:13, Lorenzo Bianconi <lorenzo@kernel.org> wrote: > > Introduce bpf_xdp_flow_offload_lookup kfunc in order to perform the > lookup of a given flowtable entry based on a fib tuple of incoming > traffic. > bpf_xdp_flow_offload_lookup can be used as building block to offload > in xdp the processing of sw flowtable when hw flowtable is not > available. > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org> > --- > include/net/netfilter/nf_flow_table.h | 10 +++ > net/netfilter/Makefile | 5 ++ > net/netfilter/nf_flow_table_bpf.c | 95 +++++++++++++++++++++++++++ > net/netfilter/nf_flow_table_inet.c | 2 + > 4 files changed, 112 insertions(+) > create mode 100644 net/netfilter/nf_flow_table_bpf.c > > diff --git a/include/net/netfilter/nf_flow_table.h b/include/net/netfilter/nf_flow_table.h > index 0bbe6ea8e0651..085660cbcd3f2 100644 > --- a/include/net/netfilter/nf_flow_table.h > +++ b/include/net/netfilter/nf_flow_table.h > @@ -312,6 +312,16 @@ unsigned int nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb, > unsigned int nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb, > const struct nf_hook_state *state); > > +#if (IS_BUILTIN(CONFIG_NF_FLOW_TABLE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \ > + (IS_MODULE(CONFIG_NF_FLOW_TABLE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)) > +extern int nf_flow_offload_register_bpf(void); > +#else > +static inline int nf_flow_offload_register_bpf(void) > +{ > + return 0; > +} > +#endif > + > #define MODULE_ALIAS_NF_FLOWTABLE(family) \ > MODULE_ALIAS("nf-flowtable-" __stringify(family)) > > diff --git a/net/netfilter/Makefile b/net/netfilter/Makefile > index 614815a3ed738..18b09cec92024 100644 > --- a/net/netfilter/Makefile > +++ b/net/netfilter/Makefile > @@ -144,6 +144,11 @@ obj-$(CONFIG_NF_FLOW_TABLE) += nf_flow_table.o > nf_flow_table-objs := nf_flow_table_core.o nf_flow_table_ip.o \ > nf_flow_table_offload.o > nf_flow_table-$(CONFIG_NF_FLOW_TABLE_PROCFS) += nf_flow_table_procfs.o > +ifeq ($(CONFIG_NF_FLOW_TABLE),m) > +nf_flow_table-$(CONFIG_DEBUG_INFO_BTF_MODULES) += nf_flow_table_bpf.o > +else ifeq ($(CONFIG_NF_FLOW_TABLE),y) > +nf_flow_table-$(CONFIG_DEBUG_INFO_BTF) += nf_flow_table_bpf.o > +endif > > obj-$(CONFIG_NF_FLOW_TABLE_INET) += nf_flow_table_inet.o > > diff --git a/net/netfilter/nf_flow_table_bpf.c b/net/netfilter/nf_flow_table_bpf.c > new file mode 100644 > index 0000000000000..836a1127e4052 > --- /dev/null > +++ b/net/netfilter/nf_flow_table_bpf.c > @@ -0,0 +1,95 @@ > +// SPDX-License-Identifier: GPL-2.0-only > +/* Unstable Flow Table Helpers for XDP hook > + * > + * These are called from the XDP programs. > + * Note that it is allowed to break compatibility for these functions since > + * the interface they are exposed through to BPF programs is explicitly > + * unstable. > + */ > + > +#include <linux/kernel.h> > +#include <linux/init.h> > +#include <linux/module.h> > +#include <net/netfilter/nf_flow_table.h> > +#include <linux/bpf.h> > +#include <linux/btf.h> > +#include <net/xdp.h> > + > +__diag_push(); > +__diag_ignore_all("-Wmissing-prototypes", > + "Global functions as their definitions will be in nf_flow_table BTF"); > + > +static struct flow_offload_tuple_rhash * > +bpf_xdp_flow_offload_tuple_lookup(struct net_device *dev, > + struct flow_offload_tuple *tuple, > + __be16 proto) > +{ > + struct flow_offload_tuple_rhash *tuplehash; > + struct nf_flowtable *flow_table; > + struct flow_offload *flow; > + > + flow_table = nf_flowtable_by_dev(dev); > + if (!flow_table) > + return ERR_PTR(-ENOENT); > + > + tuplehash = flow_offload_lookup(flow_table, tuple); > + if (!tuplehash) > + return ERR_PTR(-ENOENT); This is fine to do, but the caller should catch it using IS_ERR_PTR and return NULL. BPF side cannot distinguish ERR_PTR from normal pointer, so this will cause a bad deref in the program. > + > + flow = container_of(tuplehash, struct flow_offload, > + tuplehash[tuplehash->tuple.dir]); > + flow_offload_refresh(flow_table, flow, false); > + > + return tuplehash; > +} > + > +__bpf_kfunc struct flow_offload_tuple_rhash * > +bpf_xdp_flow_offload_lookup(struct xdp_md *ctx, > + struct bpf_fib_lookup *fib_tuple, > + u32 fib_tuple__sz) Do you think the __sz has the intended effect? It only works when the preceding parameter is a void *. If you have a type like struct bpf_fib_lookup, I think it should work fine without taking a size at all. > +{ > + struct xdp_buff *xdp = (struct xdp_buff *)ctx; > + struct flow_offload_tuple tuple = { > + .iifidx = fib_tuple->ifindex, > + .l3proto = fib_tuple->family, > + .l4proto = fib_tuple->l4_protocol, > + .src_port = fib_tuple->sport, > + .dst_port = fib_tuple->dport, > + }; > + __be16 proto; > + > + switch (fib_tuple->family) { > + case AF_INET: > + tuple.src_v4.s_addr = fib_tuple->ipv4_src; > + tuple.dst_v4.s_addr = fib_tuple->ipv4_dst; > + proto = htons(ETH_P_IP); > + break; > + case AF_INET6: > + tuple.src_v6 = *(struct in6_addr *)&fib_tuple->ipv6_src; > + tuple.dst_v6 = *(struct in6_addr *)&fib_tuple->ipv6_dst; > + proto = htons(ETH_P_IPV6); > + break; > + default: > + return ERR_PTR(-EINVAL); Likewise. While you check IS_ERR_VALUE in selftest, direct dereference will be allowed by verifier, which would crash the kernel. It's better to do something like conntrack kfuncs, where they set opts->error when returning NULL, allowing better debugging in case lookup fails. > + } > + > + return bpf_xdp_flow_offload_tuple_lookup(xdp->rxq->dev, &tuple, proto); > +} > + > +__diag_pop() > + > +BTF_KFUNCS_START(nf_ft_kfunc_set) > +BTF_ID_FLAGS(func, bpf_xdp_flow_offload_lookup) > +BTF_KFUNCS_END(nf_ft_kfunc_set) > + > +static const struct btf_kfunc_id_set nf_flow_offload_kfunc_set = { > + .owner = THIS_MODULE, > + .set = &nf_ft_kfunc_set, > +}; > + > +int nf_flow_offload_register_bpf(void) > +{ > + return register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, > + &nf_flow_offload_kfunc_set); > +} We should probably also expose it to skb? We just need net_device, so it can work with both XDP and TC progs. That would be similar to how we expose conntrack kfuncs to both XDP and TC progs. > +EXPORT_SYMBOL_GPL(nf_flow_offload_register_bpf); > diff --git a/net/netfilter/nf_flow_table_inet.c b/net/netfilter/nf_flow_table_inet.c > index 6eef15648b7b0..b13587238eceb 100644 > --- a/net/netfilter/nf_flow_table_inet.c > +++ b/net/netfilter/nf_flow_table_inet.c > @@ -98,6 +98,8 @@ static int __init nf_flow_inet_module_init(void) > nft_register_flowtable_type(&flowtable_ipv6); > nft_register_flowtable_type(&flowtable_inet); > > + nf_flow_offload_register_bpf(); > + Error checking needed here? Kfunc registration can fail. > return 0; > } > > -- > 2.45.0 > >
[...] > > + tuplehash = flow_offload_lookup(flow_table, tuple); > > + if (!tuplehash) > > + return ERR_PTR(-ENOENT); > > This is fine to do, but the caller should catch it using IS_ERR_PTR > and return NULL. > BPF side cannot distinguish ERR_PTR from normal pointer, so this will > cause a bad deref in the program. ack, I will fix it in v2. > > > + > > + flow = container_of(tuplehash, struct flow_offload, > > + tuplehash[tuplehash->tuple.dir]); > > + flow_offload_refresh(flow_table, flow, false); > > + > > + return tuplehash; > > +} > > + > > +__bpf_kfunc struct flow_offload_tuple_rhash * > > +bpf_xdp_flow_offload_lookup(struct xdp_md *ctx, > > + struct bpf_fib_lookup *fib_tuple, > > + u32 fib_tuple__sz) > > Do you think the __sz has the intended effect? It only works when the > preceding parameter is a void *. > If you have a type like struct bpf_fib_lookup, I think it should work > fine without taking a size at all. ack, I will fix it in v2. > > > +{ > > + struct xdp_buff *xdp = (struct xdp_buff *)ctx; > > + struct flow_offload_tuple tuple = { > > + .iifidx = fib_tuple->ifindex, > > + .l3proto = fib_tuple->family, > > + .l4proto = fib_tuple->l4_protocol, > > + .src_port = fib_tuple->sport, > > + .dst_port = fib_tuple->dport, > > + }; > > + __be16 proto; > > + > > + switch (fib_tuple->family) { > > + case AF_INET: > > + tuple.src_v4.s_addr = fib_tuple->ipv4_src; > > + tuple.dst_v4.s_addr = fib_tuple->ipv4_dst; > > + proto = htons(ETH_P_IP); > > + break; > > + case AF_INET6: > > + tuple.src_v6 = *(struct in6_addr *)&fib_tuple->ipv6_src; > > + tuple.dst_v6 = *(struct in6_addr *)&fib_tuple->ipv6_dst; > > + proto = htons(ETH_P_IPV6); > > + break; > > + default: > > + return ERR_PTR(-EINVAL); > > Likewise. While you check IS_ERR_VALUE in selftest, direct dereference > will be allowed by verifier, which would crash the kernel. > It's better to do something like conntrack kfuncs, where they set > opts->error when returning NULL, allowing better debugging in case > lookup fails. ack, I will fix it in v2. > > > + } > > + > > + return bpf_xdp_flow_offload_tuple_lookup(xdp->rxq->dev, &tuple, proto); > > +} > > + > > +__diag_pop() > > + > > +BTF_KFUNCS_START(nf_ft_kfunc_set) > > +BTF_ID_FLAGS(func, bpf_xdp_flow_offload_lookup) > > +BTF_KFUNCS_END(nf_ft_kfunc_set) > > + > > +static const struct btf_kfunc_id_set nf_flow_offload_kfunc_set = { > > + .owner = THIS_MODULE, > > + .set = &nf_ft_kfunc_set, > > +}; > > + > > +int nf_flow_offload_register_bpf(void) > > +{ > > + return register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, > > + &nf_flow_offload_kfunc_set); > > +} > > We should probably also expose it to skb? We just need net_device, so > it can work with both XDP and TC progs. > That would be similar to how we expose conntrack kfuncs to both XDP > and TC progs. I think we will get very similar results to sw flowtable in this case, don't you think? > > > +EXPORT_SYMBOL_GPL(nf_flow_offload_register_bpf); > > diff --git a/net/netfilter/nf_flow_table_inet.c b/net/netfilter/nf_flow_table_inet.c > > index 6eef15648b7b0..b13587238eceb 100644 > > --- a/net/netfilter/nf_flow_table_inet.c > > +++ b/net/netfilter/nf_flow_table_inet.c > > @@ -98,6 +98,8 @@ static int __init nf_flow_inet_module_init(void) > > nft_register_flowtable_type(&flowtable_ipv6); > > nft_register_flowtable_type(&flowtable_inet); > > > > + nf_flow_offload_register_bpf(); > > + > > Error checking needed here? Kfunc registration can fail. ack, I will fix it. Regards, Lorenzo > > > return 0; > > } > > > > -- > > 2.45.0 > > > > >
diff --git a/include/net/netfilter/nf_flow_table.h b/include/net/netfilter/nf_flow_table.h index 0bbe6ea8e0651..085660cbcd3f2 100644 --- a/include/net/netfilter/nf_flow_table.h +++ b/include/net/netfilter/nf_flow_table.h @@ -312,6 +312,16 @@ unsigned int nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb, unsigned int nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state); +#if (IS_BUILTIN(CONFIG_NF_FLOW_TABLE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \ + (IS_MODULE(CONFIG_NF_FLOW_TABLE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)) +extern int nf_flow_offload_register_bpf(void); +#else +static inline int nf_flow_offload_register_bpf(void) +{ + return 0; +} +#endif + #define MODULE_ALIAS_NF_FLOWTABLE(family) \ MODULE_ALIAS("nf-flowtable-" __stringify(family)) diff --git a/net/netfilter/Makefile b/net/netfilter/Makefile index 614815a3ed738..18b09cec92024 100644 --- a/net/netfilter/Makefile +++ b/net/netfilter/Makefile @@ -144,6 +144,11 @@ obj-$(CONFIG_NF_FLOW_TABLE) += nf_flow_table.o nf_flow_table-objs := nf_flow_table_core.o nf_flow_table_ip.o \ nf_flow_table_offload.o nf_flow_table-$(CONFIG_NF_FLOW_TABLE_PROCFS) += nf_flow_table_procfs.o +ifeq ($(CONFIG_NF_FLOW_TABLE),m) +nf_flow_table-$(CONFIG_DEBUG_INFO_BTF_MODULES) += nf_flow_table_bpf.o +else ifeq ($(CONFIG_NF_FLOW_TABLE),y) +nf_flow_table-$(CONFIG_DEBUG_INFO_BTF) += nf_flow_table_bpf.o +endif obj-$(CONFIG_NF_FLOW_TABLE_INET) += nf_flow_table_inet.o diff --git a/net/netfilter/nf_flow_table_bpf.c b/net/netfilter/nf_flow_table_bpf.c new file mode 100644 index 0000000000000..836a1127e4052 --- /dev/null +++ b/net/netfilter/nf_flow_table_bpf.c @@ -0,0 +1,95 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* Unstable Flow Table Helpers for XDP hook + * + * These are called from the XDP programs. + * Note that it is allowed to break compatibility for these functions since + * the interface they are exposed through to BPF programs is explicitly + * unstable. + */ + +#include <linux/kernel.h> +#include <linux/init.h> +#include <linux/module.h> +#include <net/netfilter/nf_flow_table.h> +#include <linux/bpf.h> +#include <linux/btf.h> +#include <net/xdp.h> + +__diag_push(); +__diag_ignore_all("-Wmissing-prototypes", + "Global functions as their definitions will be in nf_flow_table BTF"); + +static struct flow_offload_tuple_rhash * +bpf_xdp_flow_offload_tuple_lookup(struct net_device *dev, + struct flow_offload_tuple *tuple, + __be16 proto) +{ + struct flow_offload_tuple_rhash *tuplehash; + struct nf_flowtable *flow_table; + struct flow_offload *flow; + + flow_table = nf_flowtable_by_dev(dev); + if (!flow_table) + return ERR_PTR(-ENOENT); + + tuplehash = flow_offload_lookup(flow_table, tuple); + if (!tuplehash) + return ERR_PTR(-ENOENT); + + flow = container_of(tuplehash, struct flow_offload, + tuplehash[tuplehash->tuple.dir]); + flow_offload_refresh(flow_table, flow, false); + + return tuplehash; +} + +__bpf_kfunc struct flow_offload_tuple_rhash * +bpf_xdp_flow_offload_lookup(struct xdp_md *ctx, + struct bpf_fib_lookup *fib_tuple, + u32 fib_tuple__sz) +{ + struct xdp_buff *xdp = (struct xdp_buff *)ctx; + struct flow_offload_tuple tuple = { + .iifidx = fib_tuple->ifindex, + .l3proto = fib_tuple->family, + .l4proto = fib_tuple->l4_protocol, + .src_port = fib_tuple->sport, + .dst_port = fib_tuple->dport, + }; + __be16 proto; + + switch (fib_tuple->family) { + case AF_INET: + tuple.src_v4.s_addr = fib_tuple->ipv4_src; + tuple.dst_v4.s_addr = fib_tuple->ipv4_dst; + proto = htons(ETH_P_IP); + break; + case AF_INET6: + tuple.src_v6 = *(struct in6_addr *)&fib_tuple->ipv6_src; + tuple.dst_v6 = *(struct in6_addr *)&fib_tuple->ipv6_dst; + proto = htons(ETH_P_IPV6); + break; + default: + return ERR_PTR(-EINVAL); + } + + return bpf_xdp_flow_offload_tuple_lookup(xdp->rxq->dev, &tuple, proto); +} + +__diag_pop() + +BTF_KFUNCS_START(nf_ft_kfunc_set) +BTF_ID_FLAGS(func, bpf_xdp_flow_offload_lookup) +BTF_KFUNCS_END(nf_ft_kfunc_set) + +static const struct btf_kfunc_id_set nf_flow_offload_kfunc_set = { + .owner = THIS_MODULE, + .set = &nf_ft_kfunc_set, +}; + +int nf_flow_offload_register_bpf(void) +{ + return register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, + &nf_flow_offload_kfunc_set); +} +EXPORT_SYMBOL_GPL(nf_flow_offload_register_bpf); diff --git a/net/netfilter/nf_flow_table_inet.c b/net/netfilter/nf_flow_table_inet.c index 6eef15648b7b0..b13587238eceb 100644 --- a/net/netfilter/nf_flow_table_inet.c +++ b/net/netfilter/nf_flow_table_inet.c @@ -98,6 +98,8 @@ static int __init nf_flow_inet_module_init(void) nft_register_flowtable_type(&flowtable_ipv6); nft_register_flowtable_type(&flowtable_inet); + nf_flow_offload_register_bpf(); + return 0; }
Introduce bpf_xdp_flow_offload_lookup kfunc in order to perform the lookup of a given flowtable entry based on a fib tuple of incoming traffic. bpf_xdp_flow_offload_lookup can be used as building block to offload in xdp the processing of sw flowtable when hw flowtable is not available. Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org> --- include/net/netfilter/nf_flow_table.h | 10 +++ net/netfilter/Makefile | 5 ++ net/netfilter/nf_flow_table_bpf.c | 95 +++++++++++++++++++++++++++ net/netfilter/nf_flow_table_inet.c | 2 + 4 files changed, 112 insertions(+) create mode 100644 net/netfilter/nf_flow_table_bpf.c