Various BPF helper improvements
Subsystems (8)
Daniel Borkmann
maintainercover Various BPF helper improvements
This series adds two BPF helpers, that is, one for retrieving the classid of an skb and another one to redirect via the neigh subsystem, and improves also the cookie helpers by removing the atomic counter. I've also added the bpf_tail_call_static() helper to the libbpf API that we've been using in Cilium for a while now, and last but not least the series adds a few selftests. For details, please check individual patches, thanks!
Daniel Borkmann (6):
bpf: add classid helper only based on skb->sk
bpf, net: rework cookie generator as per-cpu one
bpf: add redirect_neigh helper as redirect drop-in bpf, libbpf: add bpf_tail_call_static helper for bpf programs bpf, selftests: use bpf_tail_call_static where appropriate bpf, selftests: add redirect_neigh selftest
include/linux/cookie.h | 41 +++ include/linux/skbuff.h | 5 + include/uapi/linux/bpf.h | 24 ++ net/core/filter.c | 277 +++++++++++++++++- net/core/net_namespace.c | 5 +- net/core/sock_diag.c | 7 +- tools/include/uapi/linux/bpf.h | 24 ++ tools/lib/bpf/bpf_helpers.h | 32 ++ tools/testing/selftests/bpf/progs/bpf_flow.c | 12 +- tools/testing/selftests/bpf/progs/tailcall1.c | 28 +- tools/testing/selftests/bpf/progs/tailcall2.c | 14 +- tools/testing/selftests/bpf/progs/tailcall3.c | 4 +- .../selftests/bpf/progs/tailcall_bpf2bpf1.c | 4 +- .../selftests/bpf/progs/tailcall_bpf2bpf2.c | 6 +- .../selftests/bpf/progs/tailcall_bpf2bpf3.c | 6 +- .../selftests/bpf/progs/tailcall_bpf2bpf4.c | 6 +- .../selftests/bpf/progs/test_tc_neigh.c | 144 +++++++++ tools/testing/selftests/bpf/test_tc_neigh.sh | 168 +++++++++++ 18 files changed, 749 insertions(+), 58 deletions(-) create mode 100644 include/linux/cookie.h create mode 100644 tools/testing/selftests/bpf/progs/test_tc_neigh.c create mode 100755 tools/testing/selftests/bpf/test_tc_neigh.sh
Signature · 1 line
2.21.0
Daniel Borkmann
maintainerPATCH 1/6 bpf: add classid helper only based on skb->sk
Similarly to 5a52ae4e32a6 ("bpf: Allow to retrieve cgroup v1 classid from v2 hooks"), add a helper to retrieve cgroup v1 classid solely based on the skb->sk, so it can be used as key as part of BPF map lookups out of tc from host ns, in particular given the skb->sk is retained these days when crossing net ns thanks to 9c4c325252c5 ("skbuff: preserve sock reference when scrubbing the skb."). This is similar to bpf_skb_cgroup_id() which implements the same for v2. Kubernetes ecosystem is still operating on v1 however, hence net_cls needs to be used there until this can be dropped in with the v2 helper of bpf_skb_cgroup_id().
--- include/uapi/linux/bpf.h | 10 ++++++++++ net/core/filter.c | 21 +++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 10 ++++++++++ 3 files changed, 41 insertions(+)
include/uapi/linux/bpf.h +10 -0
@@ -3586,6 +3586,15 @@ union bpf_attr {* the data in *dst*. This is a wrapper of **copy_from_user**\ (). * Return * 0 on success, or a negative error in case of failure. +* +* u64 bpf_skb_cgroup_classid(struct sk_buff *skb) +* Description +* See **bpf_get_cgroup_classid**\ () for the main description. +* This helper differs from **bpf_get_cgroup_classid**\ () in that +* the cgroup v1 net_cls class is retrieved only from the *skb*'s +* associated socket instead of the current process. +* Return +* The id is returned or 0 in case the id could not be retrieved. */ #define __BPF_FUNC_MAPPER (FN ) \ FN(unspec), \
@@ -3737,6 +3746,7 @@ union bpf_attr {FN (inode_storage_delete), \FN (d_path), \FN (copy_from_user), \ +FN (skb_cgroup_classid), \/* */ /* integer value in 'imm' field of BPF_CALL instruction selects which helper
net/core/filter.c +21 -0
@@ -2704,6 +2704,23 @@ static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = { .gpl_only = false, .ret_type = RET_INTEGER, }; + +BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb) +{ +struct sock *sk =skb_to_full_sk (skb); + +if (!sk || !sk_fullsock (sk)) +return 0 ; + +return sock_cgroup_classid (&sk->sk_cgrp_data); +} + +static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = { + .func = bpf_skb_cgroup_classid, + .gpl_only = false, + .ret_type =RET_INTEGER , + .arg1_type =ARG_PTR_TO_CTX , +};#endif BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
@@ -6770,6 +6787,10 @@ tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)case BPF_FUNC_skb_get_xfrm_state:return &bpf_skb_get_xfrm_state_proto;#endif +#ifdef CONFIG_CGROUP_NET_CLASSID +case BPF_FUNC_skb_cgroup_classid: +return &bpf_skb_cgroup_classid_proto; +#endif #ifdef CONFIG_SOCK_CGROUP_DATA case BPF_FUNC_skb_cgroup_id:return &bpf_skb_cgroup_id_proto;
tools/include/uapi/linux/bpf.h +10 -0
@@ -3586,6 +3586,15 @@ union bpf_attr {* the data in *dst*. This is a wrapper of **copy_from_user**\ (). * Return * 0 on success, or a negative error in case of failure. +* +* u64 bpf_skb_cgroup_classid(struct sk_buff *skb) +* Description +* See **bpf_get_cgroup_classid**\ () for the main description. +* This helper differs from **bpf_get_cgroup_classid**\ () in that +* the cgroup v1 net_cls class is retrieved only from the *skb*'s +* associated socket instead of the current process. +* Return +* The id is returned or 0 in case the id could not be retrieved. */ #define __BPF_FUNC_MAPPER (FN ) \ FN(unspec), \
@@ -3737,6 +3746,7 @@ union bpf_attr {FN (inode_storage_delete), \FN (d_path), \FN (copy_from_user), \ +FN (skb_cgroup_classid), \/* */ /* integer value in 'imm' field of BPF_CALL instruction selects which helper
Signature · 1 line
2.21.0
Daniel Borkmann
maintainerPATCH 5/6 bpf, selftests: use bpf_tail_call_static where appropriate
In reply to Daniel BorkmannFor those locations where we use an immediate tail call map index use the newly added bpf_tail_call_static() helper.
--- tools/testing/selftests/bpf/progs/bpf_flow.c | 12 ++++---- tools/testing/selftests/bpf/progs/tailcall1.c | 28 +++++++++---------- tools/testing/selftests/bpf/progs/tailcall2.c | 14 +++++----- tools/testing/selftests/bpf/progs/tailcall3.c | 4 +-- .../selftests/bpf/progs/tailcall_bpf2bpf1.c | 4 +-- .../selftests/bpf/progs/tailcall_bpf2bpf2.c | 6 ++-- .../selftests/bpf/progs/tailcall_bpf2bpf3.c | 6 ++-- .../selftests/bpf/progs/tailcall_bpf2bpf4.c | 6 ++-- 8 files changed, 40 insertions(+), 40 deletions(-)
tools/testing/selftests/bpf/progs/bpf_flow.c +6 -6
@@ -118,18 +118,18 @@ static __always_inline int parse_eth_proto(struct __sk_buff *skb, __be16 proto)switch (proto) {case bpf_htons (ETH_P_IP ): -bpf_tail_call (skb, &jmp_table,IP ); +bpf_tail_call_static (skb, &jmp_table,IP );break ;case bpf_htons (ETH_P_IPV6 ): -bpf_tail_call (skb, &jmp_table,IPV6 ); +bpf_tail_call_static (skb, &jmp_table,IPV6 );break ;case bpf_htons (ETH_P_MPLS_MC ):case bpf_htons (ETH_P_MPLS_UC ): -bpf_tail_call (skb, &jmp_table,MPLS ); +bpf_tail_call_static (skb, &jmp_table,MPLS );break ;case bpf_htons (ETH_P_8021Q ):case bpf_htons (ETH_P_8021AD ): -bpf_tail_call (skb, &jmp_table,VLAN ); +bpf_tail_call_static (skb, &jmp_table,VLAN );break ;default :/* Protocol not supported */
@@ -246,10 +246,10 @@ static __always_inline int parse_ipv6_proto(struct __sk_buff *skb, __u8 nexthdr)switch (nexthdr) {case IPPROTO_HOPOPTS :case IPPROTO_DSTOPTS : -bpf_tail_call (skb, &jmp_table,IPV6OP ); +bpf_tail_call_static (skb, &jmp_table,IPV6OP );break ;case IPPROTO_FRAGMENT : -bpf_tail_call (skb, &jmp_table,IPV6FR ); +bpf_tail_call_static (skb, &jmp_table,IPV6FR );break ;default :return parse_ip_proto (skb, nexthdr);
tools/testing/selftests/bpf/progs/tailcall1.c +14 -14
@@ -26,20 +26,20 @@ int entry(struct __sk_buff *skb)/* Multiple locations to make sure we patch * all of them. */ -bpf_tail_call (skb, &jmp_table,0 ); -bpf_tail_call (skb, &jmp_table,0 ); -bpf_tail_call (skb, &jmp_table,0 ); -bpf_tail_call (skb, &jmp_table,0 ); - -bpf_tail_call (skb, &jmp_table,1 ); -bpf_tail_call (skb, &jmp_table,1 ); -bpf_tail_call (skb, &jmp_table,1 ); -bpf_tail_call (skb, &jmp_table,1 ); - -bpf_tail_call (skb, &jmp_table,2 ); -bpf_tail_call (skb, &jmp_table,2 ); -bpf_tail_call (skb, &jmp_table,2 ); -bpf_tail_call (skb, &jmp_table,2 ); +bpf_tail_call_static (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 ); + +bpf_tail_call_static (skb, &jmp_table,1 ); +bpf_tail_call_static (skb, &jmp_table,1 ); +bpf_tail_call_static (skb, &jmp_table,1 ); +bpf_tail_call_static (skb, &jmp_table,1 ); + +bpf_tail_call_static (skb, &jmp_table,2 ); +bpf_tail_call_static (skb, &jmp_table,2 ); +bpf_tail_call_static (skb, &jmp_table,2 ); +bpf_tail_call_static (skb, &jmp_table,2 );return 3 ; }
tools/testing/selftests/bpf/progs/tailcall2.c +7 -7
@@ -13,14 +13,14 @@ struct { SEC("classifier/0")int bpf_func_0 (struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,1 ); +bpf_tail_call_static (skb, &jmp_table,1 );return 0 ; } SEC("classifier/1")int bpf_func_1 (struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,2 ); +bpf_tail_call_static (skb, &jmp_table,2 );return 1 ; }
@@ -33,25 +33,25 @@ int bpf_func_2(struct __sk_buff *skb) SEC("classifier/3")int bpf_func_3 (struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,4 ); +bpf_tail_call_static (skb, &jmp_table,4 );return 3 ; } SEC("classifier/4")int bpf_func_4 (struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,3 ); +bpf_tail_call_static (skb, &jmp_table,3 );return 4 ; } SEC("classifier")int entry (struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 );/* Check multi-prog update. */ -bpf_tail_call (skb, &jmp_table,2 ); +bpf_tail_call_static (skb, &jmp_table,2 );/* Check tail call limit. */ -bpf_tail_call (skb, &jmp_table,3 ); +bpf_tail_call_static (skb, &jmp_table,3 );return 3 ; }
tools/testing/selftests/bpf/progs/tailcall3.c +2 -2
@@ -16,14 +16,14 @@ SEC("classifier/0")int bpf_func_0 (struct __sk_buff *skb) { count++; -bpf_tail_call (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 );return 1 ; } SEC("classifier")int entry (struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 );return 0 ; }
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf1.c +2 -2
@@ -21,7 +21,7 @@ TAIL_FUNC(1)static __noinline int subprog_tail(struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 );return skb->len *2 ; }
@@ -29,7 +29,7 @@ int subprog_tail(struct __sk_buff *skb) SEC("classifier")int entry (struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,1 ); +bpf_tail_call_static (skb, &jmp_table,1 );return subprog_tail (skb); }
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf2.c +3 -3
@@ -14,9 +14,9 @@ static __noinlineint subprog_tail (struct __sk_buff *skb) {if (load_byte (skb,0 )) -bpf_tail_call (skb, &jmp_table,1 ); +bpf_tail_call_static (skb, &jmp_table,1 );else -bpf_tail_call (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 );return 1 ; }
@@ -32,7 +32,7 @@ int bpf_func_0(struct __sk_buff *skb) SEC("classifier")int entry (struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 );return 0 ; }
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf3.c +3 -3
@@ -16,9 +16,9 @@ int subprog_tail2(struct __sk_buff *skb)volatile char arr[64 ] = {};if (load_word (skb,0 ) ||load_half (skb,0 )) -bpf_tail_call (skb, &jmp_table,10 ); +bpf_tail_call_static (skb, &jmp_table,10 );else -bpf_tail_call (skb, &jmp_table,1 ); +bpf_tail_call_static (skb, &jmp_table,1 );return skb->len; }
@@ -28,7 +28,7 @@ int subprog_tail(struct __sk_buff *skb) {volatile char arr[64 ] = {}; -bpf_tail_call (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 );return skb->len *2 ; }
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf4.c +3 -3
@@ -14,21 +14,21 @@ static volatile int count;__noinline int subprog_tail_2(struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,2 ); +bpf_tail_call_static (skb, &jmp_table,2 );return skb->len *3 ; }__noinline int subprog_tail_1(struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,1 ); +bpf_tail_call_static (skb, &jmp_table,1 );return skb->len *2 ; }__noinline int subprog_tail(struct __sk_buff *skb) { -bpf_tail_call (skb, &jmp_table,0 ); +bpf_tail_call_static (skb, &jmp_table,0 );return skb->len; }
Signature · 1 line
2.21.0
Daniel Borkmann
maintainerPATCH 2/6 bpf, net: rework cookie generator as per-cpu one
In reply to Daniel BorkmannWith its use in BPF the cookie generator can be called very frequently in particular when used out of cgroup v2 hooks (e.g. connect / sendmsg) and attached to the root cgroup, for example, when used in v1/v2 mixed environments. In particular when there's a high churn on sockets in the system there can be many parallel requests to the bpf_get_socket_cookie() and bpf_get_netns_cookie() helpers which then cause contention on the shared atomic counter. As similarly done in f991bd2e1421 ("fs: introduce a per-cpu last_ino allocator"), add a small helper library that both can then use for the 64 bit counters.
--- include/linux/cookie.h | 41 ++++++++++++++++++++++++++++++++++++++++ net/core/net_namespace.c | 5 +++-- net/core/sock_diag.c | 7 ++++--- 3 files changed, 48 insertions(+), 5 deletions(-) create mode 100644 include/linux/cookie.h
include/linux/cookie.h new file +41 -0
@@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef __LINUX_COOKIE_H +#define __LINUX_COOKIE_H + +#include <linux/atomic.h> +#include <linux/percpu.h> + +struct gen_cookie { +u64 __percpu *local_last; +atomic64_t shared_last____cacheline_aligned_in_smp ; +}; + +#define COOKIE_LOCAL_BATCH 4096 + +#define DEFINE_COOKIE (name) \ + static DEFINE_PER_CPU(u64, __##name); \ + static struct gen_cookie name = { \ + .local_last = &__##name, \ + .shared_last = ATOMIC64_INIT(0), \ + } + +static inline u64 gen_cookie_next (struct gen_cookie *gc) +{ +u64 *local_last = &get_cpu_var (*gc->local_last); +u64 val = *local_last; + +if (__is_defined (CONFIG_SMP ) && +unlikely ((val & (COOKIE_LOCAL_BATCH -1 )) ==0 )) { +s64 next =atomic64_add_return (COOKIE_LOCAL_BATCH , + &gc->shared_last); + val = next -COOKIE_LOCAL_BATCH ; + } + val++; +if (unlikely (!val)) + val++; + *local_last = val; +put_cpu_var (local_last); +return val; +} + +#endif /* __LINUX_COOKIE_H */
net/core/net_namespace.c +3 -2
@@ -19,6 +19,7 @@#include <linux/net_namespace.h> #include <linux/sched/task.h> #include <linux/uidgid.h> +#include <linux/cookie.h> #include <net/sock.h> #include <net/netlink.h>
@@ -69,7 +70,7 @@ EXPORT_SYMBOL_GPL(pernet_ops_rwsem);static unsigned int max_gen_ptrs =INITIAL_NET_GEN_PTRS ; -static atomic64_t cookie_gen; +DEFINE_COOKIE (net_cookie);u64 net_gen_cookie (struct net *net) {
@@ -78,7 +79,7 @@ u64 net_gen_cookie(struct net *net)if (res)return res; - res =atomic64_inc_return (&cookie_gen); + res =gen_cookie_next (&net_cookie);atomic64_cmpxchg (&net->net_cookie,0 , res); } }
net/core/sock_diag.c +4 -3
@@ -11,7 +11,7 @@#include <linux/tcp.h> #include <linux/workqueue.h> #include <linux/nospec.h> - +#include <linux/cookie.h> #include <linux/inet_diag.h> #include <linux/sock_diag.h>
@@ -19,7 +19,8 @@ static const struct sock_diag_handler *sock_diag_handlers[AF_MAX];static int (*inet_rcv_compat)(struct sk_buff *skb,struct nlmsghdr *nlh);static DEFINE_MUTEX (sock_diag_table_mutex);static struct workqueue_struct *broadcast_wq; -static atomic64_t cookie_gen; + +DEFINE_COOKIE (sock_cookie);u64 sock_gen_cookie (struct sock *sk) {
@@ -28,7 +29,7 @@ u64 sock_gen_cookie(struct sock *sk)if (res)return res; - res =atomic64_inc_return (&cookie_gen); + res =gen_cookie_next (&sock_cookie);atomic64_cmpxchg (&sk->sk_cookie,0 , res); } }
Signature · 1 line
2.21.0
Daniel Borkmann
maintainerPATCH 4/6 bpf, libbpf: add bpf_tail_call_static helper for bpf programs
In reply to Daniel BorkmannPort of tail_call_static() helper function from Cilium's BPF code base [0] to libbpf, so others can easily consume it as well. We've been using this in production code for some time now. The main idea is that we guarantee that the kernel's BPF infrastructure and JIT (here: x86_64) can patch the JITed BPF insns with direct jumps instead of having to fall back to using expensive retpolines. By using inline asm, we guarantee that the compiler won't merge the call from different paths with potentially different content of r2/r3.
We're also using __throw_build_bug() macro in different places as a neat trick to trigger compilation errors when compiler does not remove code at compilation time. This works for the BPF backend as it does not implement the __builtin_trap().
[0] https://github.com/cilium/cilium/commit/f5537c26020d5297b70936c6b7d03a1e412a1035
--- tools/lib/bpf/bpf_helpers.h | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+)
tools/lib/bpf/bpf_helpers.h +32 -0
@@ -53,6 +53,38 @@ })#endif +/* +* Misc useful helper macros +*/ +#ifndef __throw_build_bug +# define __throw_build_bug () __builtin_trap() +#endif + +static __always_inline void +bpf_tail_call_static (void *ctx,const void *map,const __u32 slot) +{ +if (!__builtin_constant_p (slot)) +__throw_build_bug (); + +/* +* Don't gamble, but _guarantee_ that LLVM won't optimize setting +* r2 and r3 from different paths ending up at the same call insn as +* otherwise we won't be able to use the jmpq/nopl retpoline-free +* patching by the x86-64 JIT in the kernel. +* +* Note on clobber list: we need to stay in-line with BPF calling +* convention, so even if we don't end up using r0, r4, r5, we need +* to mark them as clobber so that LLVM doesn't end up using them +* before / after the call. +*/ + asmvolatile ("r1 = %[ctx]\n\t" +"r2 = %[map]\n\t" +"r3 = %[slot]\n\t" +"call 12\n\t" + :: [ctx]"r" (ctx), [map]"r" (map), [slot]"i" (slot) + :"r0" ,"r1" ,"r2" ,"r3" ,"r4" ,"r5" ); +} +/* * Helper structure used by eBPF C program * to describe BPF map attributes to libbpf loader
Signature · 1 line
2.21.0
Daniel Borkmann
maintainerPATCH 3/6 bpf: add redirect_neigh helper as redirect drop-in
In reply to Daniel BorkmannAdd a redirect_neigh() helper as redirect() drop-in replacement for the xmit side. Main idea for the helper is to be very similar in semantics to the latter just that the skb gets injected into the neighboring subsystem in order to let the stack do the work it knows best anyway to populate the L2 addresses of the packet and then hand over to dev_queue_xmit() as redirect() does.
This solves two bigger items: i) skbs don't need to go up to the stack on the host facing veth ingress side for traffic egressing the container to achieve the same for populating L2 which also has the huge advantage that ii) the skb->sk won't get orphaned in ip_rcv_core() when entering the IP routing layer on the host stack.
Given that skb->sk neither gets orphaned when crossing the netns as per 9c4c325252c5 ("skbuff: preserve sock reference when scrubbing the skb.") the helper can then push the skbs directly to the phys device where FQ scheduler can do its work and TCP stack gets proper back-pressure given we hold on to skb->sk as long as skb is still residing in queues.
With the helper used in BPF data path to then push the skb to the phys device, I observed a stable/consistent TCP_STREAM improvement on veth devices for traffic going container -> host -> host -> container from ~10Gbps to ~15Gbps for a single stream in my test environment.
--- include/linux/skbuff.h | 5 + include/uapi/linux/bpf.h | 14 ++ net/core/filter.c | 256 +++++++++++++++++++++++++++++++-- tools/include/uapi/linux/bpf.h | 14 ++ 4 files changed, 276 insertions(+), 13 deletions(-)
include/linux/skbuff.h +5 -0
@@ -2539,6 +2539,11 @@ static inline int skb_mac_header_was_set(const struct sk_buff *skb) return skb->mac_header != (typeof(skb->mac_header))~0U; } +static inline void skb_unset_mac_header (struct sk_buff *skb) +{ + skb->mac_header = (typeof(skb->mac_header))~0U; +} +static inline void skb_reset_mac_header (struct sk_buff *skb) { skb->mac_header = skb->data - skb->head;
include/uapi/linux/bpf.h +14 -0
@@ -3595,6 +3595,19 @@ union bpf_attr {* associated socket instead of the current process. * Return * The id is returned or 0 in case the id could not be retrieved. +* +* long bpf_redirect_neigh(u32 ifindex, u64 flags) +* Description +* Redirect the packet to another net device of index *ifindex* +* and fill in L2 addresses from neighboring subsystem. This helper +* is somewhat similar to **bpf_redirect**\ (), except that it +* fills in e.g. MAC addresses based on the L3 information from +* the packet. This helper is supported for IPv4 and IPv6 protocols. +* The *flags* argument is reserved and must be 0. The helper is +* currently only supported for tc BPF program types. +* Return +* The helper returns **TC_ACT_REDIRECT** on success or +* **TC_ACT_SHOT** on error. */ #define __BPF_FUNC_MAPPER (FN ) \ FN(unspec), \
@@ -3747,6 +3760,7 @@ union bpf_attr {FN (d_path), \FN (copy_from_user), \FN (skb_cgroup_classid), \ +FN (redirect_neigh), \/* */ /* integer value in 'imm' field of BPF_CALL instruction selects which helper
net/core/filter.c +243 -13
291 lines not on this page · raw message
tools/include/uapi/linux/bpf.h +14 -0
@@ -3595,6 +3595,19 @@ union bpf_attr {* associated socket instead of the current process. * Return * The id is returned or 0 in case the id could not be retrieved. +* +* long bpf_redirect_neigh(u32 ifindex, u64 flags) +* Description +* Redirect the packet to another net device of index *ifindex* +* and fill in L2 addresses from neighboring subsystem. This helper +* is somewhat similar to **bpf_redirect**\ (), except that it +* fills in e.g. MAC addresses based on the L3 information from +* the packet. This helper is supported for IPv4 and IPv6 protocols. +* The *flags* argument is reserved and must be 0. The helper is +* currently only supported for tc BPF program types. +* Return +* The helper returns **TC_ACT_REDIRECT** on success or +* **TC_ACT_SHOT** on error. */ #define __BPF_FUNC_MAPPER (FN ) \ FN(unspec), \
@@ -3747,6 +3760,7 @@ union bpf_attr {FN (d_path), \FN (copy_from_user), \FN (skb_cgroup_classid), \ +FN (redirect_neigh), \/* */ /* integer value in 'imm' field of BPF_CALL instruction selects which helper
Signature · 1 line
2.21.0
Daniel Borkmann
maintainerPATCH 6/6 bpf, selftests: add redirect_neigh selftest
In reply to Daniel BorkmannAdd a small test that excercises the new redirect_neigh() helper for the IPv4 and IPv6 case.
--- .../selftests/bpf/progs/test_tc_neigh.c | 144 +++++++++++++++ tools/testing/selftests/bpf/test_tc_neigh.sh | 168 ++++++++++++++++++ 2 files changed, 312 insertions(+) create mode 100644 tools/testing/selftests/bpf/progs/test_tc_neigh.c create mode 100755 tools/testing/selftests/bpf/test_tc_neigh.sh
tools/testing/selftests/bpf/progs/test_tc_neigh.c new file +144 -0
@@ -0,0 +1,144 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <stdint.h> +#include <stdbool.h> + +#include <linux/bpf.h> +#include <linux/stddef.h> +#include <linux/pkt_cls.h> +#include <linux/if_ether.h> +#include <linux/in.h> +#include <linux/ip.h> +#include <linux/ipv6.h> + +#include <bpf/bpf_helpers.h> +#include <bpf/bpf_endian.h> + +#ifndef barrier_data +# define barrier_data (ptr) asm volatile("": :"r"(ptr) :"memory") +#endif + +#ifndef ctx_ptr +# define ctx_ptr (field) (void *)(long)(field) +#endif + +#define dst_to_src_tmp 0xeeddddeeU +#define src_to_dst_tmp 0xeeffffeeU + +#define ip4_src 0xac100164/* 172.16.1.100 */ +#define ip4_dst 0xac100264/* 172.16.2.100 */ + +#define ip6_src { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ + 0x00, 0x01, 0xde, 0xad, 0xbe, 0xef, 0xca, 0xfe } +#define ip6_dst { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ + 0x00, 0x02, 0xde, 0xad, 0xbe, 0xef, 0xca, 0xfe } + +#ifndef v6_equal +# define v6_equal (a, b) (a.s6_addr32[0] == b.s6_addr32[0] && \ + a.s6_addr32[1] == b.s6_addr32[1] && \ + a.s6_addr32[2] == b.s6_addr32[2] && \ + a.s6_addr32[3] == b.s6_addr32[3]) +#endif + +static __always_inline bool is_remote_ep_v4(struct __sk_buff *skb, +__be32 addr) +{ +void *data_end =ctx_ptr (skb->data_end); +void *data =ctx_ptr (skb->data); +struct iphdr *ip4h; + +if (data +sizeof (struct ethhdr ) > data_end) +return false; + + ip4h = (struct iphdr *)(data +sizeof (struct ethhdr )); +if ((void *)(ip4h +1 ) > data_end) +return false; + +return ip4h->daddr == addr; +} + +static __always_inline bool is_remote_ep_v6(struct __sk_buff *skb, +struct in6_addr addr) +{ +void *data_end =ctx_ptr (skb->data_end); +void *data =ctx_ptr (skb->data); +struct ipv6hdr *ip6h; + +if (data +sizeof (struct ethhdr ) > data_end) +return false; + + ip6h = (struct ipv6hdr *)(data +sizeof (struct ethhdr )); +if ((void *)(ip6h +1 ) > data_end) +return false; + +return v6_equal (ip6h->daddr, addr); +} + +SEC("chk_neigh") int tc_chk(struct __sk_buff *skb) +{ +void *data_end =ctx_ptr (skb->data_end); +void *data =ctx_ptr (skb->data); +__u32 *raw = data; + +if (data +sizeof (struct ethhdr ) > data_end) +return TC_ACT_SHOT ; + +return !raw[0 ] && !raw[1 ] && !raw[2 ] ?TC_ACT_SHOT :TC_ACT_OK ; +} + +SEC("dst_ingress") int tc_dst(struct __sk_buff *skb) +{ +int idx = dst_to_src_tmp; +__u8 zero[ETH_ALEN *2 ]; +bool redirect = false; + +switch (skb->protocol) { +case __bpf_constant_htons (ETH_P_IP ): + redirect =is_remote_ep_v4 (skb,__bpf_constant_htonl (ip4_src)); +break ; +case __bpf_constant_htons (ETH_P_IPV6 ): + redirect =is_remote_ep_v6 (skb, (struct in6_addr )ip6_src); +break ; + } + +if (!redirect) +return TC_ACT_OK ; + +barrier_data (&idx); + idx =bpf_ntohl (idx); + +__builtin_memset (&zero,0 ,sizeof (zero)); +if (bpf_skb_store_bytes (skb,0 , &zero,sizeof (zero),0 ) <0 ) +return TC_ACT_SHOT ; + +return bpf_redirect_neigh (idx,0 ); +} + +SEC("src_ingress") int tc_src(struct __sk_buff *skb) +{ +int idx = src_to_dst_tmp; +__u8 zero[ETH_ALEN *2 ]; +bool redirect = false; + +switch (skb->protocol) { +case __bpf_constant_htons (ETH_P_IP ): + redirect =is_remote_ep_v4 (skb,__bpf_constant_htonl (ip4_dst)); +break ; +case __bpf_constant_htons (ETH_P_IPV6 ): + redirect =is_remote_ep_v6 (skb, (struct in6_addr )ip6_dst); +break ; + } + +if (!redirect) +return TC_ACT_OK ; + +barrier_data (&idx); + idx =bpf_ntohl (idx); + +__builtin_memset (&zero,0 ,sizeof (zero)); +if (bpf_skb_store_bytes (skb,0 , &zero,sizeof (zero),0 ) <0 ) +return TC_ACT_SHOT ; + +return bpf_redirect_neigh (idx,0 ); +} + +char __license[] SEC("license") = "GPL";
tools/testing/selftests/bpf/test_tc_neigh.sh new file +168 -0
168 lines not on this page · raw message
Signature · 1 line
2.21.0
Re: [PATCH bpf-next 2/6] bpf, net: rework cookie generator as per-cpu one
In reply to Daniel BorkmannQuoted 52 lines · Daniel Borkmann
With its use in BPF the cookie generator can be called very frequently in particular when used out of cgroup v2 hooks (e.g. connect / sendmsg) and attached to the root cgroup, for example, when used in v1/v2 mixed environments. In particular when there's a high churn on sockets in the system there can be many parallel requests to the
bpf_get_socket_cookie()andbpf_get_netns_cookie()helpers which then cause contention on the shared atomic counter. As similarly done inf991bd2e1421("fs: introduce a per-cpulast_inoallocator"), add a small helper library that both can then use for the 64 bit counters.Signed-off-by Daniel Borkmann <[email protected]>--- include/linux/cookie.h | 41 ++++++++++++++++++++++++++++++++++++++++ net/core/net_namespace.c | 5 +++-- net/core/sock_diag.c | 7 ++++--- 3 files changed, 48 insertions(+), 5 deletions(-) create mode 100644 include/linux/cookie.hinclude/linux/cookie.h new file +28
-0@@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef __LINUX_COOKIE_H +#define __LINUX_COOKIE_H + +#include <linux/atomic.h> +#include <linux/percpu.h> + +struct gen_cookie { +u64 __percpu *local_last; +atomic64_t shared_last____cacheline_aligned_in_smp ; +}; + +#define COOKIE_LOCAL_BATCH 4096 + +#define DEFINE_COOKIE (name) \ + static DEFINE_PER_CPU(u64, __##name); \ + static struct gen_cookie name = { \ + .local_last = &__##name, \ + .shared_last = ATOMIC64_INIT(0), \ + } + +static inline u64 gen_cookie_next (struct gen_cookie *gc) +{ +u64 *local_last = &get_cpu_var (*gc->local_last); +u64 val = *local_last; + +if (__is_defined (CONFIG_SMP ) && +unlikely ((val & (COOKIE_LOCAL_BATCH -1 )) ==0 )) {
include/linux/cookie.h line 38
Quoted 10 lines
include/linux/cookie.h new file +10
-0+s64 next =atomic64_add_return (COOKIE_LOCAL_BATCH , + &gc->shared_last); + val = next -COOKIE_LOCAL_BATCH ; + } + val++; +if (unlikely (!val)) + val++; + *local_last = val; +put_cpu_var (local_last); +return val;
This is not interrupt safe.
I think sock_gen_cookie() can be called from interrupt context.
get_next_ino() is only called from process context, that is what I used get_cpu_var() and put_cpu_var()
Re: [PATCH bpf-next 5/6] bpf, selftests: use bpf_tail_call_static where appropriate
In reply to Daniel BorkmannQuoted 4 lines · Daniel Borkmann
For those locations where we use an immediate tail call map index use the newly added
bpf_tail_call_static()helper.Signed-off-by Daniel Borkmann <[email protected]>
Quoted 10 lines
--- tools/testing/selftests/bpf/progs/bpf_flow.c | 12 ++++---- tools/testing/selftests/bpf/progs/tailcall1.c | 28 +++++++++---------- tools/testing/selftests/bpf/progs/tailcall2.c | 14 +++++----- tools/testing/selftests/bpf/progs/tailcall3.c | 4 +-- .../selftests/bpf/progs/tailcall_bpf2bpf1.c | 4 +-- .../selftests/bpf/progs/tailcall_bpf2bpf2.c | 6 ++-- .../selftests/bpf/progs/tailcall_bpf2bpf3.c | 6 ++-- .../selftests/bpf/progs/tailcall_bpf2bpf4.c | 6 ++-- 8 files changed, 40 insertions(+), 40 deletions(-)
One nit, while you're at it, maybe it would be good to also address the samples/bpf/sockex3_kern.c that is also using the immediate map index?
[...]
Re: [PATCH bpf-next 4/6] bpf, libbpf: add bpf_tail_call_static helper for bpf programs
In reply to Daniel BorkmannQuoted 27 lines · Daniel Borkmann
Port of
tail_call_static()helper function from Cilium's BPF code base [0] to libbpf, so others can easily consume it as well. We've been using this in production code for some time now. The main idea is that we guarantee that the kernel's BPF infrastructure and JIT (here:x86_64) can patch the JITed BPF insns with direct jumps instead of having to fall back to using expensive retpolines. By using inline asm, we guarantee that the compiler won't merge the call from different paths with potentially different content of r2/r3.We're also using
__throw_build_bug()macro in different places as a neat trick to trigger compilation errors when compiler does not remove code at compilation time. This works for the BPF backend as it does not implement the__builtin_trap().[0] https://github.com/cilium/cilium/commit/f5537c26020d5297b70936c6b7d03a1e412a1035
Signed-off-by Daniel Borkmann <[email protected]>--- tools/lib/bpf/bpf_helpers.h | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+)tools/lib/bpf/bpf_helpers.h +0
-0
tools/lib/bpf/bpf_helpers.h line 61
Quoted 9 lines
tools/lib/bpf/bpf_helpers.h +6
-0@@ -53,6 +53,38 @@ })#endif +/* +* Misc useful helper macros +*/ +#ifndef __throw_build_bug +# define __throw_build_bug () __builtin_trap() +#endif
this will become part of libbpf stable API, do we want/need to expose it? If we want to expose it, then we should probably provide a better description.
But also curious, how is it better than _Static_assert() (see test_cls_redirect.c), which also allows to provide a better error message?
tools/lib/bpf/bpf_helpers.h line 74
Quoted 13 lines
tools/lib/bpf/bpf_helpers.h +13
-0+ +static __always_inline void +bpf_tail_call_static (void *ctx,const void *map,const __u32 slot) +{ +if (!__builtin_constant_p (slot)) +__throw_build_bug (); + +/* +* Don't gamble, but _guarantee_ that LLVM won't optimize setting +* r2 and r3 from different paths ending up at the same call insn as +* otherwise we won't be able to use the jmpq/nopl retpoline-free +* patching by the x86-64 JIT in the kernel. +*
So the clobbering comment below is completely clear. But this one is less clear without some sort of example situation in which bad things happen. Do you mind providing some pseudo-C example in which the compiler will optimize things in such a way that the tail call patching won't happen?
Quoted 19 lines
tools/lib/bpf/bpf_helpers.h +13
-1+* Note on clobber list: we need to stay in-line with BPF calling +* convention, so even if we don't end up using r0, r4, r5, we need +* to mark them as clobber so that LLVM doesn't end up using them +* before / after the call. +*/ + asmvolatile ("r1 = %[ctx]\n\t" +"r2 = %[map]\n\t" +"r3 = %[slot]\n\t" +"call 12\n\t" + :: [ctx]"r" (ctx), [map]"r" (map), [slot]"i" (slot) + :"r0" ,"r1" ,"r2" ,"r3" ,"r4" ,"r5" ); +} +/* * Helper structure used by eBPF C program * to describe BPF map attributes to libbpf loader -- 2.21.0
Daniel Borkmann
maintainerRe: [PATCH bpf-next 2/6] bpf, net: rework cookie generator as per-cpu one
In reply to Eric DumazetQuoted 1 line · Eric Dumazet
On 9/24/20 8:21 PM, Daniel Borkmann wrote:
[...]
Quoted 47 lines
Quoted 34 lines
include/linux/cookie.h new file +28
-0@@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef __LINUX_COOKIE_H +#define __LINUX_COOKIE_H + +#include <linux/atomic.h> +#include <linux/percpu.h> + +struct gen_cookie { +u64 __percpu *local_last; +atomic64_t shared_last____cacheline_aligned_in_smp ; +}; + +#define COOKIE_LOCAL_BATCH 4096 + +#define DEFINE_COOKIE (name) \ + static DEFINE_PER_CPU(u64, __##name); \ + static struct gen_cookie name = { \ + .local_last = &__##name, \ + .shared_last = ATOMIC64_INIT(0), \ + } + +static inline u64 gen_cookie_next (struct gen_cookie *gc) +{ +u64 *local_last = &get_cpu_var (*gc->local_last); +u64 val = *local_last; + +if (__is_defined (CONFIG_SMP ) && +unlikely ((val & (COOKIE_LOCAL_BATCH -1 )) ==0 )) {include/linux/cookie.h line 38
Quoted 10 lines
include/linux/cookie.h new file +10
-0+s64 next =atomic64_add_return (COOKIE_LOCAL_BATCH , + &gc->shared_last); + val = next -COOKIE_LOCAL_BATCH ; + } + val++; +if (unlikely (!val)) + val++; + *local_last = val; +put_cpu_var (local_last); +return val;This is not interrupt safe.
Quoted 4 lines
I think
sock_gen_cookie()can be called from interrupt context.
get_next_ino()is only called from process context, that is what I usedget_cpu_var()andput_cpu_var()
Hmm, agree, good point. Need to experiment a bit more .. initial thinking potentially something like the below could do where we fall back to atomic counter iff we encounter nesting (which should be an extremely rare case normally).
BPF progs where this can be called from are non-preemptible, so we could actually move the temp preempt_disable/enable() from get/put_cpu_var() into a wrapper func for slow path non-BPF users as well.
static inline u64 gen_cookie_next(struct gen_cookie *gc)
{
u64 val;
if (likely(this_cpu_inc_return(*gc->level_nesting) == 1)) {
u64 *local_last = this_cpu_ptr(gc->local_last);
val = *local_last;
if (__is_defined(CONFIG_SMP) &&
unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0)) {
s64 next = atomic64_add_return(COOKIE_LOCAL_BATCH,
&gc->shared_last);
val = next - COOKIE_LOCAL_BATCH;
}
val++;
if (unlikely(!val))
val++;
*local_last = val;
} else {
val = atomic64_add_return(COOKIE_LOCAL_BATCH,
&gc->shared_last);
}
this_cpu_dec(*gc->level_nesting);
return val;
}Thanks,
Daniel
Daniel Borkmann
maintainerRe: [PATCH bpf-next 5/6] bpf, selftests: use bpf_tail_call_static where appropriate
In reply to Maciej FijalkowskiQuoted 16 lines · Maciej Fijalkowski
Quoted 4 lines · Daniel Borkmann
For those locations where we use an immediate tail call map index use the newly added
bpf_tail_call_static()helper.Signed-off-by Daniel Borkmann <[email protected]>Quoted 10 lines
--- tools/testing/selftests/bpf/progs/bpf_flow.c | 12 ++++---- tools/testing/selftests/bpf/progs/tailcall1.c | 28 +++++++++---------- tools/testing/selftests/bpf/progs/tailcall2.c | 14 +++++----- tools/testing/selftests/bpf/progs/tailcall3.c | 4 +-- .../selftests/bpf/progs/tailcall_bpf2bpf1.c | 4 +-- .../selftests/bpf/progs/tailcall_bpf2bpf2.c | 6 ++-- .../selftests/bpf/progs/tailcall_bpf2bpf3.c | 6 ++-- .../selftests/bpf/progs/tailcall_bpf2bpf4.c | 6 ++-- 8 files changed, 40 insertions(+), 40 deletions(-)
Quoted 2 lines
One nit, while you're at it, maybe it would be good to also address the
samples/bpf/sockex3_kern.cthat is also using the immediate map index?
Sure, np, will add.
Re: [PATCH bpf-next 3/6] bpf: add redirect_neigh helper as redirect drop-in
In reply to Daniel BorkmannQuoted 14 lines · Daniel Borkmann
net/core/filter.c +6
-0@@ -2160,6 +2160,205 @@ static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,return __bpf_redirect_no_mac (skb, dev, flags); } +#if IS_ENABLED (CONFIG_IPV6 ) +static int bpf_out_neigh_v6 (struct net *net,struct sk_buff *skb) +{ +struct dst_entry *dst =skb_dst (skb); +struct net_device *dev = dst->dev; +const struct in6_addr *nexthop;
net/core/filter.c static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, · line 2178
Quoted 10 lines
net/core/filter.c +10
-0+struct neighbour *neigh; + +if (dev_xmit_recursion ()) + goto out_rec; + skb->dev = dev; +rcu_read_lock_bh (); + nexthop =rt6_nexthop ((struct rt6_info *)dst, &ipv6_hdr (skb)->daddr); + neigh =__ipv6_neigh_lookup_noref_stub (dev, nexthop); +if (unlikely (!neigh)) + neigh =__neigh_create (ipv6_stub->nd_tbl, nexthop, dev, false);
the last 3 lines can be replaced with ip_neigh_gw6.
Quoted 21 lines
net/core/filter.c +21
-0+if (likely (!IS_ERR (neigh))) { +int ret; + +sock_confirm_neigh (skb, neigh); +dev_xmit_recursion_inc (); + ret =neigh_output (neigh, skb, false); +dev_xmit_recursion_dec (); +rcu_read_unlock_bh (); +return ret; + } +rcu_read_unlock_bh (); +IP6_INC_STATS (dev_net (dst->dev), +ip6_dst_idev (dst),IPSTATS_MIB_OUTNOROUTES ); +out_drop: +kfree_skb (skb); +return -EINVAL ; +out_rec: +net_crit_ratelimited ("bpf: recursion limit reached on datapath, buggy bpf program?\n" ); + goto out_drop; +} +
...
net/core/filter.c static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, · line 2249
Quoted 6 lines
net/core/filter.c +6
-0+ +#if IS_ENABLED (CONFIG_INET ) +static int bpf_out_neigh_v4 (struct net *net,struct sk_buff *skb) +{ +struct dst_entry *dst =skb_dst (skb); +struct rtable *rt = (struct rtable *)dst;
please use container_of here; I'd like to see the typecasts get removed.
Quoted 7 lines
net/core/filter.c +7
-0+struct net_device *dev = dst->dev; +u32 hh_len =LL_RESERVED_SPACE (dev); +struct neighbour *neigh; +bool is_v6gw = false; + +if (dev_xmit_recursion ()) + goto out_rec;
Daniel Borkmann
maintainerRe: [PATCH bpf-next 4/6] bpf, libbpf: add bpf_tail_call_static helper for bpf programs
In reply to Andrii NakryikoQuoted 42 lines · Andrii Nakryiko
Quoted 27 lines · Daniel Borkmann
Port of
tail_call_static()helper function from Cilium's BPF code base [0] to libbpf, so others can easily consume it as well. We've been using this in production code for some time now. The main idea is that we guarantee that the kernel's BPF infrastructure and JIT (here:x86_64) can patch the JITed BPF insns with direct jumps instead of having to fall back to using expensive retpolines. By using inline asm, we guarantee that the compiler won't merge the call from different paths with potentially different content of r2/r3.We're also using
__throw_build_bug()macro in different places as a neat trick to trigger compilation errors when compiler does not remove code at compilation time. This works for the BPF backend as it does not implement the__builtin_trap().[0] https://github.com/cilium/cilium/commit/f5537c26020d5297b70936c6b7d03a1e412a1035Signed-off-by Daniel Borkmann <[email protected]>--- tools/lib/bpf/bpf_helpers.h | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+)tools/lib/bpf/bpf_helpers.h +0
-0tools/lib/bpf/bpf_helpers.h line 61
Quoted 9 lines
tools/lib/bpf/bpf_helpers.h +6
-0@@ -53,6 +53,38 @@ })#endif +/* +* Misc useful helper macros +*/ +#ifndef __throw_build_bug +# define __throw_build_bug () __builtin_trap() +#endif this will become part of libbpf stable API, do we want/need to expose it? If we want to expose it, then we should probably provide a better description.
Quoted 3 lines
But also curious, how is it better than
_Static_assert()(seetest_cls_redirect.c), which also allows to provide a better error message?
Need to get back to you whether that has same semantics. We use the __throw_build_bug() also in __bpf_memzero() and friends [0] as a way to trigger a hard build bug if we hit a default switch-case [0], so we detect unsupported sizes which are not covered by the implementation yet. If _Static_assert (0, "foo") does the trick, we could also use that; will check with our code base.
[0] https://github.com/cilium/cilium/blob/master/bpf/include/bpf/builtins.h
Quoted 10 lines
Quoted 10 lines
tools/lib/bpf/bpf_helpers.h +10
-0+static __always_inline void +bpf_tail_call_static (void *ctx,const void *map,const __u32 slot) +{ +if (!__builtin_constant_p (slot)) +__throw_build_bug (); + +/* +* Don't gamble, but _guarantee_ that LLVM won't optimize setting +* r2 and r3 from different paths ending up at the same call insn as +* otherwise we won't be able to use the jmpq/nopl retpoline-free
Quoted 8 lines
tools/lib/bpf/bpf_helpers.h line 74
Quoted 2 lines
tools/lib/bpf/bpf_helpers.h +2
-0+* patching by the x86-64 JIT in the kernel. +* So the clobbering comment below is completely clear. But this one is less clear without some sort of example situation in which bad things happen. Do you mind providing some pseudo-C example in which the compiler will optimize things in such a way that the tail call patching won't happen?
The details are pretty much here [1] and mentioned at plumbers, so if we end up from different paths with different map or const key at the same tail-call call insn, then the record_func_key() will determine that we need to emit retpoline instead of patching which could have occured with two tail-call call insns, for example. This helper is just to guarantee that the latter will always happen.
Quoted 19 lines
Quoted 19 lines
tools/lib/bpf/bpf_helpers.h +13
-1+* Note on clobber list: we need to stay in-line with BPF calling +* convention, so even if we don't end up using r0, r4, r5, we need +* to mark them as clobber so that LLVM doesn't end up using them +* before / after the call. +*/ + asmvolatile ("r1 = %[ctx]\n\t" +"r2 = %[map]\n\t" +"r3 = %[slot]\n\t" +"call 12\n\t" + :: [ctx]"r" (ctx), [map]"r" (map), [slot]"i" (slot) + :"r0" ,"r1" ,"r2" ,"r3" ,"r4" ,"r5" ); +} +/* * Helper structure used by eBPF C program * to describe BPF map attributes to libbpf loader -- 2.21.0
Daniel Borkmann
maintainerRe: [PATCH bpf-next 3/6] bpf: add redirect_neigh helper as redirect drop-in
In reply to David AhernQuoted 19 lines · David Ahern
Quoted 18 lines · Daniel Borkmann
net/core/filter.c +10
-0@@ -2160,6 +2160,205 @@ static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,return __bpf_redirect_no_mac (skb, dev, flags); } +#if IS_ENABLED (CONFIG_IPV6 ) +static int bpf_out_neigh_v6 (struct net *net,struct sk_buff *skb) +{ +struct dst_entry *dst =skb_dst (skb); +struct net_device *dev = dst->dev; +const struct in6_addr *nexthop; +struct neighbour *neigh; + +if (dev_xmit_recursion ()) + goto out_rec;
Quoted 8 lines
net/core/filter.c static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, · line 2178
Quoted 6 lines
net/core/filter.c +6
-0+ skb->dev = dev; +rcu_read_lock_bh (); + nexthop =rt6_nexthop ((struct rt6_info *)dst, &ipv6_hdr (skb)->daddr); + neigh =__ipv6_neigh_lookup_noref_stub (dev, nexthop); +if (unlikely (!neigh)) + neigh =__neigh_create (ipv6_stub->nd_tbl, nexthop, dev, false);the last 3 lines can be replaced with
ip_neigh_gw6.
Ah, nice, I wasn't aware of that one. I'll take it. :)
Quoted 24 lines
Quoted 21 lines
net/core/filter.c +21
-0+if (likely (!IS_ERR (neigh))) { +int ret; + +sock_confirm_neigh (skb, neigh); +dev_xmit_recursion_inc (); + ret =neigh_output (neigh, skb, false); +dev_xmit_recursion_dec (); +rcu_read_unlock_bh (); +return ret; + } +rcu_read_unlock_bh (); +IP6_INC_STATS (dev_net (dst->dev), +ip6_dst_idev (dst),IPSTATS_MIB_OUTNOROUTES ); +out_drop: +kfree_skb (skb); +return -EINVAL ; +out_rec: +net_crit_ratelimited ("bpf: recursion limit reached on datapath, buggy bpf program?\n" ); + goto out_drop; +} +...
Quoted 7 lines
net/core/filter.c static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, · line 2249
Quoted 5 lines
net/core/filter.c +5
-0+#if IS_ENABLED (CONFIG_INET ) +static int bpf_out_neigh_v4 (struct net *net,struct sk_buff *skb) +{ +struct dst_entry *dst =skb_dst (skb); +struct rtable *rt = (struct rtable *)dst;please use
container_ofhere; I'd like to see the typecasts get removed.
Will do, thx!
Quoted 7 lines
Quoted 7 lines
net/core/filter.c +7
-0+struct net_device *dev = dst->dev; +u32 hh_len =LL_RESERVED_SPACE (dev); +struct neighbour *neigh; +bool is_v6gw = false; + +if (dev_xmit_recursion ()) + goto out_rec;
Re: [PATCH bpf-next 4/6] bpf, libbpf: add bpf_tail_call_static helper for bpf programs
In reply to Daniel BorkmannQuoted 26 lines · Daniel Borkmann
Port of
tail_call_static()helper function from Cilium's BPF code base [0] to libbpf, so others can easily consume it as well. We've been using this in production code for some time now. The main idea is that we guarantee that the kernel's BPF infrastructure and JIT (here:x86_64) can patch the JITed BPF insns with direct jumps instead of having to fall back to using expensive retpolines. By using inline asm, we guarantee that the compiler won't merge the call from different paths with potentially different content of r2/r3.We're also using
__throw_build_bug()macro in different places as a neat trick to trigger compilation errors when compiler does not remove code at compilation time. This works for the BPF backend as it does not implement the__builtin_trap().[0] https://github.com/cilium/cilium/commit/f5537c26020d5297b70936c6b7d03a1e412a1035Signed-off-by Daniel Borkmann <[email protected]>--- tools/lib/bpf/bpf_helpers.h | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+)tools/lib/bpf/bpf_helpers.h +0
-0
tools/lib/bpf/bpf_helpers.h line 60
Quoted 8 lines
tools/lib/bpf/bpf_helpers.h +5
-0@@ -53,6 +53,38 @@ })#endif +/* +* Misc useful helper macros +*/ +#ifndef __throw_build_bug +# define __throw_build_bug () __builtin_trap()
Just some general comments below. The patch itself is fine to me.
I guess we will never implement a 'trap' insn? The only possible use I know is for failed CORE relocation, which currently encoded as an illegal call insn.
Quoted 15 lines
tools/lib/bpf/bpf_helpers.h +15
-0+#endif + +static __always_inline void +bpf_tail_call_static (void *ctx,const void *map,const __u32 slot) +{ +if (!__builtin_constant_p (slot)) +__throw_build_bug (); + +/* +* Don't gamble, but _guarantee_ that LLVM won't optimize setting +* r2 and r3 from different paths ending up at the same call insn as +* otherwise we won't be able to use the jmpq/nopl retpoline-free +* patching by the x86-64 JIT in the kernel. +* +* Note on clobber list: we need to stay in-line with BPF calling
tools/lib/bpf/bpf_helpers.h line 85
Quoted 10 lines
tools/lib/bpf/bpf_helpers.h +10
-0+* convention, so even if we don't end up using r0, r4, r5, we need +* to mark them as clobber so that LLVM doesn't end up using them +* before / after the call. +*/ + asmvolatile ("r1 = %[ctx]\n\t" +"r2 = %[map]\n\t" +"r3 = %[slot]\n\t" +"call 12\n\t" + :: [ctx]"r" (ctx), [map]"r" (map), [slot]"i" (slot) + :"r0" ,"r1" ,"r2" ,"r3" ,"r4" ,"r5" );
Clever scheme to enforce a constant!
This particular case is to avoid sinking common code.
We have a similar use case in https://reviews.llvm.org/D87153
which is to avoid PHI merging of two relocation globals like
phi = [reloc_global1, reloc_global2] // reachable from two paths ... phi ...
The solution is to insert bpf internal __builtin
val = __builtin_bpf_passthrough(seq_num, val)
in proper places to prevent sinking common code.
I guess in this example, we could prevent this with compiler inserting passthrough builtin's like
...
ret = bpf_tail_call(ctx, map, slot)
ret = __builtin_bpf_passthrough(seq_num, ret)
...
If in the future, such a builtin is proved useful for bpf program writers as well. we could just define
val = __builtin_bpf_passthrough(val)
and internally, it will be transformed to
val = llvm.bpf.passthrough(seq_num, val)
Quoted 6 lines
tools/lib/bpf/bpf_helpers.h +2
-0+} +/* * Helper structure used by eBPF C program * to describe BPF map attributes to libbpf loader
Re: [PATCH bpf-next 2/6] bpf, net: rework cookie generator as per-cpu one
In reply to Daniel BorkmannQuoted 60 lines · Daniel Borkmann
Quoted 1 line · Eric Dumazet
On 9/24/20 8:21 PM, Daniel Borkmann wrote:
[...]
Quoted 47 lines
Quoted 34 lines
include/linux/cookie.h new file +28
-0@@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef __LINUX_COOKIE_H +#define __LINUX_COOKIE_H + +#include <linux/atomic.h> +#include <linux/percpu.h> + +struct gen_cookie { + u64 __percpu *local_last; + atomic64_t shared_last ____cacheline_aligned_in_smp; +}; + +#define COOKIE_LOCAL_BATCH 4096 + +#define DEFINE_COOKIE (name) \ + static DEFINE_PER_CPU(u64, __##name); \ + static struct gen_cookie name = { \ + .local_last = &__##name, \ + .shared_last = ATOMIC64_INIT(0), \ + } + +static inline u64 gen_cookie_next (struct gen_cookie *gc) +{ + u64 *local_last = &get_cpu_var(*gc->local_last); + u64 val = *local_last; + + if (__is_defined(CONFIG_SMP) && + unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0)) {include/linux/cookie.h line 38
Quoted 10 lines
include/linux/cookie.h new file +10
-0+ s64 next = atomic64_add_return(COOKIE_LOCAL_BATCH, + &gc->shared_last); + val = next - COOKIE_LOCAL_BATCH; + } + val++; + if (unlikely(!val)) + val++; + *local_last = val; + put_cpu_var(local_last); + return val;This is not interrupt safe.
Quoted 4 lines
I think
sock_gen_cookie()can be called from interrupt context.
get_next_ino()is only called from process context, that is what I usedget_cpu_var()andput_cpu_var()Hmm, agree, good point. Need to experiment a bit more .. initial thinking potentially something like the below could do where we fall back to atomic counter iff we encounter nesting (which should be an extremely rare case normally).
Quoted 8 lines
BPF progs where this can be called from are non-preemptible, so we could actually move the temp preempt_disable/enable() from get/put_cpu_var() into a wrapper func for slow path non-BPF users as well.
static inline u64
gen_cookie_next(struct gen_cookie *gc){
u64 val;
I presume you would use a single structure to hold level_nesting and local_last in the same cache line.
struct pcpu_gen_cookie {
int level_nesting;
u64 local_last;
} __aligned(16);Quoted 14 lines
if (
likely(this_cpu_inc_return(*gc->level_nesting) == 1)) { u64*local_last=this_cpu_ptr(gc->local_last);val =
*local_last;
if (__is_defined(CONFIG_SMP)&&
unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0)) { s64 next =atomic64_add_return(COOKIE_LOCAL_BATCH, &gc->shared_last); val = next -COOKIE_LOCAL_BATCH;
}
val++;
Quoted 2 lines
if (unlikely(!val))
val++;
Note that we really expect this wrapping will never happen, with 64bit value. (We had to take care of the wrapping in get_next_ino() as it was dealing with 32bit values)
Quoted 4 lines
*local_last= val;
} else {
val =atomic64_add_return(COOKIE_LOCAL_BATCH, &gc->shared_last);
Or val = atomic64_dec_return(&reverse_counter)
With reverse_counter initial value set to ATOMIC64_INIT(0) ?
This will start sending 'big cookies like 0xFFFFFFFFxxxxxxxx' to make sure applications are not breaking with them, after few months of uptime.
This would also not consume COOKIE_LOCAL_BATCH units per value, but this seems minor based on the available space.
Quoted 7 lines
}
this_cpu_dec(*gc->level_nesting);
return val;
}Thanks,
Daniel
Daniel Borkmann
maintainerRe: [PATCH bpf-next 2/6] bpf, net: rework cookie generator as per-cpu one
Quoted 69 lines · Eric Dumazet
Quoted 60 lines · Daniel Borkmann
Quoted 1 line · Eric Dumazet
On 9/24/20 8:21 PM, Daniel Borkmann wrote:
[...]
Quoted 47 lines
Quoted 34 lines
include/linux/cookie.h new file +28
-0@@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef __LINUX_COOKIE_H +#define __LINUX_COOKIE_H + +#include <linux/atomic.h> +#include <linux/percpu.h> + +struct gen_cookie { + u64 __percpu *local_last; + atomic64_t shared_last ____cacheline_aligned_in_smp; +}; + +#define COOKIE_LOCAL_BATCH 4096 + +#define DEFINE_COOKIE (name) \ + static DEFINE_PER_CPU(u64, __##name); \ + static struct gen_cookie name = { \ + .local_last = &__##name, \ + .shared_last = ATOMIC64_INIT(0), \ + } + +static inline u64 gen_cookie_next (struct gen_cookie *gc) +{ + u64 *local_last = &get_cpu_var(*gc->local_last); + u64 val = *local_last; + + if (__is_defined(CONFIG_SMP) && + unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0)) {include/linux/cookie.h line 38
Quoted 10 lines
include/linux/cookie.h new file +10
-0+ s64 next = atomic64_add_return(COOKIE_LOCAL_BATCH, + &gc->shared_last); + val = next - COOKIE_LOCAL_BATCH; + } + val++; + if (unlikely(!val)) + val++; + *local_last = val; + put_cpu_var(local_last); + return val;This is not interrupt safe.
Quoted 4 lines
I think
sock_gen_cookie()can be called from interrupt context.
get_next_ino()is only called from process context, that is what I usedget_cpu_var()andput_cpu_var()Hmm, agree, good point. Need to experiment a bit more .. initial thinking potentially something like the below could do where we fall back to atomic counter iff we encounter nesting (which should be an extremely rare case normally).
Quoted 7 lines
BPF progs where this can be called from are non-preemptible, so we could actually move the temp preempt_disable/enable() from get/put_cpu_var() into a wrapper func for slow path non-BPF users as well.
static inline u64
gen_cookie_next(struct gen_cookie *gc){
u64 val;
Quoted 7 lines
I presume you would use a single structure to hold
level_nestingandlocal_lastin the same cache line.struct pcpu_gen_cookie { int level_nesting; u64 local_last; } __aligned(16);
Yes.
Quoted 12 lines
Quoted 11 lines
if (
likely(this_cpu_inc_return(*gc->level_nesting) == 1)) { u64*local_last=this_cpu_ptr(gc->local_last);val =
*local_last;
if (__is_defined(CONFIG_SMP)&&
unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0)) { s64 next =atomic64_add_return(COOKIE_LOCAL_BATCH, &gc->shared_last); val = next -COOKIE_LOCAL_BATCH;
}
val++;
Quoted 5 lines
Quoted 2 lines
if (unlikely(!val))
val++;Note that we really expect this wrapping will never happen, with 64bit value. (We had to take care of the wrapping in
get_next_ino()as it was dealing with 32bit values)
Agree, all local counters will start off at 0, but we inc right after the batch and thus never run into it anyway and neither via overflow. Will remove.
Quoted 7 lines
Quoted 4 lines
*local_last= val;
} else {
val =atomic64_add_return(COOKIE_LOCAL_BATCH, &gc->shared_last);Or val =
atomic64_dec_return(&reverse_counter)
Quoted 7 lines
With
reverse_counterinitial value set toATOMIC64_INIT(0)?This will start sending 'big cookies like 0xFFFFFFFFxxxxxxxx' to make sure applications are not breaking with them, after few months of uptime.
This would also not consume
COOKIE_LOCAL_BATCHunits per value, but this seems minor based on the available space.
Excellent idea, I like it given it doesn't waste COOKIE_LOCAL_BATCH space. Thanks for the feedback!
Quoted 7 lines
Quoted 7 lines
}
this_cpu_dec(*gc->level_nesting);
return val;
}Thanks,
Daniel
Jakub Kicinski
maintainerRe: [PATCH bpf-next 1/6] bpf: add classid helper only based on skb->sk
In reply to Daniel BorkmannQuoted 11 lines · On Thu, 24 Sep 2020 20:21:22 +0200 Daniel Borkmann wrote:
Similarly to
5a52ae4e32a6("bpf: Allow to retrieve cgroup v1 classid from v2 hooks"), add a helper to retrieve cgroup v1 classid solely based on theskb->sk, so it can be used as key as part of BPF map lookups out of tc from host ns, in particular given theskb->skis retained these days when crossing net ns thanks to9c4c325252c5("skbuff: preserve sock reference when scrubbing the skb."). This is similar tobpf_skb_cgroup_id()which implements the same for v2. Kubernetes ecosystem is still operating on v1 however, hencenet_clsneeds to be used there until this can be dropped in with the v2 helper ofbpf_skb_cgroup_id().
Quoted 1 line
Signed-off-by Daniel Borkmann <[email protected]>
FWIW lot's of whitespace warnings from checkpatch --strict about comments having spaces before tabs here.
Jakub Kicinski
maintainerRe: [PATCH bpf-next 2/6] bpf, net: rework cookie generator as per-cpu one
In reply to Daniel BorkmannQuoted 5 lines · On Fri, 25 Sep 2020 00:03:14 +0200 Daniel Borkmann wrote:
static inline u64
gen_cookie_next(struct gen_cookie *gc){ u64 val; if (likely(this_cpu_inc_return(*gc->level_nesting) == 1)) {
Is this_cpu_inc() in itself atomic?
Is there a comparison of performance of various atomic ops and locking somewhere? I wonder how this scheme would compare to a using a cmpxchg.
Quoted 5 lines
u64 *local_last = this_cpu_ptr(gc->local_last);
val = *local_last;
if (__is_defined(CONFIG_SMP) &&
unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0)) {Can we reasonably assume we won't have more than 4k CPUs and just statically divide this space by encoding CPU id in top bits?
Quoted 15 lines
s64 next = atomic64_add_return(COOKIE_LOCAL_BATCH,
&gc->shared_last);
val = next - COOKIE_LOCAL_BATCH;
}
val++;
if (unlikely(!val))
val++;
*local_last = val;
} else {
val = atomic64_add_return(COOKIE_LOCAL_BATCH,
&gc->shared_last);
}
this_cpu_dec(*gc->level_nesting);
return val;
}Re: [PATCH bpf-next 2/6] bpf, net: rework cookie generator as per-cpu one
On 9/25/20 5:00 PM, Jakub Kicinski wrote:
unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0)) {Quoted 3 lines
Can we reasonably assume we won't have more than 4k CPUs and just statically divide this space by encoding CPU id in top bits?
This might give some food to side channel attacks, since this would give an indication of cpu that allocated the id.
Also, I hear that some distros enabled 8K cpus.
Jakub Kicinski
maintainerRe: [PATCH bpf-next 2/6] bpf, net: rework cookie generator as per-cpu one
Quoted 2 lines · On Fri, 25 Sep 2020 17:15:17 +0200 Eric Dumazet wrote:
Quoted 1 line · Jakub Kicinski
Is
this_cpu_inc()in itself atomic?
To answer my own question - it is :)
Quoted 9 lines
Quoted 4 lines
unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0))Can we reasonably assume we won't have more than 4k CPUs and just statically divide this space by encoding CPU id in top bits?
This might give some food to side channel attacks, since this would give an indication of cpu that allocated the id.
Also, I hear that some distros enabled 8K cpus.
Ok :(
Daniel Borkmann
maintainerRe: [PATCH bpf-next 1/6] bpf: add classid helper only based on skb->sk
In reply to Jakub KicinskiQuoted 12 lines · Jakub Kicinski
Quoted 11 lines · On Thu, 24 Sep 2020 20:21:22 +0200 Daniel Borkmann wrote:
Similarly to
5a52ae4e32a6("bpf: Allow to retrieve cgroup v1 classid from v2 hooks"), add a helper to retrieve cgroup v1 classid solely based on theskb->sk, so it can be used as key as part of BPF map lookups out of tc from host ns, in particular given theskb->skis retained these days when crossing net ns thanks to9c4c325252c5("skbuff: preserve sock reference when scrubbing the skb."). This is similar tobpf_skb_cgroup_id()which implements the same for v2. Kubernetes ecosystem is still operating on v1 however, hencenet_clsneeds to be used there until this can be dropped in with the v2 helper ofbpf_skb_cgroup_id().
Quoted 4 lines
Quoted 1 line
Signed-off-by Daniel Borkmann <[email protected]>FWIW lot's of whitespace warnings from checkpatch --strict about comments having spaces before tabs here.
Expected given the way the UAPI helper comment is formatted / done in order to then render the man page. So it's formatted the same way as the other helper descriptions.
Thanks,
Daniel
Daniel Borkmann
maintainerRe: [PATCH bpf-next 4/6] bpf, libbpf: add bpf_tail_call_static helper for bpf programs
In reply to Daniel BorkmannQuoted 47 lines · Daniel Borkmann
Quoted 42 lines · Andrii Nakryiko
Quoted 27 lines · Daniel Borkmann
Port of
tail_call_static()helper function from Cilium's BPF code base [0] to libbpf, so others can easily consume it as well. We've been using this in production code for some time now. The main idea is that we guarantee that the kernel's BPF infrastructure and JIT (here:x86_64) can patch the JITed BPF insns with direct jumps instead of having to fall back to using expensive retpolines. By using inline asm, we guarantee that the compiler won't merge the call from different paths with potentially different content of r2/r3.We're also using
__throw_build_bug()macro in different places as a neat trick to trigger compilation errors when compiler does not remove code at compilation time. This works for the BPF backend as it does not implement the__builtin_trap().[0] https://github.com/cilium/cilium/commit/f5537c26020d5297b70936c6b7d03a1e412a1035
Signed-off-by Daniel Borkmann <[email protected]>---
tools/lib/bpf/bpf_helpers.h| 32 ++++++++++++++++++++++++++++++++1 file changed, 32 insertions(+)tools/lib/bpf/bpf_helpers.h +0
-0tools/lib/bpf/bpf_helpers.h line 61
Quoted 9 lines
tools/lib/bpf/bpf_helpers.h +6
-0@@ -53,6 +53,38 @@ }) #endif +/* +* Misc useful helper macros +*/ +#ifndef __throw_build_bug +# define __throw_build_bug () __builtin_trap() +#endif this will become part of libbpf stable API, do we want/need to expose it? If we want to expose it, then we should probably provide a better description.
Quoted 3 lines
But also curious, how is it better than
_Static_assert()(seetest_cls_redirect.c), which also allows to provide a better error message?
Quoted 5 lines
Need to get back to you whether that has same semantics. We use the
__throw_build_bug()also in__bpf_memzero()and friends [0] as a way to trigger a hard build bug if we hit a default switch-case [0], so we detect unsupported sizes which are not covered by the implementation yet. If_Static_assert(0, "foo") does the trick, we could also use that; will check with our code base.
So _Static_assert() won't work here, for example consider:
# cat f1.c
int main(void)
{
if (0)
_Static_assert(0, "foo");
return 0;
}
# clang -target bpf -Wall -O2 -c f1.c -o f1.o
f1.c:4:3: error: expected expression
_Static_assert(0, "foo");
^
1 error generated.In order for it to work as required form the use-case, the _Static_assert() must not trigger here given the path is unreachable and will be optimized away. I'll add a comment to the __throw_build_bug() helper. Given libbpf we should probably also prefix with bpf_. If you see a better name that would fit, pls let me know.
Quoted 1 line
[0] https://github.com/cilium/cilium/blob/master/bpf/include/bpf/builtins.h
Thanks,
Daniel
Re: [PATCH bpf-next 2/6] bpf, net: rework cookie generator as per-cpu one
In reply to Jakub KicinskiQuoted 11 lines · Jakub Kicinski
Quoted 2 lines · On Fri, 25 Sep 2020 17:15:17 +0200 Eric Dumazet wrote:
Quoted 1 line · Jakub Kicinski
Is
this_cpu_inc()in itself atomic?To answer my own question - it is :)
Quoted 5 lines
Quoted 4 lines
unlikely((val & (COOKIE_LOCAL_BATCH - 1)) == 0))Can we reasonably assume we won't have more than 4k CPUs and just statically divide this space by encoding CPU id in top bits?
Quoted 7 lines
Quoted 4 lines
This might give some food to side channel attacks, since this would give an indication of cpu that allocated the id.
Also, I hear that some distros enabled 8K cpus.
Ok :(
I was not really serious about the side channel attacks, just some thought about possible implications :)
Even with 8192 max cpus, splitting space into 2^(64-13) blocks would be fine I think.
Daniel Borkmann
maintainerRe: [PATCH bpf-next 4/6] bpf, libbpf: add bpf_tail_call_static helper for bpf programs
In reply to Daniel BorkmannQuoted 56 lines · Daniel Borkmann
Quoted 47 lines · Daniel Borkmann
Quoted 42 lines · Andrii Nakryiko
Quoted 27 lines · Daniel Borkmann
Port of
tail_call_static()helper function from Cilium's BPF code base [0] to libbpf, so others can easily consume it as well. We've been using this in production code for some time now. The main idea is that we guarantee that the kernel's BPF infrastructure and JIT (here:x86_64) can patch the JITed BPF insns with direct jumps instead of having to fall back to using expensive retpolines. By using inline asm, we guarantee that the compiler won't merge the call from different paths with potentially different content of r2/r3.We're also using
__throw_build_bug()macro in different places as a neat trick to trigger compilation errors when compiler does not remove code at compilation time. This works for the BPF backend as it does not implement the__builtin_trap().[0] https://github.com/cilium/cilium/commit/f5537c26020d5297b70936c6b7d03a1e412a1035
Signed-off-by Daniel Borkmann <[email protected]>---
tools/lib/bpf/bpf_helpers.h| 32 ++++++++++++++++++++++++++++++++1 file changed, 32 insertions(+)tools/lib/bpf/bpf_helpers.h +0
-0tools/lib/bpf/bpf_helpers.h line 61
Quoted 9 lines
tools/lib/bpf/bpf_helpers.h +6
-0@@ -53,6 +53,38 @@ }) #endif +/* +* Misc useful helper macros +*/ +#ifndef __throw_build_bug +# define __throw_build_bug () __builtin_trap() +#endif this will become part of libbpf stable API, do we want/need to expose it? If we want to expose it, then we should probably provide a better description.
Quoted 3 lines
But also curious, how is it better than
_Static_assert()(seetest_cls_redirect.c), which also allows to provide a better error message?Quoted 5 lines
Need to get back to you whether that has same semantics. We use the
__throw_build_bug()also in__bpf_memzero()and friends [0] as a way to trigger a hard build bug if we hit a default switch-case [0], so we detect unsupported sizes which are not covered by the implementation yet. If_Static_assert(0, "foo") does the trick, we could also use that; will check with our code base.So
_Static_assert()won't work here, for example consider:
Quoted 12 lines
# cat f1.c
int main(void)
{
if (0)
_Static_assert(0, "foo");
return 0;
}
# clang -target bpf -Wall -O2 -c f1.c -o f1.o
f1.c:4:3: error: expected expression
_Static_assert(0, "foo");
^
1 error generated... aaand it looks like I need some more coffee. ;-) But result is the same after all:
# clang -target bpf -Wall -O2 -c f1.c -o f1.o
f1.c:4:3: error: static_assert failed "foo"
_Static_assert(0, "foo");
^ ~
1 error generated. # cat f1.c
int main(void)
{
if (0) {
_Static_assert(0, "foo");
}
return 0;
}Quoted 8 lines
In order for it to work as required form the use-case, the
_Static_assert()must not trigger here given the path is unreachable and will be optimized away. I'll add a comment to the__throw_build_bug()helper. Given libbpf we should probably also prefix withbpf_. If you see a better name that would fit, pls let me know.Quoted 1 line
[0] https://github.com/cilium/cilium/blob/master/bpf/include/bpf/builtins.h
Thanks,
Daniel
Re: [PATCH bpf-next 4/6] bpf, libbpf: add bpf_tail_call_static helper for bpf programs
Quoted 79 lines · Daniel Borkmann
Quoted 56 lines · Daniel Borkmann
Quoted 47 lines · Daniel Borkmann
Quoted 42 lines · Andrii Nakryiko
Quoted 27 lines · Daniel Borkmann
Port of
tail_call_static()helper function from Cilium's BPF code base [0] to libbpf, so others can easily consume it as well. We've been using this in production code for some time now. The main idea is that we guarantee that the kernel's BPF infrastructure and JIT (here:x86_64) can patch the JITed BPF insns with direct jumps instead of having to fall back to using expensive retpolines. By using inline asm, we guarantee that the compiler won't merge the call from different paths with potentially different content of r2/r3.We're also using
__throw_build_bug()macro in different places as a neat trick to trigger compilation errors when compiler does not remove code at compilation time. This works for the BPF backend as it does not implement the__builtin_trap().[0] https://github.com/cilium/cilium/commit/f5537c26020d5297b70936c6b7d03a1e412a1035Signed-off-by Daniel Borkmann <[email protected]>--- tools/lib/bpf/bpf_helpers.h | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+)tools/lib/bpf/bpf_helpers.h +0
-0tools/lib/bpf/bpf_helpers.h line 61
Quoted 9 lines
tools/lib/bpf/bpf_helpers.h +6
-0@@ -53,6 +53,38 @@ })#endif +/* +* Misc useful helper macros +*/ +#ifndef __throw_build_bug +# define __throw_build_bug () __builtin_trap() +#endif this will become part of libbpf stable API, do we want/need to expose it? If we want to expose it, then we should probably provide a better description.
Quoted 3 lines
But also curious, how is it better than
_Static_assert()(seetest_cls_redirect.c), which also allows to provide a better error message?Quoted 5 lines
Need to get back to you whether that has same semantics. We use the
__throw_build_bug()also in__bpf_memzero()and friends [0] as a way to trigger a hard build bug if we hit a default switch-case [0], so we detect unsupported sizes which are not covered by the implementation yet. If_Static_assert(0, "foo") does the trick, we could also use that; will check with our code base.So
_Static_assert()won't work here, for example consider:Quoted 12 lines
# cat f1.c int main(void) { if (0) _Static_assert(0, "foo"); return 0; } # clang -target bpf -Wall -O2 -c f1.c -o f1.o f1.c:4:3: error: expected expression _Static_assert(0, "foo"); ^ 1 error generated... aaand it looks like I need some more coffee. ;-) But result is the same after all:
# clang -target bpf -Wall -O2 -c f1.c -o f1.o f1.c:4:3: error: static_assert failed "foo" _Static_assert(0, "foo"); ^ ~ 1 error generated.
Quoted 8 lines
# cat f1.c
int main(void)
{
if (0) {
_Static_assert(0, "foo");
}
return 0;
}You need still more :-P. For you use case it will look like this:
$ cat test-bla.c
int bar(int x) {
_Static_assert(!__builtin_constant_p(x), "not a constant!");
return x;
}int foo() {
bar(123);
return 0;
}
$ clang -target bpf -O2 -c test-bla.c -o test-bla.o
$ echo $?
0But in general to ensure unreachable code it's probably useful anyway to have this. How about calling it __bpf_build_error() or maybe even __bpf_unreachable()?
Quoted 10 lines
Quoted 9 lines
In order for it to work as required form the use-case, the
_Static_assert()must not trigger here given the path is unreachable and will be optimized away. I'll add a comment to the__throw_build_bug()helper. Given libbpf we should probably also prefix withbpf_. If you see a better name that would fit, pls let me know.Quoted 1 line
[0] https://github.com/cilium/cilium/blob/master/bpf/include/bpf/builtins.h
Thanks,
Daniel
Daniel Borkmann
maintainerRe: [PATCH bpf-next 4/6] bpf, libbpf: add bpf_tail_call_static helper for bpf programs
Quoted 96 lines · Andrii Nakryiko
Quoted 78 lines · Daniel Borkmann
Quoted 56 lines · Daniel Borkmann
Quoted 47 lines · Daniel Borkmann
Quoted 42 lines · Andrii Nakryiko
Quoted 27 lines · Daniel Borkmann
Port of
tail_call_static()helper function from Cilium's BPF code base [0] to libbpf, so others can easily consume it as well. We've been using this in production code for some time now. The main idea is that we guarantee that the kernel's BPF infrastructure and JIT (here:x86_64) can patch the JITed BPF insns with direct jumps instead of having to fall back to using expensive retpolines. By using inline asm, we guarantee that the compiler won't merge the call from different paths with potentially different content of r2/r3.We're also using
__throw_build_bug()macro in different places as a neat trick to trigger compilation errors when compiler does not remove code at compilation time. This works for the BPF backend as it does not implement the__builtin_trap().[0] https://github.com/cilium/cilium/commit/f5537c26020d5297b70936c6b7d03a1e412a1035Signed-off-by Daniel Borkmann <[email protected]>--- tools/lib/bpf/bpf_helpers.h | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+)tools/lib/bpf/bpf_helpers.h +0
-0tools/lib/bpf/bpf_helpers.h line 61
Quoted 9 lines
tools/lib/bpf/bpf_helpers.h +6
-0@@ -53,6 +53,38 @@ })#endif +/* +* Misc useful helper macros +*/ +#ifndef __throw_build_bug +# define __throw_build_bug () __builtin_trap() +#endif this will become part of libbpf stable API, do we want/need to expose it? If we want to expose it, then we should probably provide a better description.
Quoted 3 lines
But also curious, how is it better than
_Static_assert()(seetest_cls_redirect.c), which also allows to provide a better error message?Quoted 5 lines
Need to get back to you whether that has same semantics. We use the
__throw_build_bug()also in__bpf_memzero()and friends [0] as a way to trigger a hard build bug if we hit a default switch-case [0], so we detect unsupported sizes which are not covered by the implementation yet. If_Static_assert(0, "foo") does the trick, we could also use that; will check with our code base.So
_Static_assert()won't work here, for example consider:Quoted 12 lines
# cat f1.c int main(void) { if (0) _Static_assert(0, "foo"); return 0; } # clang -target bpf -Wall -O2 -c f1.c -o f1.o f1.c:4:3: error: expected expression _Static_assert(0, "foo"); ^ 1 error generated... aaand it looks like I need some more coffee. ;-) But result is the same after all:
# clang -target bpf -Wall -O2 -c f1.c -o f1.o f1.c:4:3: error: static_assert failed "foo" _Static_assert(0, "foo"); ^ ~ 1 error generated.Quoted 8 lines
# cat f1.c int main(void) { if (0) { _Static_assert(0, "foo"); } return 0; }You need still more :-P. For you use case it will look like this:
$ cat test-bla.c int bar(int x) { _Static_assert(!__builtin_constant_p(x), "not a constant!"); return x; }
Quoted 7 lines
int foo() {
bar(123);
return 0;
}
$ clang -target bpf -O2 -c test-bla.c -o test-bla.o
$ echo $?
0Right, but that won't work for example for the use case to detect switch cases which fall into default case as mentioned with the mem* optimizations earlier in this thread.
Quoted 3 lines
But in general to ensure unreachable code it's probably useful anyway to have this. How about calling it
__bpf_build_error()or maybe even__bpf_unreachable()?
I think the __bpf_unreachable() sounds best to me, will use that.
Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
In order for it to work as required form the use-case, the
_Static_assert()must not trigger here given the path is unreachable and will be optimized away. I'll add a comment to the__throw_build_bug()helper. Given libbpf we should probably also prefix withbpf_. If you see a better name that would fit, pls let me know.Quoted 1 line
[0] https://github.com/cilium/cilium/blob/master/bpf/include/bpf/builtins.h
Thanks,
Daniel