From mboxrd@z Thu Jan 1 00:00:00 1970 From: Roman Gushchin Subject: Re: [PATCH v3 bpf-next 10/10] selftests/bpf: cgroup local storage-based network counters Date: Fri, 28 Sep 2018 11:08:29 +0100 Message-ID: <20180928100817.GC9018@castle.DHCP.thefacebook.com> References: <20180926113326.29069-1-guro@fb.com> <20180926113326.29069-11-guro@fb.com> <20180928085356.56xe7javtd6cdfz6@ast-mbp.dhcp.thefacebook.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Cc: , Song Liu , , , Daniel Borkmann , Alexei Starovoitov To: Alexei Starovoitov Return-path: Content-Disposition: inline In-Reply-To: <20180928085356.56xe7javtd6cdfz6@ast-mbp.dhcp.thefacebook.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org On Fri, Sep 28, 2018 at 10:53:58AM +0200, Alexei Starovoitov wrote: > On Wed, Sep 26, 2018 at 12:33:26PM +0100, Roman Gushchin wrote: > > This commit adds a bpf kselftest, which demonstrates how percpu > > and shared cgroup local storage can be used for efficient lookup-free > > network accounting. > > > > Cgroup local storage provides generic memory area with a very efficient > > lookup free access. To avoid expensive atomic operations for each > > packet, per-cpu cgroup local storage is used. Each packet is initially > > charged to a per-cpu counter, and only if the counter reaches certain > > value (32 in this case), the charge is moved into the global atomic > > counter. This allows to amortize atomic operations, keeping reasonable > > accuracy. > > > > The test also implements a naive network traffic throttling, mostly to > > demonstrate the possibility of bpf cgroup--based network bandwidth > > control. > > > > Expected output: > > ./test_netcnt > > test_netcnt:PASS > > > > Signed-off-by: Roman Gushchin > > Acked-by: Song Liu > > Cc: Daniel Borkmann > > Cc: Alexei Starovoitov > > --- > > tools/testing/selftests/bpf/Makefile | 6 +- > > tools/testing/selftests/bpf/netcnt_common.h | 23 +++ > > tools/testing/selftests/bpf/netcnt_prog.c | 71 +++++++++ > > tools/testing/selftests/bpf/test_netcnt.c | 153 ++++++++++++++++++++ > > 4 files changed, 251 insertions(+), 2 deletions(-) > > create mode 100644 tools/testing/selftests/bpf/netcnt_common.h > > create mode 100644 tools/testing/selftests/bpf/netcnt_prog.c > > create mode 100644 tools/testing/selftests/bpf/test_netcnt.c > > > > diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile > > index fd3851d5c079..5443399dd3a1 100644 > > --- a/tools/testing/selftests/bpf/Makefile > > +++ b/tools/testing/selftests/bpf/Makefile > > @@ -23,7 +23,8 @@ $(TEST_CUSTOM_PROGS): $(OUTPUT)/%: %.c > > TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test_progs \ > > test_align test_verifier_log test_dev_cgroup test_tcpbpf_user \ > > test_sock test_btf test_sockmap test_lirc_mode2_user get_cgroup_id_user \ > > - test_socket_cookie test_cgroup_storage test_select_reuseport > > + test_socket_cookie test_cgroup_storage test_select_reuseport \ > > + test_netcnt > > > > TEST_GEN_FILES = test_pkt_access.o test_xdp.o test_l4lb.o test_tcp_estats.o test_obj_id.o \ > > test_pkt_md_access.o test_xdp_redirect.o test_xdp_meta.o sockmap_parse_prog.o \ > > @@ -35,7 +36,7 @@ TEST_GEN_FILES = test_pkt_access.o test_xdp.o test_l4lb.o test_tcp_estats.o test > > test_get_stack_rawtp.o test_sockmap_kern.o test_sockhash_kern.o \ > > test_lwt_seg6local.o sendmsg4_prog.o sendmsg6_prog.o test_lirc_mode2_kern.o \ > > get_cgroup_id_kern.o socket_cookie_prog.o test_select_reuseport_kern.o \ > > - test_skb_cgroup_id_kern.o bpf_flow.o > > + test_skb_cgroup_id_kern.o bpf_flow.o netcnt_prog.o > > > > # Order correspond to 'make run_tests' order > > TEST_PROGS := test_kmod.sh \ > > @@ -72,6 +73,7 @@ $(OUTPUT)/test_tcpbpf_user: cgroup_helpers.c > > $(OUTPUT)/test_progs: trace_helpers.c > > $(OUTPUT)/get_cgroup_id_user: cgroup_helpers.c > > $(OUTPUT)/test_cgroup_storage: cgroup_helpers.c > > +$(OUTPUT)/test_netcnt: cgroup_helpers.c > > > > .PHONY: force > > > > diff --git a/tools/testing/selftests/bpf/netcnt_common.h b/tools/testing/selftests/bpf/netcnt_common.h > > new file mode 100644 > > index 000000000000..0e10fc276c2a > > --- /dev/null > > +++ b/tools/testing/selftests/bpf/netcnt_common.h > > @@ -0,0 +1,23 @@ > > +#ifndef __NETCNT_COMMON_H > > +#define __NETCNT_COMMON_H > > + > > +#include > > + > > +#define MAX_PERCPU_PACKETS 32 > > + > > +struct percpu_net_cnt { > > + __u64 packets; > > + __u64 bytes; > > + > > + __u64 prev_ts; > > + > > + __u64 prev_packets; > > + __u64 prev_bytes; > > +}; > > + > > +struct net_cnt { > > + __u64 packets; > > + __u64 bytes; > > +}; > > + > > +#endif > > diff --git a/tools/testing/selftests/bpf/netcnt_prog.c b/tools/testing/selftests/bpf/netcnt_prog.c > > new file mode 100644 > > index 000000000000..1198abca1360 > > --- /dev/null > > +++ b/tools/testing/selftests/bpf/netcnt_prog.c > > @@ -0,0 +1,71 @@ > > +// SPDX-License-Identifier: GPL-2.0 > > +#include > > +#include > > + > > +#include "bpf_helpers.h" > > +#include "netcnt_common.h" > > + > > +#define MAX_BPS (3 * 1024 * 1024) > > + > > +#define REFRESH_TIME_NS 100000000 > > +#define NS_PER_SEC 1000000000 > > + > > +struct bpf_map_def SEC("maps") percpu_netcnt = { > > + .type = BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE, > > + .key_size = sizeof(struct bpf_cgroup_storage_key), > > + .value_size = sizeof(struct percpu_net_cnt), > > +}; > > + > > +struct bpf_map_def SEC("maps") netcnt = { > > + .type = BPF_MAP_TYPE_CGROUP_STORAGE, > > + .key_size = sizeof(struct bpf_cgroup_storage_key), > > + .value_size = sizeof(struct net_cnt), > > +}; > > + > > +SEC("cgroup/skb") > > +int bpf_nextcnt(struct __sk_buff *skb) > > +{ > > + struct percpu_net_cnt *percpu_cnt; > > + char fmt[] = "%d %llu %llu\n"; > > + struct net_cnt *cnt; > > + __u64 ts, dt; > > + int ret; > > + > > + cnt = bpf_get_local_storage(&netcnt, 0); > > + percpu_cnt = bpf_get_local_storage(&percpu_netcnt, 0); > > + > > + percpu_cnt->packets++; > > + percpu_cnt->bytes += skb->len; > > + > > + if (percpu_cnt->packets > MAX_PERCPU_PACKETS) { > > + __sync_fetch_and_add(&cnt->packets, > > + percpu_cnt->packets); > > + percpu_cnt->packets = 0; > > + > > + __sync_fetch_and_add(&cnt->bytes, > > + percpu_cnt->bytes); > > + percpu_cnt->bytes = 0; > > + } > > + > > + ts = bpf_ktime_get_ns(); > > + dt = ts - percpu_cnt->prev_ts; > > + > > + dt *= MAX_BPS; > > + dt /= NS_PER_SEC; > > + > > + if (cnt->bytes + percpu_cnt->bytes - percpu_cnt->prev_bytes < dt) > > + ret = 1; > > + else > > + ret = 0; > > + > > + if (dt > REFRESH_TIME_NS) { > > + percpu_cnt->prev_ts = ts; > > + percpu_cnt->prev_packets = cnt->packets; > > + percpu_cnt->prev_bytes = cnt->bytes; > > + } > > + > > + return !!ret; > > +} > > + > > +char _license[] SEC("license") = "GPL"; > > +__u32 _version SEC("version") = LINUX_VERSION_CODE; > > diff --git a/tools/testing/selftests/bpf/test_netcnt.c b/tools/testing/selftests/bpf/test_netcnt.c > > new file mode 100644 > > index 000000000000..aa424f8db466 > > --- /dev/null > > +++ b/tools/testing/selftests/bpf/test_netcnt.c > > @@ -0,0 +1,153 @@ > > +// SPDX-License-Identifier: GPL-2.0 > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > + > > +#include > > +#include > > +#include > > + > > +#include "cgroup_helpers.h" > > +#include "bpf_rlimit.h" > > +#include "netcnt_common.h" > > + > > +#define BPF_PROG "./netcnt_prog.o" > > +#define TEST_CGROUP "/test-network-counters/" > > + > > +static int bpf_find_map(const char *test, struct bpf_object *obj, > > + const char *name) > > +{ > > + struct bpf_map *map; > > + > > + map = bpf_object__find_map_by_name(obj, name); > > + if (!map) { > > + printf("%s:FAIL:map '%s' not found\n", test, name); > > + return -1; > > + } > > + return bpf_map__fd(map); > > +} > > + > > +int main(int argc, char **argv) > > +{ > > + struct percpu_net_cnt *percpu_netcnt; > > + struct bpf_cgroup_storage_key key; > > + int map_fd, percpu_map_fd; > > + int error = EXIT_FAILURE; > > + struct net_cnt netcnt; > > + struct bpf_object *obj; > > + int prog_fd, cgroup_fd; > > + unsigned long packets; > > + int cpu, nproc; > > + __u32 prog_cnt; > > + > > + nproc = get_nprocs_conf(); > > + percpu_netcnt = malloc(sizeof(*percpu_netcnt) * nproc); > > + if (!percpu_netcnt) { > > + printf("Not enough memory for per-cpu area (%d cpus)\n", nproc); > > + goto err; > > + } > > + > > + if (bpf_prog_load(BPF_PROG, BPF_PROG_TYPE_CGROUP_SKB, > > + &obj, &prog_fd)) { > > + printf("Failed to load bpf program\n"); > > + goto out; > > + } > > + > > + if (setup_cgroup_environment()) { > > + printf("Failed to load bpf program\n"); > > + goto err; > > + } > > + > > + /* Create a cgroup, get fd, and join it */ > > + cgroup_fd = create_and_get_cgroup(TEST_CGROUP); > > + if (!cgroup_fd) { > > + printf("Failed to create test cgroup\n"); > > + goto err; > > + } > > + > > + if (join_cgroup(TEST_CGROUP)) { > > + printf("Failed to join cgroup\n"); > > + goto err; > > + } > > + > > + /* Attach bpf program */ > > + if (bpf_prog_attach(prog_fd, cgroup_fd, BPF_CGROUP_INET_EGRESS, 0)) { > > + printf("Failed to attach bpf program"); > > + goto err; > > + } > > + > > + assert(system("ping localhost -s 500 -c 10000 -f -q > /dev/null") == 0); > > + > > + if (bpf_prog_query(cgroup_fd, BPF_CGROUP_INET_EGRESS, 0, NULL, NULL, > > + &prog_cnt)) { > > + printf("Failed to query attached programs"); > > + goto err; > > + } > > + > > + map_fd = bpf_find_map(__func__, obj, "netcnt"); > > + if (map_fd < 0) { > > + printf("Failed to find bpf map with net counters"); > > + goto err; > > + } > > + > > + percpu_map_fd = bpf_find_map(__func__, obj, "percpu_netcnt"); > > + if (percpu_map_fd < 0) { > > + printf("Failed to find bpf map with percpu net counters"); > > + goto err; > > + } > > + > > + if (bpf_map_get_next_key(map_fd, NULL, &key)) { > > + printf("Failed to get key in cgroup storage\n"); > > + goto err; > > + } > > + > > + if (bpf_map_lookup_elem(map_fd, &key, &netcnt)) { > > + printf("Failed to lookup cgroup storage\n"); > > + goto err; > > + } > > + > > + if (bpf_map_lookup_elem(percpu_map_fd, &key, &percpu_netcnt[0])) { > > + printf("Failed to lookup percpu cgroup storage\n"); > > + goto err; > > + } > > + > > + /* Some packets can be still in per-cpu cache, but not more than > > + * MAX_PERCPU_PACKETS. > > + */ > > + packets = netcnt.packets; > > + for (cpu = 0; cpu < nproc; cpu++) { > > + if (percpu_netcnt[cpu].packets > 32) { > > pls use MAX_PERCPU_PACKETS in the above check. > could you also double check that if that #define is changed to 1k or so > the exact "!= 10000" check below still works as expected? Do you mean adding a new test with a different MAX_PERCPU_PACKETS? > > > + printf("Unexpected percpu value: %llu\n", > > + percpu_netcnt[cpu].packets); > > + goto err; > > > + } > > + > > + packets += percpu_netcnt[cpu].packets; > > + } > > + > > + /* No packets should be lost */ > > + if (packets != 10000) { > > + printf("Unexpected packet count: %lu\n", packets); > > + goto err; > > + } > > + > > + /* Let's check that bytes counter value is reasonable */ > > + if (netcnt.bytes < packets * 500 || netcnt.bytes > packets * 1500) { > > since packet count is accurate why byte count would vary ? Tbh I'm not sure if the size of the packet here can vary depending on the environment. Is there a nice way to get the expected size?