BPF updates

deleted file mode 100644
index f9b9ce3..0000000
@@ -1,258 +0,0 @@
-/*
- * eBPF user space agent part
- *
- * Simple, _self-contained_ user space agent for the eBPF kernel
- * ebpf_prog.c program, which gets all map fds passed from tc via unix
- * domain socket in one transaction and can thus keep referencing
- * them from user space in order to read out (or possibly modify)
- * map data. Here, just as a minimal example to display counters.
- *
- * The agent only uses the bpf(2) syscall API to read or possibly
- * write to eBPF maps, it doesn't need to be aware of the low-level
- * bytecode parts and/or ELF parsing bits.
- *
- * ! For more details, see header comment in bpf_prog.c !
- *
- * gcc bpf_agent.c -o bpf_agent -Wall -O2
- *
- * For example, a more complex user space agent could run on each
- * host, reading and writing into eBPF maps used by tc classifier
- * and actions. It would thus allow for implementing a distributed
- * tc architecture, for example, which would push down central
- * policies into eBPF maps, and thus altering run-time behaviour.
- *
- *   -- Happy eBPF hacking! ;)
- */
-
-#define _GNU_SOURCE
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <errno.h>
-#include <unistd.h>
-#include <stdint.h>
-#include <assert.h>
-
-#include <sys/un.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <sys/socket.h>
-
-/* Just some misc macros as min(), offsetof(), etc. */
-#include "../../include/utils.h"
-/* Common code from fd passing. */
-#include "../../include/bpf_scm.h"
-/* Common, shared definitions with ebpf_prog.c */
-#include "bpf_shared.h"
-/* Mini syscall wrapper */
-#include "bpf_sys.h"
-
-static void bpf_dump_drops(int fd)
-{
-	int cpu, max;
-
-	max = sysconf(_SC_NPROCESSORS_ONLN);
-
-	printf(" `- number of drops:");
-	for (cpu = 0; cpu < max; cpu++) {
-		long drops;
-
-		assert(bpf_lookup_elem(fd, &cpu, &drops) == 0);
-		printf("\tcpu%d: %5ld", cpu, drops);
-	}
-	printf("\n");
-}
-
-static void bpf_dump_queue(int fd)
-{
-	/* Just for the same of the example. */
-	int max_queue = 4, i;
-
-	printf("  | nic queues:");
-	for (i = 0; i < max_queue; i++) {
-		struct count_queue cq;
-		int ret;
-
-		memset(&cq, 0, sizeof(cq));
-		ret = bpf_lookup_elem(fd, &i, &cq);
-		assert(ret == 0 || (ret < 0 && errno == ENOENT));
-
-		printf("\tq%d:[pkts: %ld, mis: %ld]",
-		       i, cq.total, cq.mismatch);
-	}
-	printf("\n");
-}
-
-static void bpf_dump_proto(int fd)
-{
-	uint8_t protos[] = { IPPROTO_TCP, IPPROTO_UDP, IPPROTO_ICMP };
-	char *names[] = { "tcp", "udp", "icmp" };
-	int i;
-
-	printf("  ` protos:");
-	for (i = 0; i < ARRAY_SIZE(protos); i++) {
-		struct count_tuple ct;
-		int ret;
-
-		memset(&ct, 0, sizeof(ct));
-		ret = bpf_lookup_elem(fd, &protos[i], &ct);
-		assert(ret == 0 || (ret < 0 && errno == ENOENT));
-
-		printf("\t%s:[pkts: %ld, bytes: %ld]",
-		       names[i], ct.packets, ct.bytes);
-	}
-	printf("\n");
-}
-
-static void bpf_dump_map_data(int *tfd)
-{
-	int i;
-
-	for (i = 0; i < 30; i++) {
-		const int period = 5;
-
-		printf("data, period: %dsec\n", period);
-
-		bpf_dump_drops(tfd[BPF_MAP_ID_DROPS]);
-		bpf_dump_queue(tfd[BPF_MAP_ID_QUEUE]);
-		bpf_dump_proto(tfd[BPF_MAP_ID_PROTO]);
-
-		sleep(period);
-	}
-}
-
-static void bpf_info_loop(int *fds, struct bpf_map_aux *aux)
-{
-	int i, tfd[BPF_MAP_ID_MAX];
-
-	printf("ver: %d\nobj: %s\ndev: %lu\nino: %lu\nmaps: %u\n",
-	       aux->uds_ver, aux->obj_name, aux->obj_st.st_dev,
-	       aux->obj_st.st_ino, aux->num_ent);
-
-	for (i = 0; i < aux->num_ent; i++) {
-		printf("map%d:\n", i);
-		printf(" `- fd: %u\n", fds[i]);
-		printf("  | serial: %u\n", aux->ent[i].id);
-		printf("  | type: %u\n", aux->ent[i].type);
-		printf("  | max elem: %u\n", aux->ent[i].max_elem);
-		printf("  | size key: %u\n", aux->ent[i].size_key);
-		printf("  ` size val: %u\n", aux->ent[i].size_value);
-
-		tfd[aux->ent[i].id] = fds[i];
-	}
-
-	bpf_dump_map_data(tfd);
-}
-
-static void bpf_map_get_from_env(int *tfd)
-{
-	char key[64], *val;
-	int i;
-
-	for (i = 0; i < BPF_MAP_ID_MAX; i++) {
-		memset(key, 0, sizeof(key));
-		snprintf(key, sizeof(key), "BPF_MAP%d", i);
-
-		val = getenv(key);
-		assert(val != NULL);
-
-		tfd[i] = atoi(val);
-	}
-}
-
-static int bpf_map_set_recv(int fd, int *fds,  struct bpf_map_aux *aux,
-			    unsigned int entries)
-{
-	struct bpf_map_set_msg msg;
-	int *cmsg_buf, min_fd, i;
-	char *amsg_buf, *mmsg_buf;
-
-	cmsg_buf = bpf_map_set_init(&msg, NULL, 0);
-	amsg_buf = (char *)msg.aux.ent;
-	mmsg_buf = (char *)&msg.aux;
-
-	for (i = 0; i < entries; i += min_fd) {
-		struct cmsghdr *cmsg;
-		int ret;
-
-		min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i);
-
-		bpf_map_set_init_single(&msg, min_fd);
-
-		ret = recvmsg(fd, &msg.hdr, 0);
-		if (ret <= 0)
-			return ret ? : -1;
-
-		cmsg = CMSG_FIRSTHDR(&msg.hdr);
-		if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS)
-			return -EINVAL;
-		if (msg.hdr.msg_flags & MSG_CTRUNC)
-			return -EIO;
-
-		min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd);
-		if (min_fd > entries || min_fd <= 0)
-			return -1;
-
-		memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd);
-		memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd);
-		memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent));
-
-		if (i + min_fd == aux->num_ent)
-			break;
-	}
-
-	return 0;
-}
-
-int main(int argc, char **argv)
-{
-	int fds[BPF_SCM_MAX_FDS];
-	struct bpf_map_aux aux;
-	struct sockaddr_un addr;
-	int fd, ret, i;
-
-	/* When arguments are being passed, we take it as a path
-	 * to a Unix domain socket, otherwise we grab the fds
-	 * from the environment to demonstrate both possibilities.
-	 */
-	if (argc == 1) {
-		int tfd[BPF_MAP_ID_MAX];
-
-		bpf_map_get_from_env(tfd);
-		bpf_dump_map_data(tfd);
-
-		return 0;
-	}
-
-	fd = socket(AF_UNIX, SOCK_DGRAM, 0);
-	if (fd < 0) {
-		fprintf(stderr, "Cannot open socket: %s\n",
-			strerror(errno));
-		exit(1);
-	}
-
-	memset(&addr, 0, sizeof(addr));
-	addr.sun_family = AF_UNIX;
-	strncpy(addr.sun_path, argv[argc - 1], sizeof(addr.sun_path));
-
-	ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr));
-	if (ret < 0) {
-		fprintf(stderr, "Cannot bind to socket: %s\n",
-			strerror(errno));
-		exit(1);
-	}
-
-	memset(fds, 0, sizeof(fds));
-	memset(&aux, 0, sizeof(aux));
-
-	ret = bpf_map_set_recv(fd, fds, &aux, BPF_SCM_MAX_FDS);
-	if (ret >= 0)
-		bpf_info_loop(fds, &aux);
-
-	for (i = 0; i < aux.num_ent; i++)
-		close(fds[i]);
-
-	close(fd);
-	return 0;
-}

501 lines not on this page · raw message

deleted file mode 100644
index a24038d..0000000
@@ -1,22 +0,0 @@
-#ifndef __BPF_SHARED__
-#define __BPF_SHARED__
-
-enum {
-	BPF_MAP_ID_PROTO,
-	BPF_MAP_ID_QUEUE,
-	BPF_MAP_ID_DROPS,
-	__BPF_MAP_ID_MAX,
-#define BPF_MAP_ID_MAX	__BPF_MAP_ID_MAX
-};
-
-struct count_tuple {
-	long packets; /* type long for lock_xadd() */
-	long bytes;
-};
-
-struct count_queue {
-	long total;
-	long mismatch;
-};
-
-#endif /* __BPF_SHARED__ */

deleted file mode 100644
index 6e4f09e..0000000
@@ -1,23 +0,0 @@
-#ifndef __BPF_SYS__
-#define __BPF_SYS__
-
-#include <sys/syscall.h>
-#include <linux/bpf.h>
-
-static inline __u64 bpf_ptr_to_u64(const void *ptr)
-{
-	return (__u64) (unsigned long) ptr;
-}
-
-static inline int bpf_lookup_elem(int fd, void *key, void *value)
-{
-	union bpf_attr attr = {
-		.map_fd		= fd,
-		.key		= bpf_ptr_to_u64(key),
-		.value		= bpf_ptr_to_u64(value),
-	};
-
-	return syscall(__NR_bpf, BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
-}
-
-#endif /* __BPF_SYS__ */

new file mode 100644
index 0000000..ff0e623
@@ -0,0 +1,56 @@
+#include "../../include/bpf_api.h"
+
+#define MAP_INNER_ID	42
+
+struct bpf_elf_map __section_maps map_inner = {
+	.type		= BPF_MAP_TYPE_ARRAY,
+	.size_key	= sizeof(uint32_t),
+	.size_value	= sizeof(uint32_t),
+	.id		= MAP_INNER_ID,
+	.inner_idx	= 0,
+	.pinning	= PIN_GLOBAL_NS,
+	.max_elem	= 1,
+};
+
+struct bpf_elf_map __section_maps map_outer = {
+	.type		= BPF_MAP_TYPE_ARRAY_OF_MAPS,
+	.size_key	= sizeof(uint32_t),
+	.size_value	= sizeof(uint32_t),
+	.inner_id	= MAP_INNER_ID,
+	.pinning	= PIN_GLOBAL_NS,
+	.max_elem	= 1,
+};
+
+__section("egress")
+int emain(struct __sk_buff *skb)
+{
+	struct bpf_elf_map *map_inner;
+	int key = 0, *val;
+
+	map_inner = map_lookup_elem(&map_outer, &key);
+	if (map_inner) {
+		val = map_lookup_elem(map_inner, &key);
+		if (val)
+			lock_xadd(val, 1);
+	}
+
+	return BPF_H_DEFAULT;
+}
+
+__section("ingress")
+int imain(struct __sk_buff *skb)
+{
+	struct bpf_elf_map *map_inner;
+	int key = 0, *val;
+
+	map_inner = map_lookup_elem(&map_outer, &key);
+	if (map_inner) {
+		val = map_lookup_elem(map_inner, &key);
+		if (val)
+			printt("map val: %d\n", *val);
+	}
+
+	return BPF_H_DEFAULT;
+}
+
+BPF_LICENSE("GPL");

index 239a0f3..406c308 100644
@@ -36,6 +36,8 @@ struct bpf_elf_map {
 	__u32 flags;
 	__u32 id;
 	__u32 pinning;
+	__u32 inner_id;
+	__u32 inner_idx;
 };
 
 #endif /* __BPF_ELF__ */

216 lines not on this page · raw message

index 5361dab..6582ec8 100644
@@ -261,6 +261,8 @@ int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns,
 int bpf_prog_attach_fd(int prog_fd, int target_fd, enum bpf_attach_type type);
 int bpf_prog_detach_fd(int target_fd, enum bpf_attach_type type);
 
+void bpf_dump_prog_info(FILE *f, uint32_t id);
+
 #ifdef HAVE_ELF
 int bpf_send_map_fds(const char *path, const char *obj);
 int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux,

54 lines not on this page · raw message

index 75c44c0..2f8d12a 100644
@@ -230,6 +230,9 @@ static int bpf_print_opt(struct filter_util *qu, FILE *f,
 				      b, sizeof(b)));
 	}
 
+	if (tb[TCA_BPF_ID])
+		bpf_dump_prog_info(f, rta_getattr_u32(tb[TCA_BPF_ID]));
+
 	if (tb[TCA_BPF_POLICE]) {
 		fprintf(f, "\n");
 		tc_print_police(f, tb[TCA_BPF_POLICE]);

index 5728303..df559bc 100644
@@ -186,6 +186,9 @@ static int bpf_print_opt(struct action_util *au, FILE *f, struct rtattr *arg)
 				      b, sizeof(b)));
 	}
 
+        if (tb[TCA_ACT_BPF_ID])
+                bpf_dump_prog_info(f, rta_getattr_u32(tb[TCA_ACT_BPF_ID]));
+
 	print_action_control(f, "default-action ", parm->action, "\n");
 	fprintf(f, "\tindex %u ref %d bind %d", parm->index, parm->refcnt,
 		parm->bindcnt);