selftests/bpf: Add kfree_skb raw_tp test
authorAlexei Starovoitov <ast@kernel.org>
Wed, 16 Oct 2019 03:25:05 +0000 (20:25 -0700)
committerDaniel Borkmann <daniel@iogearbox.net>
Thu, 17 Oct 2019 14:44:36 +0000 (16:44 +0200)
Load basic cls_bpf program.
Load raw_tracepoint program and attach to kfree_skb raw tracepoint.
Trigger cls_bpf via prog_test_run.
At the end of test_run kernel will call kfree_skb
which will trigger trace_kfree_skb tracepoint.
Which will call our raw_tracepoint program.
Which will take that skb and will dump it into perf ring buffer.
Check that user space received correct packet.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20191016032505.2089704-12-ast@kernel.org
tools/testing/selftests/bpf/prog_tests/kfree_skb.c [new file with mode: 0644]
tools/testing/selftests/bpf/progs/kfree_skb.c [new file with mode: 0644]

diff --git a/tools/testing/selftests/bpf/prog_tests/kfree_skb.c b/tools/testing/selftests/bpf/prog_tests/kfree_skb.c
new file mode 100644 (file)
index 0000000..430b50d
--- /dev/null
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <test_progs.h>
+
+static void on_sample(void *ctx, int cpu, void *data, __u32 size)
+{
+       int ifindex = *(int *)data, duration = 0;
+       struct ipv6_packet *pkt_v6 = data + 4;
+
+       if (ifindex != 1)
+               /* spurious kfree_skb not on loopback device */
+               return;
+       if (CHECK(size != 76, "check_size", "size %u != 76\n", size))
+               return;
+       if (CHECK(pkt_v6->eth.h_proto != 0xdd86, "check_eth",
+                 "h_proto %x\n", pkt_v6->eth.h_proto))
+               return;
+       if (CHECK(pkt_v6->iph.nexthdr != 6, "check_ip",
+                 "iph.nexthdr %x\n", pkt_v6->iph.nexthdr))
+               return;
+       if (CHECK(pkt_v6->tcp.doff != 5, "check_tcp",
+                 "tcp.doff %x\n", pkt_v6->tcp.doff))
+               return;
+
+       *(bool *)ctx = true;
+}
+
+void test_kfree_skb(void)
+{
+       struct bpf_prog_load_attr attr = {
+               .file = "./kfree_skb.o",
+       };
+
+       struct bpf_object *obj, *obj2 = NULL;
+       struct perf_buffer_opts pb_opts = {};
+       struct perf_buffer *pb = NULL;
+       struct bpf_link *link = NULL;
+       struct bpf_map *perf_buf_map;
+       struct bpf_program *prog;
+       __u32 duration, retval;
+       int err, pkt_fd, kfree_skb_fd;
+       bool passed = false;
+
+       err = bpf_prog_load("./test_pkt_access.o", BPF_PROG_TYPE_SCHED_CLS, &obj, &pkt_fd);
+       if (CHECK(err, "prog_load sched cls", "err %d errno %d\n", err, errno))
+               return;
+
+       err = bpf_prog_load_xattr(&attr, &obj2, &kfree_skb_fd);
+       if (CHECK(err, "prog_load raw tp", "err %d errno %d\n", err, errno))
+               goto close_prog;
+
+       prog = bpf_object__find_program_by_title(obj2, "tp_btf/kfree_skb");
+       if (CHECK(!prog, "find_prog", "prog kfree_skb not found\n"))
+               goto close_prog;
+       link = bpf_program__attach_raw_tracepoint(prog, NULL);
+       if (CHECK(IS_ERR(link), "attach_raw_tp", "err %ld\n", PTR_ERR(link)))
+               goto close_prog;
+
+       perf_buf_map = bpf_object__find_map_by_name(obj2, "perf_buf_map");
+       if (CHECK(!perf_buf_map, "find_perf_buf_map", "not found\n"))
+               goto close_prog;
+
+       /* set up perf buffer */
+       pb_opts.sample_cb = on_sample;
+       pb_opts.ctx = &passed;
+       pb = perf_buffer__new(bpf_map__fd(perf_buf_map), 1, &pb_opts);
+       if (CHECK(IS_ERR(pb), "perf_buf__new", "err %ld\n", PTR_ERR(pb)))
+               goto close_prog;
+
+       err = bpf_prog_test_run(pkt_fd, 1, &pkt_v6, sizeof(pkt_v6),
+                               NULL, NULL, &retval, &duration);
+       CHECK(err || retval, "ipv6",
+             "err %d errno %d retval %d duration %d\n",
+             err, errno, retval, duration);
+
+       /* read perf buffer */
+       err = perf_buffer__poll(pb, 100);
+       if (CHECK(err < 0, "perf_buffer__poll", "err %d\n", err))
+               goto close_prog;
+       /* make sure kfree_skb program was triggered
+        * and it sent expected skb into ring buffer
+        */
+       CHECK_FAIL(!passed);
+close_prog:
+       perf_buffer__free(pb);
+       if (!IS_ERR_OR_NULL(link))
+               bpf_link__destroy(link);
+       bpf_object__close(obj);
+       bpf_object__close(obj2);
+}
diff --git a/tools/testing/selftests/bpf/progs/kfree_skb.c b/tools/testing/selftests/bpf/progs/kfree_skb.c
new file mode 100644 (file)
index 0000000..89af8a9
--- /dev/null
@@ -0,0 +1,103 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2019 Facebook
+#include <linux/bpf.h>
+#include "bpf_helpers.h"
+#include "bpf_endian.h"
+
+char _license[] SEC("license") = "GPL";
+struct {
+       __uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
+       __uint(key_size, sizeof(int));
+       __uint(value_size, sizeof(int));
+} perf_buf_map SEC(".maps");
+
+#define _(P) (__builtin_preserve_access_index(P))
+
+/* define few struct-s that bpf program needs to access */
+struct callback_head {
+       struct callback_head *next;
+       void (*func)(struct callback_head *head);
+};
+struct dev_ifalias {
+       struct callback_head rcuhead;
+};
+
+struct net_device /* same as kernel's struct net_device */ {
+       int ifindex;
+       struct dev_ifalias *ifalias;
+};
+
+typedef struct {
+        int counter;
+} atomic_t;
+typedef struct refcount_struct {
+        atomic_t refs;
+} refcount_t;
+
+struct sk_buff {
+       /* field names and sizes should match to those in the kernel */
+       unsigned int len, data_len;
+       __u16 mac_len, hdr_len, queue_mapping;
+       struct net_device *dev;
+       /* order of the fields doesn't matter */
+       refcount_t users;
+       unsigned char *data;
+       char __pkt_type_offset[0];
+};
+
+/* copy arguments from
+ * include/trace/events/skb.h:
+ * TRACE_EVENT(kfree_skb,
+ *         TP_PROTO(struct sk_buff *skb, void *location),
+ *
+ * into struct below:
+ */
+struct trace_kfree_skb {
+       struct sk_buff *skb;
+       void *location;
+};
+
+SEC("tp_btf/kfree_skb")
+int trace_kfree_skb(struct trace_kfree_skb *ctx)
+{
+       struct sk_buff *skb = ctx->skb;
+       struct net_device *dev;
+       int ifindex;
+       struct callback_head *ptr;
+       void *func;
+       int users;
+       unsigned char *data;
+       unsigned short pkt_data;
+       char pkt_type;
+
+       __builtin_preserve_access_index(({
+               users = skb->users.refs.counter;
+               data = skb->data;
+               dev = skb->dev;
+               ifindex = dev->ifindex;
+               ptr = dev->ifalias->rcuhead.next;
+               func = ptr->func;
+       }));
+
+       bpf_probe_read(&pkt_type, sizeof(pkt_type), _(&skb->__pkt_type_offset));
+       pkt_type &= 7;
+
+       /* read eth proto */
+       bpf_probe_read(&pkt_data, sizeof(pkt_data), data + 12);
+
+       bpf_printk("rcuhead.next %llx func %llx\n", ptr, func);
+       bpf_printk("skb->len %d users %d pkt_type %x\n",
+                  _(skb->len), users, pkt_type);
+       bpf_printk("skb->queue_mapping %d\n", _(skb->queue_mapping));
+       bpf_printk("dev->ifindex %d data %llx pkt_data %x\n",
+                  ifindex, data, pkt_data);
+
+       if (users != 1 || pkt_data != bpf_htons(0x86dd) || ifindex != 1)
+               /* raw tp ignores return value */
+               return 0;
+
+       /* send first 72 byte of the packet to user space */
+       bpf_skb_output(skb, &perf_buf_map, (72ull << 32) | BPF_F_CURRENT_CPU,
+                      &ifindex, sizeof(ifindex));
+       return 0;
+}