Re: [PATCH bpf-next 2/2] selftest/bpf: Implement sample UNIX domain socket iterator program.

[Yonghong Song <yhs@fb.com>]

On 7/29/21 4:36 PM, Kuniyuki Iwashima wrote:
> If there are no abstract sockets, this prog can output the same result
> compared to /proc/net/unix.
> 
>    # cat /sys/fs/bpf/unix | head -n 2
>    Num       RefCount Protocol Flags    Type St Inode Path
>    ffff9ab7122db000: 00000002 00000000 00010000 0001 01 10623 private/defer
> 
>    # cat /proc/net/unix | head -n 2
>    Num       RefCount Protocol Flags    Type St Inode Path
>    ffff9ab7122db000: 00000002 00000000 00010000 0001 01 10623 private/defer
> 
> Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
> ---
>   .../selftests/bpf/prog_tests/bpf_iter.c       | 17 +++++
>   .../selftests/bpf/progs/bpf_iter_unix.c       | 75 +++++++++++++++++++
>   2 files changed, 92 insertions(+)
>   create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_unix.c
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/bpf_iter.c b/tools/testing/selftests/bpf/prog_tests/bpf_iter.c
> index 1f1aade56504..4746bac68d36 100644
> --- a/tools/testing/selftests/bpf/prog_tests/bpf_iter.c
> +++ b/tools/testing/selftests/bpf/prog_tests/bpf_iter.c
> @@ -13,6 +13,7 @@
>   #include "bpf_iter_tcp6.skel.h"
>   #include "bpf_iter_udp4.skel.h"
>   #include "bpf_iter_udp6.skel.h"
> +#include "bpf_iter_unix.skel.h"
>   #include "bpf_iter_test_kern1.skel.h"
>   #include "bpf_iter_test_kern2.skel.h"
>   #include "bpf_iter_test_kern3.skel.h"
> @@ -313,6 +314,20 @@ static void test_udp6(void)
>   	bpf_iter_udp6__destroy(skel);
>   }
>   
> +static void test_unix(void)
> +{
> +	struct bpf_iter_unix *skel;
> +
> +	skel = bpf_iter_unix__open_and_load();
> +	if (CHECK(!skel, "bpf_iter_unix__open_and_load",
> +		  "skeleton open_and_load failed\n"))
> +		return;
> +
> +	do_dummy_read(skel->progs.dump_unix);
> +
> +	bpf_iter_unix__destroy(skel);
> +}
> +
>   /* The expected string is less than 16 bytes */
>   static int do_read_with_fd(int iter_fd, const char *expected,
>   			   bool read_one_char)
> @@ -1255,6 +1270,8 @@ void test_bpf_iter(void)
>   		test_udp4();
>   	if (test__start_subtest("udp6"))
>   		test_udp6();
> +	if (test__start_subtest("unix"))
> +		test_unix();
>   	if (test__start_subtest("anon"))
>   		test_anon_iter(false);
>   	if (test__start_subtest("anon-read-one-char"))
> diff --git a/tools/testing/selftests/bpf/progs/bpf_iter_unix.c b/tools/testing/selftests/bpf/progs/bpf_iter_unix.c
> new file mode 100644
> index 000000000000..285ec2f7944d
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/progs/bpf_iter_unix.c
> @@ -0,0 +1,75 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/* Copyright Amazon.com Inc. or its affiliates. */
> +#include "bpf_iter.h"

Could you add bpf_iter__unix to bpf_iter.h similar to bpf_iter__sockmap?
The main purpose is to make test tolerating with old vmlinux.h.

> +#include "bpf_tracing_net.h"
> +#include <bpf/bpf_helpers.h>
> +#include <bpf/bpf_endian.h>
> +
> +char _license[] SEC("license") = "GPL";
> +
> +#define __SO_ACCEPTCON		(1 << 16)
> +#define UNIX_HASH_SIZE		256
> +#define UNIX_ABSTRACT(unix_sk)	(unix_sk->addr->hash < UNIX_HASH_SIZE)

Could you add the above three define's in bpf_tracing_net.h?
We try to keep all these common defines in a common header for
potential reusability.

> +
> +static long sock_i_ino(const struct sock *sk)
> +{
> +	const struct socket *sk_socket = sk->sk_socket;
> +	const struct inode *inode;
> +	unsigned long ino;
> +
> +	if (!sk_socket)
> +		return 0;
> +
> +	inode = &container_of(sk_socket, struct socket_alloc, socket)->vfs_inode;
> +	bpf_probe_read_kernel(&ino, sizeof(ino), &inode->i_ino);
> +	return ino;
> +}
> +
> +SEC("iter/unix")
> +int dump_unix(struct bpf_iter__unix *ctx)
> +{
> +	struct unix_sock *unix_sk = ctx->unix_sk;
> +	struct sock *sk = (struct sock *)unix_sk;
> +	struct seq_file *seq;
> +	__u32 seq_num;
> +
> +	if (!unix_sk)
> +		return 0;
> +
> +	seq = ctx->meta->seq;
> +	seq_num = ctx->meta->seq_num;
> +	if (seq_num == 0)
> +		BPF_SEQ_PRINTF(seq, "Num       RefCount Protocol Flags    "
> +			       "Type St Inode Path\n");
> +
> +	BPF_SEQ_PRINTF(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
> +		       unix_sk,
> +		       sk->sk_refcnt.refs.counter,
> +		       0,
> +		       sk->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
> +		       sk->sk_type,
> +		       sk->sk_socket ?
> +		       (sk->sk_state == TCP_ESTABLISHED ?
> +			SS_CONNECTED : SS_UNCONNECTED) :
> +		       (sk->sk_state == TCP_ESTABLISHED ?
> +			SS_CONNECTING : SS_DISCONNECTING),
> +		       sock_i_ino(sk));
> +
> +	if (unix_sk->addr) {
> +		if (UNIX_ABSTRACT(unix_sk))
> +			/* Abstract UNIX domain socket can contain '\0' in
> +			 * the path, and it should be escaped.  However, it
> +			 * requires loops and the BPF verifier rejects it.
> +			 * So here, print only the escaped first byte to
> +			 * indicate it is an abstract UNIX domain socket.
> +			 * (See: unix_seq_show() and commit e7947ea770d0d)
> +			 */
> +			BPF_SEQ_PRINTF(seq, " @");
> +		else
> +			BPF_SEQ_PRINTF(seq, " %s", unix_sk->addr->name->sun_path);
> +	}

I looked at af_unix.c, for the above "if (unix_sk->addr) { ... }" code,
the following is the kernel source code,

                 if (u->addr) {  // under unix_table_lock here
                         int i, len;
                         seq_putc(seq, ' ');

                         i = 0;
                         len = u->addr->len - sizeof(short);
                         if (!UNIX_ABSTRACT(s))
                                 len--;
                         else {
                                 seq_putc(seq, '@');
                                 i++;
                         }
                         for ( ; i < len; i++)
                                 seq_putc(seq, u->addr->name->sun_path[i] ?:
                                          '@');
                 }

It does not seem to match bpf program non UNIX_ABSTRACT case.
I am not familiar with unix socket so it would be good if you can 
explain a little more.

For verifier issue with loops, do we have a maximum upper bound for 
u->addr->len? If yes, does bounded loop work?

> +
> +	BPF_SEQ_PRINTF(seq, "\n");
> +
> +	return 0;
> +}
>