Skip to contentSkip to conversation

bpf: Batching iter for AF_UNIX sockets.

PATCH Applied Kuniyuki Iwashima 2 participants 7 messages Jan 13, 2022 – Jan 18, 2022

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

v2 cover bpf: Batching iter for AF_UNIX sockets.

Last year the commit afd20b9290e1 ("af_unix: Replace the big lock with small locks.") landed on bpf-next. Now we can use a batching algorithm for AF_UNIX bpf iter as TCP bpf iter.

Changelog:
- Add the 1st patch.
- Call unix_get_first() in .start()/.next() to always acquire a lock in each iteration in the 2nd patch.

Kuniyuki Iwashima (5):
af_unix: Refactor unix_next_socket().
bpf: af_unix: Use batching algorithm in bpf unix iter.
bpf: Support bpf_(get|set)sockopt() in bpf unix iter. selftest/bpf: Test batching and bpf_(get|set)sockopt in bpf unix iter. selftest/bpf: Fix a stale comment.

 net/unix/af_unix.c                            | 250 ++++++++++++++++--
 .../bpf/prog_tests/bpf_iter_setsockopt_unix.c | 100 +++++++
 .../bpf/progs/bpf_iter_setsockopt_unix.c      |  60 +++++
 .../selftests/bpf/progs/bpf_iter_unix.c       |   2 +-
 .../selftests/bpf/progs/bpf_tracing_net.h     |   2 +
 5 files changed, 385 insertions(+), 29 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/bpf_iter_setsockopt_unix.c
 create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_setsockopt_unix.c
Signature · 1 line
2.30.2
#1 lore raw

PATCH v2 1/5 af_unix: Refactor unix_next_socket().

Currently, unix_next_socket() is overloaded depending on the 2nd argument. If it is NULL, unix_next_socket() returns the first socket in the hash. If not NULL, it returns the next socket in the same hash list or the first socket in the next non-empty hash list.

This patch refactors unix_next_socket() into two functions unix_get_first() and unix_get_next(). unix_get_first() newly acquires a lock and returns the first socket in the list. unix_get_next() returns the next socket in a list or releases a lock and falls back to unix_get_first().

In the following patch, bpf iter holds entire sockets in a list and always releases the lock before .show(). It always calls unix_get_first() to acquire a lock in each iteration. So, this patch makes the change easier to follow.

---
 net/unix/af_unix.c | 51 +++++++++++++++++++++++++++-------------------
 1 file changed, 30 insertions(+), 21 deletions(-)
net/unix/af_unix.c +30 -21
index c19569819866..e1c4082accdb 100644
@@ -3240,49 +3240,58 @@ static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
 	return sk;
 }
 
-static struct sock *unix_next_socket(struct seq_file *seq,
-				     struct sock *sk,
-				     loff_t *pos)
+static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
 {
 	unsigned long bucket = get_bucket(*pos);
+	struct sock *sk;
 
-	while (sk > (struct sock *)SEQ_START_TOKEN) {
-		sk = sk_next(sk);
-		if (!sk)
-			goto next_bucket;
-		if (sock_net(sk) == seq_file_net(seq))
-			return sk;
-	}
-
-	do {
+	while (bucket < ARRAY_SIZE(unix_socket_table)) {
 		spin_lock(&unix_table_locks[bucket]);
+
 		sk = unix_from_bucket(seq, pos);
 		if (sk)
 			return sk;
 
-next_bucket:
-		spin_unlock(&unix_table_locks[bucket++]);
-		*pos = set_bucket_offset(bucket, 1);
-	} while (bucket < ARRAY_SIZE(unix_socket_table));
+		spin_unlock(&unix_table_locks[bucket]);
+
+		*pos = set_bucket_offset(++bucket, 1);
+	}
 
 	return NULL;
 }
 
+static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
+				  loff_t *pos)
+{
+	unsigned long bucket = get_bucket(*pos);
+
+	for (sk = sk_next(sk); sk; sk = sk_next(sk))
+		if (sock_net(sk) == seq_file_net(seq))
+			return sk;
+
+	spin_unlock(&unix_table_locks[bucket]);
+
+	*pos = set_bucket_offset(++bucket, 1);
+
+	return unix_get_first(seq, pos);
+}
+
 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
 {
 	if (!*pos)
 		return SEQ_START_TOKEN;
 
-	if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
-		return NULL;
-
-	return unix_next_socket(seq, NULL, pos);
+	return unix_get_first(seq, pos);
 }
 
 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 {
 	++*pos;
-	return unix_next_socket(seq, v, pos);
+
+	if (v == SEQ_START_TOKEN)
+		return unix_get_first(seq, pos);
+
+	return unix_get_next(seq, v, pos);
 }
 
 static void unix_seq_stop(struct seq_file *seq, void *v)
Signature · 1 line
2.30.2
#2 lore raw

PATCH v2 2/5 bpf: af_unix: Use batching algorithm in bpf unix iter.

In reply to Kuniyuki Iwashima

The commit 04c7820b776f ("bpf: tcp: Bpf iter batching and lock_sock") introduces the batching algorithm to iterate TCP sockets with more consistency.

This patch uses the same algorithm to iterate AF_UNIX sockets.

---
 net/unix/af_unix.c | 184 +++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 177 insertions(+), 7 deletions(-)
net/unix/af_unix.c +177 -7
index e1c4082accdb..d383d5f63b6b 100644
@@ -3356,6 +3356,15 @@ static const struct seq_operations unix_seq_ops = {
 };
 
 #if IS_BUILTIN(CONFIG_UNIX) && defined(CONFIG_BPF_SYSCALL)
+struct bpf_unix_iter_state {
+	struct seq_net_private p;
+	unsigned int cur_sk;
+	unsigned int end_sk;
+	unsigned int max_sk;
+	struct sock **batch;
+	bool st_bucket_done;
+};
+
 struct bpf_iter__unix {
 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
 	__bpf_md_ptr(struct unix_sock *, unix_sk);
@@ -3374,24 +3383,156 @@ static int unix_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
 	return bpf_iter_run_prog(prog, &ctx);
 }
 
+static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
+
+{
+	struct bpf_unix_iter_state *iter = seq->private;
+	unsigned int expected = 1;
+	struct sock *sk;
+
+	sock_hold(start_sk);
+	iter->batch[iter->end_sk++] = start_sk;
+
+	for (sk = sk_next(start_sk); sk; sk = sk_next(sk)) {
+		if (sock_net(sk) != seq_file_net(seq))
+			continue;
+
+		if (iter->end_sk < iter->max_sk) {
+			sock_hold(sk);
+			iter->batch[iter->end_sk++] = sk;
+		}
+
+		expected++;
+	}
+
+	spin_unlock(&unix_table_locks[start_sk->sk_hash]);
+
+	return expected;
+}
+
+static void bpf_iter_unix_put_batch(struct bpf_unix_iter_state *iter)
+{
+	while (iter->cur_sk < iter->end_sk)
+		sock_put(iter->batch[iter->cur_sk++]);
+}
+
+static int bpf_iter_unix_realloc_batch(struct bpf_unix_iter_state *iter,
+				       unsigned int new_batch_sz)
+{
+	struct sock **new_batch;
+
+	new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
+			     GFP_USER | __GFP_NOWARN);
+	if (!new_batch)
+		return -ENOMEM;
+
+	bpf_iter_unix_put_batch(iter);
+	kvfree(iter->batch);
+	iter->batch = new_batch;
+	iter->max_sk = new_batch_sz;
+
+	return 0;
+}
+
+static struct sock *bpf_iter_unix_batch(struct seq_file *seq,
+					loff_t *pos)
+{
+	struct bpf_unix_iter_state *iter = seq->private;
+	unsigned int expected;
+	bool resized = false;
+	struct sock *sk;
+
+	if (iter->st_bucket_done)
+		*pos = set_bucket_offset(get_bucket(*pos) + 1, 1);
+
+again:
+	/* Get a new batch */
+	iter->cur_sk = 0;
+	iter->end_sk = 0;
+
+	sk = unix_get_first(seq, pos);
+	if (!sk)
+		return NULL; /* Done */
+
+	expected = bpf_iter_unix_hold_batch(seq, sk);
+
+	if (iter->end_sk == expected) {
+		iter->st_bucket_done = true;
+		return sk;
+	}
+
+	if (!resized && !bpf_iter_unix_realloc_batch(iter, expected * 3 / 2)) {
+		resized = true;
+		goto again;
+	}
+
+	return sk;
+}
+
+static void *bpf_iter_unix_seq_start(struct seq_file *seq, loff_t *pos)
+{
+	if (!*pos)
+		return SEQ_START_TOKEN;
+
+	/* bpf iter does not support lseek, so it always
+	 * continue from where it was stop()-ped.
+	 */
+	return bpf_iter_unix_batch(seq, pos);
+}
+
+static void *bpf_iter_unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
+{
+	struct bpf_unix_iter_state *iter = seq->private;
+	struct sock *sk;
+
+	/* Whenever seq_next() is called, the iter->cur_sk is
+	 * done with seq_show(), so advance to the next sk in
+	 * the batch.
+	 */
+	if (iter->cur_sk < iter->end_sk)
+		sock_put(iter->batch[iter->cur_sk++]);
+
+	++*pos;
+
+	if (iter->cur_sk < iter->end_sk)
+		sk = iter->batch[iter->cur_sk];
+	else
+		sk = bpf_iter_unix_batch(seq, pos);
+
+	return sk;
+}
+
 static int bpf_iter_unix_seq_show(struct seq_file *seq, void *v)
 {
 	struct bpf_iter_meta meta;
 	struct bpf_prog *prog;
 	struct sock *sk = v;
 	uid_t uid;
+	bool slow;
+	int ret;
 
 	if (v == SEQ_START_TOKEN)
 		return 0;
 
+	slow = lock_sock_fast(sk);
+
+	if (unlikely(sk_unhashed(sk))) {
+		ret = SEQ_SKIP;
+		goto unlock;
+	}
+
 	uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
 	meta.seq = seq;
 	prog = bpf_iter_get_info(&meta, false);
-	return unix_prog_seq_show(prog, &meta, v, uid);
+	ret = unix_prog_seq_show(prog, &meta, v, uid);
+unlock:
+	unlock_sock_fast(sk, slow);
+	return ret;
 }
 
 static void bpf_iter_unix_seq_stop(struct seq_file *seq, void *v)
 {
+	struct bpf_unix_iter_state *iter = seq->private;
 	struct bpf_iter_meta meta;
 	struct bpf_prog *prog;
 
@@ -3402,12 +3543,13 @@ static void bpf_iter_unix_seq_stop(struct seq_file *seq, void *v)
 			(void)unix_prog_seq_show(prog, &meta, v, 0);
 	}
 
-	unix_seq_stop(seq, v);
+	if (iter->cur_sk < iter->end_sk)
+		bpf_iter_unix_put_batch(iter);
 }
 
 static const struct seq_operations bpf_iter_unix_seq_ops = {
-	.start	= unix_seq_start,
-	.next	= unix_seq_next,
+	.start	= bpf_iter_unix_seq_start,
+	.next	= bpf_iter_unix_seq_next,
 	.stop	= bpf_iter_unix_seq_stop,
 	.show	= bpf_iter_unix_seq_show,
 };
@@ -3456,11 +3598,39 @@ static struct pernet_operations unix_net_ops = {
 DEFINE_BPF_ITER_FUNC(unix, struct bpf_iter_meta *meta,
 		     struct unix_sock *unix_sk, uid_t uid)
 
+#define INIT_BATCH_SZ 16
+
+static int bpf_iter_init_unix(void *priv_data, struct bpf_iter_aux_info *aux)
+{
+	struct bpf_unix_iter_state *iter = priv_data;
+	int err;
+
+	err = bpf_iter_init_seq_net(priv_data, aux);
+	if (err)
+		return err;
+
+	err = bpf_iter_unix_realloc_batch(iter, INIT_BATCH_SZ);
+	if (err) {
+		bpf_iter_fini_seq_net(priv_data);
+		return err;
+	}
+
+	return 0;
+}
+
+static void bpf_iter_fini_unix(void *priv_data)
+{
+	struct bpf_unix_iter_state *iter = priv_data;
+
+	bpf_iter_fini_seq_net(priv_data);
+	kvfree(iter->batch);
+}
+
 static const struct bpf_iter_seq_info unix_seq_info = {
 	.seq_ops		= &bpf_iter_unix_seq_ops,
-	.init_seq_private	= bpf_iter_init_seq_net,
-	.fini_seq_private	= bpf_iter_fini_seq_net,
-	.seq_priv_size		= sizeof(struct seq_net_private),
+	.init_seq_private	= bpf_iter_init_unix,
+	.fini_seq_private	= bpf_iter_fini_unix,
+	.seq_priv_size		= sizeof(struct bpf_unix_iter_state),
 };
 
 static struct bpf_iter_reg unix_reg_info = {
Signature · 1 line
2.30.2
#3 lore raw

PATCH v2 3/5 bpf: Support bpf_(get|set)sockopt() in bpf unix iter.

In reply to Kuniyuki Iwashima

This patch makes bpf_(get|set)sockopt() available when iterating AF_UNIX sockets.

---
 net/unix/af_unix.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
net/unix/af_unix.c +15 -0
index d383d5f63b6b..3e0d6281fd1e 100644
@@ -3633,6 +3633,20 @@ static const struct bpf_iter_seq_info unix_seq_info = {
 	.seq_priv_size		= sizeof(struct bpf_unix_iter_state),
 };
 
+static const struct bpf_func_proto *
+bpf_iter_unix_get_func_proto(enum bpf_func_id func_id,
+			     const struct bpf_prog *prog)
+{
+	switch (func_id) {
+	case BPF_FUNC_setsockopt:
+		return &bpf_sk_setsockopt_proto;
+	case BPF_FUNC_getsockopt:
+		return &bpf_sk_getsockopt_proto;
+	default:
+		return NULL;
+	}
+}
+
 static struct bpf_iter_reg unix_reg_info = {
 	.target			= "unix",
 	.ctx_arg_info_size	= 1,
@@ -3640,6 +3654,7 @@ static struct bpf_iter_reg unix_reg_info = {
 		{ offsetof(struct bpf_iter__unix, unix_sk),
 		  PTR_TO_BTF_ID_OR_NULL },
 	},
+	.get_func_proto         = bpf_iter_unix_get_func_proto,
 	.seq_info		= &unix_seq_info,
 };
 
Signature · 1 line
2.30.2
#4 lore raw

PATCH v2 4/5 selftest/bpf: Test batching and bpf_(get|set)sockopt in bpf unix iter.

In reply to Kuniyuki Iwashima

This patch adds a test for the batching and bpf_(get|set)sockopt in bpf unix iter.

It does the following.

1. Creates an abstract UNIX domain socket
2. Call bpf_setsockopt()
3. Call bpf_getsockopt() and save the value
4. Call setsockopt()
5. Call getsockopt() and save the value
6. Compare the saved values

---
 .../bpf/prog_tests/bpf_iter_setsockopt_unix.c | 100 ++++++++++++++++++
 .../bpf/progs/bpf_iter_setsockopt_unix.c      |  60 +++++++++++
 .../selftests/bpf/progs/bpf_tracing_net.h     |   2 +
 3 files changed, 162 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/bpf_iter_setsockopt_unix.c
 create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_setsockopt_unix.c
tools/testing/selftests/bpf/prog_tests/bpf_iter_setsockopt_unix.c new file +100 -0
new file mode 100644
index 000000000000..ee725d4d98a5
@@ -0,0 +1,100 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright Amazon.com Inc. or its affiliates. */
+#include <sys/socket.h>
+#include <sys/un.h>
+#include <test_progs.h>
+#include "bpf_iter_setsockopt_unix.skel.h"
+
+#define NR_CASES 5
+
+static int create_unix_socket(struct bpf_iter_setsockopt_unix *skel)
+{
+	struct sockaddr_un addr = {
+		.sun_family = AF_UNIX,
+		.sun_path = "",
+	};
+	socklen_t len;
+	int fd, err;
+
+	fd = socket(AF_UNIX, SOCK_STREAM, 0);
+	if (!ASSERT_NEQ(fd, -1, "socket"))
+		return -1;
+
+	len = offsetof(struct sockaddr_un, sun_path);
+	err = bind(fd, (struct sockaddr *)&addr, len);
+	if (!ASSERT_OK(err, "bind"))
+		return -1;
+
+	len = sizeof(addr);
+	err = getsockname(fd, (struct sockaddr *)&addr, &len);
+	if (!ASSERT_OK(err, "getsockname"))
+		return -1;
+
+	memcpy(&skel->bss->sun_path, &addr.sun_path,
+	       len - offsetof(struct sockaddr_un, sun_path));
+
+	return fd;
+}
+
+static void test_sndbuf(struct bpf_iter_setsockopt_unix *skel, int fd)
+{
+	socklen_t optlen;
+	int i, err;
+
+	for (i = 0; i < NR_CASES; i++) {
+		if (!ASSERT_NEQ(skel->data->sndbuf_getsockopt[i], -1,
+				"bpf_(get|set)sockopt"))
+			return;
+
+		err = setsockopt(fd, SOL_SOCKET, SO_SNDBUF,
+				 &(skel->data->sndbuf_setsockopt[i]),
+				 sizeof(skel->data->sndbuf_setsockopt[i]));
+		if (!ASSERT_OK(err, "setsockopt"))
+			return;
+
+		optlen = sizeof(skel->bss->sndbuf_getsockopt_expected[i]);
+		err = getsockopt(fd, SOL_SOCKET, SO_SNDBUF,
+				 &(skel->bss->sndbuf_getsockopt_expected[i]),
+				 &optlen);
+		if (!ASSERT_OK(err, "getsockopt"))
+			return;
+
+		if (!ASSERT_EQ(skel->data->sndbuf_getsockopt[i],
+			       skel->bss->sndbuf_getsockopt_expected[i],
+			       "bpf_(get|set)sockopt"))
+			return;
+	}
+}
+
+void test_bpf_iter_setsockopt_unix(void)
+{
+	struct bpf_iter_setsockopt_unix *skel;
+	int err, unix_fd, iter_fd;
+	char buf;
+
+	skel = bpf_iter_setsockopt_unix__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "open_and_load"))
+		return;
+
+	unix_fd = create_unix_socket(skel);
+	if (!ASSERT_NEQ(unix_fd, -1, "create_unix_server"))
+		goto destroy;
+
+	skel->links.change_sndbuf = bpf_program__attach_iter(skel->progs.change_sndbuf, NULL);
+	if (!ASSERT_OK_PTR(skel->links.change_sndbuf, "bpf_program__attach_iter"))
+		goto destroy;
+
+	iter_fd = bpf_iter_create(bpf_link__fd(skel->links.change_sndbuf));
+	if (!ASSERT_GE(iter_fd, 0, "bpf_iter_create"))
+		goto destroy;
+
+	while ((err = read(iter_fd, &buf, sizeof(buf))) == -1 &&
+	       errno == EAGAIN)
+		;
+	if (!ASSERT_OK(err, "read iter error"))
+		goto destroy;
+
+	test_sndbuf(skel, unix_fd);
+destroy:
+	bpf_iter_setsockopt_unix__destroy(skel);
+}
tools/testing/selftests/bpf/progs/bpf_iter_setsockopt_unix.c new file +60 -0

60 lines not on this page · raw message

tools/testing/selftests/bpf/progs/bpf_tracing_net.h +2 -0
index e0f42601be9b..1c1289ba5fc5 100644
@@ -5,6 +5,8 @@
 #define AF_INET			2
 #define AF_INET6		10
 
+#define SOL_SOCKET		1
+#define SO_SNDBUF		7
 #define __SO_ACCEPTCON		(1 << 16)
 
 #define SOL_TCP			6
Signature · 1 line
2.30.2
#5 lore raw

PATCH v2 5/5 selftest/bpf: Fix a stale comment.

In reply to Kuniyuki Iwashima

The commit b8a58aa6fccc ("af_unix: Cut unix_validate_addr() out of unix_mkname().") moved the bound test part into unix_validate_addr().

---
 tools/testing/selftests/bpf/progs/bpf_iter_unix.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/bpf/progs/bpf_iter_unix.c +1 -1
index c21e3f545371..e6aefae38894 100644
@@ -63,7 +63,7 @@ int dump_unix(struct bpf_iter__unix *ctx)
 			BPF_SEQ_PRINTF(seq, " @");
 
 			for (i = 1; i < len; i++) {
-				/* unix_mkname() tests this upper bound. */
+				/* unix_validate_addr() tests this upper bound. */
 				if (i >= sizeof(struct sockaddr_un))
 					break;
 
Signature · 1 line
2.30.2
#6 lore raw
In reply to Kuniyuki Iwashima
Quoted 10 lines · Kuniyuki Iwashima

Last year the commit afd20b9290e1 ("af_unix: Replace the big lock with small locks.") landed on bpf-next. Now we can use a batching algorithm for AF_UNIX bpf iter as TCP bpf iter.

Changelog:
- Add the 1st patch.
- Call unix_get_first() in .start()/.next() to always acquire a lock in each iteration in the 2nd patch.

Applied. Thanks