Skip to contentSkip to conversation

transition sockmap testing to test_progs

PATCH In review John Fastabend 2 participants 15 messages Jan 24, 2024 – Jan 26, 2024

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

v2 cover transition sockmap testing to test_progs

Its much easier to write and read tests than it was when sockmap was originally created. At that time we created a test_sockmap prog that did sockmap tests. But, its showing its age now. For example it reads user vars out of maps, is hard to run targetted tests, has a different format from the familiar test_progs and so on.

I recently thought there was an issue with pop helpers so I created some tests to try and track it down. It turns out it was a bug in the BPF program we had not the kernel. But, I think it makes sense to start deprecating test_sockmap and converting these to the nicer test_progs.

So this is a first round of test_prog tests for sockmap cork and pop helpers. I'll add push and pull tests shortly. I think its fine, maybe preferred to review smaller patchsets, to send these incrementally as I get them created.

Thanks!

v2: fix unint vars in some branches from make RELEASE=1

John Fastabend (4):
bpf: sockmap, add test for sk_msg prog pop msg helper
bpf: sockmap, add a sendmsg test so we can check that path
bpf: sockmap, add a cork to force buffering of the scatterlist
bpf: sockmap test cork and pop combined

 .../bpf/prog_tests/sockmap_helpers.h          |  18 +
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 353 ++++++++++++++++++
 .../bpf/progs/test_sockmap_msg_helpers.c      |  67 ++++
 3 files changed, 438 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c
Signature · 1 line
2.33.0
#1 lore raw

PATCH v2 1/4 bpf: sockmap, add test for sk_msg prog pop msg helper

For msg_pop sk_msg helpers we only have older tests in test_sockmap, but these are showing their age. They don't use any of the newer style BPF and also require running test_sockmap. Lets use the prog_test framework and add a test for msg_pop.

This is a much nicer test env using newer style BPF. We can extend this to support all the other helpers shortly.

The bpf program is a template that lets us run through all the helpers so we can cover not just pop, but all the other helpers as well.

---
 .../bpf/prog_tests/sockmap_helpers.h          |  10 +
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 210 ++++++++++++++++++
 .../bpf/progs/test_sockmap_msg_helpers.c      |  52 +++++
 3 files changed, 272 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c
tools/testing/selftests/bpf/prog_tests/sockmap_helpers.h +10 -0
index e880f97bc44d..781cbdf01d7b 100644
@@ -112,6 +112,16 @@
 		__ret;                                                         \
 	})
 
+#define xrecv_nonblock(fd, buf, len, flags)                                    \
+	({                                                                     \
+		ssize_t __ret = recv_timeout((fd), (buf), (len), (flags),      \
+					     IO_TIMEOUT_SEC);                  \
+		if (__ret == -1)                                               \
+			FAIL_ERRNO("recv");                                    \
+		__ret;                                                         \
+	})
+
+
 #define xsocket(family, sotype, flags)                                         \
 	({                                                                     \
 		int __ret = socket(family, sotype, flags);                     \
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c new file +210 -0

210 lines not on this page · raw message

tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c new file +52 -0

52 lines not on this page · raw message

Signature · 1 line
2.33.0
#2 lore raw

Re: [PATCH bpf-next v2 1/4] bpf: sockmap, add test for sk_msg prog pop msg helper

Quoted 20 lines · John Fastabend

For msg_pop sk_msg helpers we only have older tests in test_sockmap, but these are showing their age. They don't use any of the newer style BPF and also require running test_sockmap. Lets use the prog_test framework and add a test for msg_pop.

This is a much nicer test env using newer style BPF. We can extend this to support all the other helpers shortly.

The bpf program is a template that lets us run through all the helpers so we can cover not just pop, but all the other helpers as well.

---
 .../bpf/prog_tests/sockmap_helpers.h          |  10 +
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 210 ++++++++++++++++++
 .../bpf/progs/test_sockmap_msg_helpers.c      |  52 +++++
 3 files changed, 272 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c

[...]

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c line 2

Quoted 8 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c new file +2 -0
new file mode 100644
index 000000000000..9ffe02f45808
@@ -0,0 +1,210 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2020 Cloudflare

Thanks, but we can't take the credit for this brand new file ;-)

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c line 14

Quoted 12 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c new file +12 -0
+#include <error.h>
+#include <netinet/tcp.h>
+#include <sys/epoll.h>
+
+#include "test_progs.h"
+#include "test_sockmap_msg_helpers.skel.h"
+#include "sockmap_helpers.h"
+
+#define TCP_REPAIR		19	/* TCP sock is under repair right now */
+
+#define TCP_REPAIR_ON		1
+#define TCP_REPAIR_OFF_NO_WP	-1	/* Turn off without window probes */

These defines are not unused by this module. Copy-pasted by mistake?

[...]

tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c line 2

Quoted 8 lines
tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c new file +2 -0
new file mode 100644
index 000000000000..c721a00b6001
@@ -0,0 +1,52 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2020 Cloudflare

^ :-)

[...]

#3 lore raw

Re: [PATCH bpf-next v2 1/4] bpf: sockmap, add test for sk_msg prog pop msg helper

Quoted 24 lines · Jakub Sitnicki
Quoted 20 lines · John Fastabend

For msg_pop sk_msg helpers we only have older tests in test_sockmap, but these are showing their age. They don't use any of the newer style BPF and also require running test_sockmap. Lets use the prog_test framework and add a test for msg_pop.

This is a much nicer test env using newer style BPF. We can extend this to support all the other helpers shortly.

The bpf program is a template that lets us run through all the helpers so we can cover not just pop, but all the other helpers as well.

---
 .../bpf/prog_tests/sockmap_helpers.h          |  10 +
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 210 ++++++++++++++++++
 .../bpf/progs/test_sockmap_msg_helpers.c      |  52 +++++
 3 files changed, 272 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c

[...]

Quoted 10 lines

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c line 2

Quoted 8 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c new file +2 -0
new file mode 100644
index 000000000000..9ffe02f45808
@@ -0,0 +1,210 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2020 Cloudflare

Thanks, but we can't take the credit for this brand new file ;-)

Ah yep seems I cut'n'pasted it from one of your files. I'll update it.

Quoted 15 lines

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c line 14

Quoted 13 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c new file +12 -0
+#include <error.h>
+#include <netinet/tcp.h>
+#include <sys/epoll.h>
+
+#include "test_progs.h"
+#include "test_sockmap_msg_helpers.skel.h"
+#include "sockmap_helpers.h"
+
+#define TCP_REPAIR		19	/* TCP sock is under repair right now */
+
+#define TCP_REPAIR_ON		1
+#define TCP_REPAIR_OFF_NO_WP	-1	/* Turn off without window probes */

These defines are not unused by this module. Copy-pasted by mistake?

Yep first iteration I used these will drop them. Thanks.

Quoted 15 lines

[...]

tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c line 2

Quoted 8 lines
tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c new file +2 -0
new file mode 100644
index 000000000000..c721a00b6001
@@ -0,0 +1,52 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2020 Cloudflare

^ :-)

[...]

#4 lore raw

PATCH v2 2/4 bpf: sockmap, add a sendmsg test so we can check that path

In reply to John Fastabend

Sendmsg path with multiple buffers is slightly different from a single send in how we have to handle and walk the sg when doing pops. Lets ensure this walk is correct.

---
 .../bpf/prog_tests/sockmap_helpers.h          |  8 +++
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 53 +++++++++++++++++++
 2 files changed, 61 insertions(+)
tools/testing/selftests/bpf/prog_tests/sockmap_helpers.h +8 -0
index 781cbdf01d7b..4d8d24482032 100644
@@ -103,6 +103,14 @@
 		__ret;                                                         \
 	})
 
+#define xsendmsg(fd, msg, flags)                                               \
+	({                                                                     \
+		ssize_t __ret = sendmsg((fd), (msg), (flags));                 \
+		if (__ret == -1)                                               \
+			FAIL_ERRNO("sendmsg");                                 \
+		__ret;                                                         \
+	})
+
 #define xrecv_nonblock(fd, buf, len, flags)                                    \
 	({                                                                     \
 		ssize_t __ret = recv_timeout((fd), (buf), (len), (flags),      \
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +53 -0

65 lines not on this page · raw message

Signature · 1 line
2.33.0
#5 lore raw

Re: [PATCH bpf-next v2 2/4] bpf: sockmap, add a sendmsg test so we can check that path

Quoted 38 lines · John Fastabend

Sendmsg path with multiple buffers is slightly different from a single send in how we have to handle and walk the sg when doing pops. Lets ensure this walk is correct.

---
 .../bpf/prog_tests/sockmap_helpers.h          |  8 +++
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 53 +++++++++++++++++++
 2 files changed, 61 insertions(+)
tools/testing/selftests/bpf/prog_tests/sockmap_helpers.h +8 -0
index 781cbdf01d7b..4d8d24482032 100644
@@ -103,6 +103,14 @@
 		__ret;                                                         \
 	})
 
+#define xsendmsg(fd, msg, flags)                                               \
+	({                                                                     \
+		ssize_t __ret = sendmsg((fd), (msg), (flags));                 \
+		if (__ret == -1)                                               \
+			FAIL_ERRNO("sendmsg");                                 \
+		__ret;                                                         \
+	})
+
 #define xrecv_nonblock(fd, buf, len, flags)                                    \
 	({                                                                     \
 		ssize_t __ret = recv_timeout((fd), (buf), (len), (flags),      \
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +1 -0
index 9ffe02f45808..e5e618e84950 100644
@@ -52,6 +52,50 @@ static void pop_simple_send(struct msg_test_opts *opts, int start, int len)
 	ASSERT_OK(cmp, "pop cmp end bytes failed");
 }
 
+static void pop_complex_send(struct msg_test_opts *opts, int start, int len)

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c static void pop_simple_send(struct msg_test_opts *opts, int start, int len) · line 65

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +10 -0
+{
+	struct test_sockmap_msg_helpers *skel = opts->skel;
+	char buf[] = "abcdefghijklmnopqrstuvwxyz";
+	size_t sent, recv, total = 0;
+	struct msghdr msg = {0};
+	struct iovec iov[15];
+	char *recvbuf;
+	int i;
+
+	for (i = 0; i < 15; i++) {

Always nice to use ARRAY_SIZE.

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c static void pop_simple_send(struct msg_test_opts *opts, int start, int len) · line 73

Quoted 8 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +8 -0
+		iov[i].iov_base = buf;
+		iov[i].iov_len = sizeof(buf);
+		total += sizeof(buf);
+	}
+
+	recvbuf = malloc(total);
+	if (!recvbuf)
+		FAIL("pop complex send malloc failure\n");

390 bytes, why not have it on stack?

Quoted 30 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +25 -0
+
+	msg.msg_iov = iov;
+	msg.msg_iovlen = 15;
+
+	skel->bss->pop = true;
+
+	if (start == -1)
+		start = sizeof(buf) - len - 1;
+
+	skel->bss->pop_start = start;
+	skel->bss->pop_len = len;
+
+	sent = xsendmsg(opts->client, &msg, 0);
+	if (sent != total)
+		FAIL("xsend failed");
+
+	ASSERT_OK(skel->bss->err, "pop error");
+
+	recv = xrecv_nonblock(opts->server, recvbuf, total, 0);
+	if (recv != sent - skel->bss->pop_len)
+		FAIL("Received incorrect number number of bytes after pop");
+
+	free(recvbuf);
+}
+
 static void test_sockmap_pop(void)
 {
 	struct msg_test_opts opts;
@@ -92,6 +136,15 @@ static void test_sockmap_pop(void)
 	/* Pop from end */

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c static void test_sockmap_pop(void) · line 146

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +8 -0
 	pop_simple_send(&opts, POP_END, 5);
 
+	/* Empty pop from start of sendmsg */
+	pop_complex_send(&opts, 0, 0);
+	/* Pop from start of sendmsg */
+	pop_complex_send(&opts, 0, 10);
+	/* Pop from middle of sendmsg */
+	pop_complex_send(&opts, 100, 10);
+	/* Pop from end of sendmsg */
+	pop_complex_send(&opts, 394, 10);

Isn't the start offset here past the end? 15*26=390?

Quoted 4 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +1 -0
+
 close_sockets:
 	close(client);
 	close(server);
#6 lore raw

Re: [PATCH bpf-next v2 2/4] bpf: sockmap, add a sendmsg test so we can check that path

Quoted 1 line · Jakub Sitnicki

On Wed, Jan 24, 2024 at 10:54 AM -08, John Fastabend wrote:

[...]

Quoted 4 lines
Quoted 4 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +0 -0
@@ -92,6 +136,15 @@ static void test_sockmap_pop(void)
 	/* Pop from end */
 	pop_simple_send(&opts, POP_END, 5);
 
Quoted 10 lines

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c static void test_sockmap_pop(void) · line 146

Quoted 8 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +8 -0
+	/* Empty pop from start of sendmsg */
+	pop_complex_send(&opts, 0, 0);
+	/* Pop from start of sendmsg */
+	pop_complex_send(&opts, 0, 10);
+	/* Pop from middle of sendmsg */
+	pop_complex_send(&opts, 100, 10);
+	/* Pop from end of sendmsg */
+	pop_complex_send(&opts, 394, 10);

Isn't the start offset here past the end? 15*26=390?

Plus terminating null bytes. Nevermind. I can't count.

#7 lore raw

Re: [PATCH bpf-next v2 2/4] bpf: sockmap, add a sendmsg test so we can check that path

In reply to Jakub Sitnicki
Quoted 38 lines · Jakub Sitnicki
Quoted 37 lines · John Fastabend

Sendmsg path with multiple buffers is slightly different from a single send in how we have to handle and walk the sg when doing pops. Lets ensure this walk is correct.

---
 .../bpf/prog_tests/sockmap_helpers.h          |  8 +++
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 53 +++++++++++++++++++
 2 files changed, 61 insertions(+)
tools/testing/selftests/bpf/prog_tests/sockmap_helpers.h +8 -0
index 781cbdf01d7b..4d8d24482032 100644
@@ -103,6 +103,14 @@
 		__ret;                                                         \
 	})
 
+#define xsendmsg(fd, msg, flags)                                               \
+	({                                                                     \
+		ssize_t __ret = sendmsg((fd), (msg), (flags));                 \
+		if (__ret == -1)                                               \
+			FAIL_ERRNO("sendmsg");                                 \
+		__ret;                                                         \
+	})
+
 #define xrecv_nonblock(fd, buf, len, flags)                                    \
 	({                                                                     \
 		ssize_t __ret = recv_timeout((fd), (buf), (len), (flags),      \
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +0 -0
index 9ffe02f45808..e5e618e84950 100644
@@ -52,6 +52,50 @@ static void pop_simple_send(struct msg_test_opts *opts, int start, int len)
 	ASSERT_OK(cmp, "pop cmp end bytes failed");
 }
 
Quoted 13 lines

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c static void pop_simple_send(struct msg_test_opts *opts, int start, int len) · line 65

Quoted 11 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +11 -0
+static void pop_complex_send(struct msg_test_opts *opts, int start, int len)
+{
+	struct test_sockmap_msg_helpers *skel = opts->skel;
+	char buf[] = "abcdefghijklmnopqrstuvwxyz";
+	size_t sent, recv, total = 0;
+	struct msghdr msg = {0};
+	struct iovec iov[15];
+	char *recvbuf;
+	int i;
+
+	for (i = 0; i < 15; i++) {

Always nice to use ARRAY_SIZE.

Agree will do.

Quoted 11 lines

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c static void pop_simple_send(struct msg_test_opts *opts, int start, int len) · line 73

Quoted 9 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +8 -0
+		iov[i].iov_base = buf;
+		iov[i].iov_len = sizeof(buf);
+		total += sizeof(buf);
+	}
+
+	recvbuf = malloc(total);
+	if (!recvbuf)
+		FAIL("pop complex send malloc failure\n");

390 bytes, why not have it on stack?

Sure one less thing that could fail seems like a win.

#8 lore raw

PATCH v2 3/4 bpf: sockmap, add a cork to force buffering of the scatterlist

In reply to John Fastabend

By using cork we can force multiple sends into a single scatterlist and test that first the cork gives us the correct number of bytes, but then also test the pop over the corked data.

---
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 81 +++++++++++++++++++
 .../bpf/progs/test_sockmap_msg_helpers.c      |  3 +
 2 files changed, 84 insertions(+)
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +81 -0
index e5e618e84950..8ced54fe1a0b 100644
@@ -21,6 +21,85 @@ struct msg_test_opts {
 
 #define POP_END -1
 
+static void cork_send(struct msg_test_opts *opts, int cork)
+{
+	struct test_sockmap_msg_helpers *skel = opts->skel;
+	char buf[] = "abcdefghijklmnopqrstuvwxyz";
+	size_t sent, total = 0, recv;
+	char *recvbuf;
+	int i;
+
+	skel->bss->pop = false;
+	skel->bss->cork = cork;
+
+	/* Send N bytes in 27B chunks */
+	for (i = 0; i < cork / sizeof(buf); i++) {
+		sent = xsend(opts->client, buf, sizeof(buf), 0);
+		if (sent < sizeof(buf))
+			FAIL("xsend failed");
+		total += sent;
+	}
+
+	recvbuf = malloc(total);
+	if (!recvbuf)
+		FAIL("cork send malloc failure\n");
+
+	ASSERT_OK(skel->bss->err, "cork error");
+	ASSERT_EQ(skel->bss->size, cork, "cork did not receive all bytes");
+
+	recv = xrecv_nonblock(opts->server, recvbuf, total, 0);
+	if (recv != total)
+		FAIL("Received incorrect number of bytes");
+
+	free(recvbuf);
+}
+
+static void test_sockmap_cork()
+{
+	struct test_sockmap_msg_helpers *skel;
+	struct msg_test_opts opts;
+	int s, client, server;
+	int err, map, prog;
+
+	skel = test_sockmap_msg_helpers__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "open_and_load"))
+		return;
+
+	map = bpf_map__fd(skel->maps.sock_map);
+	prog = bpf_program__fd(skel->progs.msg_helpers);
+	err = bpf_prog_attach(prog, map, BPF_SK_MSG_VERDICT, 0);
+	if (!ASSERT_OK(err, "bpf_prog_attach"))
+		goto out;
+
+	s = socket_loopback(AF_INET, SOCK_STREAM);
+	if (s < 0)
+		goto out;
+
+	err = create_pair(s, AF_INET, SOCK_STREAM, &client, &server);
+	if (err < 0)
+		goto close_loopback;
+
+	err = add_to_sockmap(map, client, server);
+	if (err < 0)
+		goto close_sockets;
+
+	opts.client = client;
+	opts.server = server;
+	opts.skel = skel;
+
+	/* Small cork */
+	cork_send(&opts, 54);
+	/* Full cork */
+	cork_send(&opts, 270);
+close_sockets:
+	close(client);
+	close(server);
+close_loopback:
+	close(s);
+out:
+	test_sockmap_msg_helpers__destroy(skel);
+}
+
 static void pop_simple_send(struct msg_test_opts *opts, int start, int len)
 {
 	struct test_sockmap_msg_helpers *skel = opts->skel;
@@ -260,4 +339,6 @@ void test_sockmap_msg_helpers(void)
 		test_sockmap_pop();
 	if (test__start_subtest("sockmap pop errors"))
 		test_sockmap_pop_errors();
+	if (test__start_subtest("sockmap cork"))
+		test_sockmap_cork();
 }
tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c +3 -0
index c721a00b6001..9622f154d016 100644
@@ -30,10 +30,13 @@ int pop_start = 0;
 int pop_len = 0;
 
 int err;
+int size;
 
 SEC("sk_msg")
 int msg_helpers(struct sk_msg_md *msg)
 {
+	size = msg->size;
+
 	if (cork)
 		err = bpf_msg_cork_bytes(msg, cork);
 
Signature · 1 line
2.33.0
#9 lore raw

Re: [PATCH bpf-next v2 3/4] bpf: sockmap, add a cork to force buffering of the scatterlist

Quoted 17 lines · John Fastabend

By using cork we can force multiple sends into a single scatterlist and test that first the cork gives us the correct number of bytes, but then also test the pop over the corked data.

---
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 81 +++++++++++++++++++
 .../bpf/progs/test_sockmap_msg_helpers.c      |  3 +
 2 files changed, 84 insertions(+)
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +0 -0
index e5e618e84950..8ced54fe1a0b 100644
@@ -21,6 +21,85 @@ struct msg_test_opts {
 
 #define POP_END -1

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c struct msg_test_opts { · line 32

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +9 -0
 
+static void cork_send(struct msg_test_opts *opts, int cork)
+{
+	struct test_sockmap_msg_helpers *skel = opts->skel;
+	char buf[] = "abcdefghijklmnopqrstuvwxyz";
+	size_t sent, total = 0, recv;
+	char *recvbuf;
+	int i;
+
+	skel->bss->pop = false;

Why reset it? Every subtest loads new program & creates new maps.

Quoted 23 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +23 -0
+	skel->bss->cork = cork;
+
+	/* Send N bytes in 27B chunks */
+	for (i = 0; i < cork / sizeof(buf); i++) {
+		sent = xsend(opts->client, buf, sizeof(buf), 0);
+		if (sent < sizeof(buf))
+			FAIL("xsend failed");
+		total += sent;
+	}
+
+	recvbuf = malloc(total);
+	if (!recvbuf)
+		FAIL("cork send malloc failure\n");
+
+	ASSERT_OK(skel->bss->err, "cork error");
+	ASSERT_EQ(skel->bss->size, cork, "cork did not receive all bytes");
+
+	recv = xrecv_nonblock(opts->server, recvbuf, total, 0);
+	if (recv != total)
+		FAIL("Received incorrect number of bytes");
+
+	free(recvbuf);
+}

[...]

#10 lore raw

Re: [PATCH bpf-next v2 3/4] bpf: sockmap, add a cork to force buffering of the scatterlist

Quoted 19 lines · Jakub Sitnicki
Quoted 18 lines · John Fastabend

By using cork we can force multiple sends into a single scatterlist and test that first the cork gives us the correct number of bytes, but then also test the pop over the corked data.

---
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 81 +++++++++++++++++++
 .../bpf/progs/test_sockmap_msg_helpers.c      |  3 +
 2 files changed, 84 insertions(+)
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +0 -0
index e5e618e84950..8ced54fe1a0b 100644
@@ -21,6 +21,85 @@ struct msg_test_opts {
 
 #define POP_END -1
 
Quoted 11 lines

tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c struct msg_test_opts { · line 32

Quoted 9 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +9 -0
+static void cork_send(struct msg_test_opts *opts, int cork)
+{
+	struct test_sockmap_msg_helpers *skel = opts->skel;
+	char buf[] = "abcdefghijklmnopqrstuvwxyz";
+	size_t sent, total = 0, recv;
+	char *recvbuf;
+	int i;
+
+	skel->bss->pop = false;

Why reset it? Every subtest loads new program & creates new maps.

Agree I'lld elete it.

Quoted 26 lines
Quoted 24 lines
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +23 -0
+	skel->bss->cork = cork;
+
+	/* Send N bytes in 27B chunks */
+	for (i = 0; i < cork / sizeof(buf); i++) {
+		sent = xsend(opts->client, buf, sizeof(buf), 0);
+		if (sent < sizeof(buf))
+			FAIL("xsend failed");
+		total += sent;
+	}
+
+	recvbuf = malloc(total);
+	if (!recvbuf)
+		FAIL("cork send malloc failure\n");
+
+	ASSERT_OK(skel->bss->err, "cork error");
+	ASSERT_EQ(skel->bss->size, cork, "cork did not receive all bytes");
+
+	recv = xrecv_nonblock(opts->server, recvbuf, total, 0);
+	if (recv != total)
+		FAIL("Received incorrect number of bytes");
+
+	free(recvbuf);
+}

[...]

#11 lore raw

PATCH v2 4/4 bpf: sockmap test cork and pop combined

In reply to John Fastabend

Its possible to cork data for some N bytes and then pop a some bytes off that scatterlist. Test combining cork and pop here.

---
 .../bpf/prog_tests/sockmap_msg_helpers.c      | 19 ++++++++++++++-----
 .../bpf/progs/test_sockmap_msg_helpers.c      | 14 +++++++++++++-
 2 files changed, 27 insertions(+), 6 deletions(-)
tools/testing/selftests/bpf/prog_tests/sockmap_msg_helpers.c +14 -5
index 8ced54fe1a0b..cfb965f6832f 100644
@@ -21,7 +21,7 @@ struct msg_test_opts {
 
 #define POP_END -1
 
-static void cork_send(struct msg_test_opts *opts, int cork)
+static void cork_send(struct msg_test_opts *opts, int cork, int start, int len)
 {
 	struct test_sockmap_msg_helpers *skel = opts->skel;
 	char buf[] = "abcdefghijklmnopqrstuvwxyz";
@@ -29,9 +29,12 @@ static void cork_send(struct msg_test_opts *opts, int cork)
 	char *recvbuf;
 	int i;
 
-	skel->bss->pop = false;
+	skel->bss->pop = !!len;
 	skel->bss->cork = cork;
 
+	skel->bss->pop_start = start;
+	skel->bss->pop_len = len;
+
 	/* Send N bytes in 27B chunks */
 	for (i = 0; i < cork / sizeof(buf); i++) {
 		sent = xsend(opts->client, buf, sizeof(buf), 0);
@@ -48,7 +51,7 @@ static void cork_send(struct msg_test_opts *opts, int cork)
 	ASSERT_EQ(skel->bss->size, cork, "cork did not receive all bytes");
 
 	recv = xrecv_nonblock(opts->server, recvbuf, total, 0);
-	if (recv != total)
+	if (recv != total - len)
 		FAIL("Received incorrect number of bytes");
 
 	free(recvbuf);
@@ -88,9 +91,15 @@ static void test_sockmap_cork()
 	opts.skel = skel;
 
 	/* Small cork */
-	cork_send(&opts, 54);
+	cork_send(&opts, 54, 0, 0);
 	/* Full cork */
-	cork_send(&opts, 270);
+	cork_send(&opts, 270, 0, 0);
+
+	/* Combine cork and pop small */
+	cork_send(&opts, 54, 0, 10);
+	/* Full cork and pop */
+	cork_send(&opts, 270, 200, 50);
+
 close_sockets:
 	close(client);
 	close(server);
tools/testing/selftests/bpf/progs/test_sockmap_msg_helpers.c +13 -1
index 9622f154d016..4c7e70367e35 100644
@@ -37,8 +37,19 @@ int msg_helpers(struct sk_msg_md *msg)
 {
 	size = msg->size;
 
-	if (cork)
+	/* If message is not yet fully cork'ed skip push, pull, pop */
+	if (cork && cork > msg->size) {
 		err = bpf_msg_cork_bytes(msg, cork);
+		goto out;
+	} else if (cork) {
+	/* If we previously corked the msg we need to clear the cork
+	 * otherwise next pop would cause datapath to wait for the
+	 * popped bytes to actually do the send.
+	 */
+		err = bpf_msg_cork_bytes(msg, 0);
+		if (err)
+			goto out;
+	}
 
 	if (pull)
 		err = bpf_msg_pull_data(msg, pull_start, pull_end, 0);
@@ -49,6 +60,7 @@ int msg_helpers(struct sk_msg_md *msg)
 	if (pop)
 		err = bpf_msg_pop_data(msg, pop_start, pop_len, 0);
 
+out:
 	return SK_PASS;
 }
 
Signature · 1 line
2.33.0
#12 lore raw
In reply to John Fastabend
Quoted 12 lines · John Fastabend

Its much easier to write and read tests than it was when sockmap was originally created. At that time we created a test_sockmap prog that did sockmap tests. But, its showing its age now. For example it reads user vars out of maps, is hard to run targetted tests, has a different format from the familiar test_progs and so on.

I recently thought there was an issue with pop helpers so I created some tests to try and track it down. It turns out it was a bug in the BPF program we had not the kernel. But, I think it makes sense to start deprecating test_sockmap and converting these to the nicer test_progs.

Quoted 8 lines

So this is a first round of test_prog tests for sockmap cork and pop helpers. I'll add push and pull tests shortly. I think its fine, maybe preferred to review smaller patchsets, to send these incrementally as I get them created.

Thanks!

v2: fix unint vars in some branches from make RELEASE=1

I'll wait a bit to see if there is any additional feedback, but on bpf-next these tests were stable. When we backported to 6.1 they became a bit flaky because recv() would sometimes only get part of the msg. I'll take a look, but this should be fine adding a retry logic to the recv() so it does a few recv's before giving up allows it to recv partial messages, but still pass the test.

Thanks,
John

#13 lore raw
In reply to John Fastabend
Quoted 12 lines · John Fastabend

Its much easier to write and read tests than it was when sockmap was originally created. At that time we created a test_sockmap prog that did sockmap tests. But, its showing its age now. For example it reads user vars out of maps, is hard to run targetted tests, has a different format from the familiar test_progs and so on.

I recently thought there was an issue with pop helpers so I created some tests to try and track it down. It turns out it was a bug in the BPF program we had not the kernel. But, I think it makes sense to start deprecating test_sockmap and converting these to the nicer test_progs.

Quoted 4 lines

So this is a first round of test_prog tests for sockmap cork and pop helpers. I'll add push and pull tests shortly. I think its fine, maybe preferred to review smaller patchsets, to send these incrementally as I get them created.

Cool to see this transition starting.

#14 lore raw
Quoted 13 lines · Jakub Sitnicki
Quoted 12 lines · John Fastabend

Its much easier to write and read tests than it was when sockmap was originally created. At that time we created a test_sockmap prog that did sockmap tests. But, its showing its age now. For example it reads user vars out of maps, is hard to run targetted tests, has a different format from the familiar test_progs and so on.

I recently thought there was an issue with pop helpers so I created some tests to try and track it down. It turns out it was a bug in the BPF program we had not the kernel. But, I think it makes sense to start deprecating test_sockmap and converting these to the nicer test_progs.

Quoted 6 lines
Quoted 4 lines

So this is a first round of test_prog tests for sockmap cork and pop helpers. I'll add push and pull tests shortly. I think its fine, maybe preferred to review smaller patchsets, to send these incrementally as I get them created.

Cool to see this transition starting.

Thanks for the review.