tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'selftests-net-packetdrill-netns-and-two-imports'
Willem de Bruijn says:
====================
selftests/net: packetdrill: netns and two imports
From: Willem de Bruijn <willemb@google.com>
1/3: run in nets, as discussed, and add missing CONFIGs
2/3: import tcp/zerocopy
3/3: import tcp/slow_start
====================
Link: https://patch.msgid.link/20240912005317.1253001-1-willemdebruijn.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
+1108
-2
+1
tools/testing/selftests/net/packetdrill/Makefile
+1
tools/testing/selftests/net/packetdrill/Makefile
+6
tools/testing/selftests/net/packetdrill/config
+6
tools/testing/selftests/net/packetdrill/config
+2
-2
tools/testing/selftests/net/packetdrill/ksft_runner.sh
+2
-2
tools/testing/selftests/net/packetdrill/ksft_runner.sh
···
33
33
ktap_print_header
34
34
ktap_set_plan 2
35
35
36
-
packetdrill ${ipv4_args[@]} $(basename $script) > /dev/null \
36
+
unshare -n packetdrill ${ipv4_args[@]} $(basename $script) > /dev/null \
37
37
&& ktap_test_pass "ipv4" || ktap_test_fail "ipv4"
38
-
packetdrill ${ipv6_args[@]} $(basename $script) > /dev/null \
38
+
unshare -n packetdrill ${ipv6_args[@]} $(basename $script) > /dev/null \
39
39
&& ktap_test_pass "ipv6" || ktap_test_fail "ipv6"
40
40
41
41
ktap_finished
+38
tools/testing/selftests/net/packetdrill/set_sysctls.py
+38
tools/testing/selftests/net/packetdrill/set_sysctls.py
···
1
+
#!/usr/bin/env python3
2
+
# SPDX-License-Identifier: GPL-2.0
3
+
4
+
"""Sets sysctl values and writes a file that restores them.
5
+
6
+
The arguments are of the form "<proc-file>=<val>" separated by spaces.
7
+
The program first reads the current value of the proc-file and creates
8
+
a shell script named "/tmp/sysctl_restore_${PACKETDRILL_PID}.sh" which
9
+
restores the values when executed. It then sets the new values.
10
+
11
+
PACKETDRILL_PID is set by packetdrill to the pid of itself, so a .pkt
12
+
file could restore sysctls by running `/tmp/sysctl_restore_${PPID}.sh`
13
+
at the end.
14
+
"""
15
+
16
+
import os
17
+
import subprocess
18
+
import sys
19
+
20
+
filename = '/tmp/sysctl_restore_%s.sh' % os.environ['PACKETDRILL_PID']
21
+
22
+
# Open file for restoring sysctl values
23
+
restore_file = open(filename, 'w')
24
+
print('#!/bin/bash', file=restore_file)
25
+
26
+
for a in sys.argv[1:]:
27
+
sysctl = a.split('=')
28
+
# sysctl[0] contains the proc-file name, sysctl[1] the new value
29
+
30
+
# read current value and add restore command to file
31
+
cur_val = subprocess.check_output(['cat', sysctl[0]], universal_newlines=True)
32
+
print('echo "%s" > %s' % (cur_val.strip(), sysctl[0]), file=restore_file)
33
+
34
+
# set new value
35
+
cmd = 'echo "%s" > %s' % (sysctl[1], sysctl[0])
36
+
os.system(cmd)
37
+
38
+
os.system('chmod u+x %s' % filename)
+56
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-1pkt.pkt
+56
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-1pkt.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start when not application-limited, so that
3
+
// the cwnd continues to grow.
4
+
// In this variant, the receiver ACKs every packet.
5
+
6
+
// Set up config. To keep things simple, disable the
7
+
// mechanism that defers sending in order to send bigger TSO packets.
8
+
`./defaults.sh
9
+
sysctl -q net.ipv4.tcp_tso_win_divisor=100`
10
+
11
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
12
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
13
+
+0 bind(3, ..., ...) = 0
14
+
+0 listen(3, 1) = 0
15
+
16
+
+.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
17
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
18
+
+.1 < . 1:1(0) ack 1 win 257
19
+
+0 accept(3, ..., ...) = 4
20
+
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
21
+
22
+
+0 write(4, ..., 30000) = 30000
23
+
+0 > P. 1:10001(10000) ack 1
24
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
25
+
26
+
+.105 < . 1:1(0) ack 1001 win 257
27
+
+0 > P. 10001:12001(2000) ack 1
28
+
29
+
+0 < . 1:1(0) ack 2001 win 257
30
+
+0 > P. 12001:14001(2000) ack 1
31
+
32
+
+.005 < . 1:1(0) ack 3001 win 257
33
+
+0 > P. 14001:16001(2000) ack 1
34
+
35
+
+0 < . 1:1(0) ack 4001 win 257
36
+
+0 > P. 16001:18001(2000) ack 1
37
+
38
+
+.005 < . 1:1(0) ack 5001 win 257
39
+
+0 > P. 18001:20001(2000) ack 1
40
+
41
+
+0 < . 1:1(0) ack 6001 win 257
42
+
+0 > P. 20001:22001(2000) ack 1
43
+
44
+
+.005 < . 1:1(0) ack 7001 win 257
45
+
+0 > P. 22001:24001(2000) ack 1
46
+
47
+
+0 < . 1:1(0) ack 8001 win 257
48
+
+0 > P. 24001:26001(2000) ack 1
49
+
50
+
+.005 < . 1:1(0) ack 9001 win 257
51
+
+0 > P. 26001:28001(2000) ack 1
52
+
53
+
+0 < . 1:1(0) ack 10001 win 257
54
+
+0 > P. 28001:30001(2000) ack 1
55
+
56
+
+0 %{ assert tcpi_snd_cwnd == 20, tcpi_snd_cwnd }%
+33
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-2pkt-send-5pkt.pkt
+33
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-2pkt-send-5pkt.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start when an outstanding flight of packets is
3
+
// less than the current cwnd, and not big enough to bump up cwnd.
4
+
//
5
+
// In this variant, the receiver ACKs every other packet,
6
+
// approximating standard delayed ACKs.
7
+
8
+
// Set up config.
9
+
`./defaults.sh`
10
+
11
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
12
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
13
+
+0 bind(3, ..., ...) = 0
14
+
+0 listen(3, 1) = 0
15
+
16
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
17
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
18
+
+0 < . 1:1(0) ack 1 win 257
19
+
+0 accept(3, ..., ...) = 4
20
+
21
+
// Only send 5 packets.
22
+
+0 write(4, ..., 5000) = 5000
23
+
+0 > P. 1:5001(5000) ack 1
24
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
25
+
26
+
+0 < . 1:1(0) ack 2001 win 257
27
+
+0 %{ assert tcpi_snd_cwnd == 10, 'cwnd=%d' % tcpi_snd_cwnd }%
28
+
29
+
+0 < . 1:1(0) ack 4001 win 257
30
+
+0 %{ assert tcpi_snd_cwnd == 10, 'cwnd=%d' % tcpi_snd_cwnd }%
31
+
32
+
+0 < . 1:1(0) ack 5001 win 257
33
+
+0 %{ assert tcpi_snd_cwnd == 10, 'cwnd=%d' % tcpi_snd_cwnd }%
+34
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-2pkt-send-6pkt.pkt
+34
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-2pkt-send-6pkt.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start when an outstanding flight of packets is
3
+
// less than the current cwnd, but still big enough that in slow
4
+
// start we want to increase our cwnd a little.
5
+
//
6
+
// In this variant, the receiver ACKs every other packet,
7
+
// approximating standard delayed ACKs.
8
+
9
+
// Set up config.
10
+
`./defaults.sh`
11
+
12
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
13
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
14
+
+0 bind(3, ..., ...) = 0
15
+
+0 listen(3, 1) = 0
16
+
17
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
18
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
19
+
+0 < . 1:1(0) ack 1 win 257
20
+
+0 accept(3, ..., ...) = 4
21
+
22
+
// Only send 6 packets.
23
+
+0 write(4, ..., 6000) = 6000
24
+
+0 > P. 1:6001(6000) ack 1
25
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
26
+
27
+
+0 < . 1:1(0) ack 2001 win 257
28
+
+0 %{ assert tcpi_snd_cwnd == 12, 'cwnd=%d' % tcpi_snd_cwnd }%
29
+
30
+
+0 < . 1:1(0) ack 4001 win 257
31
+
+0 %{ assert tcpi_snd_cwnd == 12, 'cwnd=%d' % tcpi_snd_cwnd }%
32
+
33
+
+0 < . 1:1(0) ack 6001 win 257
34
+
+0 %{ assert tcpi_snd_cwnd == 12, 'cwnd=%d' % tcpi_snd_cwnd }%
+42
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-2pkt.pkt
+42
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-2pkt.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start when not application-limited, so that
3
+
// the cwnd continues to grow.
4
+
// In this variant, the receiver ACKs every other packet,
5
+
// approximating standard delayed ACKs.
6
+
7
+
// Set up config. To keep things simple, disable the
8
+
// mechanism that defers sending in order to send bigger TSO packets.
9
+
`./defaults.sh
10
+
sysctl -q net.ipv4.tcp_tso_win_divisor=100`
11
+
12
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
13
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
14
+
+0 bind(3, ..., ...) = 0
15
+
+0 listen(3, 1) = 0
16
+
17
+
+.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
18
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
19
+
+.1 < . 1:1(0) ack 1 win 257
20
+
+0 accept(3, ..., ...) = 4
21
+
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
22
+
23
+
+0 write(4, ..., 30000) = 30000
24
+
+0 > P. 1:10001(10000) ack 1
25
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
26
+
27
+
+.105 < . 1:1(0) ack 2001 win 257
28
+
+0 > P. 10001:14001(4000) ack 1
29
+
30
+
+.005 < . 1:1(0) ack 4001 win 257
31
+
+0 > P. 14001:18001(4000) ack 1
32
+
33
+
+.005 < . 1:1(0) ack 6001 win 257
34
+
+0 > P. 18001:22001(4000) ack 1
35
+
36
+
+.005 < . 1:1(0) ack 8001 win 257
37
+
+0 > P. 22001:26001(4000) ack 1
38
+
39
+
+.005 < . 1:1(0) ack 10001 win 257
40
+
+0 > P. 26001:30001(4000) ack 1
41
+
42
+
+0 %{ assert tcpi_snd_cwnd == 20, tcpi_snd_cwnd }%
+35
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-4pkt.pkt
+35
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-ack-per-4pkt.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start when not application-limited, so that
3
+
// the cwnd continues to grow.
4
+
// In this variant, the receiver sends one ACK per 4 packets.
5
+
6
+
// Set up config. To keep things simple, disable the
7
+
// mechanism that defers sending in order to send bigger TSO packets.
8
+
`./defaults.sh
9
+
sysctl -q net.ipv4.tcp_tso_win_divisor=100`
10
+
11
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
12
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
13
+
+0 bind(3, ..., ...) = 0
14
+
+0 listen(3, 1) = 0
15
+
16
+
+.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
17
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
18
+
+.1 < . 1:1(0) ack 1 win 257
19
+
+0 accept(3, ..., ...) = 4
20
+
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
21
+
22
+
+0 write(4, ..., 30000) = 30000
23
+
+0 > P. 1:10001(10000) ack 1
24
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
25
+
26
+
+.11 < . 1:1(0) ack 4001 win 257
27
+
+0 > P. 10001:18001(8000) ack 1
28
+
29
+
+.01 < . 1:1(0) ack 8001 win 257
30
+
+0 > P. 18001:26001(8000) ack 1
31
+
32
+
+.005 < . 1:1(0) ack 10001 win 257
33
+
+0 > P. 26001:30001(4000) ack 1
34
+
35
+
+0 %{ assert tcpi_snd_cwnd == 20, tcpi_snd_cwnd }%
+39
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-after-idle.pkt
+39
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-after-idle.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start after idle
3
+
// This test expects tso size to be at least initial cwnd * mss
4
+
5
+
`./defaults.sh
6
+
./set_sysctls.py /proc/sys/net/ipv4/tcp_slow_start_after_idle=1 \
7
+
/proc/sys/net/ipv4/tcp_min_tso_segs=10`
8
+
9
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
10
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
11
+
+0 bind(3, ..., ...) = 0
12
+
+0 listen(3, 1) = 0
13
+
14
+
+0 < S 0:0(0) win 65535 <mss 1000,sackOK,nop,nop,nop,wscale 7>
15
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
16
+
+.1 < . 1:1(0) ack 1 win 511
17
+
+0 accept(3, ..., ...) = 4
18
+
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
19
+
20
+
+0 write(4, ..., 26000) = 26000
21
+
+0 > P. 1:5001(5000) ack 1
22
+
+0 > P. 5001:10001(5000) ack 1
23
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
24
+
25
+
+.1 < . 1:1(0) ack 10001 win 511
26
+
+0 %{ assert tcpi_snd_cwnd == 20, tcpi_snd_cwnd }%
27
+
+0 > P. 10001:20001(10000) ack 1
28
+
+0 > P. 20001:26001(6000) ack 1
29
+
30
+
+.1 < . 1:1(0) ack 26001 win 511
31
+
+0 %{ assert tcpi_snd_cwnd == 36, tcpi_snd_cwnd }%
32
+
33
+
+2 write(4, ..., 20000) = 20000
34
+
// If slow start after idle works properly, we should send 5 MSS here (cwnd/2)
35
+
+0 > P. 26001:31001(5000) ack 1
36
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
37
+
38
+
// Reset sysctls
39
+
`/tmp/sysctl_restore_${PPID}.sh`
+50
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-after-win-update.pkt
+50
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-after-win-update.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start after window update
3
+
// This test expects tso size to be at least initial cwnd * mss
4
+
5
+
`./defaults.sh
6
+
./set_sysctls.py /proc/sys/net/ipv4/tcp_slow_start_after_idle=1 \
7
+
/proc/sys/net/ipv4/tcp_min_tso_segs=10`
8
+
9
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
10
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
11
+
+0 bind(3, ..., ...) = 0
12
+
+0 listen(3, 1) = 0
13
+
14
+
+0 < S 0:0(0) win 65535 <mss 1000,sackOK,nop,nop,nop,wscale 7>
15
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
16
+
+.1 < . 1:1(0) ack 1 win 511
17
+
+0 accept(3, ..., ...) = 4
18
+
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
19
+
20
+
+0 write(4, ..., 26000) = 26000
21
+
+0 > P. 1:5001(5000) ack 1
22
+
+0 > P. 5001:10001(5000) ack 1
23
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
24
+
25
+
+.1 < . 1:1(0) ack 10001 win 511
26
+
+0 %{ assert tcpi_snd_cwnd == 20, tcpi_snd_cwnd }%
27
+
+0 > P. 10001:20001(10000) ack 1
28
+
+0 > P. 20001:26001(6000) ack 1
29
+
30
+
+.1 < . 1:1(0) ack 26001 win 0
31
+
+0 %{ assert tcpi_snd_cwnd == 36, tcpi_snd_cwnd }%
32
+
33
+
+0 write(4, ..., 20000) = 20000
34
+
// 1st win0 probe
35
+
+.3~+.310 > . 26000:26000(0) ack 1
36
+
+0 %{ assert tcpi_snd_cwnd == 36, tcpi_snd_cwnd }%
37
+
38
+
// 2nd win0 probe
39
+
+.6~+.620 > . 26000:26000(0) ack 1
40
+
+0 %{ assert tcpi_snd_cwnd == 36, tcpi_snd_cwnd }%
41
+
42
+
// 3rd win0 probe
43
+
+1.2~+1.240 > . 26000:26000(0) ack 1
44
+
+0 %{ assert tcpi_snd_cwnd == 36, tcpi_snd_cwnd }%
45
+
46
+
+.9 < . 1:1(0) ack 26001 win 511
47
+
+0 > P. 26001:31001(5000) ack 1
48
+
49
+
// Reset sysctls
50
+
`/tmp/sysctl_restore_${PPID}.sh`
+38
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-app-limited-9-packets-out.pkt
+38
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-app-limited-9-packets-out.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start when application-limited: in this case,
3
+
// with IW10, if we don't fully use our cwnd but instead
4
+
// send just 9 packets, then cwnd should grow to twice that
5
+
// value, or 18 packets.
6
+
7
+
// Set up config.
8
+
`./defaults.sh`
9
+
10
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
11
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
12
+
+0 bind(3, ..., ...) = 0
13
+
+0 listen(3, 1) = 0
14
+
15
+
+.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
16
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
17
+
+.1 < . 1:1(0) ack 1 win 257
18
+
+0 accept(3, ..., ...) = 4
19
+
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
20
+
21
+
+0 write(4, ..., 9000) = 9000
22
+
+0 > P. 1:9001(9000) ack 1
23
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
24
+
25
+
+.105 < . 1:1(0) ack 2001 win 257
26
+
+0 %{ assert tcpi_snd_cwnd == 12, tcpi_snd_cwnd }%
27
+
28
+
+.005 < . 1:1(0) ack 4001 win 257
29
+
+0 %{ assert tcpi_snd_cwnd == 14, tcpi_snd_cwnd }%
30
+
31
+
+.005 < . 1:1(0) ack 6001 win 257
32
+
+0 %{ assert tcpi_snd_cwnd == 16, tcpi_snd_cwnd }%
33
+
34
+
+.005 < . 1:1(0) ack 8001 win 257
35
+
+0 %{ assert tcpi_snd_cwnd == 18, tcpi_snd_cwnd }%
36
+
37
+
+.005 < . 1:1(0) ack 9001 win 257
38
+
+0 %{ assert tcpi_snd_cwnd == 18, tcpi_snd_cwnd }%
+36
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-app-limited.pkt
+36
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-app-limited.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start when application-limited: in this case,
3
+
// with IW10, if we send exactly 10 packets then cwnd should grow to 20.
4
+
5
+
// Set up config.
6
+
`./defaults.sh`
7
+
8
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
9
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
10
+
+0 bind(3, ..., ...) = 0
11
+
+0 listen(3, 1) = 0
12
+
13
+
+.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
14
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
15
+
+.1 < . 1:1(0) ack 1 win 257
16
+
+0 accept(3, ..., ...) = 4
17
+
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
18
+
19
+
+0 write(4, ..., 10000) = 10000
20
+
+0 > P. 1:10001(10000) ack 1
21
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
22
+
23
+
+.105 < . 1:1(0) ack 2001 win 257
24
+
+0 %{ assert tcpi_snd_cwnd == 12, tcpi_snd_cwnd }%
25
+
26
+
+.005 < . 1:1(0) ack 4001 win 257
27
+
+0 %{ assert tcpi_snd_cwnd == 14, tcpi_snd_cwnd }%
28
+
29
+
+.005 < . 1:1(0) ack 6001 win 257
30
+
+0 %{ assert tcpi_snd_cwnd == 16, tcpi_snd_cwnd }%
31
+
32
+
+.005 < . 1:1(0) ack 8001 win 257
33
+
+0 %{ assert tcpi_snd_cwnd == 18, tcpi_snd_cwnd }%
34
+
35
+
+.005 < . 1:1(0) ack 10001 win 257
36
+
+0 %{ assert tcpi_snd_cwnd == 20, tcpi_snd_cwnd }%
+63
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-fq-ack-per-2pkt.pkt
+63
tools/testing/selftests/net/packetdrill/tcp_slow_start_slow-start-fq-ack-per-2pkt.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Test of slow start when not application-limited, so that
3
+
// the cwnd continues to grow, even if TSQ triggers.
4
+
// In this variant, the receiver ACKs every other packet,
5
+
// approximating standard delayed ACKs.
6
+
7
+
// Note we use FQ/pacing to check if TCP Small Queues is not hurting
8
+
9
+
`./defaults.sh
10
+
tc qdisc replace dev tun0 root fq
11
+
sysctl -q net/ipv4/tcp_pacing_ss_ratio=200
12
+
sysctl -e -q net.ipv4.tcp_min_tso_segs=2`
13
+
14
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
15
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
16
+
+0 bind(3, ..., ...) = 0
17
+
+0 listen(3, 1) = 0
18
+
19
+
+.1 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
20
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
21
+
+.1 < . 1:1(0) ack 1 win 500
22
+
+0 accept(3, ..., ...) = 4
23
+
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
24
+
25
+
+0 write(4, ..., 40000) = 40000
26
+
// This might change if we cook the initial packet with 10 MSS.
27
+
+0 > P. 1:2921(2920) ack 1
28
+
+0 > P. 2921:5841(2920) ack 1
29
+
+0 > P. 5841:8761(2920) ack 1
30
+
+0 > P. 8761:11681(2920) ack 1
31
+
+0 > P. 11681:14601(2920) ack 1
32
+
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
33
+
34
+
+.105 < . 1:1(0) ack 2921 win 500
35
+
+0 %{ assert tcpi_snd_cwnd == 12, tcpi_snd_cwnd }%
36
+
37
+
// Note: after this commit : "net_sched: sch_fq: account for schedule/timers drifts"
38
+
// FQ notices that this packet missed the 'time to send next packet' computed
39
+
// when prior packet (11681:14601(2920)) was sent.
40
+
// So FQ will allow following packet to be sent a bit earlier (quantum/2)
41
+
// (FQ commit allows an application/cwnd limited flow to get at most quantum/2 extra credit)
42
+
+0 > P. 14601:17521(2920) ack 1
43
+
44
+
+.003 < . 1:1(0) ack 5841 win 500
45
+
+0 %{ assert tcpi_snd_cwnd == 14, tcpi_snd_cwnd }%
46
+
47
+
+.001 > P. 17521:20441(2920) ack 1
48
+
49
+
+.001 < . 1:1(0) ack 8761 win 500
50
+
+0 %{ assert tcpi_snd_cwnd == 16, tcpi_snd_cwnd }%
51
+
52
+
// remaining packets are delivered at a constant rate.
53
+
+.007 > P. 20441:23361(2920) ack 1
54
+
55
+
+.002 < . 1:1(0) ack 11681 win 500
56
+
+0 %{ assert tcpi_snd_cwnd == 18, tcpi_snd_cwnd }%
57
+
+.001 < . 1:1(0) ack 14601 win 500
58
+
59
+
+.004 > P. 23361:26281(2920) ack 1
60
+
61
+
+.007 > P. 26281:29201(2920) ack 1
62
+
63
+
+0 %{ assert tcpi_snd_cwnd == 20, 'cwnd=%d' % tcpi_snd_cwnd }%
+55
tools/testing/selftests/net/packetdrill/tcp_zerocopy_basic.pkt
+55
tools/testing/selftests/net/packetdrill/tcp_zerocopy_basic.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// basic zerocopy test:
3
+
//
4
+
// send a packet with MSG_ZEROCOPY and receive the notification ID
5
+
// repeat and verify IDs are consecutive
6
+
7
+
`./defaults.sh`
8
+
9
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
10
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
11
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
12
+
+0 bind(3, ..., ...) = 0
13
+
+0 listen(3, 1) = 0
14
+
15
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
16
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
17
+
+0 < . 1:1(0) ack 1 win 257
18
+
19
+
+0 accept(3, ..., ...) = 4
20
+
21
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
22
+
+0 > P. 1:4001(4000) ack 1
23
+
+0 < . 1:1(0) ack 4001 win 257
24
+
25
+
+0 recvmsg(4, {msg_name(...)=...,
26
+
msg_iov(1)=[{...,0}],
27
+
msg_flags=MSG_ERRQUEUE,
28
+
msg_control=[
29
+
{cmsg_level=CMSG_LEVEL_IP,
30
+
cmsg_type=CMSG_TYPE_RECVERR,
31
+
cmsg_data={ee_errno=0,
32
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
33
+
ee_type=0,
34
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
35
+
ee_info=0,
36
+
ee_data=0}}
37
+
]}, MSG_ERRQUEUE) = 0
38
+
39
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
40
+
+0 > P. 4001:8001(4000) ack 1
41
+
+0 < . 1:1(0) ack 8001 win 257
42
+
43
+
+0 recvmsg(4, {msg_name(...)=...,
44
+
msg_iov(1)=[{...,0}],
45
+
msg_flags=MSG_ERRQUEUE,
46
+
msg_control=[
47
+
{cmsg_level=CMSG_LEVEL_IP,
48
+
cmsg_type=CMSG_TYPE_RECVERR,
49
+
cmsg_data={ee_errno=0,
50
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
51
+
ee_type=0,
52
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
53
+
ee_info=1,
54
+
ee_data=1}}
55
+
]}, MSG_ERRQUEUE) = 0
+41
tools/testing/selftests/net/packetdrill/tcp_zerocopy_batch.pkt
+41
tools/testing/selftests/net/packetdrill/tcp_zerocopy_batch.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// batch zerocopy test:
3
+
//
4
+
// send multiple packets, then read one range of all notifications.
5
+
6
+
`./defaults.sh`
7
+
8
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
9
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
10
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
11
+
+0 bind(3, ..., ...) = 0
12
+
+0 listen(3, 1) = 0
13
+
14
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
15
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
16
+
+0 < . 1:1(0) ack 1 win 257
17
+
18
+
+0 accept(3, ..., ...) = 4
19
+
+0 setsockopt(4, SOL_SOCKET, SO_MARK, [666], 4) = 0
20
+
21
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
22
+
+0 > P. 1:4001(4000) ack 1
23
+
+0 < . 1:1(0) ack 4001 win 257
24
+
25
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
26
+
+0 > P. 4001:8001(4000) ack 1
27
+
+0 < . 1:1(0) ack 8001 win 257
28
+
29
+
+0 recvmsg(4, {msg_name(...)=...,
30
+
msg_iov(1)=[{...,0}],
31
+
msg_flags=MSG_ERRQUEUE,
32
+
msg_control=[
33
+
{cmsg_level=CMSG_LEVEL_IP,
34
+
cmsg_type=CMSG_TYPE_RECVERR,
35
+
cmsg_data={ee_errno=0,
36
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
37
+
ee_type=0,
38
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
39
+
ee_info=0,
40
+
ee_data=1}}
41
+
]}, MSG_ERRQUEUE) = 0
+30
tools/testing/selftests/net/packetdrill/tcp_zerocopy_client.pkt
+30
tools/testing/selftests/net/packetdrill/tcp_zerocopy_client.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Minimal client-side zerocopy test
3
+
4
+
`./defaults.sh`
5
+
6
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 4
7
+
+0 setsockopt(4, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
8
+
+0...0 connect(4, ..., ...) = 0
9
+
10
+
+0 > S 0:0(0) <mss 1460,sackOK,TS val 0 ecr 0,nop,wscale 8>
11
+
+0 < S. 0:0(0) ack 1 win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
12
+
+0 > . 1:1(0) ack 1
13
+
14
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
15
+
+0 > P. 1:4001(4000) ack 1
16
+
+0 < . 1:1(0) ack 4001 win 257
17
+
18
+
+0 recvmsg(4, {msg_name(...)=...,
19
+
msg_iov(1)=[{...,0}],
20
+
msg_flags=MSG_ERRQUEUE,
21
+
msg_control=[
22
+
{cmsg_level=CMSG_LEVEL_IP,
23
+
cmsg_type=CMSG_TYPE_RECVERR,
24
+
cmsg_data={ee_errno=0,
25
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
26
+
ee_type=0,
27
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
28
+
ee_info=0,
29
+
ee_data=0}}
30
+
]}, MSG_ERRQUEUE) = 0
+44
tools/testing/selftests/net/packetdrill/tcp_zerocopy_closed.pkt
+44
tools/testing/selftests/net/packetdrill/tcp_zerocopy_closed.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// send with MSG_ZEROCOPY on a non-established socket
3
+
//
4
+
// verify that a send in state TCP_CLOSE correctly aborts the zerocopy
5
+
// operation, specifically it does not increment the zerocopy counter.
6
+
//
7
+
// First send on a closed socket and wait for (absent) notification.
8
+
// Then connect and send and verify that notification nr. is zero.
9
+
10
+
`./defaults.sh`
11
+
12
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 4
13
+
+0 setsockopt(4, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
14
+
15
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = -1 EPIPE (Broken pipe)
16
+
17
+
+0.1 recvmsg(4, {msg_name(...)=...,
18
+
msg_iov(1)=[{...,0}],
19
+
msg_flags=MSG_ERRQUEUE,
20
+
msg_control=[]}, MSG_ERRQUEUE) = -1 EAGAIN (Resource temporarily unavailable)
21
+
22
+
+0...0 connect(4, ..., ...) = 0
23
+
24
+
+0 > S 0:0(0) <mss 1460,sackOK,TS val 0 ecr 0,nop,wscale 8>
25
+
+0 < S. 0:0(0) ack 1 win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
26
+
+0 > . 1:1(0) ack 1
27
+
28
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
29
+
+0 > P. 1:4001(4000) ack 1
30
+
+0 < . 1:1(0) ack 4001 win 257
31
+
32
+
+0 recvmsg(4, {msg_name(...)=...,
33
+
msg_iov(1)=[{...,0}],
34
+
msg_flags=MSG_ERRQUEUE,
35
+
msg_control=[
36
+
{cmsg_level=CMSG_LEVEL_IP,
37
+
cmsg_type=CMSG_TYPE_RECVERR,
38
+
cmsg_data={ee_errno=0,
39
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
40
+
ee_type=0,
41
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
42
+
ee_info=0,
43
+
ee_data=0}}
44
+
]}, MSG_ERRQUEUE) = 0
+61
tools/testing/selftests/net/packetdrill/tcp_zerocopy_epoll_edge.pkt
+61
tools/testing/selftests/net/packetdrill/tcp_zerocopy_epoll_edge.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// epoll zerocopy test:
3
+
//
4
+
// EPOLLERR is known to be not edge-triggered unlike EPOLLIN and EPOLLOUT but
5
+
// it is not level-triggered either.
6
+
//
7
+
// fire two sends with MSG_ZEROCOPY and receive the acks. confirm that EPOLLERR
8
+
// is correctly fired only once, when EPOLLET is set. send another packet with
9
+
// MSG_ZEROCOPY. confirm that EPOLLERR is correctly fired again only once.
10
+
`./defaults.sh`
11
+
12
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
13
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
14
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
15
+
+0 bind(3, ..., ...) = 0
16
+
+0 listen(3, 1) = 0
17
+
18
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
19
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
20
+
+0 < . 1:1(0) ack 1 win 257
21
+
22
+
+0 accept(3, ..., ...) = 4
23
+
24
+
+0 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
25
+
+0 epoll_create(1) = 5
26
+
+0 epoll_ctl(5, EPOLL_CTL_ADD, 4, {events=EPOLLOUT|EPOLLET, fd=4}) = 0
27
+
+0 epoll_wait(5, {events=EPOLLOUT, fd=4}, 1, 0) = 1
28
+
29
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
30
+
+0 > P. 1:4001(4000) ack 1
31
+
+0 < . 1:1(0) ack 4001 win 257
32
+
33
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
34
+
+0 > P. 4001:8001(4000) ack 1
35
+
+0 < . 1:1(0) ack 8001 win 257
36
+
37
+
// receive only one EPOLLERR for the two sends above.
38
+
+0 epoll_wait(5, {events=EPOLLERR|EPOLLOUT, fd=4}, 1, 0) = 1
39
+
+0 epoll_wait(5, {events=0, ptr=0}, 1, 0) = 0
40
+
41
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
42
+
+0 > P. 8001:12001(4000) ack 1
43
+
+0 < . 1:1(0) ack 12001 win 257
44
+
45
+
// receive only one EPOLLERR for the third send above.
46
+
+0 epoll_wait(5, {events=EPOLLERR|EPOLLOUT, fd=4}, 1, 0) = 1
47
+
+0 epoll_wait(5, {events=0, ptr=0}, 1, 0) = 0
48
+
49
+
+0 recvmsg(4, {msg_name(...)=...,
50
+
msg_iov(1)=[{...,0}],
51
+
msg_flags=MSG_ERRQUEUE,
52
+
msg_control=[
53
+
{cmsg_level=CMSG_LEVEL_IP,
54
+
cmsg_type=CMSG_TYPE_RECVERR,
55
+
cmsg_data={ee_errno=0,
56
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
57
+
ee_type=0,
58
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
59
+
ee_info=0,
60
+
ee_data=2}}
61
+
]}, MSG_ERRQUEUE) = 0
+63
tools/testing/selftests/net/packetdrill/tcp_zerocopy_epoll_exclusive.pkt
+63
tools/testing/selftests/net/packetdrill/tcp_zerocopy_epoll_exclusive.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// epoll zerocopy test:
3
+
//
4
+
// EPOLLERR is known to be not edge-triggered unlike EPOLLIN and EPOLLOUT but
5
+
// it is not level-triggered either. this tests verify that the same behavior is
6
+
// maintained when we have EPOLLEXCLUSIVE.
7
+
//
8
+
// fire two sends with MSG_ZEROCOPY and receive the acks. confirm that EPOLLERR
9
+
// is correctly fired only once, when EPOLLET is set. send another packet with
10
+
// MSG_ZEROCOPY. confirm that EPOLLERR is correctly fired again only once.
11
+
`./defaults.sh`
12
+
13
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
14
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
15
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
16
+
+0 bind(3, ..., ...) = 0
17
+
+0 listen(3, 1) = 0
18
+
19
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
20
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
21
+
+0 < . 1:1(0) ack 1 win 257
22
+
23
+
+0 accept(3, ..., ...) = 4
24
+
25
+
+0 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
26
+
+0 epoll_create(1) = 5
27
+
+0 epoll_ctl(5, EPOLL_CTL_ADD, 4,
28
+
{events=EPOLLOUT|EPOLLET|EPOLLEXCLUSIVE, fd=4}) = 0
29
+
+0 epoll_wait(5, {events=EPOLLOUT, fd=4}, 1, 0) = 1
30
+
31
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
32
+
+0 > P. 1:4001(4000) ack 1
33
+
+0 < . 1:1(0) ack 4001 win 257
34
+
35
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
36
+
+0 > P. 4001:8001(4000) ack 1
37
+
+0 < . 1:1(0) ack 8001 win 257
38
+
39
+
// receive only one EPOLLERR for the two sends above.
40
+
+0 epoll_wait(5, {events=EPOLLERR|EPOLLOUT, fd=4}, 1, 0) = 1
41
+
+0 epoll_wait(5, {events=0, ptr=0}, 1, 0) = 0
42
+
43
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
44
+
+0 > P. 8001:12001(4000) ack 1
45
+
+0 < . 1:1(0) ack 12001 win 257
46
+
47
+
// receive only one EPOLLERR for the third send above.
48
+
+0 epoll_wait(5, {events=EPOLLERR|EPOLLOUT, fd=4}, 1, 0) = 1
49
+
+0 epoll_wait(5, {events=0, ptr=0}, 1, 0) = 0
50
+
51
+
+0 recvmsg(4, {msg_name(...)=...,
52
+
msg_iov(1)=[{...,0}],
53
+
msg_flags=MSG_ERRQUEUE,
54
+
msg_control=[
55
+
{cmsg_level=CMSG_LEVEL_IP,
56
+
cmsg_type=CMSG_TYPE_RECVERR,
57
+
cmsg_data={ee_errno=0,
58
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
59
+
ee_type=0,
60
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
61
+
ee_info=0,
62
+
ee_data=2}}
63
+
]}, MSG_ERRQUEUE) = 0
+66
tools/testing/selftests/net/packetdrill/tcp_zerocopy_epoll_oneshot.pkt
+66
tools/testing/selftests/net/packetdrill/tcp_zerocopy_epoll_oneshot.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// epoll zerocopy test:
3
+
//
4
+
// This is a test to confirm that EPOLLERR is only fired once for an FD when
5
+
// EPOLLONESHOT is set.
6
+
//
7
+
// fire two sends with MSG_ZEROCOPY and receive the acks. confirm that EPOLLERR
8
+
// is correctly fired only once, when EPOLLONESHOT is set. send another packet
9
+
// with MSG_ZEROCOPY. confirm that EPOLLERR is not fired. Rearm the FD and
10
+
// confirm that EPOLLERR is correctly set.
11
+
`./defaults.sh`
12
+
13
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
14
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
15
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
16
+
+0 bind(3, ..., ...) = 0
17
+
+0 listen(3, 1) = 0
18
+
19
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
20
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
21
+
+0 < . 1:1(0) ack 1 win 257
22
+
23
+
+0 accept(3, ..., ...) = 4
24
+
25
+
+0 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
26
+
+0 epoll_create(1) = 5
27
+
+0 epoll_ctl(5, EPOLL_CTL_ADD, 4,
28
+
{events=EPOLLOUT|EPOLLET|EPOLLONESHOT, fd=4}) = 0
29
+
30
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
31
+
+0 > P. 1:4001(4000) ack 1
32
+
+0 < . 1:1(0) ack 4001 win 257
33
+
34
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
35
+
+0 > P. 4001:8001(4000) ack 1
36
+
+0 < . 1:1(0) ack 8001 win 257
37
+
38
+
// receive only one EPOLLERR for the two sends above.
39
+
+0 epoll_wait(5, {events=EPOLLERR|EPOLLOUT, fd=4}, 1, 0) = 1
40
+
+0 epoll_wait(5, {events=0, ptr=0}, 1, 0) = 0
41
+
42
+
+0 send(4, ..., 4000, MSG_ZEROCOPY) = 4000
43
+
+0 > P. 8001:12001(4000) ack 1
44
+
+0 < . 1:1(0) ack 12001 win 257
45
+
46
+
// receive no EPOLLERR for the third send above.
47
+
+0 epoll_wait(5, {events=0, ptr=0}, 1, 0) = 0
48
+
49
+
// rearm the FD and verify the EPOLLERR is fired again.
50
+
+0 epoll_ctl(5, EPOLL_CTL_MOD, 4, {events=EPOLLOUT|EPOLLONESHOT, fd=4}) = 0
51
+
+0 epoll_wait(5, {events=EPOLLERR|EPOLLOUT, fd=4}, 1, 0) = 1
52
+
+0 epoll_wait(5, {events=0, ptr=0}, 1, 0) = 0
53
+
54
+
+0 recvmsg(4, {msg_name(...)=...,
55
+
msg_iov(1)=[{...,0}],
56
+
msg_flags=MSG_ERRQUEUE,
57
+
msg_control=[
58
+
{cmsg_level=CMSG_LEVEL_IP,
59
+
cmsg_type=CMSG_TYPE_RECVERR,
60
+
cmsg_data={ee_errno=0,
61
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
62
+
ee_type=0,
63
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
64
+
ee_info=0,
65
+
ee_data=2}}
66
+
]}, MSG_ERRQUEUE) = 0
+56
tools/testing/selftests/net/packetdrill/tcp_zerocopy_fastopen-client.pkt
+56
tools/testing/selftests/net/packetdrill/tcp_zerocopy_fastopen-client.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Fastopen client zerocopy test:
3
+
//
4
+
// send data with MSG_FASTOPEN | MSG_ZEROCOPY and verify that the
5
+
// kernel returns the notification ID.
6
+
//
7
+
// Fastopen requires a stored cookie. Create two sockets. The first
8
+
// one will have no data in the initial send. On return 0 the
9
+
// zerocopy notification counter is not incremented. Verify this too.
10
+
11
+
`./defaults.sh`
12
+
13
+
// Send a FastOpen request, no cookie yet so no data in SYN
14
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
15
+
+0 fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK) = 0
16
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
17
+
+0 sendto(3, ..., 500, MSG_FASTOPEN|MSG_ZEROCOPY, ..., ...) = -1 EINPROGRESS (Operation now in progress)
18
+
+0 > S 0:0(0) <mss 1460,sackOK,TS val 1000 ecr 0,nop,wscale 8,FO,nop,nop>
19
+
+.01 < S. 123:123(0) ack 1 win 14600 <mss 940,TS val 2000 ecr 1000,sackOK,nop,wscale 6, FO abcd1234,nop,nop>
20
+
+0 > . 1:1(0) ack 1 <nop,nop,TS val 1001 ecr 2000>
21
+
22
+
// Read from error queue: no zerocopy notification
23
+
+1 recvmsg(3, {msg_name(...)=...,
24
+
msg_iov(1)=[{...,0}],
25
+
msg_flags=MSG_ERRQUEUE,
26
+
msg_control=[]}, MSG_ERRQUEUE) = -1 EAGAIN (Resource temporarily unavailable)
27
+
28
+
+.01 close(3) = 0
29
+
+0 > F. 1:1(0) ack 1 <nop,nop,TS val 1002 ecr 2000>
30
+
+.01 < F. 1:1(0) ack 2 win 92 <nop,nop,TS val 2001 ecr 1002>
31
+
+0 > . 2:2(0) ack 2 <nop,nop,TS val 1003 ecr 2001>
32
+
33
+
// Send another Fastopen request, now SYN will have data
34
+
+.07 `sysctl -q net.ipv4.tcp_timestamps=0`
35
+
+.1 socket(..., SOCK_STREAM, IPPROTO_TCP) = 5
36
+
+0 fcntl(5, F_SETFL, O_RDWR|O_NONBLOCK) = 0
37
+
+0 setsockopt(5, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
38
+
+0 sendto(5, ..., 500, MSG_FASTOPEN|MSG_ZEROCOPY, ..., ...) = 500
39
+
+0 > S 0:500(500) <mss 1460,nop,nop,sackOK,nop,wscale 8,FO abcd1234,nop,nop>
40
+
+.05 < S. 5678:5678(0) ack 501 win 14600 <mss 1460,nop,nop,sackOK,nop,wscale 6>
41
+
+0 > . 501:501(0) ack 1
42
+
43
+
// Read from error queue: now has first zerocopy notification
44
+
+0.5 recvmsg(5, {msg_name(...)=...,
45
+
msg_iov(1)=[{...,0}],
46
+
msg_flags=MSG_ERRQUEUE,
47
+
msg_control=[
48
+
{cmsg_level=CMSG_LEVEL_IP,
49
+
cmsg_type=CMSG_TYPE_RECVERR,
50
+
cmsg_data={ee_errno=0,
51
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
52
+
ee_type=0,
53
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
54
+
ee_info=0,
55
+
ee_data=0}}
56
+
]}, MSG_ERRQUEUE) = 0
+44
tools/testing/selftests/net/packetdrill/tcp_zerocopy_fastopen-server.pkt
+44
tools/testing/selftests/net/packetdrill/tcp_zerocopy_fastopen-server.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// Fastopen server zerocopy test:
3
+
//
4
+
// send data with MSG_FASTOPEN | MSG_ZEROCOPY and verify that the
5
+
// kernel returns the notification ID.
6
+
7
+
`./defaults.sh
8
+
./set_sysctls.py /proc/sys/net/ipv4/tcp_fastopen=0x207`
9
+
10
+
// Set up a TFO server listening socket.
11
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
12
+
+.1 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
13
+
+0 bind(3, ..., ...) = 0
14
+
+0 listen(3, 1) = 0
15
+
+0 setsockopt(3, SOL_TCP, TCP_FASTOPEN, [2], 4) = 0
16
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
17
+
18
+
// Client sends a SYN with data.
19
+
+.1 < S 0:1000(1000) win 32792 <mss 1460,sackOK,nop,nop>
20
+
+0 > S. 0:0(0) ack 1001 <mss 1460,nop,nop,sackOK>
21
+
22
+
// Server accepts and replies with data.
23
+
+.005 accept(3, ..., ...) = 4
24
+
+0 read(4, ..., 1024) = 1000
25
+
+0 sendto(4, ..., 1000, MSG_ZEROCOPY, ..., ...) = 1000
26
+
+0 > P. 1:1001(1000) ack 1001
27
+
+.05 < . 1001:1001(0) ack 1001 win 32792
28
+
29
+
// Read from error queue: now has first zerocopy notification
30
+
+0.1 recvmsg(4, {msg_name(...)=...,
31
+
msg_iov(1)=[{...,0}],
32
+
msg_flags=MSG_ERRQUEUE,
33
+
msg_control=[
34
+
{cmsg_level=CMSG_LEVEL_IP,
35
+
cmsg_type=CMSG_TYPE_RECVERR,
36
+
cmsg_data={ee_errno=0,
37
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
38
+
ee_type=0,
39
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
40
+
ee_info=0,
41
+
ee_data=0}}
42
+
]}, MSG_ERRQUEUE) = 0
43
+
44
+
`/tmp/sysctl_restore_${PPID}.sh`
+118
tools/testing/selftests/net/packetdrill/tcp_zerocopy_maxfrags.pkt
+118
tools/testing/selftests/net/packetdrill/tcp_zerocopy_maxfrags.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// tcp_MAX_SKB_FRAGS test
3
+
//
4
+
// Verify that sending an iovec of tcp_MAX_SKB_FRAGS + 1 elements will
5
+
// 1) fit in a single packet without zerocopy
6
+
// 2) spill over into a second packet with zerocopy,
7
+
// because each iovec element becomes a frag
8
+
// 3) the PSH bit is set on an skb when it runs out of fragments
9
+
10
+
`./defaults.sh`
11
+
12
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
13
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
14
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
15
+
16
+
// Each pinned zerocopy page is fully accounted to skb->truesize.
17
+
// This test generates a worst case packet with each frag storing
18
+
// one byte, but increasing truesize with a page (64KB on PPC).
19
+
+0 setsockopt(3, SOL_SOCKET, SO_SNDBUF, [2000000], 4) = 0
20
+
21
+
+0 bind(3, ..., ...) = 0
22
+
+0 listen(3, 1) = 0
23
+
24
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
25
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
26
+
+0 < . 1:1(0) ack 1 win 257
27
+
28
+
+0 accept(3, ..., ...) = 4
29
+
30
+
// send an iov of 18 elements: just becomes a linear skb
31
+
+0 sendmsg(4, {msg_name(...)=...,
32
+
msg_iov(18)=[{..., 1}, {..., 1}, {..., 1}, {..., 1},
33
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
34
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
35
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
36
+
{..., 1}, {..., 1}],
37
+
msg_flags=0}, 0) = 18
38
+
39
+
+0 > P. 1:19(18) ack 1
40
+
+0 < . 1:1(0) ack 19 win 257
41
+
42
+
// send a zerocopy iov of 18 elements:
43
+
+1 sendmsg(4, {msg_name(...)=...,
44
+
msg_iov(18)=[{..., 1}, {..., 1}, {..., 1}, {..., 1},
45
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
46
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
47
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
48
+
{..., 1}, {..., 1}],
49
+
msg_flags=0}, MSG_ZEROCOPY) = 18
50
+
51
+
// verify that it is split in one skb of 17 frags + 1 of 1 frag
52
+
// verify that both have the PSH bit set
53
+
+0 > P. 19:36(17) ack 1
54
+
+0 < . 1:1(0) ack 36 win 257
55
+
56
+
+0 > P. 36:37(1) ack 1
57
+
+0 < . 1:1(0) ack 37 win 257
58
+
59
+
+1 recvmsg(4, {msg_name(...)=...,
60
+
msg_iov(1)=[{...,0}],
61
+
msg_flags=MSG_ERRQUEUE,
62
+
msg_control=[
63
+
{cmsg_level=CMSG_LEVEL_IP,
64
+
cmsg_type=CMSG_TYPE_RECVERR,
65
+
cmsg_data={ee_errno=0,
66
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
67
+
ee_type=0,
68
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
69
+
ee_info=0,
70
+
ee_data=0}}
71
+
]}, MSG_ERRQUEUE) = 0
72
+
73
+
// send a zerocopy iov of 64 elements:
74
+
+0 sendmsg(4, {msg_name(...)=...,
75
+
msg_iov(64)=[{..., 1}, {..., 1}, {..., 1}, {..., 1},
76
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
77
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
78
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
79
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
80
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
81
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
82
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
83
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
84
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
85
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
86
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
87
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
88
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
89
+
{..., 1}, {..., 1}, {..., 1}, {..., 1},
90
+
{..., 1}, {..., 1}, {..., 1}, {..., 1}],
91
+
msg_flags=0}, MSG_ZEROCOPY) = 64
92
+
93
+
// verify that it is split in skbs with 17 frags
94
+
+0 > P. 37:54(17) ack 1
95
+
+0 < . 1:1(0) ack 54 win 257
96
+
97
+
+0 > P. 54:71(17) ack 1
98
+
+0 < . 1:1(0) ack 71 win 257
99
+
100
+
+0 > P. 71:88(17) ack 1
101
+
+0 < . 1:1(0) ack 88 win 257
102
+
103
+
+0 > P. 88:101(13) ack 1
104
+
+0 < . 1:1(0) ack 101 win 257
105
+
106
+
+1 recvmsg(4, {msg_name(...)=...,
107
+
msg_iov(1)=[{...,0}],
108
+
msg_flags=MSG_ERRQUEUE,
109
+
msg_control=[
110
+
{cmsg_level=CMSG_LEVEL_IP,
111
+
cmsg_type=CMSG_TYPE_RECVERR,
112
+
cmsg_data={ee_errno=0,
113
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
114
+
ee_type=0,
115
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
116
+
ee_info=1,
117
+
ee_data=1}}
118
+
]}, MSG_ERRQUEUE) = 0
+57
tools/testing/selftests/net/packetdrill/tcp_zerocopy_small.pkt
+57
tools/testing/selftests/net/packetdrill/tcp_zerocopy_small.pkt
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
// small packet zerocopy test:
3
+
//
4
+
// verify that SO_EE_CODE_ZEROCOPY_COPIED is set on zerocopy
5
+
// packets of all sizes, including the smallest payload, 1B.
6
+
7
+
`./defaults.sh`
8
+
9
+
0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
10
+
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
11
+
+0 setsockopt(3, SOL_SOCKET, SO_ZEROCOPY, [1], 4) = 0
12
+
+0 bind(3, ..., ...) = 0
13
+
+0 listen(3, 1) = 0
14
+
15
+
+0 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
16
+
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
17
+
+0 < . 1:1(0) ack 1 win 257
18
+
19
+
+0 accept(3, ..., ...) = 4
20
+
21
+
// send 1B
22
+
+0 send(4, ..., 1, MSG_ZEROCOPY) = 1
23
+
+0 > P. 1:2(1) ack 1
24
+
+0 < . 1:1(0) ack 2 win 257
25
+
26
+
+1 recvmsg(4, {msg_name(...)=...,
27
+
msg_iov(1)=[{...,0}],
28
+
msg_flags=MSG_ERRQUEUE,
29
+
msg_control=[
30
+
{cmsg_level=CMSG_LEVEL_IP,
31
+
cmsg_type=CMSG_TYPE_RECVERR,
32
+
cmsg_data={ee_errno=0,
33
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
34
+
ee_type=0,
35
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
36
+
ee_info=0,
37
+
ee_data=0}}
38
+
]}, MSG_ERRQUEUE) = 0
39
+
40
+
// send 1B again
41
+
+0 send(4, ..., 1, MSG_ZEROCOPY) = 1
42
+
+0 > P. 2:3(1) ack 1
43
+
+0 < . 1:1(0) ack 3 win 257
44
+
45
+
+1 recvmsg(4, {msg_name(...)=...,
46
+
msg_iov(1)=[{...,0}],
47
+
msg_flags=MSG_ERRQUEUE,
48
+
msg_control=[
49
+
{cmsg_level=CMSG_LEVEL_IP,
50
+
cmsg_type=CMSG_TYPE_RECVERR,
51
+
cmsg_data={ee_errno=0,
52
+
ee_origin=SO_EE_ORIGIN_ZEROCOPY,
53
+
ee_type=0,
54
+
ee_code=SO_EE_CODE_ZEROCOPY_COPIED,
55
+
ee_info=1,
56
+
ee_data=1}}
57
+
]}, MSG_ERRQUEUE) = 0