xfrm: Reinject transport-mode packets through workqueue
authorLiu Jian <liujian56@huawei.com>
Sat, 24 Sep 2022 08:01:57 +0000 (16:01 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 26 Oct 2022 10:34:46 +0000 (12:34 +0200)
commit7d98b26684cb2390729525b341ea099f0badbe18
tree94703440d6e437e41d6fcfdc46b24855253ebc33
parent397e880acf441dd15e4bdbb14b4c36d3f358afab
xfrm: Reinject transport-mode packets through workqueue

[ Upstream commit 4f4920669d21e1060b7243e5118dc3b71ced1276 ]

The following warning is displayed when the tcp6-multi-diffip11 stress
test case of the LTP test suite is tested:

watchdog: BUG: soft lockup - CPU#0 stuck for 22s! [ns-tcpserver:48198]
CPU: 0 PID: 48198 Comm: ns-tcpserver Kdump: loaded Not tainted 6.0.0-rc6+ #39
Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
pstate: 80400005 (Nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : des3_ede_encrypt+0x27c/0x460 [libdes]
lr : 0x3f
sp : ffff80000ceaa1b0
x29: ffff80000ceaa1b0 x28: ffff0000df056100 x27: ffff0000e51e5280
x26: ffff80004df75030 x25: ffff0000e51e4600 x24: 000000000000003b
x23: 0000000000802080 x22: 000000000000003d x21: 0000000000000038
x20: 0000000080000020 x19: 000000000000000a x18: 0000000000000033
x17: ffff0000e51e4780 x16: ffff80004e2d1448 x15: ffff80004e2d1248
x14: ffff0000e51e4680 x13: ffff80004e2d1348 x12: ffff80004e2d1548
x11: ffff80004e2d1848 x10: ffff80004e2d1648 x9 : ffff80004e2d1748
x8 : ffff80004e2d1948 x7 : 000000000bcaf83d x6 : 000000000000001b
x5 : ffff80004e2d1048 x4 : 00000000761bf3bf x3 : 000000007f1dd0a3
x2 : ffff0000e51e4780 x1 : ffff0000e3b9a2f8 x0 : 00000000db44e872
Call trace:
 des3_ede_encrypt+0x27c/0x460 [libdes]
 crypto_des3_ede_encrypt+0x1c/0x30 [des_generic]
 crypto_cbc_encrypt+0x148/0x190
 crypto_skcipher_encrypt+0x2c/0x40
 crypto_authenc_encrypt+0xc8/0xfc [authenc]
 crypto_aead_encrypt+0x2c/0x40
 echainiv_encrypt+0x144/0x1a0 [echainiv]
 crypto_aead_encrypt+0x2c/0x40
 esp6_output_tail+0x1c8/0x5d0 [esp6]
 esp6_output+0x120/0x278 [esp6]
 xfrm_output_one+0x458/0x4ec
 xfrm_output_resume+0x6c/0x1f0
 xfrm_output+0xac/0x4ac
 __xfrm6_output+0x130/0x270
 xfrm6_output+0x60/0xec
 ip6_xmit+0x2ec/0x5bc
 inet6_csk_xmit+0xbc/0x10c
 __tcp_transmit_skb+0x460/0x8c0
 tcp_write_xmit+0x348/0x890
 __tcp_push_pending_frames+0x44/0x110
 tcp_rcv_established+0x3c8/0x720
 tcp_v6_do_rcv+0xdc/0x4a0
 tcp_v6_rcv+0xc24/0xcb0
 ip6_protocol_deliver_rcu+0xf0/0x574
 ip6_input_finish+0x48/0x7c
 ip6_input+0x48/0xc0
 ip6_rcv_finish+0x80/0x9c
 xfrm_trans_reinject+0xb0/0xf4
 tasklet_action_common.constprop.0+0xf8/0x134
 tasklet_action+0x30/0x3c
 __do_softirq+0x128/0x368
 do_softirq+0xb4/0xc0
 __local_bh_enable_ip+0xb0/0xb4
 put_cpu_fpsimd_context+0x40/0x70
 kernel_neon_end+0x20/0x40
 sha1_base_do_update.constprop.0.isra.0+0x11c/0x140 [sha1_ce]
 sha1_ce_finup+0x94/0x110 [sha1_ce]
 crypto_shash_finup+0x34/0xc0
 hmac_finup+0x48/0xe0
 crypto_shash_finup+0x34/0xc0
 shash_digest_unaligned+0x74/0x90
 crypto_shash_digest+0x4c/0x9c
 shash_ahash_digest+0xc8/0xf0
 shash_async_digest+0x28/0x34
 crypto_ahash_digest+0x48/0xcc
 crypto_authenc_genicv+0x88/0xcc [authenc]
 crypto_authenc_encrypt+0xd8/0xfc [authenc]
 crypto_aead_encrypt+0x2c/0x40
 echainiv_encrypt+0x144/0x1a0 [echainiv]
 crypto_aead_encrypt+0x2c/0x40
 esp6_output_tail+0x1c8/0x5d0 [esp6]
 esp6_output+0x120/0x278 [esp6]
 xfrm_output_one+0x458/0x4ec
 xfrm_output_resume+0x6c/0x1f0
 xfrm_output+0xac/0x4ac
 __xfrm6_output+0x130/0x270
 xfrm6_output+0x60/0xec
 ip6_xmit+0x2ec/0x5bc
 inet6_csk_xmit+0xbc/0x10c
 __tcp_transmit_skb+0x460/0x8c0
 tcp_write_xmit+0x348/0x890
 __tcp_push_pending_frames+0x44/0x110
 tcp_push+0xb4/0x14c
 tcp_sendmsg_locked+0x71c/0xb64
 tcp_sendmsg+0x40/0x6c
 inet6_sendmsg+0x4c/0x80
 sock_sendmsg+0x5c/0x6c
 __sys_sendto+0x128/0x15c
 __arm64_sys_sendto+0x30/0x40
 invoke_syscall+0x50/0x120
 el0_svc_common.constprop.0+0x170/0x194
 do_el0_svc+0x38/0x4c
 el0_svc+0x28/0xe0
 el0t_64_sync_handler+0xbc/0x13c
 el0t_64_sync+0x180/0x184

Get softirq info by bcc tool:
./softirqs -NT 10
Tracing soft irq event time... Hit Ctrl-C to end.

15:34:34
SOFTIRQ          TOTAL_nsecs
block                 158990
timer               20030920
sched               46577080
net_rx             676746820
tasklet           9906067650

15:34:45
SOFTIRQ          TOTAL_nsecs
block                  86100
sched               38849790
net_rx             676532470
timer             1163848790
tasklet           9409019620

15:34:55
SOFTIRQ          TOTAL_nsecs
sched               58078450
net_rx             475156720
timer              533832410
tasklet           9431333300

The tasklet software interrupt takes too much time. Therefore, the
xfrm_trans_reinject executor is changed from tasklet to workqueue. Add add
spin lock to protect the queue. This reduces the processing flow of the
tcp_sendmsg function in this scenario.

Fixes: acf568ee859f0 ("xfrm: Reinject transport-mode packets through tasklet")
Signed-off-by: Liu Jian <liujian56@huawei.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
net/xfrm/xfrm_input.c