net/smc: Send directly when TCP_CORK is cleared
authorTony Lu <tonylu@linux.alibaba.com>
Sun, 30 Jan 2022 18:02:55 +0000 (02:02 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 13 Apr 2022 18:59:02 +0000 (20:59 +0200)
commit35380262304ff5fc9432c0d9d6793d5d39634cc6
treeeec147a7595c2009207c180f174acc638f4a4e90
parent20d01a11efde2e05e47d5c66101f5c26eaca68e2
net/smc: Send directly when TCP_CORK is cleared

[ Upstream commit ea785a1a573b390a150010b3c5b81e1ccd8c98a8 ]

According to the man page of TCP_CORK [1], if set, don't send out
partial frames. All queued partial frames are sent when option is
cleared again.

When applications call setsockopt to disable TCP_CORK, this call is
protected by lock_sock(), and tries to mod_delayed_work() to 0, in order
to send pending data right now. However, the delayed work smc_tx_work is
also protected by lock_sock(). There introduces lock contention for
sending data.

To fix it, send pending data directly which acts like TCP, without
lock_sock() protected in the context of setsockopt (already lock_sock()ed),
and cancel unnecessary dealyed work, which is protected by lock.

[1] https://linux.die.net/man/7/tcp

Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
net/smc/af_smc.c
net/smc/smc_tx.c
net/smc/smc_tx.h