[PATCH 4.19 085/264] pty: do tty_flip_buffer_push without port->lock in pty_write

From: Greg Kroah-Hartman
Date: Tue Oct 27 2020 - 10:22:51 EST


From: Artem Savkov <asavkov@xxxxxxxxxx>

[ Upstream commit 71a174b39f10b4b93223d374722aa894b5d8a82e ]

b6da31b2c07c "tty: Fix data race in tty_insert_flip_string_fixed_flag"
puts tty_flip_buffer_push under port->lock introducing the following
possible circular locking dependency:

[30129.876566] ======================================================
[30129.876566] WARNING: possible circular locking dependency detected
[30129.876567] 5.9.0-rc2+ #3 Tainted: G S W
[30129.876568] ------------------------------------------------------
[30129.876568] sysrq.sh/1222 is trying to acquire lock:
[30129.876569] ffffffff92c39480 (console_owner){....}-{0:0}, at: console_unlock+0x3fe/0xa90

[30129.876572] but task is already holding lock:
[30129.876572] ffff888107cb9018 (&pool->lock/1){-.-.}-{2:2}, at: show_workqueue_state.cold.55+0x15b/0x6ca

[30129.876576] which lock already depends on the new lock.

[30129.876577] the existing dependency chain (in reverse order) is:

[30129.876578] -> #3 (&pool->lock/1){-.-.}-{2:2}:
[30129.876581] _raw_spin_lock+0x30/0x70
[30129.876581] __queue_work+0x1a3/0x10f0
[30129.876582] queue_work_on+0x78/0x80
[30129.876582] pty_write+0x165/0x1e0
[30129.876583] n_tty_write+0x47f/0xf00
[30129.876583] tty_write+0x3d6/0x8d0
[30129.876584] vfs_write+0x1a8/0x650

[30129.876588] -> #2 (&port->lock#2){-.-.}-{2:2}:
[30129.876590] _raw_spin_lock_irqsave+0x3b/0x80
[30129.876591] tty_port_tty_get+0x1d/0xb0
[30129.876592] tty_port_default_wakeup+0xb/0x30
[30129.876592] serial8250_tx_chars+0x3d6/0x970
[30129.876593] serial8250_handle_irq.part.12+0x216/0x380
[30129.876593] serial8250_default_handle_irq+0x82/0xe0
[30129.876594] serial8250_interrupt+0xdd/0x1b0
[30129.876595] __handle_irq_event_percpu+0xfc/0x850

[30129.876602] -> #1 (&port->lock){-.-.}-{2:2}:
[30129.876605] _raw_spin_lock_irqsave+0x3b/0x80
[30129.876605] serial8250_console_write+0x12d/0x900
[30129.876606] console_unlock+0x679/0xa90
[30129.876606] register_console+0x371/0x6e0
[30129.876607] univ8250_console_init+0x24/0x27
[30129.876607] console_init+0x2f9/0x45e

[30129.876609] -> #0 (console_owner){....}-{0:0}:
[30129.876611] __lock_acquire+0x2f70/0x4e90
[30129.876612] lock_acquire+0x1ac/0xad0
[30129.876612] console_unlock+0x460/0xa90
[30129.876613] vprintk_emit+0x130/0x420
[30129.876613] printk+0x9f/0xc5
[30129.876614] show_pwq+0x154/0x618
[30129.876615] show_workqueue_state.cold.55+0x193/0x6ca
[30129.876615] __handle_sysrq+0x244/0x460
[30129.876616] write_sysrq_trigger+0x48/0x4a
[30129.876616] proc_reg_write+0x1a6/0x240
[30129.876617] vfs_write+0x1a8/0x650

[30129.876619] other info that might help us debug this:

[30129.876620] Chain exists of:
[30129.876621] console_owner --> &port->lock#2 --> &pool->lock/1

[30129.876625] Possible unsafe locking scenario:

[30129.876626] CPU0 CPU1
[30129.876626] ---- ----
[30129.876627] lock(&pool->lock/1);
[30129.876628] lock(&port->lock#2);
[30129.876630] lock(&pool->lock/1);
[30129.876631] lock(console_owner);

[30129.876633] *** DEADLOCK ***

[30129.876634] 5 locks held by sysrq.sh/1222:
[30129.876634] #0: ffff8881d3ce0470 (sb_writers#3){.+.+}-{0:0}, at: vfs_write+0x359/0x650
[30129.876637] #1: ffffffff92c612c0 (rcu_read_lock){....}-{1:2}, at: __handle_sysrq+0x4d/0x460
[30129.876640] #2: ffffffff92c612c0 (rcu_read_lock){....}-{1:2}, at: show_workqueue_state+0x5/0xf0
[30129.876642] #3: ffff888107cb9018 (&pool->lock/1){-.-.}-{2:2}, at: show_workqueue_state.cold.55+0x15b/0x6ca
[30129.876645] #4: ffffffff92c39980 (console_lock){+.+.}-{0:0}, at: vprintk_emit+0x123/0x420

[30129.876648] stack backtrace:
[30129.876649] CPU: 3 PID: 1222 Comm: sysrq.sh Tainted: G S W 5.9.0-rc2+ #3
[30129.876649] Hardware name: Intel Corporation 2012 Client Platform/Emerald Lake 2, BIOS ACRVMBY1.86C.0078.P00.1201161002 01/16/2012
[30129.876650] Call Trace:
[30129.876650] dump_stack+0x9d/0xe0
[30129.876651] check_noncircular+0x34f/0x410
[30129.876653] __lock_acquire+0x2f70/0x4e90
[30129.876656] lock_acquire+0x1ac/0xad0
[30129.876658] console_unlock+0x460/0xa90
[30129.876660] vprintk_emit+0x130/0x420
[30129.876660] printk+0x9f/0xc5
[30129.876661] show_pwq+0x154/0x618
[30129.876662] show_workqueue_state.cold.55+0x193/0x6ca
[30129.876664] __handle_sysrq+0x244/0x460
[30129.876665] write_sysrq_trigger+0x48/0x4a
[30129.876665] proc_reg_write+0x1a6/0x240
[30129.876666] vfs_write+0x1a8/0x650

It looks like the commit was aimed to protect tty_insert_flip_string and
there is no need for tty_flip_buffer_push to be under this lock.

Fixes: b6da31b2c07c ("tty: Fix data race in tty_insert_flip_string_fixed_flag")
Signed-off-by: Artem Savkov <asavkov@xxxxxxxxxx>
Acked-by: Jiri Slaby <jirislaby@xxxxxxxxxx>
Link: https://lore.kernel.org/r/20200902120045.3693075-1-asavkov@xxxxxxxxxx
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
drivers/tty/pty.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/tty/pty.c b/drivers/tty/pty.c
index 00099a8439d21..c6a1d8c4e6894 100644
--- a/drivers/tty/pty.c
+++ b/drivers/tty/pty.c
@@ -120,10 +120,10 @@ static int pty_write(struct tty_struct *tty, const unsigned char *buf, int c)
spin_lock_irqsave(&to->port->lock, flags);
/* Stuff the data into the input queue of the other end */
c = tty_insert_flip_string(to->port, buf, c);
+ spin_unlock_irqrestore(&to->port->lock, flags);
/* And shovel */
if (c)
tty_flip_buffer_push(to->port);
- spin_unlock_irqrestore(&to->port->lock, flags);
}
return c;
}
--
2.25.1